Introduction

The olive (Olea europaea L.) is one of the major tree crops in the Mediterranean Basin, where more than 90% of the world’s olive oil production is realized (IOC 2016). Olive cultivation has a history of approximately 5000–6000 years (Vossen 2007) with intense presence in nutrition, culture and commerce of many different civilizations throughout history. The dominant theory asserts that the first olive domestication occurred in the Middle East. However, other scenarios of local domestication events, as well as admixture shaping olive germplasm, have also been proposed (Diez et al. 2015). Based on the post-glacial history of oleaster populations, Greece is considered as one of the main regions of primary domestication and secondary diversification of the olive tree (Besnard and Rubio de Casas 2016). Archaeological evidence reveals that the cultivated olive was initially domesticated from a Near Eastern and Aegean gene pool by Phoenicians and Greeks (Newton et al. 2014). Subsequently, olive was introduced in other Mediterranean areas such as Marseille, France, 2500 years ago (Breton et al. 2009), contributing to the evolution of existing olive germplasm in recent times. The long-standing occurrence of the olive in the Mediterranean is also supported by molecular phylogenetic analyses (Besnard et al. 2013).

Greece is the third largest olive oil-producing country with an average annual production of 320,000 t and over one million hectares of olive trees (IOC 2016). Turkey is the second largest table olive-producing country with an average annual production of 225,000 t and approximately 800,000 ha of olive trees (IOC 2016). Crete is considered among the first areas of olive cultivation in ancient times (Lanza 2011). Indeed, the trade of olive products contributed to the wealth of the Minoan Civilization approximately 3000 bc. Currently, more than 30 million trees are cultivated on the island, the vast majority of which are cvs. ‘Koroneiki’, ‘Mastoidis’ and ‘Throubolia’.

The genetic characterization of olive germplasm is a fundamental task for establishing the origin of the currently grown cultivars, for selecting cultivars with important agronomic traits, and for developing new cultivars customized to modern plantation systems or adapted to adverse environmental conditions. Additionally, it is a prerequisite for the certification of high-quality true-to-type propagation material. Morphological descriptors and molecular markers have been the main tools used for this purpose. Indeed, microsatellites, or simple sequence repeats (SSRs), are the most widely used molecular markers for the identification of cultivars in gene banks (Trujillo et al. 2013) as well as for establishing the relationships between cultivated and wild olives (Belaj et al. 2011). This is due to their higher reproducibility and non-anonymous and co-dominant nature, especially when compared to previously employed markers such as random amplified polymorphic DNA (RAPD) and amplified fragment length polymorphism (AFLP; discussed in Belaj et al. 2003).

Greece and Turkey have a long history of interactions resulting in similarities in cultural and economic aspects of their modern societies, including agricultural production patterns and cultivars. Consequently, two major Turkish olive cultivars were included in the present study in order to investigate their potential relationship with Greek germplasm. For one of them, cv. ‘Gemlik’, Yoruk and Taskin (2014) indicated that it is not related to any cultivars from other regions in Turkey. In Greece, studies assessing genetic diversity of olive tree germplasm usually involve only one or a very small number of independent genotypes per cultivar derived from germplasm banks and focus on inter-cultivar rather than on intra-cultivar diversity (Linos et al. 2014; Xanthopoulou et al. 2014; Roubos et al. 2010; Owen et al. 2005; Terzopoulos et al. 2005; Hagidimitriou et al. 2005; Nikoloudakis et al. 2003). At present, the genetic constitution of traditionally employed autochthonous materials in olive orchards, especially before the government-directed introductions of few selected clones, is still critically poorly understood. Indeed, one of the aims of the present paper is to provide an initial genetic characterization of such autochthonous materials. This could, in turn, provide initial insights into the origin of the currently grown cultivars in Crete, Greece, as well as into possible exchanges of olive tree germplasm across the two sides of the Aegean Sea during historical times. Additionally, and to the best of our knowledge, there exist no data on the genetic characterization of wild olives in Greece. However, given the long history of olive growing in Greece, as well as the traditional germplasm selection practices applied, wild olives are expected to represent a valuable source of genetic variability. In fact, some important agronomic traits that are hardly found in olive cultivars have been observed in the wild germplasm (Colella et al. 2008; Mkize et al. 2008; Belaj et al. 2011).

