Cryptic diversity, polyploidy and intraspecific divergence in Erysimum (Brassicaceae) species endemic to the western Balkans

Abstract

Background and Aims

Several evolutionary drivers have been proposed to explain the exceptional species and genetic diversity of the Balkan flora, including the moderate impact of glaciations enabling long-term in situ persistence, ecological and topographic heterogeneity promoting vicariant differentiation, and polyploidy. Despite these shared drivers, plant lineages often show contrasting evolutionary histories, suggesting lineage-specific rather than broadly generalizable responses. The genus Erysimum, notable for its high endemism, karyological variation, recent and rapid speciation, and signatures of reticulate evolution, provides an excellent model for exploring this diversity of evolutionary trajectories and speciation drivers in the Balkans.

Methods

Focusing on five Balkan endemics of the E. odoratum species complex sampled from 47 populations, we examined chromosomal and genome size variation and applied two complementary genomic approaches, target enrichment (Hyb-Seq) and RADseq, along with morphometric analyses. These datasets were used to circumscribe species, uncover intraspecific variation patterns and assess the extent and evolutionary consequences of interspecific gene flow and polyploidy.

Key Results

Genetic and morphological evidence was largely congruent, clarifying species boundaries but also revealing cryptic diversity, leading to the recognition of eight entities. Most diversity was concentrated in the southern Dinarides, where we detected multiple cytotypes, an undescribed lineage, a polyphyletic species resulting from two independent autopolyploidization events, and allotetraploids arising in a contact zone. Geographical distance and biogeographical barriers across the mountainous Balkan landscape are likely to have shaped patterns of divergence and promoted speciation. Discrepancies between chromosome numbers and genome size indicate dynamic chromosomal evolution.

Conclusions

Vicariant differentiation, auto- and allopolyploidy, genome size and chromosomal dynamics together account for the pronounced diversity of Erysimum in the Balkans, illustrating how speciation is driven within this biodiversity hotspot. Importantly, substantial evolutionary potential is retained within species, because the coexistence of multiple intraspecific genetic lineages and cytotypes can reinforce reproductive barriers and promote future diversification.

INTRODUCTION

The Balkan Peninsula is one of the major biodiversity centres in Europe (Griffiths et al., 2004; Nieto Feliner, 2014), with its central and southern regions particularly notable for high species richness and endemism (Stevanović et al., 2007; Georghiou and Delipetrou, 2010; Lubarda et al., 2014; Tomović et al., 2014). Phylogeographical studies likewise revealed greater genetic diversity in the southern than in the northern regions of the peninsula (Španiel and Rešetnik, 2022). This concentration of species and genetic diversity appears to reflect the interplay between the complex palaeogeographical history and heterogeneous environmental conditions in the Balkans (Médail and Diadema, 2009; Nieto Feliner, 2014; Thompson, 2020). The relatively moderate impact of Pleistocene glaciations (Hughes et al., 2006) contributed to long-term environmental stability across much of the area, enabling the persistence of many plant lineages in situ (Hewitt, 1996, 2011; Nieto Feliner, 2011). At the same time, pronounced topographic complexity, such as high mountain ridges and deep river gorges, especially in the central Balkans, promoted population isolation and vicariant differentiation (Španiel and Rešetnik, 2022). Polyploidization, a major driver of angiosperm diversification, has also been shown to be common and recurrent in the Balkans (Marques et al., 2018, see e.g. Niketić et al., 2022; Rešetnik et al., 2023; Đurović et al., 2024; Šlenker et al., 2024; Berisha et al., 2025). Despite these shared evolutionary drivers, phylogenetic and phylogeographical studies consistently reveal considerable heterogeneity among taxa, underscoring that no single evolutionary scenario applies uniformly across different plant groups (Španiel and Rešetnik, 2022). Within this context, the genus Erysimum (Brassicaceae) provides an excellent system to study the mosaic of diverse evolutionary trajectories in the Balkan flora. This genus comprises >200 species distributed mainly in Eurasia, with its centre of diversity extending from the Mediterranean Basin to Central Asia (German and Al-Shehbaz, 2008; Moazzeni et al., 2014; German et al., 2023). Only a few Erysimum species occupy broad geographical ranges, whereas the majority are narrow endemics (Jalas and Suominen, 1994; Polatschek, 2013; Moazzeni et al., 2014; Czarna et al., 2016). In the Balkan Peninsula, ∼30 mainland species (excluding several Aegean endemics) are currently recognized (Jalas and Suominen, 1994), yet their phylogenetic relationships and evolutionary history remain unexplored.

Genus-level phylogenetic studies indicate that Erysimum underwent a recent and rapid radiation, with speciation beginning in the late Pliocene to early Pleistocene (5.2–2.3 Mya). The diversification of European clades occurred later, primarily during the early to middle Pleistocene (Moazzeni et al., 2014; Züst et al., 2020). Pronounced phylogenetic discordance has been attributed to extensive incomplete lineage sorting and reticulate evolutionary processes, which are common during recent species radiations (Züst et al., 2020). Previous studies of Iberian and Carpathian species further suggest multiple allopolyploidization events and signatures of introgression even among diploids, patterns consistent with weak prezygotic post-pollination barriers documented in several Iberian species (Osuna-Mascaró, 2020; Osuna-Mascaró et al., 2022, 2023; Bačák et al., 2026). Although traditional taxonomic treatments grouped some species into complexes based on morphological resemblance, growth form, life cycles and biogeographical patterns (Ball et al., 1993; Jalas and Suominen, 1994; Polatschek, 2013; Abdelaziz et al., 2014; Peccenini and Polatschek, 2016), molecular evidence demonstrates that these assemblages do not necessarily reflect natural evolutionary lineages (Abdelaziz et al., 2014; Bačák et al., 2026). Nevertheless, published phylogenies consistently reveal a strong geographical signal, suggesting that spatial separation has been one of major drivers of genus diversification (Moazzeni et al., 2014; Züst et al., 2020; Bačák et al., 2026). Rapid radiation in response to allopatric or ecological isolation is also thought to have generated cryptic species (Abdelaziz et al., 2011; Czarna et al., 2016). Furthermore, Erysimum exhibits remarkable karyological diversity, encompassing multiple base chromosome numbers (x = 6–17) and ploidy levels, with chromosome numbers ranging from 2n = 12 to 2n ≈ 90 (Warwick and Al-Shehbaz, 2006; Polatschek, 2013; Kiefer et al., 2014).

Our recent study of the Erysimum odoratum complex, comprising Balkan and Carpathian biennial species, demonstrated its polyphyletic nature (Bačák et al., 2026). Despite morphological similarity, Balkan and Carpathian species formed distinct clades. Carpathian diploids were proposed to represent a single species, Erysimum witmannii Zaw., within which biogeographical barriers and cytotype variation have driven population differentiation. However, this differentiation did not align with morphological patterns and contradicted earlier taxonomic treatments that recognized up to four diploid species in the Carpathians (Jalas and Suominen, 1994; Kliment, 1999; Michalková, 1999). The more broadly distributed Erysimum odoratum Ehrh., ranging from north-eastern France to eastern Ukraine and extending south into the northern Balkans (Jalas and Suominen, 1994; Polatschek, 2013), was found to consist of sibling polyploids that had arisen independently (Bačák et al., 2026).

In contrast, five Balkan species traditionally assigned to the E. odoratum complex were previously addressed only marginally (Bačák et al., 2026). These species occur in the western Balkans and Peloponnese, where they inhabit screes, rock crevices, xerothermic rocky grasslands and forest edges on limestone, in addition to ultramafic and silicate substrates (Jalas and Suominen, 1994; Polatschek, 2013). They include Erysimum carniolicum Dolliner (2n = 14; Slovenia, Croatia, Bosnia and Herzegovina, south-western Serbia and Montenegro), Erysimum kuemmerlei Jáv. (2n = 14; Kosovo, North Macedonia and Albania) and Erysimum pectinatum Bory & Chaub. (2n = 12, 14; Peloponnese), all three occurring in the montane to (sub)alpine belt. Two additional southern Dinaric endemics were described more recently (Polatschek, 2013), both growing from coastal to subalpine habitats: Erysimum vitekii Polatschek (2n = 12; southern Bosnia and Herzegovina, southern Croatia, Montenegro and northern Albania) and Erysimum croaticum Polatschek (2n = 42; southern Croatia and Kosovo). Several taxonomic and evolutionary challenges surround these taxa. Despite their disjunct occurrence, E. kuemmerlei and E. pectinatum show striking morphological resemblance, particularly in their pinnatifid to pinnatisect stem leaves. They were even synonymized in the past (Josifović et al., 1972; Vangjeli, 2017), or it was suggested that E. kuemmerlei might be morphologically intermediate between E. carniolicum and E. pectinatum (Jalas and Suominen, 1994). Ambiguous morphological distinction between E. carniolicum and E. odoratum has also been reported from western Serbia and Montenegro (Jalas and Suominen, 1994). Furthermore, when describing E. croaticum, Polatschek (2013) noted morphological differentiation between populations from the Adriatic coast in southern Croatia and those from Kosovo but suggested that these differences did not justify separate taxonomic recognition. The presence of two diploid cytotypes, 2n = 12 and 2n = 14 (see above) indicated two basic chromosome numbers (x = 6 and x = 7) in these western Balkan species (Polatschek, 2013). In addition, recent findings revealed remarkable variation in genome size and the presence of another cytotype (2n = 16), although based on limited sampling (Bačák et al., 2026). Overall, the previous findings in the E. odoratum complex highlight the need for an integrative approach that evaluates patterns of variations from multiple comparative perspectives (Dayrat, 2005). Focusing on the Balkan lineage traditionally assigned to this species complex, and in line with the unified species concept (de Queiroz, 2007; Karbstein et al., 2024), our aims were as follows: (1) to clarify species circumscription and assess species divergence by integrating genomic, cytotype, morphological and geographical evidence; (2) to characterize genome size and cytotype variation within species and test whether these patterns correspond to genetic structure or geographic distribution; (3) to assess the extent and evolutionary significance of interspecific gene flow, both within the target group and with sympatric or parapatric congeners; and (4) to evaluate specifically whether disjunctly distributed entities (such as E. kuemmerlei vs. E. pectinatum, or E. croaticum from Adriatic vs. inland regions) exhibit significant divergence to be supported as different species.

