Variation in olive tree chloroplast genomes

The cpDNA genome sizes vary between 155,531 base pairs (bp; lineage C2; Almhiwit 5.1) and 155,896 bp (lineage M; Imouzzer S1). As suspected by Besnard & Bervillé [30] based on RFLPs, two long indels were observed in the seven olive tree cpDNA genomes: a 342-bp deletion (in the ycf1 gene) was observed in lineage E3 (Gué de Constantine 20), while a 225-bp deletion (in the trnQ-rps16 intergenic spacer) was detected in both individuals from South Asia (lineages C1 and C2). In addition, 15 smaller indels (i.e., inferior or equal to 12 bp, excluding microsatellite motifs) were also detected. Five of these indels correspond to the presence/absence of a repeated motif of seven to 12 bp (i.e., composed of one or two motifs; located at nucleotide 7,328, 9,526, 14,693, 83,196 and 85,059 in the 'Manzanilla de Sevilla' sequence; see GenBank/EMBL accession no FN996972).

Sequence variation was low, with a total of 218 substitutions on the seven olive plastid genomes. A maximum of 106 substitutions (0.07%) was detected between Gué de Constantine 20 (Algeria) and Almhiwit 5.1 (Yemen), while cpDNA genomes of Guangzhou 1 (China) and Almhiwit 5.1 (Yemen) only showed 34 substitutions (Additional file 1). The plastid genome of O. woodiana displays between 417 and 432 substitutions (< 0.28%) when compared to the seven O. europaea genomes. Again, this level of variation is surprisingly low if we consider that the divergence between sections Olea (O. europaea) and Ligustroides (O. woodiana) is estimated to be between 14 and 22 million years (My; [29]). Based on these results, the cpDNA substitution rate was estimated to be between 1.2 × 10^-10 and 2 × 10^-10 in the Olea subgenus, which is about ten times lower than the typical mutation rate reported for the plastid genome [7]. This slow molecular evolution might be related to the long generation time of the olive tree [23, 24].

Twelve differences (i.e., three length polymorphisms and nine SNPs, of which one is located in the inverted repeat) were observed between the genomes of 'Frantoio' (GenBank/EMBL accession GU931818; Italy; [18]) and 'Manzanilla de Sevilla' (Spain; this study). According to our approach, we re-sequenced the variable regions in 'Frantoio', from the Olive World Germplasm Bank (OWGB) at Córdoba, Spain (GenBank/EMBL accessions no. FR754486 to FR754495), but these polymorphisms were not confirmed. These 12 differences are not located in the cpDNA regions screened for sequence variation by Mariotti et al. [18] and may be seen as putative sequencing mistakes in accession GU931818. Considering this fact, our analyses indicate that 'Frantoio' and 'Manzanilla de Sevilla' display the same plastid genome, supporting a common maternal origin for these two cultivars.

Based only on nucleotide substitutions (i.e., only 65 out of 218 substitutions were parsimony-informative in the olive tree complex), phylogenetic relationships were depicted from the complete cpDNA genomes using both maximum parsimony (MP) and maximum likelihood (ML) techniques (Figure 2). The resulting topologies confirm results from Besnard et al. [15, 29] through the recovery of two main clades: a Mediterranean/North African clade (clade Cp-II) including lineages E1, E2, E3 and M, and a cuspidata clade (clade Cp-I) including lineages C1, C2 and A. In clade Cp-II, moderate bootstrap support for an early-diverging position of lineage E3 (Gué de Constantine 20) agrees with results based on a few cpDNA microsatellites, indels and CAPS [15]. A moderate level of support was also recovered for the clustering of lineages E1 and E2. Only nine informative substitutions were detected in clade Cp-II, three of them being non-synonymous (Table 1). The information brought by these sites does not strongly support any relationship, suggesting that some sites may be homoplastic. Indeed, two of the three non-synonymous substitutions (52,165 and 83,304) are polymorphic in both clades Cp-I and Cp-II, suggesting that these sites could be under selective pressures, either maintaining polymorphism or contributing to the recurrent appearance of the same substitution (see also [18]). Understanding the molecular variation at these non-synonymous sites would deserve the design of an experiment to test their origin and their adaptive significance.