In the present study in Crete, one of the centres of ancient olive cultivation, nuclear SSR markers were employed for phylogenetic inferences as well as for the evaluation of the degree of genetic variation (i) within and between the three major autochthonous cultivars grown in Crete, (ii) between the same cultivars and wild olive trees in Crete, and (iii) between the same cultivars and two autochthonous cultivars of major significance for olive growing in Turkey.

Materials and Methods

Plant Material

Five olive tree cultivars of high economic importance in Greece and in Turkey and wild individuals growing in Crete, Greece, were included in the present study (Table 1). Representative trees of the Greek cvs. ‘Koroneiki’, ‘Mastoidis’ and ‘Throubolia’ were identified and leaves were collected from the major traditional cultivation areas of each cultivar in the olive growing regions on the island of Crete. The focus of the present work lies in autochthonous germplasm. Consequently, only mature trees (i.e., over 40 years old) from old olive orchards were sampled based on morphological identification. This was done to avoid young plants massively distributed by nurseries during the recent two to three decades following clonal propagation of reference materials. The cv. ‘Mastoidis’ is mainly cultivated in mountainous areas of western Crete, cv. ‘Throubolia’ is grown in mountainous areas of central Crete and cv. ‘Koroneiki’ dominates agricultural lands across the island (Fig. 1). Morphological characteristics of individual trees (mostly of leaves, fruit and crown form) grouped within previously defined ranges for each cultivar according to the RESGEN project of the International Olive Council as presented in brief in the World Catalogue of Olive Varieties (2000). Reference genotypes for each cultivar from the National Germplasm Collection of Olive Tree of Greece at the Institute for Olive Tree, Subtropical Plants and Viticulture (HAO Demeter) were also included in the analysis. In addition, leaf samples from mature wild olive trees from Crete were collected and analysed in a way identical to cultivars. Wild olive trees were sampled in the vicinity of old olive orchards (those employed for cultivar characterization) on uncultivated lands with natural vegetation with free bush shape characterized by small endocarp and thin mesocarp. Similar to the Greek cultivars, samples from mature trees over 40 years old of the two major Turkish cvs. ‘Gemlik’ and ‘Samanli’ were selected from their main areas of cultivation located in the broader areas of Bursa and Yalova, Turkey.

Table 1 Olive cultivars and wild populations (AGR) analysed, regions of cultivation and uses
Full size table
Fig. 1
figure 1

Main areas of cultivation of olive cvs. ‘Koroneiki’, ‘Mastoidis’ and ‘Throubolia’ in the island of Crete, Greece

Full size image

DNA Extraction

Total genomic DNA was isolated from leaf material using the DNeasy Plant Mini kit (Qiagen, cat. no 69104) according to manufacturer’s instructions. Initial grinding was conducted using the automated grinder TissueLyser (Retsch, Germany) in the presence of liquid nitrogen. DNA was quantified using the Hoechst 33258 fluorescence dye (Sigma, No. B2883) on a computerized TD 700 Fluorimeter (Turner Designs, Sunnyvale, CA, USA) against a series of calf thymus DNA standards (Sigma, No. D4764).