MATERIALS AND METHODS

Sampling design

This study comprised 47 populations of the target species, i.e. Balkan species traditionally assigned to the E. odoratum complex: E. carniolicum, E. kuemmerlei, E. pectinatum, E. vitekii and E. croaticum, including their type localities (Fig. 1A; Supplementary Data Table S1). Eleven of these populations (one to three per species) were previously sampled and analysed in the study focused on the Carpathians (Bačák et al., 2026), from which the relevant flow cytometric, sequence and chromosome-count data were obtained (for details, see Supplementary Data Table S1). In the present study, we substantially expanded both the sampling and datasets to cover the entire distribution ranges of these Balkan species and to capture potential intraspecific patterns. All 47 populations were subjected to flow cytometry (Doležel et al., 2007a) to estimate relative nuclear DNA content, and two next-generation sequencing approaches: Restriction site-Associated DNA sequencing (RADseq) (Andrews et al., 2016), and Hyb-Seq (Weitemier et al., 2014). A subset of populations was sampled for morphometric analyses (30 populations) and determination of the number of chromosomes (31 populations). Chromosomes were counted for each species and each genome size category (cytotype) within species, as inferred from flow cytometry data, to identify all cytotypes present across the sampled populations. Given that E. croaticum was found to be divided into two geographically structured phylogenetic clades (see Results), we refer to the species in a broad sense as E. croaticum s.l. For the clade including samples from the type locality (Brela: Zadvarje, Croatia), we use the name E. croaticum, whereas the clade encompassing samples from Kosovo is referred to as E. croaticum auct.

Fig. 1.

Balkan species of the traditionally circumscribed Erysimum odoratum complex and their cytotype variation. (A) Geographical distribution of sampled populations used for flow cytometry, RADseq and Hyb-Seq. A subset of these populations was examined for chromosome number and morphometrics (see Supplementary Data Table S1). Symbols correspond to those in B. Different symbol colours represent different species (or entities), and the symbol shapes distinguish cytotypes within the species. The two circle types used for E. carniolicum denote cytotypes that share the same chromosome number (2n = 12) but differ in genome size. Coloured dashed outlines in the map indicate genetic clades following Fig. 3 and Supplementary Data Fig. S2. Small grey circles on the map indicate sample sites of additional, non-target species occurring within the mapped area (all listed in Supplementary Data Table S1). (B) Photographs of the target species and mitotic chromosomes from selected accessions. Sample origin (population codes as in Supplementary Data Table S1): E. carniolicum (Sadikovac), 2n = 12 (Kalnik, chromosome spread prepared from anthers), 2n = 16 (MGLC, chromosome spread prepared from root tips); E. vitekii (Rilja), 2n = 12 (CRKV, root tips), 2n = 14 (Podvranić, root tips); E. pectinatum (MSVN), 2n = 12 (ELAT, anthers), 2n = 22 (MSVN, anthers); E. kuemmerlei (GJSL), 2n = 12 (MVRV, root tips), 2n = 14 (GLCA, root tips); E. croaticum (OMIS), 2n = 30 (OMIS, root tips), 2n = 38 (GBRE, root tips); E. croaticum auct. (GLSH), 2n = 36 (PRIZ, anthers); taxonomically unassigned entity from Mt. Ovčar, 2n = 12 (OVCA, root tips); allotetraploids E. carniolicum × E. vitekii (VRMAC), 2n = 22, 24 (VRMAC, anthers). Scale bars: 5 μm. Additional chromosomal plates are shown in Supplementary Data Fig. S1.

To place the target species in a broader phylogenetic context and assess potential gene flow with congeners, we included 26 additional Erysimum species represented by 99 populations. Of these, ten species (45 populations) were sampled and analysed by Bačák et al. (2026), and we integrated those data into the present analyses. All remaining species and populations were newly sampled and analysed here (for details, see Supplementary Data Table S1). For E. odoratum and E. witmannii, we selected 21 populations (from the 120 populations examined in detail by Bačák et al., 2026) that capture different cytotypes and geographical regions; for E. odoratum, this included particularly the southern margin of its distribution (Croatia, Serbia, Bosnia and Herzegovina), where it overlaps with the ranges of the target Balkan species (Supplementary Data Table S1). Two population samples initially identified as E. sylvestre from the south-eastern Alps and Dinarides are treated here as E. sylvestre auct., because this species appears to be polyphyletic (R. Bačák, unpublished research). It comprises an Eastern Alpine clade (not sampled in the present study) and a clade distributed in the south-eastern Alps and Dinarides (E. sylvestre auct.), requiring further taxonomic studies. Malcolmia graeca (Supplementary Data Table S1), representing the genus sister to Erysimum (Hendriks et al., 2023), was selected as the outgroup for the phylogenetic analyses. Voucher specimens from all sampled populations are deposited in the herbarium SAV.

Flow cytometric measurements

Flow cytometry was used to estimate relative nuclear DNA content following Doležel et al. (2007a, b) with a CyFlow Space flow cytometer (Partec, Münster, Germany) equipped with an ultraviolet light-emitting diode excitation source. A total of 296 individuals were measured for this study, supplemented by measurements of 200 individuals from previous work (Bačák et al., 2026; for details, see above and Supplementary Data Table S1). Multiple individuals per population were analysed: three to ten for the target species and mostly two to four for the additional species (Supplementary Data Table S1). Silica gel-dried leaves were processed together with leaf tissue of internal standards in every flow cytometric run. As standards, we used either Solanum lycopersicum ‘Stupické polní rané’ (2C = 1.96 pg; Doležel et al., 1992) or Solanum pseudocapsicum (2C = 2.59 pg; Temsch et al., 2010), selected to avoid overlap in genome size with the sample. Each tube measured by flow cytometry contained both the sample and the internal standard tissue, and no samples were pooled. A protocol using Otto buffers (Doležel et al., 2007b) and DAPI (4′,6-diamidino-2-phenylindole) fluorochrome was followed, as described by Bačák et al. (2026). The fluorescence intensity of 5000 stained nuclei was measured, and flow cytometric histograms were analysed using FloMax software v.2.7d (Partec, Münster, Germany). Relative nuclear DNA content (RGS; 2C value) was calculated from the ratio of the G0/G1 fluorescence peak positions of the sample and the internal standard (sample/standard ratio), expressed in arbitrary units (a.u.), with S. lycopersicum used as the primary standard. Only well-resolved histograms with coefficients of variation of <5 % were retained for analysis. RGS values were visualized in scatter plots to identify genome size categories (cytotypes) and summarized as boxplots by species and by cytotypes within species. To link these RGS values unambiguously to chromosome numbers, we counted chromosomes for selected individuals of each cytotype within each species (Fig. 2; Supplementary Data Table S1).

Fig. 2.

Variation in nuclear DNA content [sample/standard ratio, with Solanum lycopersicum used as the primary standard, given in arbitrary units (a.u.)] of the target Erysimum species as assessed by flow cytometry. The boxes represent the interquartile range, the horizontal lines within the boxes are the median values, and the whiskers extend to the minimum and maximum values. Symbols indicate different cytotypes (colours refer to species or entities, and shapes refer to cytotypes within the species differing in chromosome numbers and/or nuclear DNA content), as also depicted in Fig. 1. Insets show simultaneous flow cytometric analyses of three and two accessions of different cytotypes within E. carniolicum (accessions DURM, TUSBR and ZLAT) and E. kuemmerlei (accessions SHTR and Tomorrit; see Supplementary Data Table S1), respectively.

Determination of the number of chromosomes

Chromosome counts were performed on mitotic spreads. Depending on material availability, mitotic spreads were prepared either from anther tapetum (young anthers collected and fixed either in the field or from plants grown from seeds) or from root tips (harvested from seedlings or cultivated plants) in a total of 98 accessions from 31 populations (Supplementary Data Table S1; Fig. S1). For each accession, 20 mitotic figures were examined, and one representative image was selected for presentation (Fig. 1; Supplementary Data Fig. S1). All chromosome preparations followed the protocol of Mandáková and Lysak (2016), after which slides were stained with DAPI (2 mg mL⁻¹) in Vectashield antifade. Preparations were examined using a Zeiss Axioimager Z2 epifluorescence microscope equipped with a CoolCube camera (MetaSystems).

Next-generation sequencing library preparation

Two next-generation sequencing methods were used: double-digest RAD sequencing (ddRADseq; Andrews et al., 2016) and Hyb-Seq, which combines target enrichment with genome skimming (Weitemier et al., 2014). Applying both methods in parallel has proved effective for elucidating recent, highly reticulate evolutionary histories involving polyploids (Karbstein et al., 2022; Bačák et al., 2026). RADseq provides numerous short loci and large numbers of genome-wide single-nucleotide polymorphisms (SNPs), whereas Hyb-Seq yields alignments of longer loci that allow allele phasing, homeologue identification and subgenomic tree reconstruction. For the target species, the same populations (Fig. 1A) were analysed with both methods, although RADseq included more individuals per population (two to five individuals) than Hyb-Seq (one or two individuals). For the additional, non-target species, only a subset of species and populations was included in Hyb-Seq, guided by preliminary RADseq results and excluding some species outside the main study area (for details, see Supplementary Data Table S1). Overall, the RADseq dataset comprised 234 individuals, and the Hyb-Seq dataset included 88 individuals. Both datasets integrated newly generated sequences with data previously published by Bačák et al. (2026), the latter including 92 RADseq and 30 Hyb-Seq samples.

Total genomic DNA was extracted from silica gel-dried leaves using the Exgene Plant SV kit (GeneAll, Seoul, Korea) and purified with 0.4× AMPure XP beads (Beckman Coulter, CA, USA). The ddRADseq libraries were prepared according to the published protocol (Peterson et al., 2012), with some modifications as described in detail by Šlenker et al. (2024). The Hyb-Seq libraries were prepared using the NEBNext Ultra II FS DNA Library Prep Kit and NEBNext Multiplex Oligos Kit for Illumina (New England Biolabs, MA, USA) following the manufacturer’s protocols, with some modifications as described by Bačák et al. (2026). For the target enrichment step, we used a custom set of Hyb-Seq probes developed for Erysimum, targeting 1004 exons from 955 genes, ranging in length from 695 to 1197 bp. The probe design followed the protocols outlined by Weitemier et al. (2014) and Schmickl et al. (2016), with details provided by Bačák et al. (2026). Both Hyb-Seq and RADseq libraries were sequenced with 150 bp paired-end reads on Illumina systems (MiSeq at BIOCEV, Czechia for Hyb-Seq, and NovaSeq at Novogene, UK for RADseq).

