Introduction

The diploid species (2n = 18) Brassica oleracea L. (Brassicaceae), the most diverse species in the genus Brassica, includes several vegetable crops with a long history of cultivation and domestication worldwide, such as cauliflower, cabbage, broccoli, Brussels sprouts, kohlrabi, and kale (Golicz et al. 2016). In particular, kale leaves (B. oleracea var. acephala) has attracted great interest due to its excellent nutritional value compared to other crops. It is considered an ideal source of vitamins, essential minerals, and fatty acids (Pathirana et al. 2017). Therefore, it occupies a privileged position in the cuisine and diet of European, Asian and American populations (Gonçalves et al. 2012). Moreover, it is widely used in traditional medicine for the treatment of rheumatism, liver diseases, eye problems, bone weakness, anaemia, and obesity (Gonçalves et al. 2012; Kuerban et al. 2017; Thavarajah et al. 2016) and also as an animal feed (Cartea et al. 2003).

Brassica species are known as a highly diverse group of plants due to their ability of spontaneous cross-pollination and gene flow among sexually compatible relatives (Meglič and Pipan 2018; Pipan et al. 2013, 2011). B. oleracea var. acephala is classified into different morphotypes based on morphological characteristics. For example, kale (B. oleracea L. var. acephala DC.) has dark green and crinkled leaves, Scotch kale (B. oleracea L. var. acephala (DC.) Alef. var. sabellica L.) is characterized by grey-green and highly crinkled and wrinkled leaves, while Marrow stem kale (B. oleracea L. var. acephala (DC.) Alef. var. medullosa L.) is characterized by a soft and thick stem and different leaf types. In addition to variation among morphotypes, populations/cultivars within the same morphotype may also exhibit wide morphological variation due to a long history of domestication and adaptation to environmental conditions (Hahn et al. 2022). According to Hahn et al. (2022), three types of kale are distinguished according to their origin: curly kale (Scotch type, Brassica oleracea covar. acephala var. sabellica), Italian kale (Lacinato type, Brassica oleracea covar. acephala var. palmifolia), and collard (Brassica oleracea covar. acephala var. viridis). Indeed, in B. oleracea var. acephala, large morphological variation could occur both within populations as a result of cross-pollination and between populations due to extensive selection by farmers and/or adaptation to local environments (Cartea et al. 2003; Šamec et al. 2019a; Thavarajah et al. 2016).

Detailed agro-morphological characterization and assessment of patterns of genetic variability based on quantitative and qualitative traits is the first and crucial step in describing available genetic resources and their efficient use in breeding programs (Balkaya and Yanmaz 2005; Oumata et al. 2023). However, although agro-morphological traits can be controlled at the genetic level, they may be nonspecific and non-polymorphic, and their variation could be strongly influenced by environmental factors (Choudhury et al. 2022; Petit et al. 2020; Terlević et al. 2023). In contrast, DNA-based molecular markers are stable, intact, abundant in the genome, and free from environmental influences. Therefore, they are a powerful tool not only for comprehensive depiction the genetic diversity patterns, but also for other advanced analyses such as gene mapping and molecular marker-assisted breeding (Zhu et al. 2019). Although various molecular markers have been developed to date, simple requence Repeats (SSRs) or microsatellites remain the most attractive for genetic variability research and plant breeding because they are codominantly inherited, abundant, multi-allelic and highly polymorphic (Riangwong et al. 2020; Rivera et al. 2016), and are also easy to handle in the laboratory or automated by capillary sequencers (Schuelke 2000). Therefore, assessment of genetic variation using agro-morphological traits and SSR markers has proven to be a very powerful approach for genetic resources management, utilization, and conservation in many crops, including chickpea (Ghaffari et al. 2014), tomato (Mercati et al. 2015), pepper (Rivera et al. 2016), leafy mustard (Sharma et al. 2020), and cauliflower (Rakshita et al. 2021).

In the present study, a large set of quantitative and qualitative agro-morphological traits and SSRs markers were used to evaluate the genetic diversity and pattern of genetic structure of B. oleracea L. var. acephala collection from South-East Europe. The results of the study will be very useful for further use of the analysed accessions in ongoing and future breeding programs.

Materials and methods

Plant material

The plant material used in the study consists of 26 accessions of B. oleracea var. acephala, of which 23 accessions were B. oleracea L. var. acephala DC obtained from the two national plant gene banks, i.e. Croatia (six accessions) and Bosnia and Herzegovina (17 accessions). The remaining three accessions were commercial cultivars developed by three different seed producers. 'Kodrolistni ohrovt' and 'Krmni ohrovt' were two cultivars of Scotch kale and Marrow stem kale developed by Semenarna Ljubljana (Slovenia) and Semina Royal Seeds (Slovenia), respectively, while 'Ohrovt Nero di Toscana' was a cultivar of Lacinato kale, B. oleracea var. palmifolia developed by Franchi Sementi (Italy). All available passport data for the studied accessions are summarised in Supplementary Table S1.