Microsatellite Assay

Seven microsatellite loci that have been used in the past for similar analyses of olive germplasm were selected on the basis of their informativeness (values of polymorphic information content; PIC), ease of scoring and observed heterozygosity. These were OeUA-DCA3, OeUA-DCA15 (Sefc et al. 2000), UDO99-019, UDO99-031 (Cipriani et al. 2002), IAS-Oli23, IAS-Oli26 and IAS-Oli27 (Díaz et al. 2006). Polymerase chain reactions were carried out in a 20-μl reaction in a Perkin Elmer 9600 thermocycler including 25 ng of template DNA, 0.2 mM of each dNTP, 0.2 μM of each primer, 2.5 mM MgCl2 and 1 U of DiaTaqDD Polymerase (HyTest, Finland). Thermal cycling included initial denaturation for 5 min at 95 °C, followed by 35 cycles of 95 °C for 30 s, the corresponding annealing temperature for 45 s and 72 °C for 45 s, with a final extension at 72 °C for 10 min. Amplicons were denatured at 97 °C in the presence of formamide for 3 min and loaded on an 8% polyacrylamide gel automated sequencer IR2 (LICOR, NE, USA) at the Microchemistry Laboratory of the Foundation for Research and Technology (FO.R.T.H.), Heraklion, Crete, Greece. The LICOR proprietary software SAGA GT® was employed for IR2 gel characterization. Genotyping was initiated by employing loci with high discriminating power according to the existing literature and adding one locus at a time in a step-wise fashion. The optimal tally of loci (locus1 + locus2 + …) whereupon no further discrimination between genotypes could be attained was retained for subsequent genetic analyses.

Data Analysis

Within SAGA GT, SSR allele sizing and binning were manually performed or verified while additional measures were taken to ensure high accuracy and reproducibility of gel-based SSR data. Such measures involved the use of multiple ladder lanes within a single gel, use of common samples between gels and employing the “de-smiling” function of SAGA GT.

Per locus, allele sizing was based on published repeat patterns. Data matrices were produced, and, as a first step, genetic diversity measures were determined for each employed locus across all fingerprinted genotypes (n = 112) as shown in Table 2. These measures included (i) individual locus polymorphic information content (PIC; Botstein et al. 1980), (ii) observed heterozygosity (HO) and (iii) expected heterozygosity (HE). PIC, HO and HE were calculated using CERVUS ver. 3.0.3 (Kalinowski et al. 2007). Probability of identity (PI) was calculated using the software Identity 1.0 (Sefc et al. 1997). To test for the overall significance of variance partitioning within and between Cretan and Turkish olive tree cultivars (including the wild populations from Crete), an analysis of molecular variance (AMOVA) was performed using GenAlEx ver. 6.5 as a plug-in module within Microsoft Excel (Peakall and Smouse 2006). The number of alleles per locus (Na), effective number of alleles (Ne) and Shannon information index (I) and inbreeding coefficient (FIS) were also calculated using GenAlEx. A similarity matrix was produced employing the Lynch distance metric (Lynch 1990), which is a simple band-sharing measure [termed the “Band” coefficient within NTSYSpc ver. 2.21L (Rohlf 2008)]. Subsequently, a genetic similarity tree was constructed employing hierarchical clustering and the agglomerative un-weighted pair group method with arithmetic mean (UPGMA) algorithm (Sneath and Sokal 1973) in NTSYSpc ver. 2.21L (Rohlf 2008).

Table 2 Size ranges, number of observed alleles (Na), observed (HO), expected heterozygosity (HE), polymorphic information content (PIC), probability of null alleles (r) and probability of identity (PI) for seven microsatellite loci used in 112 olive accessions
Full size table