RADseq data processing

The RADseq reads were demultiplexed using FASTX-Toolkit v.0.0.14 (http://hannonlab.cshl.edu/fastx_toolkit/index.html), quality-filtered using fastp v.0.20.1 (Chen, 2023) and deduplicated with clumpify.sh (BBTools; https://archive.jgi.doe.gov/data-and-tools/software-tools/bbtools/). The processed reads were mapped to the genome of Erysimum cheiranthoides (GenBank: GCA_011420285.1) using BWA v.0.7.5a (Li, 2013). Variant calling and filtering were performed using GATK v.4.4.0.0 (McKenna et al., 2010), applying hard-filtering thresholds following GATK Best Practices (Van der Auwera et al., 2013). Only biallelic SNPs with a minimum sequencing depth of 8× and ≤60 % missing genotypes were retained.

Phylogenetic relationships were reconstructed using a maximum likelihood (ML) approach implemented in RAxML-NG v.0.9.0 (Kozlov et al., 2019), based on inputs in PHYLIP format with heterozygous sites encoded using IUPAC ambiguities. The GTR substitution model with Felsenstein’s ascertainment bias correction was applied. Conversion of the VCF file to PHYLIP format was carried out with the vcf2phylip.py script (Ortiz, 2019), and invariant sites were removed using the ascbias.py script (https://github.com/btmartin721/raxml_ascbias). Branch support was evaluated using 500 bootstrap replicates and the quartet sampling method (Pease et al., 2018). Quartet sampling differentiates low support attributable to insufficient phylogenetic signal from that caused by phylogenetic discordance and further quantifies asymmetry in discordance that might indicate introgression. Additional insight into the overall genetic structure was obtained using NeighbourNet (NN) constructed in SplitsTree4 (Huson and Bryant, 2006) based on Nei’s genetic distances (Nei, 1972) calculated in the R package (Pembleton et al., 2013) using R v.4.4.0 (R Core Team, 2024). Distances were estimated from a pseudo-diploidized VCF file retaining both homozygous and heterozygous genotypes.

Potential introgression events between the studied species were examined using the ABBA–BABA and related statistics (Durand et al., 2011) as implemented in Dsuite (Malinsky et al., 2021). Because the topology of the ML tree, particularly its backbone, showed low support (see Results), these tests were restricted to the clade comprising the target species. To generate the required input tree, we first identified RADseq loci spaced ≥1000 bp apart using the identifiRadLoci.workflow (Šlenker, 2024). For each RADseq locus and sample, FASTA sequences were generated with the FastaAlternateReferenceMaker tool in GATK, encoding heterozygous SNPs with IUPAC ambiguity codes. We selected RADseq loci ≥80 bp in length and containing a minimum of six phylogenetically informative sites (7168 loci in total) to infer gene trees in RAxML-NG (as described above). These gene trees were subsequently summarized into a species tree using ASTRAL-III (Zhang et al., 2018). Two populations from Montenegro inferred to be of allotetraploid origin (see Results) were excluded.

To investigate possible origins of the polyploid populations from the target species (E. croaticum s.l. and the two tetraploid populations identified in Montenegro), the relatedness coefficients between individuals from these polyploids and all other sampled species were calculated, using the method-of-moments estimator implemented in PolyRelatedness (Huang et al., 2014). Relatedness patterns were visualized using violin plots (R Core Team, 2020; Adler and Kelly, 2021).

Hyb-Seq data processing

The Hyb-Seq reads were processed with HybPiper v.2.2.0 (Johnson et al., 2016) to extract consensus sequences of the targeted exons and flanking regions (supercontigs). Potentially paralogous sequences were identified with HybPhaser (https://github.com/LarsNauheimer/HybPhaser) and excluded when the proportion of SNPs exceeded 5 %. This filtering resulted in a final dataset of 914 supercontigs. Consensus sequences were aligned using MAFFT v.7.450 (Katoh and Standley, 2013). Flanking regions and sites with gaps in >25 % of sequences were trimmed using the R package ips (Heibl, 2008) in R v.4.4.0 (R Core Team, 2024). For species-tree inference, gene trees were initially reconstructed in RAxML-NG using the best-fitting substitution models selected by the ModelFinder function in IQ-TREE v.1.6.12 (Chernomor et al., 2016; Kalyaanamoorthy et al., 2017) under the Bayesian information criterion. Branch support was assessed with 500 bootstrap replicates, and internal branches with bootstrap support ≤20 % were collapsed using Newick-Utilities v.1.6 (Junier and Zdobnov, 2010). The species tree was then inferred under the multispecies coalescent model implemented in ASTRAL-III (Zhang et al., 2018), with branch support evaluated using local posterior probabilities (Sayyari and Mirarab, 2016). A NeighbourNet network was constructed following the same procedure as for the RADseq data, except that Nei’s genetic distances were calculated from VCF files generated by remapping Hyb-Seq reads to consensus exon sequences.

In addition, to gain insights into the polyploid genomes of E. croaticum s.l. and two tetraploid populations from Montenegro, their supercontig sequences were subjected to read-backed phasing to infer allele sequences, following the procedure described by Šlenker et al. (2021). Two complementary approaches were then used to identify distinct homeologues and determine their phylogenetic placement, thereby revealing the most likely parental species or lineages: EPA-ng (Barbera et al., 2019) and AlleleSorting (Šlenker et al., 2021). EPA-ng is a re-implementation of the evolutionary placement algorithm (EPA) that places allelic sequences onto a reference phylogenetic tree using maximum likelihood. To avoid bias from unequal sampling, a single centroid sequence per species was selected using usearch (Edgar, 2010). Only placements with a likelihood weight ratio >0.9 were considered reliable and retained for further processing. Major homeologue types, defined as those representing ≥15 % of the polyploid genome, were subsequently used for subgenomic species tree inference in ASTRAL-III.

In the AlleleSorting approach (applicable only to tetraploid samples), alleles were sorted into two homeologues based on sequence divergence, then attributed to putative parental lineages [setting the threshold of between_homeolog_distance to 1 and between_parents_distance to 0.5; see Šlenker et al. (2021) and https://github.com/MarekSlenker/AlleleSorting for more details]. The labelled homeologues, representing the two subgenomes of each tetraploid, were treated as independent accessions in the ASTRAL-III species tree inference.

Morphometric analyses

Morphometric measurements were conducted on field-collected individuals at the flowering stage that were preserved as herbarium specimens. From each plant, one flower was detached. Its parts were spread out and fixed to a sheet of paper with transparent tape to maintain their original size and shape. In total, 31 vegetative and floral characters were measured, and 24 ratio characters were derived. One original character used in the ratio calculation was excluded from the final data matrix. To identify pairs of highly correlated traits (using a threshold of r > 0.95) that could bias discriminant analyses, non-parametric Spearman rank correlations were calculated for all characters. When such correlations were detected, one character from each pair was removed from the final matrices. A detailed list of all characters is provided in Table 1. Sheets with floral parts were scanned and measured using ImageJ v.1.53e software (Schneider et al., 2012). Trichomes were examined and measured with a stereomicroscope (Olympus SZ61) and QuickPHOTO Micro v.3.2 software. The final data matrix for multivariate analyses comprised 42 characters and 495 specimens representing E. carniolicum, E. kuemmerlei, E. pectinatum, E. vitekii and E. croaticum s.l. (Supplementary Data Table S1). Subsets of this matrix were used for specific partial analyses (see Results). Cluster analyses (based on population means) and canonical discriminant analyses (based on individual plants) were performed using the MorphoTools2 package (Šlenker et al., 2022) in R v.4.0.0 (R Core Team, 2020). In addition to the characters included in the multivariate analyses, we recorded the presence of different trichome types (trichomes with two to six rays) on petals, stems, leaves and siliques from the same set of herbarium specimens.

Table 1.

List of morphological characters measured or scored in the present study.

Floral characters
LP1	Length of petal (including claw and blade)
LP2a	Length of petal blade
LP3	Length from the widest point of the petal blade to its tip
WPmax	Max. width of petal blade
LS	Length of sepal
LSty	Length of style
WSti	Width of stigma
LF	Length of longer filaments
Ratios of floral characters: LP2/LP1, LP2/WPmax, LP3/LP2
Stem and leaf characters
LL1	Length of lower stem leaf (including petiole if present)
LL2	Length of middle stem leaf
LL3a	Length of upper stem leaf
WLmax1a	Max. width of lower stem leaf
WLmax2a	Max. width of middle stem leaf
WLmax3a	Max. width of upper stem leaf
LLw1a	Length from the widest part of the lower stem leaf to its tip
LLw2a	Length from the widest part of the middle stem leaf to its tip
LLw3a	Length from the widest part of the upper stem leaf to its tip
NLb	Number of stem leaves in the upper half of the leafy part of stem
NT1a	Number of teeth on one side of the lower stem leaf margin
NT2a	Number of teeth on one side of the middle stem leaf margin
NT3a	Number of teeth on one side of the upper stem leaf margin
DT1	Max. depth of teeth on lower stem leaf margin
DT2	Max. depth of teeth on middle stem leaf margin
DT3	Max. depth of teeth on upper stem leaf margin
NTF1c	Number of primary teeth bearing secondary teeth on lower stem leaf margin
NTF2c	Number of primary teeth bearing secondary teeth on middle stem leaf margin
NTF3c	Number of primary teeth bearing secondary teeth on upper stem leaf margin
HStema	Height of the leafy part of the stem
Ratios of stem and leaf characters: LL1/LL2, LL2/LL3, NL/HStem, LL1/HStem, LL2/HStem, LL3/HStem, LL1/WLmax1, LL2/WLmax2, LL3/WLmax3, LLw1/LL1, LLw2/LL2, LLw3/LL3, NT1/LL1, NT2/LL2, NT3/LL3, DT1/WLmax1, DT2/WLmax2, DT3/WLmax3, NTF1/NT1c, NTF2/NT2c, NTF3/NT3c
Trichome characters
NrTrL	Number of trichomes on the upper surface of the middle stem leaf per area 0.5 × 0.5 mm2
LTrRayd	Max. length of the trichome ray on the upper surface of the middle stem leaf (μm), measured for ≥3-fid trichomes
LTrRay2d	Max. length of the trichome ray on the upper surface of the middle stem leaf (μm), measured for 2-fid trichomes

^aCharacters excluded from the matrix owing to a high correlation with another character (ρ > 0.95) and/or used only for computing ratios.