Experimental design and agro-morphological characterization

In March 2019, four individuals from each accession were initially cultivated in seedling trays filled with fertilized peat. Two months after germination, seedlings were transplanted to open-field conditions and grown to maturity. Adjacent plants were spaced 50 cm apart and a black polyethylene cover was placed on the soil before planting. Each plant was phenotypically characterised using 11 quantitative and 15 qualitative non-destructive descriptors for Brassica spp. The whole plant, entire leaf, leaf blade and petiole were characterised according to the European Cooperative Programme for Plant Genetic Resources developed by the Community Plant Variety Office (CPVO 2011) and the International Board for Plant Genetic Resources (IBPGR 1990). Quantitative traits included plant height [PLANTHEIGHT], plant diameter [PLANTDIAM], plant height/diameter ratio [PLANTRATIO], leaf length [LEAFLENGTH], leaf area [LEAFAREA], leaf blade length [BLADELENGTH], leaf blade width [BLADEWIDTH], leaf blade width/leaf length ratio [BLADERATIO], petiole length [PETIOLELENGTH], petiole width [PETIOLEWIDTH], and petiole length/width ratio [PETIOLERATIO]. For qualitative traits, plant shape [PLANTSHAPE] and position of growing point relative to the top of the plant [PLANTPOSIT] were used to characterise plant shape. Leaf division (margin) [LEAFMARGIN], leaf division (incision) [LEAFINCISION], leaf apex shape [LEAFAPEX], leaf tip attitude [LEAFTIP], leaf lamina attitude [LEAFLAMINA], leaf anthocyanin coloration [LEAFANTHO], distribution of anthocyanin coloration on the leaf [LEAFDISTRIB], leaf colour [LEAFCOLOUR], and intensity of the colour of the fully developed leaf [LEAFINTENS] were evaluated to describe leaf morphology. The blade shape [BLADESHAPE] and blade blistering [BLADEBLISTER] were also used to characterise the leaf blade. Finally, the petiole by assessing petiole and/or midvein enlargement [PETIOLEMEDVEIN] and the colour of the petiole and/or midvein colour [PETIOLECOLOUR] were characterised. The full description of all analysed traits is given in the Supplementary Table S2. In addition to agro-morphological characterization, a young healthy leaf was taken from 2–3 plants per accession and stored at –80 °C for further molecular characterization. Non-healthy plants were excluded from agro-morphological and molecular characterization. The experiment was conducted at the experimental fields of the Agricultural Institute of Slovenia in Jablje (304 m a.s.l.; 46.151°N 14.562°E).

Molecular analysis

First, genomic DNA was extracted from 100 mg young and healthy leaves from 2 to 3 individuals per accession as described in (Pipan et al. 2017). DNA from all individuals was then checked for quality and quantity using a fluorimeter (Qubit 3.0; ThermoFisher Scientific, MA, USA) and then diluted to a uniform final concentration of 5.6 ng/μL and stored at –20 °C for further analysis.

DNAs from a total of 71 individuals were used separately (2–3 individuals per accession) and then amplified with a set of 12 genome-specific SSR markers previously used to analyse the genetic diversity of different B. oleracea crops (kale, cabbage, cauliflower, and Brussels sprouts) and found to be highly polymorphic (El-Esawi et al. 2016). PCR amplification was performed using fluorescently labelled universal primers according to the protocol described by (Schuelke 2000). First, for each SSR marker, the forward primer was labelled with the M13 sequence (5'-CACGACGTTGTAAAACGAC-3'). The PCR reaction was performed in a final volume of 11 μL containing 5.6 ng genomic DNA, 1 μL 10 × PCR buffer (Biotools, Spain), 0.5 μL 50 mM MgCl2 (Biotools, Spain), 0.2 μL each 10 mM dNTPs (Sigma-Aldrich, USA), 0.1 μL 10 μM forward primer (Sigma-Aldrich, USA), 0.25 μL 10 μM reverse primer (Sigma-Aldrich, USA), 0.183 μL 10 μM 5'-fluorescently labelled universal primer (with 6-FAM, NED or HEX; Omega, Slovenia), and 0.5 μL 5 U Taq DNA polymerase (Biotools, Spain). PCR amplification was performed in a thermal cycler (Veriti, ThermoFisher Scientific) using the following primer-specific touchdown profile: 94 °C for 4 min; 15 cycles at 94 °C for 1 min; temperature reduction from 60 (62) °C to 49.5 (51.5) °C at 0.7 °C per cycle for 30 s; 72 °C for 1 min; followed by 23 cycles at 94 °C for 30 s; 53 °C for 30 s; 72 °C for 1 min; and final extension for 5 min at 72 °C.

Ultimately, an Applied Biosystems ABI3130XL genetic analyser was used to resolve the DNA fragments. The allele sizes were then determined by comparison with an internal size standard (GeneScan-500 ROX; Applied Biosystems) using the GeneMapper 6.0 software (Applied Biosystems).