Classification Binary Tree Analysis

To reveal the relationships among unique genotypes, the data set was trimmed to include all molecularly differentiated individuals by eliminating all but one identical individual. The resulting matrix of 68 genotypes by seven SSR loci was then subjected to classification binary tree (CBT) analysis (Breiman et al. 1984). Similar to the genetic analyses, the CBT input dataset consisted of SSR loci allelic sizes. The output of CBT, called a mobile, initially entails the split of the original sample set into two parts on the basis of a criterion involving one or two discriminatory loci in a simple algebraic expression. All polymorphic loci retained in the construction of a mobile comprise the best diagnostic variable set for the groups already defined prior to the analysis. The groups at the beginning of a CBT construction were the five olive tree cultivars and one wild olive population. The tree of a mobile grows according to splits that produce maximally informative and ‘pure’ groups according to an impurity function (Breiman et al. 1984). The reduction of error in the classification is monitored by means of an information-theory-compatible Gini-Simpson index in the form of a ‘towing’ loss function, using the towing coefficient originally proposed by Breiman et al. (1984). CBT analysis was performed using routines and packages within the R environment (R Development Core Team 2008) and used the packages ‘rpart’ (Therneau and Atkinson 1997; R package version 3.1-52, http://mayoresearch.mayo.edu/mayo/research/biostat/splusfunctions.cfm) and ‘vegan’ (Dixon 2003) as implemented by R package version 2.0-3 (http://CRAN.R-project.org/package=vegan). Some algorithms and graphics were applied using SYSTAT.

Results

Overall Microsatellite Diversity

One hundred twelve independent genotypes were fingerprinted using seven SSR loci. Size ranges, number of detected alleles (Na), observed (HO) and expected heterozygosity (HE), polymorphic information content (PIC), probability of null alleles (r) and probability of identity (PI) for seven microsatellite loci used in 112 olive genotypes (cultivars as well as wild olive populations from Crete) are presented in Table 2. All seven SSR loci employed herein were highly polymorphic, yielding a total of 81 alleles, and were able to uniquely distinguish 68 out of the 112 genotypes analysed (the former number representing the number of distinct SSR profiles produced in the framework of the present study). Further, all loci were characterized as highly informative since PIC > 0.5.

The number of alleles produced per locus ranged from 6 (OeUA-DCA15) to 19 (IAS-Oli27), with a mean value of 11.6 alleles per locus (Table 2). The overall HO ranged from 0.674 (UDO99-031) to 0.991 (IAS-Oli23), with an average of 0.852 (Table 2). Expected heterozygosity (HE) ranged from 0.637 (UDO99-019) to 0.894 (IAS-Oli27), with a mean value of 0.789 (Table 2). Furthermore, the polymorphic information content of each locus (PIC) ranged from 0.575 (UDO99-019) to 0.881 (IAS-Oli27), with a mean value of 0.755. Both HE and PI values indicated that IAS-Oli27 was the most informative and UDO99-019 was the least informative marker. The probability of null alleles (r) had a mean value of − 0.0425 and ranged from + 0.0979 (UDO99-031) to − 0.258 (UDO99-019), while two loci (OeUA-DCA15 and UDO99-031) exhibited an r > 0.05 (Table 2). The overall probability of identity (PI) was 6.73 × 10−09.

When cultivars and wild olive genotypes were considered, 81 alleles were found. When only cultivar genotypes were considered, the number of detected alleles decreased to 58 (data not shown), yielding an average value of 8.3 detected alleles per locus. This represents 23 fewer alleles (28.4%) compared to a combined pool of all cultivars and wild genotypes. Detected private alleles and their sizes are presented in Table 3. A substantial number of private alleles was detected, especially in wild olives (23), followed by Greek and Turkish cultivar genotypes with eight and three private alleles, respectively.

Table 3 Private allele lengths per SSR locus and per cultivar and wild olive tree group (AGR)
Full size table

Genetic Differentiation and Relationships Between Wild and Cultivated Olives From Greece and Turkey