^bLower stem leaves can be shed at the flowering stage; therefore, the total number of stem leaves could not be determined reliably, and only those in the upper stem half were counted.

^cSecondary teeth are absent on the leaves of E. croaticum, E. croaticum auct., E. odoratum and E. witmannii; therefore, these characters were omitted from the analyses including these species.

^dMeasured within a single randomly selected visual field of the stereomicroscope. Because the rays of 2-fid trichomes are substantially longer than those of trichomes with three or more rays (≥3-fid trichomes), these two trichome types were treated separately. In E. carniolicum, E. vitekii, E. kuemmerlei, E. pectinatum, E. odoratum and E. witmannii, which possess 2- to 5(6)-fid trichomes but predominantly ≥3-fid forms, only the latter were measured (character LTrRay). In contrast, for E. croaticum s.l., only the rays of 2-fid trichomes were measured (character LTrRay2), because these are dominant and 3-fid trichomes occur only rarely.

Finally, we examined the morphological differentiation of the target western Balkan species from E. odoratum and E. witmannii, given the reported ambiguity in distinguishing E. odoratum from E. carniolicum (see Introduction). For this purpose, the data matrix generated here was supplemented with eight population samples of E. witmannii and eight population samples of E. odoratum from the previous study (Bačák et al., 2026), selected to represent different intraspecific lineages.

RESULTS

Nuclear DNA content and chromosome numbers

Flow cytometric estimates of nuclear DNA content, together with chromosome counts, revealed substantial variation across all focal species, E. carniolicum, E. kuemmerlei, E. pectinatum, E. vitekii and E. croaticum s.l., with each species comprising at least two distinct cytotypes (Figs 1 and 2; Supplementary Data Fig. S1).

In E. carniolicum, relative nuclear DNA content (relative genome size, RGS) values clustered into three well-separated size groups (0.23 ± 0.004, 0.44 ± 0.019 and 0.64 ± 0.023 RGS units; Fig. 2). The two smaller-genome cytotypes, despite clearly distinct RGS values, shared the same number of chromosomes (2n = 12; Fig. 1; Supplementary Data Table S1; Fig. S1). The largest-genome cytotype displayed chromosomal variation, with 2n = 16 in the Jasik and MGLC populations and with 2n = 18 and 2n = 30 in the ZLAT population (Supplementary Data Table S1; Fig. S1). All three cytotypes occurred in the southern part of the species range (Serbia, Montenegro, Bosnia and Herzegovina), whereas only one (2n = 12; 0.44 RGS units) was observed in northern populations in Slovenia and Croatia (Fig. 1).

Both E. vitekii and E. kuemmerlei comprised two cytotypes differing in genome size (0.21–0.22 vs. 0.40–0.41 RGS units) and number of chromosomes (2n = 12 vs. 2n = 14; Figs 1 and 2; Supplementary Data Fig. S1). No geographical structure was apparent in E. vitekii, whereas E. kuemmerlei showed a clear north–south pattern (Fig. 1).

In E. pectinatum, two cytotypes were also detected (0.23 ± 0.003 vs. 0.43 ± 0.013 RGS units). The smaller-genome cytotype had 2n = 12 chromosomes, and the larger-genome cytotype exhibited either 2n = 16 (population PTER) or 2n = 22 (population MSVN; Supplementary Data Fig. S1). Their spatial distribution indicates a north–south pattern with overlap in the central Peloponnese (Fig. 1).

The taxonomically unassigned population from Mt. Ovčar (Ovčarsko-Kablarska klisura gorge, western Serbia) showed a single cytotype (0.22 ± 0.004 RGS units; 2n = 12). In contrast, two Montenegrin populations (codes OBGL and VRMAC), shown to be of hybrid origin (see below), displayed a cytotype with higher genome size (0.47 ± 0.011 RGS units) and variable numbers of chromosomes (2n = 20, 22 and 24; Fig. 1; Supplementary Data Table S1; Fig. S1).

Finally, E. croaticum from Croatia and Bosnia and Herzegovina comprised two cytotypes differing in genome size (0.71 ± 0.017 vs. 0.92 ± 0.022 RGS units) and number of chromosomes (2n = 30 vs. 2n = 36, 38 and 40). Populations from Kosovo (E. croaticum auct.) had a single cytotype (0.68 ± 0.019 RGS units; 2n = 36; Fig. 1; Supplementary Data Table S1; Fig. S1).

Sequencing data characteristics

Sequencing of the RADseq libraries yielded an average of 5 509 109 (1 024 913–22 985 571; Supplementary Data Table S2) quality-filtered and deduplicated read pairs per sample. Mapping rates to the reference genome ranged from 63.69 to 97.43 %, with a mean of 90.00 %. The final VCF file contained 763 107 SNPs distributed across 18 076 putatively unlinked genomic regions (RADseq loci).

Sequencing of the Hyb-Seq libraries yielded an average of 708 902 (341 667–1 250 979) quality-filtered and deduplicated read pairs per individual. Of these, an average of 56.12 % (43.56–61.78 %; Supplementary Data Table S2) mapped to the 1004 targeted nuclear loci. Targeted loci were assembled using the HybPiper2 pipeline (Johnson et al., 2016). To minimize missing data, we retained only loci recovered in >96 % of sampled individuals, corresponding to 94 % of the total targeted loci. Sequence recovery across these loci was high, reaching 99.9 %. After excluding outlier sequences with excessive variability, 914 sequences were retained, with lengths ranging from 774 to 1414 bp and a mean length of 1076.45 bp.

Phylogenetic relationships and intraspecific genetic patterns

The three western Balkan endemics examined here, E. carniolicum, E. kuemmerlei and E. vitekii, formed a well-supported phylogenetic clade in both RADseq- and Hyb-Seq-based inferences [Fig. 3A, B; Supplementary Data Fig. S2A; bootstrap support (BS) = 92 %; quartet concordance (QC) = 0.4; and local posterior probability (LPP) = 0.88]. The taxonomically unassigned population from Mt. Ovčar in Serbia was consistently recovered as a separate lineage placed within this clade, strongly supported as sister to E. kuemmerlei (BS = 100 %, QC = 0.63 and LPP = 0.98). The placement of E. vitekii within this western Balkan clade, however, remained ambiguous. Depending on the dataset, it was inferred as sister to either E. carniolicum (RADseq; Fig. 3A; Supplementary Data Fig. S2A) or E. kuemmerlei (including the Ovčar branch; Hyb-Seq; Fig. 3B). However, both alternative topologies received weak support (BS = 27 %, QC = 0.29 and LPP = 0.6), and the low quartet differential score (QD = 0.11; Supplementary Data Fig. S2A) indicates an imbalance in the frequencies of the two discordant topologies. Additionally, the Alpine–Dinaric E. sylvestre auct. was nested within this western Balkan clade (Fig. 3A, B), appearing closest to E. carniolicum. Yet, its position was poorly supported in the RADseq tree (BS = 43 % and QC = −0.09) and was associated with a QD score of zero (Supplementary Data Fig. S2A).

Fig. 3.

Phylogenetic relationships and genetic structure of the target Erysimum species (highlighted with colours and/or symbols, following Figs 1 and 2) and relatives. (A) Maximum likelihood tree inferred in RAxML based on RADseq data. Clades of non-target species are shown collapsed. See Supplementary Data Fig. S2 for the complete tree including all terminal labels. Branch support (quartet concordance symbols in the nodes and bootstrap values above the branches) is shown, except for the terminal intraspecific clades. (B) Species tree inferred in ASTRAL-III from 914 gene trees based on Hyb-Seq sequences, omitting the allotetraploid accessions of E. vitekii × E. carniolicum (OBGL, VRMAC, shown in the subgenomic tree in Fig. 4B). Branches with low support (LPP < 0.8) are depicted with dotted lines. (C) NeighbourNet analysis of the target clade (outlined in A and B), including the allotetraploids (OBGL and VRMAC), based on Hyb-Seq data. Population codes follow Supplementary Data Table S1.

The Peloponnese endemic E. pectinatum formed a clade clearly distinct from E. kuemmerlei (Fig. 3). It was resolved as sister to E. parnassi, an endemic of Mt. Parnassos, and together these two lineages were connected to the western Balkan clade described above. However, this relationship was supported only weakly in both RADseq (Fig. 3A; Supplementary Data Fig. S2; BS = 51 %, QC = 0.29 and QD = 0.11) and Hyb-Seq trees (Fig. 3B; LPP = 0.42). Finally, E. croaticum s.l. was divided into two distinct lineages, neither of which showed a close affinity to the other western Balkan endemics. Populations from Croatia (including the type locality) and Bosnia and Herzegovina, representing E. croaticum s.s., formed a well-supported clade whose phylogenetic placement remained uncertain (Hyb-Seq; Fig. 3B) or appeared sister to a clade comprising several other Balkan species (RADseq; Fig. 3A). In contrast, populations from Kosovo (E. croaticum auct.) formed a different, strongly supported clade together with the Carpathian–Balkan species E. comatum (BS = 99 %, QC = 0.26 and LPP = 0.97; Fig. 3).

All four western Balkan endemics, E. carniolicum, E. kuemmerlei, E. vitekii and E. pectinatum, formed well-supported, species-specific clades, despite pronounced intraspecific variation in genome size and number of chromosomes (see above). Within each species, both RADseq and Hyb-Seq analyses reveal marked intraspecific genetic structure (Fig. 3A, C; Supplementary Data Fig. S2B), which corresponded primarily to geographical origin and only secondarily to cytotype patterns. In E. carniolicum, three clades/clusters aligned with distinct geographical regions as follows: (1) northern populations from Slovenia and Croatia (0.44 RGS units, 2n = 12); (2) populations from Montenegro together with a geographically distant population from Mt. Kopaonik in Serbia but bearing the same cytotype (all 0.23 RGS units and 2n = 12); and (3) populations from the border region of Serbia, Montenegro, and Bosnia and Herzegovina. This last group included populations with the largest genomes (0.64 RGS units, 2n = 16–30) but also the geographically close population (code TUSBR, western Serbia), despite possessing the cytotype typical for northern populations (0.44 RGS units, 2n = 12). In E. vitekii, three clades/clusters likewise corresponded to the three sampled geographical regions but only partly to cytotype variation. Two geographically proximate populations in Bosnia and Herzegovina (population codes RILJA and NEVES) clustered together despite differing in cytotypes (0.21 vs. 0.40 RGS units). Finally, in E. kuemmerlei, the three clades/clusters resolved matched three geographical regions, with the two northern groups sharing the same cytotype (0.22 RGS units, 2n = 12).