Agro-morphological data analysis

The obtained data were statistically analysed with a series of univariate and multivariate approaches using R statistical programming environment version 3.4.4 R Core Team (2021). First, the package "ggplot2" (Villanueva and Chen 2019) was used to visualise the frequency of distribution of all agro-morphological traits across accessions using histograms. To describe patterns of the agro-morphological variation, the package "summarytools"(Comtois and Comtois 2016) was used to compute basic descriptive statistics, including maximum and minimum values (max and min), standard deviation (SD), and coefficient of variation (CV) for the quantitative traits. Analysis of variance (ANOVA) was used to partition variation within and between accessions using the "aov" function of the "Stats" package (R Core Team 2021). For qualitative traits, the "vegan" package (Oksanen et al. 2019) was used to compute the Shannon–Weaver diversity index (Shannon and Weaver 1949) as a measure of phenotypic diversity. Patterns of correlations between pairwise of quantitative and qualitative traits was estimated separately via Pearson (r) and Cramer’s V (v) coefficients, with P < 0.05 set as significance level, using packages "corrplot" (Wei et al. 2017) and "vcd" (Meyer et al. 2021), respectively.

A multifactorial analysis (MFA) was conducted using the "FactoMineR" (Lê et al. 2008) and "factoextra" packages (Kassambara and Mundt 2020) to examine patterns of the agro-morphological differentiation and determine the factors and quantitative and qualitative agro-morphological traits that explain variability among accessions. Gower distance (Gower 1971) was calculated using the "daisy" function in the "cluster" package (Maechler et al. 2012) to create a dissimilarity matrix between accessions based on all traits, which was eventually used to visualise the relationship between accessions UPGMA clustering using the "hclust" function.

SSR data analysis

The data matrix resulting from allele sizing was first analysed using Power Marker version 3.25 with the default parameters (Liu and Muse 2005) to calculate several measures describing the genetic diversity at the loci and accession level, including number of different alleles (Na), effective number of alleles (Ne), major allele frequency (MAF), and polymorphic information content (PIC). We also used the default parameter settings in GenAlEx 6.5.2 software (Peakall and Smouse 2006) to calculate the Shannon information index (I), observed heterozygosity (Ho), and expected heterozygosity (He), as well as the fixation index (F) and the number of private alleles (No. PA). The same software was also used for the analysis of molecular variance (AMOVA) to estimate the proportion of variation explained within accessions, between accessions, and between geographic origins of accessions. In addition, to examine the genetic structure of the studied accessions using the data from the 12 SSR loci, Bayesian clustering was performed using the software STRUCTURE 2.3.4 (Pritchard et al. 2000). The analysis was performed using the Admixture model by adjusting the number of populations (K) from 1 to 10 based on 10 independent simulations for each K. Each simulation consisted of a burn-in period of 10.000 steps and 100.000 Markov Chain Monte Carlo (MCMC). The presumed number of clusters (optimal K) was then determined using the software STRUCTURE HARVESTER (Earl and VonHoldt 2012). Using the R package StructuRly 0.1.0 (Criscuolo and Angelini 2020), the Bayesian clusters of the studied accessions were visualised in the form of admixture bar plots. The genetic relationship between accessions was visualised via an unweighted pair group with arithmetic average (UPGMA) clustering using Poptree2 (Takezaki et al. 2010) based on the matrix of Nei genetic distance between accessions pairwise generated by GenAlEx 6.5.2 software (Peakall and Smouse 2006). Genetic relatedness among kale accessions was further analysed using UPGMA (Unweighted Pair Group Method Arithmetic Average) cluster analysis based on Nei genetic distance, and clustering accuracy was tested using 1000 bootstrap resamples with the R package "poppr" (Kamvar et al. 2014). Finally, the matrix of Nei’s genetic distance between accessions pairwise was used along with the dissimilarity matrix between accessions based on Gower distance to perform a Mantel test with the package "vegan" (Oksanen et al. 2019) to test the degree of association between agro-morphological and molecular variation of the studied kale accessions.