The distance metric of Lynch (1990) was utilized for clustering of the analysed genotypes (Fig. 2). All five investigated cultivars (‘Koroneiki’, ‘Throubolia’, ‘Mastoidis’, ‘Gemlik’ and ‘Samanli’) clearly separated into cultivar-specific groups. Interestingly, the wild olive genotypes did not cluster into a single group but rather as three sub-groups, each joining one of the local cultivar groups of Crete (‘Koroneiki’, ‘Mastoidis’ or ‘Throubolia’). In fact, the Cretan cv. ‘Throubolia’ forms one of the two major sub-clusters together with the non-Cretan cvs. ‘Gemlik’ and ‘Samanli’. The other major sub-cluster includes the Cretan cvs. ‘Koroneiki’ and ‘Mastoidis’. For the Cretan cultivars, the average number of alleles per locus was 4.57 for ‘Throubolia’, 4.14 for ‘Koroneiki’ and 3.43 for ‘Mastoidis’. A higher average number of alleles were detected in wild olives (9.14), and lower values were found in the two Turkish cvs. ‘Gemlik’ and ‘Samanli’ (2.86) (Table 4). The average number of effective alleles (Ne) was similar for the Greek cultivars (2.22–2.55) and was higher in wild olives (6.09) and lower in cultivars from Turkey (2.27–2.45). The Shannon information index (I) was higher in wild olives (1.91), followed by Cretan (0.92–0.98) and Turkish cultivars (0.84–0.85). In general, higher values were obtained for observed (HO = 0.789–1) than for expected (HE = 0.498–0.809) heterozygosity. For observed heterozygosity, cvs. ‘Mastoidis’ and ‘Gemlik’ exhibited the highest (H O  = 1) while cv. ‘Throubolia’ exhibited the lowest (HO = 0.738). Wild olives exhibited the highest expected heterozygosity (HE = 0.809) while cv. ‘Throubolia’ exhibited the lowest (HE = 0.498). On a per-cultivar basis, all HO were higher than HE. Specifically, cvs. ‘Gemlik’, ‘Koroneiki’, ‘Samanli’, ‘Throubolia’ and ‘Mastoidis’ showed 83.5, 50.6, 69.0, 48.2 and 69.8% increases, respectively. Average values of HO across all loci were slightly higher (7.3%) in comparison to their respective HE, indicating the presence of out-breeding. Higher HO than HE, determined across all loci, is in turn reflected in the negative inbreeding coefficient (FIS) values. FIS values ranked in diminishing order as follows: (i) Turkish cultivars, (ii) Greek cultivars and (iii) wild olives (Table 4).

Fig. 2
figure 2

UPGMA phenogram obtained from 7 SSR loci (81 alleles) using the band similarity coefficient for 112 olive samples of Cretan and Turkish olive and wild olive tree genotypes

Full size image
Table 4 Genetic diversity indexes of olive cultivars and wild olive populations (AGR) in Crete, Greece, as well as of two major Turkish olive cultivars
Full size table

Analysis of molecular variance (AMOVA) significantly partitioned genetic variance into two hierarchical levels (Table 5), one among genotypes belonging to the same cultivar (within cultivars; 48%), and another among genotypes belonging to different cultivars (between cultivars; 52%).

Table 5 AMOVA summary table for Cretan and Turkish cultivars
Full size table

The mobile of the CBT is shown in Fig. 3 and includes all seven microsatellite loci. The root of the mobile tree includes all five local cultivars as well as the wild population (defined in the ab initio conditions as a singular group), while at the leaf (top) of the tree all five local cultivars and the wild population are recovered. The mobile efficiently describes the loci data in the cultivars while the overall error reduction is 0.97. Table 6 indicates the proportional reduction in error and the improvement achieved at the different steps of the CBT scheme. Six of the loci were used only once, while one microsatellite locus (OeUA-DCA16) appeared twice (once in the separation of cv. ‘Koroneiki’ from the rest and subsequently to separate cv. ‘Throubolia’ from cvs. ‘Gemlik’ and ‘Samanli’ (see Fig. 3)).