Development of cpDNA markers

The low cpDNA substitution rate combined with possible selective effects (which can be problematic for phylogenetic reconstructions [31]) led us to focus on "length polymorphisms". Such polymorphisms were either the result of a variable number of repeats in a microsatellite motif (referred as "microsatellites"), or another type of insertion/deletion (referred as "indel"). Sixty-two regions, of which 51 display variable microsatellite motifs, were investigated (Additional file 2). These sites are located in non-coding regions (except for loci 61 in ycf1) and can thus be considered as mostly neutral. The list of polymerase chain reaction (PCR) primers to amplify the 62 regions is given in Additional file 2. Two CAPS loci (located in rpl14 and the petA-psbJ intergenic spacer) were also characterised to allow the distinction of new haplotypes in lineage E1 (see Methods). After the characterisation of 315 cultivated and wild trees, a multilocus profile (or cpDNA haplotype) was defined for each individual (Additional file 3a). Also, an 88-year old herbarium leaf sample was successfully characterised, suggesting that our method is appropriate for investigating cpDNA variation even on poorly preserved DNA. A total of 40 loci were polymorphic in the Mediterranean/North African olive tree (Additional file 3b). We hope that data generated using this method by different laboratories could be compared to generate a reference dataset for the Mediterranean olive tree. In this way, it should be possible to reconstruct a detailed phylogeography of the species based on a large number of populations, as has been done, for instance, for the European white oaks [32].

Polymorphism assessment in the Mediterranean olive

Some olive tree varieties are used to produce high-quality (and thus more expensive) extra virgin olive oil. Therefore, they may be granted a label of protected designation of origin (PDO; a European Union label referring to food products specific to a particular region or town, conveying a particular quality or characteristic of the specified area). Our markers could find some applications in the traceability of such high quality olive oils, but their discriminating power needs to be determined for assessing their putative utility. Using our cpDNA loci, 12 haplotypes were detected in cultivars (Table 2, Figure 3a and Additional file 3): hence our approach permitted a two-fold increase of the number of detected variants compared to previous studies [17, 18]. The most frequent haplotype (E1.1) was detected in 77% of cultivars, including 'Frantoio' and 'Manzanilla de Sevilla'. Two other haplotypes (E1.2 and E3.2) displayed a frequency superior to 5%, but the remaining haplotypes were rare, and sometimes detected only once (i.e., L1.1, E2.3, E2.5 and E2.6) or twice (i.e., E1.3, E2.2 and E3.1). Several of these rare haplotypes were detected in local cultivars with a limited economic importance (e.g., E2.5, E2.6 and L1.1). The probability that two samples chosen at random display a different haplotype was low (D = 0.40) when compared to nuclear markers, especially nuclear microsatellites for which the discriminating power per locus generally exceeds 0.70 [33–35]. This indicates that the utility of the cpDNA variation for forensic analysis is restricted to rare haplotypes such as the ones detected for 'Picholine' (E2.1) and 'Olivière' (E3.1) in France, 'Villalonga'-'Blanqueta' (E1.3), 'Farga' (E3.1) and 'Lechín de Sevilla' (E2.3) in Spain, or 'Megaritiki' (E2.2) in Greece. These varieties are used to produce high quality extra virgin olive oil (e.g., for Spanish cultivars see [36]). The cpDNA variation, which is a priori easily analysable compared to nuclear single-copy genes, should thus be helpful to complement other procedures for olive traceability based on nuclear polymorphisms [e.g., [37]].