Results

Morpho-agronomic variability of kales from South-East Europe

A total of 26 quantitative and qualitative morpho-agronomic descriptors related to plants, leaves, leaf blades, and petioles were evaluated for these 26 kale accessions originating from South-East Europe. Descriptive statistics revealed that the 11 quantitative traits had moderate (10% ≤ CV ≤ 20%) to very high (CV ≥ 30%) morpho-agronomic variation for all traits with an average coefficient of variation (CV) of 25.33% (Table 1). The frequencies of distribution of the 26 analysed accessions for the 11 quantitative traits are shown in Fig. 1. Plant height was ≤ 60 cm in 16 accessions with a mean value of 61.34 cm and a CV of 24.03%, plant diameter exceeded 60 cm in 18 accessions with an average of 57.97 cm and a CV of 19.46%, while height/diameter was above the average (1.09) in 11 accessions with a CV value of 17.38%. As for leaf characteristics, the length in most accessions (18) was 31–40 cm with a mean value of 36.39 cm and a CV of 19.22%, while the area in 17 accessions was 200–400 cm2 with a mean value of 261.94 cm2 and a CV of 42.42%. Phenotypic evaluation of leaf blade showed that leaf blade length ranged from 16 to 21 cm in 13 accessions, with an average of 19.19 cm and a CV of 19.07% for all collection. Leaf blade width was ≤ 15 cm in 8 accessions and 18–19 cm in 8 accessions with an average of 16.77 cm and a CV of 22.68%, while most accessions (16) had a width/length ratio above the average (0.46) with a CV of 17.26%. As for petiole, length in most accessions was between 14 and 19 cm with an average of 17.02 cm (CV = 32.07%), width in 15 accessions was above average (1.14) with CV of 25.19%, while length/width ratio in 14 accessions was above average (16.06) with CV of 39.86%. Correlations between pairwise quantitative traits (Fig. 3A) showed that, with the exception of the ratio between plant, leaf blade, and petiole, most of the remaining traits had strong positive associations with each other (r = 0.39, P < 0.05). The highest correlations were observed between leaf and petiole length (r = 0.86), leaf area and the blade width (r = 0.82), plant diameter and leaf length (r = 0.79), and plant diameter and leaf petiole length (r = 0.76), most likely reflecting the strong developmental associations among traits related to plant vigour. Interestingly, analysis of variance revealed a significant difference (P < 0.0001) among accessions for all quantitative traits. As shown in Fig. 4A, variation between accessions far exceeded variation within accessions, with a level of variation greater than 50% and an average of 73.98% for all traits. In addition, as indicated by the Shannon–Weaver index (H'), a wide variation was observed in all qualitative morphological descriptors. For instance, plant shape (H' = 0.69) was either inverted pyramid or a dome (13 accessions). In the latter, the position of the growing point (H' = 0.61) was deeply below the plant top for 10 accessions and slightly below in only 3 accessions (Fig. 2). Regarding leaf shape, the leaf margin (H' = 1.41) was mostly crenate and serrate (9 and 6 accessions, respectively), while the incision (H' = 1.02) was mostly lyrate and sinuate (14 and 9 accessions, respectively). Moreover, most accessions exhibited either a rounded (11) or broadly rounded (8) leaf apex (H' = 1.07), the tip was mostly dropping (17) (H' = 0.76), and the lamina was mostly either concave drooping (12) or straight (12) lamina (H' = 0.91). The blade shape (H' = 1.79) was mostly obovate (11) or oblong (5), while blade blistering (H' = 1.54) was mostly low (11) or intermediate (7) (Fig. 2). In addition, anthocyanin (H' = 0.68) was visible in the leaves of only 11 accessions, 10 of which had partial distribution in the leaf (H' = 0.49). In most accessions, leaf colour (H' = 1.03) was mostly green (13) or dark green (9) and intensity (H' = 1.01) was mostly medium (12) or dark (10). Petiole colour (H' = 1.21) was mostly light green (9), green (8), or purple (8) (Fig. 2), while petiole and/or midvein enlargement (H' = 0.82) was either intermediate (13) or narrow (12) in most accessions. Moreover, as revealed by the Cramer’s coefficient (Fig. 3B), plant shape and growing point position, as well as anthocyanin coloration and its distribution in the leaf, were completely interdependent (r = 1). Some other traits appeared to be strongly (r ≥ 0.75) interdependent, such as leaf colour and its intensity, as well as the blade shape and the distribution of anthocyanin coloration on the leaf. In contrast, no association was detected between leaf anthocyanin coloration and plant shape, and a weak association was found between leaf apex shape and anthocyanin coloration and its distribution on the leaf (r ≤ 0.15).

Table 1 Summary of descriptive statistics of the 11 quantitative traits of the studied kale accessions
Full size table
Fig. 1
figure 1

Frequency distribution of the 26 kale accessions for the 11 quantitative traits. a Plant height [PLANTHEIGHT]. b Plant diameter [PLANTDIAM]. c Plant height/diameter ratio [PLANTRATIO]. d Leaf length [LEAFLENGTH]. e Leaf area [LEAFAREA]. f Leaf blade length [BLADELENGTH] g Leaf blade width [BLADEWIDTH]. h Leaf blade width/leaf length ratio [BLADERATIO]. i Petiole length [PETIOLELENGTH]. j Petiole width [PETIOLEWIDTH]. k Petiole length/width ratio [PETIOLERATIO]