Fig. 3
figure 3

Mobile constructed by CBT classification of three Greek cultivars (‘Mastoides’/MAS, ‘Koroneiki’/KOR and ‘Throubolia’/THR), two Turkish cultivars (‘Gemlik’/GEM, ‘Samanli’/SAM) and wild olive genotypes/AGR in Crete. All terminal leaves of the mobile have zero impurities, that is, any cultivar is contained within its own leaf. Moreover the Turkish cultivars (GEM and SAM) which are geographically distantiated from the rest are contained in neighbouring leaves of a single terminal branch, are separated on the basis of the OeUA-DCA3 locus and are distinguished from cv. THR on the basis of the UDO99032 locus

Full size image
Table 6 CBT analysis of Greek cultivars and wild olive populations in Crete and reference Turkish olive cultivars using the Gini-Simpson index
Full size table

Discussion

The main target of this study was the introduction of seven polymorphic SSR markers in order to (1) assess genetic polymorphism and (2) establish phylogenetic relationships within and between wild and cultivated olives in Crete employing as out-group representative genotypes from two major Turkish cvs., ‘Samanli’ and ‘Gemlik’. The high degree of molecular polymorphism (as evidenced by overall PI), the proper grouping of genotypes into cultivar-specific clusters and the fact that gradual increasing of employed loci up to seven did not allow for any further separation of genotypes indicated that seven loci were adequate for the purposes of the present study. Although important, the number of loci used is not always crucial, as few polymorphic SSR markers may discriminate more than 100 olive genotypes (Baldoni et al. 2009). Bracci et al. (2009) initially employed 12 SSR loci and identified 60 of the 63 olive tree accessions from Liguria (Italy) and the Mediterranean. Nevertheless, they also found that the same resolution could be obtained by employing a subset of six properly selected SSR loci. Previously, Díaz et al. (2006) had found that all pairs of 51 cultivars, originating from different countries and analysed within their study, could be distinguished using just three selected microsatellites (IAS-oli23, IAS-oli26 and IAS-oli27), which were also employed within the present study.

One indication of cultivar genetic isolation and distinctness is the number of private alleles (i.e. alleles uniquely detected among individuals of a specific cultivar/population). Our approach allowed for the identification of novel as well as private allelic variants. The latter may be useful for the classification of unknown individuals into a cultivar/population, although private alleles are not detected in all of the individuals in a specific cultivar/population (floating frequency < 1). The number of detected alleles per locus compares favourably with the highest encountered in the literature (ranging from 3 to 16; Alba et al. 2009; Bandelj et al. 2002; Belaj et al. 2010; Cipriani et al. 2002; Muzzalupo et al. 2009; Noormohammadi et al. 2014). The present sampling and analytical approach allowed for detection of a high total number of alleles, in turn permitting the calculation of valid population genetic parameters and producing a significant potential for the discrimination of unknown genotypes. The high average number of detected alleles can be further attributed to the pre-existing genetic diversity of the rather diverse olive tree germplasm cultivated in Crete, Greece or Turkey. The average number of detected alleles per locus within wild olive genotypes is higher compared to the average value of each cultivar. For example, in the case of the widely employed locus OeUA-DCA3 (DCA3), the average value of detected alleles across all cultivars was 4.5 while in wild olives a total of 9.1 alleles were detected. In a similar work, Ganino et al. (2007), found six DCA3 alleles analysing eight ancient cultivars from the Emilia region (Italy). On the other hand, Yoruk and Taskin (2014) analysed 200 and four oleaster trees and 27 local cultivars from Turkey identified 21 DCA3 alleles in total (present study: 13) with a range of 230 to 268 bases and 6 UDO099-019 alleles in total (present study: 8) with a range of 121 to 165 bases. Because of the inter-fertile nature of wild olives, it may be proposed that the number of alleles from the original genetic pool of wild olives is higher than that retained within cultivars. It appears that during the historical development of cultivars, selection, inbreeding and clonal selection have all been applied, resulting in the lower genetic diversity within cultivars.