Assessing polyploid origins and the presence of interspecific gene flow

Two populations from Montenegro (codes OBGL and VRMAC; Supplementary Data Table S1) possessed a unique cytotype (0.47 ± 0.011 RGS units, 2n = 20–24) that clearly distinguished them from the nearby populations of E. vitekii and E. carniolicum (all with 0.21–0.23 RGS units, 2n = 12). In the ML tree inferred from RADseq data, these populations were placed within the clade of E. vitekii (Fig. 3A), yet the NeighbourNet analyses of both RADseq (Supplementary Data Fig. S2B) and Hyb-Seq (Fig. 3B) data consistently indicated an intermediate position between the clusters of E. vitekii and E. carniolicum. This pattern, combined with their ∼2-fold DNA content and number of chromosomes relative to adjacent populations, suggests a possible allotetraploid origin, which we tested explicitly. The relatedness estimator (PolyRelatedness) based on RADseq data showed the highest pairwise values between these Montenegrin populations and both E. carniolicum and E. vitekii (Fig. 4A). Hyb-Seq-based analyses using the EPA-ng and AlleleSorting approaches each identified two distinct homeologs in their genomes, phylogenetically placed as sister to E. carniolicum and E. vitekii, respectively, confirming their allotetraploid origins (Fig. 4B). More specifically, EPA-ng unambiguously classified 32–34 % of alleles, of which 29–33 % were assigned to the ‘vitekii’ type and 20–24 % to the ‘carniolicum’ type. The remaining alleles were distributed at low frequencies (<10 %) across a broad set of species (Supplementary Data Table S3). In the AlleleSorting analysis, 26–28 % of exon sequences passed the filtering thresholds (see the Materials and Methods), and their alleles could be sorted unambiguously into homeologue A (labelling of the alleles with shorter distance to the branch of E. vitekii, one of the suspected progenitors based on the topology of the RADseq tree; Fig. 3A) and homeologue B. In the ASTRAL species tree, homeologue A was resolved as sister to E. vitekii, whereas homeologue B was placed as sister to E. carniolicum (Fig. 4B).

Fig. 4.

Origin of polyploids assessed from RADseq (A) and Hyb-Seq (B) data. (A) Violin plots showing the distribution of pairwise relatedness values between the tetraploids from Montenegro (populations VRMAC and OBGL; see Fig. 1A; Supplementary Data Table S1) and the sampled species (only those with relatedness mean values >0.0 are shown) computed from RADseq data using the PolyRelatedness tool. The boxes within the violin plots represent the s.d., and the horizontal lines within the boxes are the mean values. Violin plots for the polyploids of E. croaticum s.l. are shown in Supplementary Data Fig. S3. (B) Species tree inferred in ASTRAL-III from subgenomic gene trees constructed from Hyb-Seq alignments of diploid species and the homeologues identified in the polyploid accessions (tetraploid accessions OBGL and VRMAC, accessions of E. croaticum and E. croaticum auct.), labelled and assigned to subgenomes using the phylogenetic placement algorithm implemented in EPA-ng. The homeologues are labelled as vit, carn, lin and com according to their phylogenetic positions. Only the major homeologue types with a frequency of >15 % were processed (see Supplementary Data Table S3). Branches with low support (LPP < 0.8) are depicted with dotted lines. The inset in frame shows the clade of the ASTRAL-III species tree inferred following homeologue sorting and labelling using the AlleleSorting tool (applicable to tetraploids only).

We further investigated the polyploid (presumably hexaploid) origin of E. croaticum s.l. using both the relatedness estimator in PolyRelatedness based on RADseq data and the EPA-ng approach applied to Hyb-Seq sequences. The relatedness estimator yielded equivocal results (Supplementary Data Fig. S3): elevated relatedness values were detected with several Balkan species. For accessions of E. croaticum from Croatia and Bosnia and Herzegovina, these included the polyploids E. cephalonicum, E. diffusum s.l., E. graecum and E. janchenii, in addition to the diploid E. linariifolium. For accessions from Kosovo, elevated relatedness was observed with the polyploids E. cephalonicum, E. diffusum s.l. and E. janchenii and with the diploid E. comatum. EPA-ng analysis of Hyb-Seq data provided a clearer signal. For E. croaticum accessions from Croatia and Bosnia and Herzegovina, 23–25 % of alleles were classified, of which 18–22 % were assigned to the ‘linariifolium’ type, and the remaining alleles were distributed among other species at lower frequencies (<11 %; Supplementary Data Table S3). For E. croaticum auct. from Kosovo, 27–30 % of alleles were classified, of which 28–35 % were assigned to the ‘comatum’ type, with the rest assigned to other species at lower frequencies (<15 %; Supplementary Data Table S3). The subgenomic species tree (Fig. 4B) indicated that E. croaticum s.l. does not share a single polyploid origin. Instead, the data support independent autopolyploid origins from different progenitors in the Adriatic (Croatia, Bosnia and Herzegovina) vs. inland (Kosovo) lineages.

Next, we used ABBA–BABA tests to investigate signals of interspecific gene flow within the clade comprising the target species E. carniolicum, E. kuemmerlei, E. pectinatum and E. vitekii, including the unassigned population from Mt. Ovčar but excluding the two allotetraploid populations from Montenegro (see above). The f-branch statistics (Supplementary Data Fig. S4) revealed strong introgression signals between E. sylvestre auct. and E. carniolicum, particularly involving the southern populations of the latter, which is consistent with the quartet differential (QD) score of zero in the ML tree (see above). Introgression was also identified between the unassigned population from Mt. Ovčar and E. carniolicum, especially involving the populations of the same cytotype, characterized by the lowest genome size and a chromosome number of 2n = 12, distributed in Montenegro and in the Kopaonik Mts. in Serbia. Furthermore, strong signals of interspecific gene flow were detected between E. vitekii and E. kuemmerlei and between E. vitekii and E. carniolicum (notably their Montenegrin populations). These results are congruent with the uncertain phylogenetic placement of E. vitekii, its low QD score in the ML tree (see above) and the existence of allotetraploid populations involving the latter two species (see above).

Multivariate morphometrics

Cluster analysis based on population means of 30 morphometrically analysed populations revealed three major clusters, corresponding to: (1) E. pectinatum together with E. kuemmerlei; (2) E. croaticum s.l.; and (3) E. carniolicum and E. vitekii, along with the two populations of allotetraploid origin (see above) and the taxonomically unassigned population from Mt. Ovčar (Fig. 5A). CDA of 495 individual specimens, with seven a priori defined groups (excluding allotetraploids), identified E. croaticum s.l. as the most morphologically distinct (Fig. 5B). Canonical structure coefficients indicate that this distinctiveness is mainly attributable to long, narrow leaves with sparse and shallow marginal teeth (Supplementary Data Fig. S5B; see also Fig. 6G). A subsequent CDA excluding E. croaticum s.l. resolved three clearly distinct clusters of E. carniolicum, E. vitekii and E. kuemmerlei intermingled with E. pectinatum (Fig. 5C). Specimens of E. kuemmerlei and E. pectinatum are characterized by conspicuously pinnatisect leaves (densely and deeply dentate leaf margins, with secondary teeth) and comparatively small floral parts (Fig. 6D, E; Supplementary Data Fig. S5C). Erysimum carniolicum and E. vitekii were separated from each other along the second canonical axis, mainly reflecting differences in floral size (larger in E. carniolicum), density of teeth on leaf margins (higher in E. carniolicum) and traits such as number of leaves, the length of leaves and trichome rays on leaf surface (all higher in E. vitekii; Supplementary Data Fig. S5C; see also Fig. 6A, B). Specimens from the two allotetraploid populations, passively projected into this CDA, occupied intermediate positions between E. carniolicum and E. vitekii. Finally, the specimens from Mt. Ovčar overlapped mostly with E. vitekii, consistent with the placement of this population in the clustering analysis (Fig. 5A). A separate CDA (Fig. 5D) including only E. vitekii, E. kuemmerlei and specimens from Mt. Ovčar, supported the distinct morphology of this taxonomically unassigned population. Relative to E. vitekii, plants in this population have narrower petals and wider basal leaves.

Fig. 5.

Morphometric analyses of Balkan species of the traditionally defined Erysimum odoratum complex. (A) UPGMA cluster analysis based on the matrix of 30 sampled populations (listed in Supplementary Data Table S1) and population means of 41 characters (excluding the LTrRay and LTrRay2 characters; see Table 1). (B) Canonical discriminant analysis (CDA) based on 495 specimens and 41 characters (as in A), with seven a priori defined entities as indicated (excluding the tetraploid populations of E. vitekii × E. carniolicum). (C) CDA based on 451 specimens and 42 characters (including the LTrRay character), with six entities as indicated. Unlike the previous CDA in B, the most distinct E. croaticum and E. croaticum auct. were omitted here, and specimens of E. vitekii × E. carniolicum were included as passively projected objects. (D) CDA based on 192 specimens and 42 characters (as in C), with three defined entities as indicated. The taxonomically unassigned population from Mt. Ovčar is differentiated here from the morphologically closest species. In the graphs in B–D, the specimen points are connected to the centroid of the given group, producing a spider-like diagram. (E) CDA based on 79 specimens and 42 characters (including the LTrRay2 character), with two entities, E. croaticum (Croatia, Bosnia and Herzegovina) and E. croaticum auct. (Kosovo). (F) CDA based on 146 specimens and 42 characters (including the LTrRay character), with two entities, E. kuemmerlei and E. pectinatum. Total canonical structure coefficients, expressing the contributions of characters to the axes, are shown in Supplementary Data Fig. S5.