Full size image
Fig. 2
figure 2

Frequency distribution of the 26 kale accessions for the 15 qualitative traits. a Plant shape [PLANTSHAPE]. b Position of growing point in relation to top of plant (only for those of dome shape) [PLANTPOSIT]. c Leaf division (margin) [LEAFMARGIN]. d Leaf division (incision) [LEAFINCISION]. (e) Leaf apex shape [LEAFAPEX]. f Leaf tip attitude [LEAFTIP]. g Leaf lamina attitude [LEAFLAMINA]. h Leaf blade shape [BLADESHAPE]. i Leaf blade blistering [BLADEBLISTER]. j Anthocyanin coloration of leaf [LEAFANTHO]. k Distribution of anthocyanin coloration on leaf [LEAFDISTRIB]. l Leaf colour [LEAFCOLOUR]. m Intensity of colour of fully developed leaf [LEAFINTENS]. n Petiole and/or midvein enlargement [PETIOLEMEDVEIN]. o Petiole and/or midvein colour [PETIOLECOLOUR]

Full size image
Fig. 3
figure 3

Patterns of correlations between pairwise quantitative A and qualitative traits B used for agro-morphological characterization of the 26 kale accessions

Full size image

Patterns of the agro-morphological differentiation among accessions

To examine patterns of the agro-morphological variation and determine the major sources of variation among kale accessions, data from both quantitative and qualitative traits as well as the morphotype and country of origin were analysed together using multifactorial analysis (MFA). The first three MFA components cumulatively explained 43.49% of the total variance. As shown in Fig. 4B, the traits that contributed most to the variation explained by the first three MFA dimensions, and therefore generally differed among the studied accessions, were the leaf margin (22.08%), leaf apex (20.85%), intensity of colour of fully developed leaf (19.72%), leaf blade blistering (16.38%) and leaf incision (15.53%). Among the quantitative traits, the petiole ratio (11.14%) and the leaf area (10.99%) and the petiole width (10.04%) were the main contributors to differentiation. Overall, the contribution of qualitative traits to the agro-morphological differentiation of the studied accessions slightly exceeded that of quantitative traits (11.28% and 10.88%, respectively). However, morphotype seems to be the main driver of agro-morphological diversification of kale accessions, contributing with 32.61% (Fig. 4B). This pattern was evidenced by the 2D plot of the first two MFA dimensions, which showed that the two Slovenian commercial cultivars of Scotch Kale (KIS19_R1) and Marrow stem Kale (KIS19_R3) and the Italian commercial cultivar of B. oleracea L. (KIS19_R2) formed three outgroups clearly separated from the other kale accessions (Fig. 5A). When only the accessions of B. oleracea L. var. acephala DC. were examined, the 2D MFA plot did not show a clear separation between the Croatian and Bosnian kales (Fig. 5B), suggesting that some variation may occur between geographically close sites within the same country.

Fig. 4
figure 4

A Variation decomposition within and among accessions for the 11 quantitative traits and B the contribution of morphotype, the 15 qualitative (filled bar plots) and the 11 quantitative (non-filled bar plots) agro-morphological traits to the total variance explained by the first three dimensions of the multifactorial analysis

Full size image
Fig. 5
figure 5

Two-dimensional multifactorial analysis plots showing the relationship patterns between all 26 kale accessions (A) and only the B. oleracea var. acephala accession based on the data of the 26 agro-morphological traits(B). C The optimal number of clusters and D the UPGMA dendrogram depicting the relationship between the 26 kale accessions based on the Gower distance matrix between pairwise accessions

Full size image

To further investigate the agro-morphological differentiation among kale accessions, the degree of dissimilarity was estimated using Gower distance (GD) based on phenotypic data of all traits. The results showed that KIS19_R1 and KIS19_R3 were the most divergent accessions (GD = 0.87), while the two Bosnian accessions KIS19_R16 and KIS19_R26 (GD = 0.22) were the most phenotypically similar, with an average GD of 0.45 between all kale accessions. Cluster analysis based on Gower distance divided the 26 accessions into 8 groups (Fig. 5C). Consistent with the MFA results, accessions KIS19_R1, KIS19_R2, and KIS19_R3 diverged from the other accessions in three different groups (C6, C1, and C2). Cluster C2 included only Bosnian accessions, while the four remaining clusters were a mixture of Croatian and Bosnian accessions (Fig. 5D).

Microsatellite polymorphism and genetic diversity level

The twelve SSR markers used to characterise the genetic diversity of the 26 kale accessions yielded a total of 108 alleles, with the observed number of alleles (Na) ranging from three (Na14-C12) to 14 (Ol10-A03a and Ol13-C12), with an average of nine alleles per locus. The effective number of alleles (Ne) ranged from 2.01 to 3.16 in Na14-C12 and Ol10-A03a, respectively, with a mean of 2.62. The 12 SSR markers had a high polymorphism rate, with a polymorphism information content (PIC) of 0.4 (Na14-C12) to 0.86 (Ol12-F02) and a mean PIC of 0.68. Major allele frequencies were lowest in Ol12-F02 (MAF = 0.27) and highest in Na12-C08 (MAF = 0.54), with a mean frequency of 0.39. In addition, observed heterozygosity (Ho) varied from 0.76 (Ol13-C12) to 1 (Ra2-E11, Ol12-F11, and Ra2-E03) with a mean of 0.95. Importantly, high gene diversity expressed by expected heterozygosity (He) was observed in all SSR markers, with values ranging from 0.52 in Na14-C12 to 0.87 in Ol12-F02 and He of 0.73 in all markers. This pattern is reflected in the negative Fis values obtained for all markers, indicating a general excess of heterozygosity (Table 2).