Only two previous works employed some of the same SSR loci while fingerprinting the same cultivars analysed in the framework of the present study. Baldoni et al. (2009) evaluated cvs. ‘Koroneiki’ and ‘Mastoidis’ composed of single genotypes over the locus OeUA-DCA3. They concluded that locus DCA3 was not polymorphic over cv. ‘Koroneiki’ in contrast to our results detecting two different private alleles and five alleles in total (data not shown) across the entire cv. ‘Koroneiki’ germplasm, while for cv. ‘Mastoidis’ Baldoni et al. (2009) detected two alleles (232 and 243) compared to three alleles detected in the present study across the entire cv. ‘Mastoidis’ range. On the other hand, Roubos et al. (2010) analysed 26 different Greek cultivars each represented by four clonal genotypes while they employed agarose gel for allele fragmentation and sizing. For locus DCA3, they determined an allelic range of 226–253 (present study 220–254) and 7 alleles in total (present study 13). It is worth mentioning that ample molecular genetic polymorphisms were detected at the intra-cultivar level in the present study. This is expected since sampling exclusively focused on aged and autochthonous trees by simultaneously avoiding recently established and clonally propagated trees. On-farm selection by farmers of clones with superior characteristics (e.g. high oil content, drupe size, resistance to drought in areas with low precipitation or resistance to fungal infections in areas of high air humidity) might be the source of this intra-cultivar variation. These genotypes could have derived from seedlings by intra-cultivar self-pollination or from long-term epigenetic adaptation of olive trees to local microenvironments of isolated habitats. Similarly, Lopes et al. (2004) identified several Iberian olive cultivars with high levels of intra-varietal polymorphism. This is in sharp contrast with an extensive analysis of cv. Picholine marocaine in Morocco (Khadari et al. 2008).

Our results exhibit lower diversity indexes when compared to the study of Bracci et al. (2009), who analysed cultivars with heterogeneous origins coming from 10 Mediterranean countries and to Sarri et al. (2006) who studied 112 cultivars from 14 Mediterranean countries.

An increased number of private alleles as well as higher diversity indexes were detected in wild Cretan olives (in comparison to cultivated genotypes) in agreement with Belaj et al. (2010). Higher HO of wild olives is in agreement with the findings of Lumaret et al. (2004) but in contrast with those of Belaj et al. (2010). The latter authors proposed that continuous crossings among existing traditional and newly introduced cultivars allowed for the maintenance of higher HO in cultivated rather in wild olives. Since all cultivars showed higher HO than HE, some heterozygotic divergence could be inferred.