Fig. 6.

Scans and photographs of selected morphological features of the Balkan species and entities recognized here within the traditionally defined Erysimum odoratum complex, including petals, middle stem leaves, and trichomes on the upper surface of middle stem leaves. Scale bars: 1 cm for petals and leaves, 0.25 cm for trichomes. (A) Erysimum carniolicum, population ZLAT. (B) Erysimum vitekii, population Podvranić. (C) Erysimum vitekii × E. carniolicum, population OBGL, two petals are shown to illustrate large variation in their shape. (D) Erysimum kuemmerlei, population GJSL. (E) Erysimum pectinatum, population PTER. (F) Taxonomically unassigned entity from Mt. Ovčar, population OVCA. (G) Erysimum croaticum, population Omis. (H) Erysimum odoratum, populations Zukici (petals, middle stem leaf), V.Ucka (trichomes). Population codes follow Supplementary Data Table S1.

Two additional CDAs examined differentiation within two species pairs that appeared intermingled in the above analyses: E. croaticum (Croatia, Bosnia and Herzegovina) vs. E. croaticum auct. (Kosovo) and E. kuemmerlei vs. E. pectinatum. CDA of E. croaticum s.l. (Fig. 5E; Supplementary Data Fig. S5E) showed that the population from Kosovo differs by narrower petals, longer upper stem leaves and shorter trichome rays on leaves. In the comparison of E. kuemmerlei and E. pectinatum (Fig. 5F; Supplementary Data Fig. S5F), CDA showed that E. kuemmerlei can be distinguished by having more stem leaves, higher trichome density on leaf surface and larger floral parts (see also Fig. 6D, E).

Variation in trichome types on stems, leaves, siliques and petals also contributed to species differentiation (Fig. 6; Supplementary Data Table S4). Erysimum croaticum s.l. is distinct in possessing exclusively 2- and 3-fid trichomes (Fig. 6G) on leaves and siliques, and predominantly 2-fid trichomes on petals and stems (with 3-fid trichomes occurring only rarely). In contrast, E. carniolicum, E. vitekii and E. kuemmerlei bear 2- to 5(6)-fid trichomes (Fig. 6A, B, D) on leaves, siliques and petals, and 2- to 4-fid trichomes on stems. Erysimum pectinatum differs by having only 2- and 3-fid trichomes on siliques, stems and petals (thus resembling E. croaticum s.l.), while having 2- to 4-fid trichomes on leaves (Fig. 6E; Supplementary Data Table S4).

Finally, we also assessed morphological differentiation of E. carniolicum and E. vitekii relative to two other species of the traditional E. odoratum complex, E. odoratum and E. witmannii (Supplementary Data Fig. S6). Erysimum witmannii (Carpathian diploids) differs from both Balkan species particularly by its longer, narrower stem leaves; from E. carniolicum additionally by less dentate leaves; and from E. vitekii by wider petals and longer sepals. Erysimum odoratum (widespread polyploids) is clearly separable from E. carniolicum by its smaller floral parts, less dentate leaves, more stem leaves and longer trichome rays on the leaf surface. Although E. odoratum and E. vitekii showed some morphological overlap, E. odoratum typically has more leaves, less dentate leaf margins, smaller upper stem leaves relative to middle stem leaves (LL2/LL3 ratio), longer sepals and a wider stigma (Supplementary Data Fig. S6).

An identification key to the members of the traditional E. odoratum complex in the Balkans is presented. In addition, Fig. 6 illustrates some of the diagnostic characters that aid species identification, namely variation in trichome types, variably dentate margins of middle stem leaves, and differences in petal size and shape. Descriptive statistics for all quantitative characters measured (Table 1) per species are presented in Supplementary Data Table S5.

DISCUSSION

Taxonomic perspectives: morphological and genetic distinction of the studied Balkan endemics

Our results show that the three Balkan species traditionally assigned to the Erysimum odoratum complex, E. carniolicum, E. kuemmerlei and E. pectinatum, together with the more recently described E. vitekii (Polatschek, 2013), represent distinct, separately evolving lineages, supporting their recognition as species. In our analyses, all four lineages were resolved as monophyletic and morphologically distinguishable (Figs 3, 5 and 6; see also Identification key). Published occurrence data (Strid, 1989; Jalas and Suominen, 1994; Polatschek and Snogerup, 2002; Polatschek, 2013) and our present sampling (Fig. 1) further indicate largely allopatric ranges, with only narrow zones of contact or sympatry. Such geographical separation generates extrinsic reproductive barriers, which, however, may break down in sympatry. This was observed between E. carniolicum and E. vitekii, but such events appear rare, spatially restricted, and in this case led to a new allopolyploid entity, rather than eroding species integrity through extensive introgression. No clear differentiation among species was observed in chromosome numbers and genome size variation, because several cytotypes (2n = 12, 14 and 16) were shared across species. In contrast, we detected substantial variation within species, highlighting pronounced genome size and chromosomal dynamics in the genus, and particularly in this species complex (see below). Overall, following the unified species concept (de Queiroz, 2007; Karbstein et al., 2024), these four lineages exhibit several properties (lines of evidence) to consider them as separately evolving metapopulation lineages and to treat them as distinct species. Finally, E croaticum s.l. appears to comprise two independently arisen, genetically divergent lineages despite their morphological resemblance. Additional sampling and analyses are required to resolve them taxonomically (see also below).

Despite earlier ambiguities and misidentifications (Jalas and Suominen, 1994; Vassilev et al., 2012, 2016, 2018), we have demonstrated that E. carniolicum, the most widespread of these Balkan endemics, is well differentiated from the broadly distributed and parapatric E. odoratum. Several diagnostic characters reported for E. carniolicum (Polatschek, 2013), such as large petals, absence of sterile axillary shoots, persistence of lower stem leaves during flowering and mostly glabrous petals (see Supplementary Data Table S4), are supported by our observations. Additional differences in the number of stem leaves, leaf margin and trichome-ray length (Supplementary Data Table S5; Fig. S6) further distinguish these two species. We assume that past confusion between E. carniolicum and E. odoratum might have resulted from two main factors: their parapatric distribution and the distinct species E. vitekii, described only relatively recently from Montenegro (Polatschek, 2013). Range overlap between E. carniolicum and E. odoratum occurs mainly in north-western Croatia (Polatschek, 2013; Nikolić, 2015), where they may co-occur even within the same locality (e.g. in the Učka Mts., Istria; see Supplementary Data Table S1). In Bosnia and Herzegovina and in Serbia, their distributions seem to be separated in a north–south direction (Jalas and Suominen, 1994; Polatschek, 2013; Stupar et al., 2018; Bačák et al., 2026; and present records), with some zone of sympatry but with elevational differentiation: E. carniolicum typically grows above 1000 m a.s.l., whereas E. odoratum inhabits lower elevations. Our southernmost record of E. odoratum in Serbia is from the Kavlar peak (790 m a.s.l.; Supplementary Data Table S1), previously misidentified as E. carniolicum (Lakušić, 2019).

In Montenegro, confusion in classification has probably been caused by populations later described as E. vitekii (Polatschek, 2013), which are shown here to be morphologically closest to E. odoratum (Supplementary Data Fig. S6). The previously proposed absence of trichomes on petals in E. vitekii (Polatschek, 2013) was found to be an unreliable trait (see Supplementary Data Table S4). Both E. odoratum and E. vitekii typically bear 2- and 3-fid trichomes on petals, but differ in other characters, including the number of stem leaves, leaf margin and size of some floral parts (Supplementary Data Table S5; Fig. S6; see also Identification key). Despite their overall morphological similarity, these two species are phylogenetically divergent and allopatric. Erysimum vitekii appears confined to Montenegro and the southernmost parts of Bosnia and Herzegovina, reported also from southern Croatia (Pelješac) and northern Albania (Polatschek, 2013; and present records).

The geographically disjunct E. kuemmerlei and E. pectinatum are morphologically similar, as previously noted, and sometimes even merged into a single species (Josifović et al., 1972; Vangjeli, 2017). However, we identified differences in the number of stem leaves, trichome density on leaf surface, trichome type frequency and size of floral parts (Supplementary Data Table S4; Fig. S5) and, importantly, proved their phylogenetic divergence, finally supporting their recognition as two separate species. Their resemblance, particularly in pinnatifid to pinnatisect stem leaves (Fig. 6D, E), might reflect parallel evolution, perhaps by selection on ancestral standing genetic variation (Thompson et al., 2019; James et al., 2021).

The morphological differentiation presented and reported here was based on field-collected plants, which might partly reflect environmentally induced plastic variation. However, our recent parallel datasets of cultivated and field-collected samples in E. odoratum and E. witmannii (Bačák et al., 2026) indicated that traits of the floral organs (shape and size), trichomes, and leaf shape and margin are generally stable, whereas the number of leaves and the size of vegetative structures (stems and leaves) can be affected by environmental conditions. Growing plants in a common garden would therefore be valuable in future work for the Balkan endemics, as it would help to disentangle environmentally driven variation from genetically based morphological differences.

Phylogenetic resolution, impact of interspecific gene flow, and major evolutionary drivers

Our phylogenetic reconstruction provides initial insights into the evolutionary relationships among Balkan Erysimum species, although it covers only part of the species diversity in the peninsula (Supplementary Data Fig. S2). Several internal nodes have low support, as indicated by low bootstrap and quartet concordance values, along with both high and low quartet differential (QD) values. This suggests that incomplete lineage sorting and reticulate evolution both contribute to these patterns, consistent with previous phylogenetic studies in the genus (Züst et al., 2020). For the five focal species, the western Balkan endemics E. carniolicum, E. kuemmerlei, E. vitekii and E. croaticum s.l. and the Peloponnese endemic E. pectinatum, our results largely confirm the phylogenetic patterns inferred in an earlier study based on limited sampling (one to three populations per species; Bačák et al., 2026). A notable exception concerns E. vitekii. Its position was previously well resolved as sister to E. kuemmerlei based on the single known cytotype (2n = 12), but with the discovery of an additional cytotype (2n = 14) and inclusion of both cytotypes in the present analyses, its placement becomes ambiguous (Fig. 3). Low QD values, together with ABBA–BABA tests (Supplementary Data Fig. S4), point towards introgression rather than incomplete lineage sorting as a likely explanation for conflicting branching patterns, although weak phylogenetic support limits the confidence with which these processes can be distinguished. Novel and well-supported phylogenetic findings of the present study include the identification of a taxonomically unrecognized population from Mt. Ovčar and two distinct lineages within E. croaticum s.l. (see below).