Table 2 Genetic diversity parameters of the 12 simple sequence repeat (SSR) markers used for molecular characterization of kale accessions
Full size table

Genetic diversity estimates obtained at the accession level are shown in Table 3. The lowest number of observed (Na) and effective (Ne) alleles was observed in accessions KIS19_R21 and KIS19_R27 (Na = 1.750; Ne = 1,750), whereas the highest allelic richness was observed in accessions KIS19_R1 (Na = 3.667; Ne = 3.264), KIS19_R5 (Na = 3.5; Ne = 3.160), and KIS19_R11 (Na = 3.667; Ne = 3.124). Across all analysed accessions, the mean values for Na and Ne were 2.881 and 2.616, respectively. Moreover, the Shannon diversity index (I) and expected heterozygosity (He), with mean values of 0.953 and 0.578 across accessions, respectively, indicated that accessions KIS19_R11 were the most genetically diverse (I = 1.887 and He = 0.662), whereas accessions KIS19_R21 and KIS19_R27 were the least diverse (I = 0.587 and He = 0.417). Remarkably, the observed heterozygosity was higher than the expected heterozygosity in all accessions, with an average value of 0.939, resulting in negative inbreeding coefficients (Fis) that averaged -0.671, confirming the high level of heterozygosity in the entire collection.

Table 3 Sample size (N) and genetic diversity estimates among the 26 kale accessions
Full size table

Genetic differentiation and population structure

Analysis of molecular variance (AMOVA) revealed that 94% of the genetic variation occurred within accessions. Genetic differentiation among accessions (Fst) ranged from 0.01 to 0.294, with an average of 0.06 and a level of gene flow of 3.55 migrants per generation. However, no genetic differentiation was observed between countries of origin (Fst = 0). Bayesian clustering using STRUCTURE revealed that ΔK reached its maximum (ΔK = 3.84) at K = 4, indicating that the accessions could be divided into four main clusters (Fig. 6A) without a geographic clustering pattern. With the exception of the Italian commercial cultivar (KIS19_R2), the Croatian accession KIS19_R9, and the three Bosnian accessions KIS19_R20, KIS19_R25, and KIS19_R27, which were assigned to cluster 3 with a membership coefficient Q > 75%, the remaining accessions exhibited mixed clustering with partial membership in multiple clusters (Fig. 6B). Remarkably, very little divergence in allele frequency (DAF) was detected among the four clusters (< 5%), with completely identical genetic composition in Cluster 2 and Cluster 4 (DAF = 0), indicating little genetic differentiation among kale accessions. Similar to the STRUCTURE based clustering, UPGMA clustering based on Nei’s genetic distance did not reveal any geographic grouping patterns. As shown in Fig. 7, the 26 kale accessions could be divided into two main groups with 100% bootstrap support. The first group included six Bosnian accessions, while the 20 remaining accessions were mixed in a second major group that was divided into several subgroups differentiated with low bootstraps support. Within the second group, the three commercial cultivars were separated from the remaining accessions in a single group.

Fig. 6
figure 6

A A graph showing the relationship between ΔK and K values; B the inferred genetic structure of the 26 kale accessions based on Bayesian clustering of the 12 simple sequence repeat (SSR) marker data

Full size image
Fig. 7
figure 7

UPGMA clustering showing the relationship between the 26 kale accessions based on Nei’s genetic distances derived from the 12 simple sequence repeat (SSR) markers data. Values in the branches indicate the clustering support at 1000 bootstraps resamples

Full size image

Association between agro-morphological and molecular variation

The Mantel test revealed a very low and non-significant association between the Nei’s genetic distance matrix generated from the molecular data and the Gower distance matrix among accessions, with a cophenetic coefficient (r) of 0.06 (P = 0.213). When only the qualitative trait data were used, r was 0.06 (P = 0.249) and dropped to 0.03 (P = 0.520) when only the quantitative traits were used. This result suggests that molecular differentiation is not correlated with agro-morphological differentiation among kale accessions, as clearly evidenced by the different patterns of clustering in our study.

Discussion

Detailed genetic characterization of a particular germplasm is instrumental for establishing appropriate management, preservation and breeding practices. Within the Brassicaceae family, B. oleracea var. acephala L. (leafy kale) has attracted great interest worldwide as a 'superfood' and ornamental plant (Šamec et al. 2018). Therefore, several investigations have already been conducted to characterize the genetic diversity of local accessions in different countries based on agro-morphological or nutritional traits, e.g. in Portugal (Dias et al. 1994), Spain (Cartea et al. 2003; Padilla et al. 2007), Turkey (Balkaya and Yanmaz 2005), Croatia (Šamec et al. 2019a) and Italy (Lotti et al. 2018). Some other investigations have been conducted using molecular markers such as random amplified polymorphic DNA (RAPD) (Farnham 1996; Okumus and Balkaya 2007), amplified fragment length polymorphism (AFLP) (Christensen et al. 2011), single nucleotide polymorphism (SNP) (Hahn et al. 2022) and SSR markers (Šutković et al. 2021). In addition to these valuable reports, detailed characterization of kale genetic resources will be very useful to breed superior accessions, promote their valorisation and management to halt genetic erosion and economic losses due to climate change.