In our study, the mean calculated HO value (0.852) was higher than those calculated for accessions in Italy (0.621 on average), Spain (0.74), Slovenia (0.769) and Iran (0.61) (Alba et al. 2009; Bandelj et al. 2002; Belaj et al. 2010; Muzzalupo et al. 2009; Noormohammadi et al. 2007). Further, all genotypes studied clearly clustered according to their expected cultivar group. Wild olives were separated into three sub-groups, each assigned to one of the three Cretan cultivars implying in turn that these are not genuine wild olives but rather feral forms resulting from productive cross-pollination between oleasters and domesticated individuals (also discussed by Khadari et al. 2008). Domesticated forms coexist in most of the olive-producing regions of the Mediterranean basin. This link between wild and cultivated olives was realized perhaps through some type of reticulate evolution (i.e. frequent introgressive hybridizations among the two types with traditional on-farm selection providing the evolutionary direction to the cultivated individuals). A contributing factor may be that true seedlings of the examined cultivars produced new fertile olive trees. Before clonal propagation by cuttings or tissue culture became popular, local nurseries and farmers used trunk pieces of either wild olive trees collected from forest areas or cultivated trees to produce new plants, which would be either grafted or grown self-rooted according to fruit yield, size and oil content. Erre et al. (2010), in a study similar to the present one, determined that most Sardinian cultivars shared the same allelic profiles with the ancient cultivated trees, while Bayesian analysis pointed to an autochthonous origin of Sardinian cultivars. Nevertheless, the wild olive trees from Sardinia formed a separate genetic pool. Similar results to ours were found by Yoruk and Taskin (2014) and by Ipek et al. (2009) by comparing SSR profiles of local Turkish cultivars with those of local wild olive trees. The CBT mobile is in agreement with the similarity dendrogram in that cvs. ‘Koroneiki’ and ‘Mastoidis’ exhibit higher genetic similarity between them when compared to cv. ‘Throubolia’ and the two Turkish cultivars. Additionally, cvs. ‘Koroneiki’, ‘Mastoidis’ or ‘Throubolia’ are genetically closer to wild individuals than to the Turkish cultivars. Wild olives are included in the similarity dendrogram and CBT because they may have served as the original gene pool (true oleasters) from which all cultivars were derived. Conversely, many wild olive trees originated from cultivars through the process of gene flow and introgressive hybridization (feral forms), which results in some individual trees being drawn from more than one group (Mekuria et al. 2002). Indeed, these researchers found that the molecular genetic variation of three groups of olive trees—i.e., (i) feral trees in nearby hills, (ii) fence and (iii) cultivated trees—is not significantly different and that the three clusters formed in a RAPD UPGMA phenogram were homogenous. Nevertheless, this grouping provided evidence that gene flow among the three groups is significant (Mekuria et al. 2002). Further, in the present study, inclusion of wild trees together with cultivars in the CBT better shapes the classification space by coagulating cultivars that are similar to wild olive trees. In fact, the entire data space is partitioned into smaller sections—i.e. recursive partitioning of cultivars and/or wild types—where mathematical interactions of loci are more clear (Breiman et al. 1984). The CBT methodology was also successfully employed in works with virgin oil produced by different O. europaea cultivars and different oil adulteration treatments (Petrakis et al. 2008; Agiomyrgianaki et al. 2010) by classifying oils characterized by individual compound contents detected with NMR spectroscopy. The decision process produced according to the present CBT provides a method for sorting any new unknown material purportedly originating from any of the presently analysed cultivars. It appears that, based on just seven SSR loci, we have devised a system for sorting the present cultivar collection according to an a priori phenotypic classification and to predict membership of future unknown samples to any of the presently analysed cultivars.

Ample genetic diversity was detected between and within cultivars. Ganino et al. (2007) analysed ancient trees in the Emilia region in Northern Italy and concluded that the longevity of this species and the low breeding pressure has contributed to the conservation of its variability and that the reduced extent of genetic erosion within its germplasm has allowed the persistence of olive diversity. The elevated genetic diversity is apparent in the similarity dendrogram, whereupon inclusion of all analysed genotypes creates a complex network of linked genotypes. Groups of identical genotypes were also discovered with all analysed cultivars (with the exception of cv. ‘Samanli’) which can likely be attributed to the presence of clonal materials. This is especially true for the establishment of new olive orchards in Crete over the past 50 years given the preference of farmers for cv. ‘Koroneiki’ due to its suitability for modern intensive farming. On the other hand, the Cretan cv. ‘Throubolia’ exhibits higher genetic affinity to the Turkish cvs. ‘Gemlik’ and ‘Samanli’ than to the other two Cretan cvs. ‘Koroneiki’ and ‘Mastoidis’ in agreement with Owen et al. (2005). Yoruk and Taskin (2014) examined the genetic diversity of cvs. ‘Gemlik’ and ‘Samanli’ within the broad genetic pool of Turkish olive cultivars and wild olive trees. They concluded that cv. ‘Gemlik’ is a widespread cultivar in the Marmara region but that this cultivar is not related to any of the studied cultivars from other regions in Turkey. Further, Yoruk and Taskin (2014) found that the majority of Turkish cultivars were not clearly clustered according to their geographic origin and proposed that humans have moved cultivars to different sites in the country during the past several 1000 years of olive cultivation. Some interesting scenarios regarding the post-glacial persistence of oleasters in Greece and Turkey and the history of subsequent domestication and selection of cultivars were presented by Besnard and Rubio de Casas (2016). Nevertheless, further detailed research is needed to ascertain the exact paths of material flow in the area.