Although gene flow was inferred between several of the species studied here, their mostly allopatric distributions currently limit opportunities for hybridization and might reinforce their genetic divergence. Geographical distance and biogeographical barriers have repeatedly been shown to shape both intra- and interspecific differentiation in this genus (Moazzeni et al., 2014; Züst et al., 2020; Bačák et al., 2026). The topographically complex landscapes of the Balkans facilitate geographical structuring, a pattern also documented in other plant groups (Španiel and Rešetnik, 2022). Erysimum carniolicum and E. kuemmerlei are clearly allopatric, with their ranges separated by the well-known biogeographical barrier formed by the Drin river basin and Kosovo Polje, which separates the Dinaric mountains from the Scardo-Pindhic range (Griffiths et al., 2004; Stevanović et al., 2009). This barrier marks distribution limits for many Balkan species (Španiel and Rešetnik, 2022) and also separates subspecies (e.g. Cerastium decalvans, Niketić et al., 2022) or intraspecific lineages (Salvia officinalis, Rešetnik et al., 2016a; Knautia drymeia, Rešetnik et al., 2016b; Euphorbia epithymoides s.l., Đurović et al., 2024). Ecological adaptation, whether climatic, elevational or edaphic, may also contribute to speciation in the target complex, but probably in concert with geographical isolation. The range of E. vitekii appears restricted to the southern Dinaric mountains adjacent to the Adriatic coast, which resembles the distribution of some other, mostly thermophilous western Balkan endemics (Euphorbia hercegovina, Sharovikj Ivanova et al., 2025) or intraspecific lineages (Edraianthus tenuifolius, Surina et al., 2011; Viola suavis, Mereďa et al., 2011; Salvia officinalis, Rešetnik et al., 2016a). Populations of E. vitekii rarely ascend above 1000 m, which separates them from the high-elevation species E. carniolicum. Furthermore, E. vitekii grows on calcareous bedrock, whereas both E. kuemmerlei and E. carniolicum have been recorded from calcareous, siliceous and serpentine substrates (Polatschek, 2013).

Polyploidy has also been a significant evolutionary driver in the study complex, including autopolyploidization inferred for E. croaticum s.l. and allopolyploidy (E. carniolicum × E. vitekii) confirmed in populations from Montenegro (Fig. 4). This latter case illustrates how allopolyploids can arise in secondary contact zones of related species or genetic lineages, a scenario also documented in other plant groups in the Balkans (see e.g. Zozomová-Lihová et al., 2020; López-González et al., 2021; Lazarević et al., 2022; Šlenker et al., 2024). A comparable parapatric distribution along the elevational gradient resulting in a hybrid zone was observed in Erysimum in the Sierra Nevada in south-eastern Spain; however, in that case hybridization occurred without an increase in ploidy level and was characterized by extensive backcrossing (Abdelaziz et al., 2021). In contrast, the tetraploid level of the hybrids in our study appears to act as a reproductive barrier to the diploid parental species, preventing the formation of hybrid swarms. The allopolyploid plants occupy a relatively large area, spreading a few kilometres along the roadside, and produce viable seeds (J. Zozomová-Lihová, personal observation), which indicate successful establishment and the potential for long-term persistence and allopolyploid speciation (Soltis and Soltis, 2009; Edger et al., 2025).

The diversity hotspot and cryptic diversification in the southern Dinarides

The results presented here reveal a pronounced diversity hotspot in the southern Dinarides, spanning south-western Serbia, Kosovo, southern Bosnia and Herzegovina, and Montenegro (Fig. 1A). This region forms a major contact zone where several of the studied species meet and multiple cytotypes co-occur, including previously unrecognized cryptic entities. We recorded a heterogeneous, polyphyletic E. croaticum s.l., allotetraploids derived from E. vitekii and E. carniolicum, and a taxonomically unassigned population from Mt. Ovčar. At the cytotype level, all three cytotypes of E. carniolicum and both cytotypes of E. vitekii occur here. These findings are consistent with other cases of remarkable and often cryptic diversity in the region, documented at the genetic, cytotypic or species levels (Kučera et al., 2010; Kutnjak et al., 2014; Španiel et al., 2017; Grdiša et al., 2019; Đurović et al., 2021; Kuzmanović et al., 2021; Sharovikj Ivanova et al., 2025). All these patterns suggest that the southern Dinarides provided favourable conditions for long-term diversity accumulation. The environmentally complex landscapes forming a mosaic of diverse habitats support high species richness, while strong topographic barriers restrict migration and gene flow, fostering genetic divergence. Estimates of divergence times in the genus Erysimum (Moazzeni et al., 2014; Huang et al., 2020; Hendriks et al., 2023) allow us to link speciation and diversification of the focal species to the Pleistocene epoch. Therefore, we also suggest that the moderate impact of Pleistocene glaciations in this area and the climatic buffering effect of the nearby Adriatic Sea contributed to environmental stability and population persistence in (micro)refugia (Griffiths et al., 2004; Hughes et al., 2006; Španiel and Rešetnik, 2022). This concentration of local diversity, in turn, increased opportunities for secondary contacts, hybridization and polyploidization events (e.g. Lakušić et al., 2009; Kučera et al., 2010; Šlenker et al., 2024).

One of the cryptic entities identified here is the population from Mt. Ovčar in the Ovčarsko-Kablarska klisura gorge in western Serbia, previously recorded and treated by Polatschek (2013) as E. carniolicum. Our results show that this population is morphologically closest to E. vitekii (Fig. 5), yet phylogenetically it is placed as sister to E. kuemmerlei (Fig. 3). Despite this discrepancy, a hybrid origin appears unlikely: the population is diploid, it exhibits high seed set and germination rates (J. Zozomová-Lihová, unpublished research), and our attempts to identify different subgenomes from Hyb-Seq data (not shown) disproved them. Although E. kuemmerlei has been reported from south-western Serbia (Lakušić, 2019), our sampling shows that these records represent misidentifications of E. carniolicum. In contrast, the northernmost confirmed localities of E. kuemmerlei are in Kosovo (Polatschek, 2013; and present results). We therefore suggest that the entity from Mt. Ovčar might have arisen from a common ancestor with E. kuemmerlei either through past range fragmentation (vicariance) or through long-distance dispersal, followed by isolation and divergence; hypotheses that require further testing. Its morphological and genetic divergence, in addition to geographical separation, suggest its treatment as a distinct species, but given the limited data available on its distribution, we refrain from proposing a formal taxonomic treatment at this stage. So far, only a single population of this entity has been confirmed, but it occupies a relatively large area on the southern slopes of Mt. Ovčar at elevations ∼700–900 m a.s.l. The possibility that it represents a stenoendemic taxon should also be considered, as illustrated by Edraianthus canescens D. Lakušić, a recently described stenoendemic restricted to the calcareous cliffs of the Ovčarsko-Kablarska klisura (Lakušić et al., 2013).

Another previously unrecognized entity is represented by the tetraploids found along the roadside in the Sinjajevina Mts. in Montenegro. Both genomic and karyological data support their allotetraploid origin (Figs 2 and 4), consistent with their morphological intermediacy (Fig. 5C) and their location in the contact zone between populations of E. carniolicum and E. vitekii. These plants also document remarkable chromosomal dynamics and rapid reduction in the number of chromosomes, which appears characteristic of the entire species complex. Here, we found counts of 2n = 20 and 2n = 22 occurring alongside the regular tetraploid number 2n = 24, while maintaining a uniform genome size (Fig. 2; Supplementary Data Fig. S1).

Finally, our results support Polatschek’s (2013) treatment of E. croaticum as a distinct entity, in addition to morphological differentiation between populations from Kosovo and the Adriatic coast. Although the morphological differences are subtle (Fig. 5), the phylogenetic divergence between these disjunct populations is likely to reflect independent polyploidization events (Fig. 4B), supporting separate taxonomic treatment. Further sampling in western Kosovo, however, is needed, particularly to clarify the distinction from E. janchenii Fritsch, which was proposed as a distinct tetraploid species extending to western Kosovo (Polatschek, 2013) but otherwise synonymized with diploid E. linariifolium (Jalas and Suominen, 1994; Caković and Stešević, 2021).

Intraspecific divergence: geographical structuring and remarkable cytotype variation

The pronounced genetic structuring observed within the studied Balkan endemics reflects both geographical regions and cytotype variation. This suggests that intraspecific divergence is shaped by a combination of extrinsic gene flow barriers, such as geographical distance, biogeographical barriers or historical colonization patterns, as documented for other plant species in the Balkans (reviewed by Španiel and Rešetnik, 2022), and intrinsic barriers associated with differences in chromosome number and genome size. Nevertheless, the geographical and cytotype patterns were not fully congruent. This is particularly evident in the southern Dinarides, where we identified cytotypically mixed genetic lineages in both E. vitekii and E. carniolicum (Figs 1 and 3). A key to understanding these patterns is likely to lie in the origin and evolution of these cytotypes; however, discrepancies between genome sizes and chromosome numbers point to complex patterns of chromosomal changes. Within species and among close relatives, genome size generally increases proportionally with chromosome number and ploidy level (Sliwinska et al., 2022) and has therefore been widely used as a ploidy estimator (e.g. Suda et al., 2007; Balao et al., 2009; Kobrlová et al., 2024). However, this relationship is not always straightforward. In polyploids, a reduction in monoploid genome size, known as genome downsizing, has frequently been documented and is associated with the gradual process of diploidization (Leitch and Bennett, 2004; Mandáková and Lysak, 2018). Moreover, the activity of repetitive DNA can modify genome size in both directions, further decoupling genome size from chromosome number or ploidy level (McCann et al., 2020; Pellicer et al., 2021). As a result, the relationship between chromosome number and genome size can become complex, because both traits can evolve independently across different phylogenetic lineages of a genus (Macas et al., 2015; Campos-Domínguez et al., 2022). Such complexity is also evident in our target species, where we observe discrepancies between chromosome number and genome size. What is particularly striking is that the degree of mismatch we detect would typically be expected among different phylogenetic clades, yet in our case it occurs among closely related species and even within them.