In the current study, a large number of qualitative and quantitative agro-morphological traits and SSR markers was combined to characterise the genetic variation of 23 Croatian and Bosnian leafy kale accessions for inclusion in future breeding programmes. Similar to previous research on Spanish (Cartea et al. 2003) and Indian (Gorka et al. 2018; Singh et al. 2017) accessions, our study showed considerable agro-morphological variation in quantitative traits among the studied kale accessions. Despite the similarity in leaf morphology in terms of length and width, the Indian kale accessions had shorter plant height (average 47.71 cm versus 61.34 cm) compared to the accessions analysed in our study (Singh et al. 2017). In contrast, the Spanish accessions were characterised by outstanding plant vigour with plant height exceeding 160 cm. Interestingly, most of the quantitative traits analysed in our study showed strong positive correlation among themselves, indicating polygenic control through the pleiotropic effect of genes on different traits or linkage disequilibrium among different loci (Chebib and Guillaume 2021). Similar results were also observed in B. oleracea L. var. botrytis (Kumar et al. 2019), B. rapa var. rapa L. (Khadivi et al. 2022) and B. carinata A. Braun (Yimer et al. 2021) accessions. Moreover, our results revealed that most of the phenotypic variation in quantitative traits resided among accessions. Overall, the strong correlations among traits and the large phenotypic diversity among accessions provide a great opportunity to use both traits and accessions analysed in our study as a source of genetic diversity in future breeding programmes. In addition, a large phenotypic diversity was found in qualitative traits (H' = 0.49 – 1.54). Similar to the Indian accessions analysed by (Gorka et al. 2018), most of our accessions were either an inverted pyramid or a dome shaped with green leaves. In contrast to the low variation in leaf blade shape among Indian accessions, which were mainly characterised by elliptic blade shape, this trait was most variable among the accessions analysed in our study (H' = 1.54).

Previous research has demonstrated that leaf shape is strongly influenced by environmental factors in addition to its genetic control (Dkhar and Pareek 2014) and that its variation may be the result of strong adaptation to environmental stresses (Kidner and Umbreen 2010). Šamec et al. (2019a) reported that kale exhibits broad tolerance to different temperatures. On the other hand, variation in leaf shape in kale is thought to be the result of selection by the breeder (Arias et al. 2021). Thus, the variation in leaf shape observed in our study could be due to the adaptation of the studied accessions to their local environment, since they originated from different locations. Cartea et al. (2003) also considered that the agro-morphological variation among Spanish kale accessions could be due to both adaptation to local ecological conditions and selection by farmers. In addition, the MFA analysis revealed that qualitative rather than quantitative traits contributed most to agro-morphological diversification among the studied kale accessions. The MFA analysis and UPGMA clustering based on the Gower distance also revealed that the three commercial cultivars were classified as independent outliers that deviated from the other kale accessions, with geographic origin being an important contributor to diversification. A similar pattern of agro-morphological differentiation was observed in cabbage and Galega kale landraces from Portugal (Dias et al. 1994) and Spanish kale populations (Cartea et al. 2003). The considerable effect of geographic origin on differentiation and the evidence of adaptation to the local environment suggest that the influence of genotype by environment on several agro-morphological traits is important. Moreover, the divergence among the three commercial cultivars with each other and with the kale accessions was expected, as they belong to different morphotypes and indicate that they were selected for different agro-morphological traits. Interestingly, in agreement with our results, previous research has shown that qualitative traits such as leaf colour and division, leaf blade blistering, anthocyanin coloration, and anthocyanin coloration distribution were the main drivers of variation in Croatian accessions of B. oleracea var. acephala with different seed source (Batelja et al. 2009; Šamec et al. 2019a).

In addition to agro-morphological characterization, molecular characterization, particularly via codominant markers such as microsatellites (SSRs), is required to provide a robust appraisal of the genetic diversity of kale accessions. The joint analysis maximizes the comprehensive assessment of the diversity of available genetic material, facilitates its organization, and enables the identification of desired traits and distinctive alleles, allowing the accurate selection of promising genetic resources for breeding programs (Li et al. 2020). In our study, the twelve SSR markers used for genetic characterization of kale accessions, previously shown to be highly informative (El-Esawi et al. 2016), were found to be highly polymorphic with a mean PIC value (0.68) and an average of nine alleles. These values were higher than those reported for the same SSR set in Irish accessions (El-Esawi et al. 2016) and for other SSR markers in different commercial cultivars of B. oleracea L. (Izzah et al. 2013; Tonguç and Griffiths 2004). This result indicates the high reliability of these SSR markers for the analysis of the genetic structure of kale accessions.