The discrepancy between chromosome number and genome size is absent in the lowest genome-size category (sample/standard ratio ≈ 0.2 RGS units; Fig. 2), where all examined species (E. carniolicum, E. vitekii, E. kuemmerlei and E. pectinatum) consistently exhibit 2n = 12. In contrast, discrepancies emerge at genome size values ≈ 0.4–0.5 RGS units, where we observe 2n = 12 (in E. carniolicum), 2n = 14 (E. vitekii and E. kuemmerlei) and 2n = 16–22 (E. pectinatum). These patterns might reflect autotetraploidy undergoing rapid structural diploidization manifested by a reduction in the number of chromosomes (Mandáková and Lysak, 2018). The most extreme case occurs in E. carniolicum, where some populations share the same number of chromosomes but differ 2-fold in genome size (Fig. 2). This highlights the need for future cytogenomic and genomic approaches to determine whether this mismatch between genome size and chromosome number results from recent autopolyploidy followed by accelerated reduction in the number of chromosomes, great variation in repetitive DNA content or a combination of both processes operating in parallel.

Nevertheless, whether arising from polyploidy, chromosomal rearrangements, repetitive DNA evolution or their combination, the co-existence of multiple cytotypes within the study species might limit gene flow between populations (Ramsey and Schemske, 1998; Kolář et al., 2017). When such populations are also geographically separated, such as in E. kuemmerlei and E. carniolicum, both extrinsic and intrinsic barriers may operate together, potentially accelerating divergence and promoting speciation. Nevertheless, different ploidy levels or cytotypes do not always constitute complete reproductive barriers (Kolář et al., 2017), as also shown in Iberian Erysimum (Osuna-Mascaró et al., 2023). Furthermore, here we also detected unique cases of intrapopulational cytotype variation. Some populations (Supplementary Data Table S1) comprise individuals with different chromosome number yet exhibit uniform genome size. This represents an unusual pattern of intraspecific dysploidy, similar to that documented in Cardamine pratensis (Melichárková et al., 2020) and recently reported in E. witmannii (Bačák et al., 2026). Taken together, these findings reinforce the view that Erysimum is among the most cytogenetically dynamic and variable genera in the Brassicaceae (Moazzeni et al., 2014).

Conclusion

Our study of Balkan Erysimum endemics revealed predominantly allopatric patterns and congruence between genetic and morphological data, clarifying species boundaries while also uncovering interspecific gene flow and cryptic diversity. The southern Dinarides emerged as a prominent diversity hotspot, where we found multiple cytotypes, inferred both auto- and allopolyploidization events and identified a previously unrecognized diploid entity. Overall, these findings underscore how geographical isolation, polyploidy and dynamic chromosomal evolution interact to drive diversification. This combination of processes helps to explain the exceptional diversity and complex evolutionary histories characteristic of the Balkan flora.

IDENTIFICATION KEY TO THE MEMBERS OF THE ERYSIMUM ODORATUM COMPLEX IN THE BALKANS

1a. Trichomes on leaves 2-fid (trichomes with two rays)¹, occasionally also 3-fid (trichomes with three rays), but never ≥4-fid; stem leaves narrow: middle stem leaves (2–)3(–4) mm wide, 9–23-times longer than wide, and long relative to stem height (reaching 1/5 to 1/3 of stem height), with only sparse and shallow teeth on the leaf margin … E. croaticum s.l.²

1b. Trichomes on leaves of multiple types, including 2-, 3-, 4-, 5-fid¹, and occasionally also 6-fid; stem leaves wider: middle stem leaves (4–)7(–12) mm wide, three to eight times longer than wide, and shorter relative to stem height (reaching 1/8 to 1/5 of stem height), with several conspicuous teeth on the leaf margin … 2

2a. Stem leaves with deeply dentate leaf margins¹, with primary teeth reaching up to the midrib (pinnatisect leaves) and with secondary teeth commonly present¹; petals (9–)12(–15) mm long³ and (2–)4(–6) mm wide … 3

2b. Stem leaves with conspicuous teeth on the leaf margin¹, reaching up to one-third of the leaf width, rarely with secondary teeth present; petals larger, (13–)19(–25) mm long³ and (3–)7(–10) mm wide … 4

3a. Trichomes on leaves 3-, 4- and 5-fid¹, occasionally also 2-fid; trichomes on siliques 3-, 4- and 5-fid, rarely also 6-fid; number of stem leaves in the upper stem half⁴ (5–)10(–17); petals ≤16 mm long³ and 6 mm wide … E. kuemmerlei (Albania, Kosovo, North Macedonia)

3b. Trichomes on leaves 2-, 3- and 4-fid¹; trichomes on siliques only 2- and 3-fid; number of stem leaves in the upper stem half⁴ (4–)7(–11); petals ≤13 mm long³ and 4 mm wide … E. pectinatum (Peloponnese)

4a. Petals (16–)21(–25) mm long³ and (6–)8(–10) mm wide; stems mostly simple, rarely branched … E. carniolicum

4b. Petals (13–)16(–19) mm long³ and (3–)5(–7) mm wide; stems mostly branched … 5

5a. Petal blades oblong or narrowly obovate¹, (5–)6(–7) mm long and (3–)4(–5) mm wide, ratio of petal blade length/width (1.3–)1.6(–2.1); lower stem leaf margin with (4–)6(–9) teeth on one side, occasionally with secondary teeth present … taxonomically unassigned population from the Ovčarsko-Kablarska klisura gorge (Mt. Ovčar) in Serbia

5b. Petal blades obovate¹, (5–)7(–8) mm long and (4–)6(–7) mm wide, ratio of petal blade length/width (0.9–)1.2(–1.5); lower stem leaf margin with (2–)5(–7) teeth on one side, secondary teeth absent … 6

6a. Sepals (7–)8(–9) mm long, style (8–)10(–12) mm long, stigma (0.6–)0.9(–1.2) mm wide; number of stem leaves in the upper stem half⁴ (6–)13(–23); lower stem leaf margin with teeth ≤4.5 mm long, those on the middle stem leaf margin ≤3 mm long; upper stem leaves (3–)6(–10) mm wide, middle stem leaves ∼1.5 times longer than upper stem leaves … E. vitekii (southern Bosnia and Herzegovina⁵, southern Croatia (Pelješac), Montenegro, northern Albania)

6b. Sepals (7–)9(–11) mm long, style (8–)11(–15) mm long, stigma (0.7–)1(–1.6) mm wide; number of stem leaves in the upper stem half⁴ (9–)20(–31); lower stem leaf margin with teeth ≤3.5 mm long, those on the middle stem leaf margin ≤2.5 mm long; upper stem leaves (2–)4(–8) mm wide, middle stem leaves ∼2 times longer than upper stem leaves … E. odoratum (widespread, extending to southern Bosnia and Herzegovina⁵ and central Serbia, but absent from Montenegro and Albania)

Notes. Erysimum witmannii, a diploid endemic of the Carpathians that does not extend to the Balkans, is not included in the key. The allotetraploids (E. carniolicum × E. vitekii), which combine the characters of the parental species and show a large range of variation in some characters (see petal shape variation in Fig. 6C), are also excluded. In contrast, the taxonomically unassigned population from the Ovčarsko-Kablarska klisura gorge is included in the key but requires further chorological and nomenclatural study to resolve its taxonomic treatment. For quantitative characters, means with 5th and 95th percentiles in parentheses are shown.

¹Trichomes, leaves and petals of selected herbarium specimens are shown in Fig. 6.

²Including E. croaticum s.s. (Croatia, Bosnia and Herzegovina) and E. croaticum auct. from Kosovo (see Polatschek, 2013), because the latter requires further sampling and study to reach a taxonomic conclusion and is therefore not treated separately here.

³Length of petals including blade and claw.

⁴The number of stem leaves is given only for the upper half of the stem owing to unreliable counting in its lower part; lower stem leaves can be shed at the flowering stage (see Table 1).

⁵The distribution ranges of E. vitekii and E. odoratum could marginally overlap in southern Bosnia and Herzegovina (for distribution records of E. odoratum, see Bačák et al., 2026), but their precise range limits still need to be determined.

Supplementary Data

Supplementary data are available at Annals of Botany online and consist of the following. Figure S1: mitotic chromosomes of selected Erysimum accessions. Figure S2: maximum likelihood tree with quartet sampling and bootstrap support, and NeighbourNet based on RADseq data. Figure S3: violin plots showing the distribution of pairwise relatedness values between the polyploid Erysimum croaticum s.l. and other sampled species, computed from RADseq data using the PolyRelatedness tool. Figure S4: results of f-branch analysis computed from RADseq data in Dsuite. Figure S5: morphometric analyses as shown in Fig. 5, but also depicting the total canonical structure coefficients, expressing the contributions of characters to the canonical axes. Figure S6: canonical discriminant analyses depicting morphological differentiation between Balkan species Erysimum carniolicum and E. vitekii, and the core species of the E. odoratum complex, E. odoratum and E. witmannii. Table S1: list of the Erysimum populations analysed in present study, including relative nuclear DNA content values and chromosome numbers. Table S2: number of quality-filtered and deduplicated read pairs per sample, and percentage of reads mapped for Hyb-Seq and RADseq data. Table S3: assignment of alleles obtained from phased exon sequences (Hyb-Seq data) in the polyploids to diploid species using the EPA-ng algorithm. Table S4: variation in trichome types on petals, stems, leaves and siliques in the studied Erysimum species. Table S5: descriptive statistics for quantitative morphological characters measured in the studied Erysimum species.

Funding

This work was supported by the Slovak Research and Development Agency (grant number APVV-21-0044) and the Grant Agency VEGA, Bratislava, Slovak Republic (grant number 2/0010/25).

Author contributions

J.Z.-L. conceived and designed the study and wrote the paper with contributions by M.Š., R.B. and T.M. R.B., M.Š., H.M., B.Š., A.K., K.S., I.T., P.T. and J.Z.-L. participated in plant sampling. R.B., M.Š., H.M., B.Š., T.M. and J.Z.-L. generated, analysed and interpreted data. All authors commented on and approved the manuscript.

Data availability

The Illumina sequencing reads generated for this study are available from the NCBI Sequence Read Archive under BioProject PRJNA1284580, with BioSample accession numbers specified in Supplementary Data Table S1.