Our results revealed that the accessions analysed in this study had high genetic diversity (He = 0.578), which exceeded the level of Irish kale (He = 0.459), cabbage (He = 0.548), cauliflower (He = 0.446) and sprouts (He = 0.536) using SSR markers (El-Esawi et al. 2016). Moreover, similar to the above study, a significant excess of heterozygosity was observed in our study as reflected by the high observed heterozygosity and negative inbreeding coefficient (Fis). This was to be expected considering the small number of individuals within the studied accessions (2 to 3 individuals) and the outcrossing breeding system of B. oleracea L.

On the other hand, in contrast to the results we obtained for quantitative traits, the SSR data revealed that most of the genetic variation occurred at the level within accessions (95%), indicating low genetic differentiation and weak genetic structure in the studied accessions (Fst = 0.05). Although genetic differentiation was greater, a similar pattern was observed in Irish kale (Fst = 11%), cauliflower (Fst = 11%), and Brussels sprouts (Fst = 9%) (El-Esawi et al. 2016), as well as in closely related cultivars such as B. oleracea L. var. italica (Fst = 13%) (Shen et al. 2021) and B. rapa subsp. rapa L (Fst = 10%) (Soengas et al. 2011). The low genetic differentiation observed in our study was supported by the relatively high gene flow observed between accessions (Nm = 3.55), which was higher than the level observed for B. rapa subsp. rapa (Nm = 2.25) (Soengas et al. 2011) and B. oleracea L. var. italica (Nm = 1.74) (Shen et al. 2021). Such excellent gene flow can occur naturally or via seed exchange between farmers given the open pollination in B. oleracea L. In agreement with the above research, as well as with the study of Shen et al. (2021) on B. oleracea L. var. italica, the weak genetic structure of the studied accessions was confirmed by UPGMA clustering and Bayesian analysis, which did not reveal a clear relationship between clustering of accessions and geographical origin. A similar pattern was also observed in wild populations of B. oleracea L. (Christensen et al. 2011; Lanner-Herrera et al. 1996; Mittell et al. 2020) and B. oleracea L. var. italica (Shen et al. 2021). In contrast, in a large collection of Spanish B. oleracea L. wild accessions, restricted gene flow (Nm = 0.26) and large genetic differentiation (Fst = 55.1%) were observed, mainly promoted by geographic isolation (Tortosa et al. 2017). However, in our study, UPGMA clustering showed that the three commercial standard cultivars, although belonging to three different subspecies, were clearly differentiated from kale accessions in a single group, suggesting that their genomes harbour common genomic regions and could be bred for similar agronomic traits. Moreover, our results revealed that the grouping of accessions based on the agro-morphological data did not match the grouping determined from the SSR data. This was confirmed by the weak and non-significant association between the dissimilarity matrices between the accessions determined by the Mantel test for both pairs of data. The weak association between the phenotypic and genotypic matrices was also reported in broccoli (Li et al. 2019) and cabbage (Kang et al. 2011) and several other crops (Hartings et al. 2008; Roldan-Ruiz et al. 2000; Soriano et al. 2016). Previous research indicated that such incongruence highlights the complementarity and non-overlap between the agro-morphological and genotypic data to investigate the extent and nature of genetic variability in crops (Da Silva et al. 2017; Darkwa et al. 2020; Geleta et al. 2006). Importantly, this pattern indicates that the polymorphism detected by SSR markers does not fully contribute to the agro-morphological variability among kale accessions, suggesting that only a few genes are involved in the desired agro-morphological traits. Another plausible explanation could be the massive influence of genotype-environment interaction on the quantitatively inherited complex agro-morphological traits controlled by several genomic regions, including SSR loci which were not used in the present study.

Conclusion

In conclusion, the present study presented a detailed assessment of the genetic diversity and structure of kale accessions using a large set of agro-morphological traits and SSR markers. The results revealed that the kale accessions exhibit wide variation in several economically important agro-morphological traits. The agro-morphological characterization revealed that qualitative traits contributed more than quantitative traits to phenotypic differentiation among accessions. The diversity among accessions in some particular traits such as leaf shape, leaf colour and distribution of anthocyanin coloration on the leaf is useful to select a subset of accessions for breeding programs. The SSR markers used in our study, which proved to be very useful for molecular charcterization, showed great genetic diversity at the molecular level. As expected, the SSR data showed a large intra-accession variation and a weak genetic structure non-associated with the geographic origin of the kale accessions. Importantly, the weak and nonsignificant association between dissmilarity matrices among accessions generated by agro-morphological and SSR data indicates the complementarity of the multidisciplianry approaches used in our study. Overall, the results of our study pointed to the importance of the kale collection studied as a valuable reservoir of agro-morphological and genetic variability that could be exploited in future breeding programs.