	Haplotype frequency (%)
Haplotype *	Cultivars	Bin El Ouidane	Minorca	Pugnochiuso	Gialova	Rajo
E1.1	77.0	42.9	-	4.5	21.6	46.2
E1.2	7.0	-	-	-	-	26.9
E1.3	1.1	-	-	-	-	3.8
E1.4	-	-	-	-	-	19.2
E1.5	-	-	-	-	-	3.8
E1.6	-	-	-	-	8.1	-
E1.7	-	-	-	-	10.8	-
E1.8	-	-	-	-	13.5	-
E1.9	-	-	-	-	13.5	-
L1.1	0.5	-	-	-	-	-
E2.1	3.2	4.8	-	68.2	-	-
E2.2	1.1	-	52.2	27.3	-	-
E2.3	0.5	4.8	4.3	-	-	-
E2.4	2.1	-	-	-	-	-
E2.5	0.5	14.3	-	-	-	-
E2.6	0.5	23.8	-	-	-	-
E2.7	-	-	-	-	32.4	-
E2.8	-	14.3	-	-	-	-
E3.1	1.1	-	26.1	-	-	-
E3.2	5.3	-	17.4	-	-	-

* See Additional file 2 for the haplotype profile definition.

In the five populations of oleasters, 18 cpDNA haplotypes were detected, ten of which were shared with cultivars (Table 2, Figure 3b and Additional file 3). The discriminating power of cpDNA was high in these populations (D = 0.89) compared to the cultivated olive tree. Fourteen haplotypes were unique to one population, while the four remaining haplotypes were shared between at least two populations: E1.1 (Rajo, Gialova, Pugnochiuso and Bin El Ouidane), E2.1 and E2.2 (Bin El Ouidane and Pugnochiuso) and E2.3 (Minorca and Bin El Ouidane). These four haplotypes have been detected in cultivated olive trees and could reflect long-distance gene flow mediated by humans [15, 38]. In this way, the most frequent haplotype in cultivars (E1.1) is also the most frequent and widespread haplotype in oleasters (22%; Figure 3b).

Implications for phylogeography

Previous cpDNA phylogeographic studies of the Mediterranean olive tree have been limited due to the low number of haplotypes detected [17, 18]. Here, we demonstrate that a genomic profiling approach of the plastid DNA mostly based on microsatellites and indels can solve this problem. The high variation detected in five distant wild populations indicates a high potential of our approach for resolving the Mediterranean olive tree history. One putative limitation is the level of homoplasy on microsatellite motifs, reported by different authors [39–42], and which could prove problematic when accurately identifying evolutionary relationships between haplotypes. We reconstructed a reduced median network based on molecular markers (Figure 3c). The Mediterranean haplotypes clustered into three lineages (E1, E2 and E3), while the haplotype of subsp. maroccana formed a fourth lineage (M) in northern Africa. This topology is fully congruent with Besnard et al. [15, 29], who used different cpDNA data (i.e., microsatellites, indels and CAPS, or nucleotides). Each lineage displays at least one specific indel, with the exception of lineage M (Figure 3c). Phylogenetic relationships remain unresolved at the base of lineages E1 and E2, as well as in the centre of the network, as a consequence of homoplasy between haplotypes belonging to different lineages (e.g., shared length polymorphisms between clades Cp-I and Cp-II at loci 1, 2, 9, 17, 25, 38, 47, 48, 49, 50 and 58; Additional file 3). Such a difficulty for determining the ancestral state hampers the correct identification of historical links between divergent lineages. In contrast, we expect that homoplasy will not be a serious limitation to resolve phylogenetic relationships among lineages, since their haplotypes have diverged more recently [42]. In any case, for an optimal analysis of the cpDNA variation at the population level, possible length homoplasy will need to be considered and the use of appropriate models will be necessary [41, 43].

The partial or complete cpDNA sequencing of new individuals may reveal nucleotide substitutions that would be of interest [18] for the development of new molecular markers like SNPs (or CAPS). Such SNPs could be used to improve our approach. Nevertheless, the homoplasy is not restricted to repetitive sequences as illustrated with non-synonymous sites in genes under selection, such as the polymorphism detected at the CAPS-XapI locus (in rpl14; Table 1). In the present study, we found restriction polymorphism at this locus in lineages E1 and E2 (clade Cp-II) and also in clade Cp-I (for which we analysed only three accessions; Figure 3c) indicating that this site is highly homoplastic (see also Mariotti et al. [18]). Thus, this site should be used with caution for phylogeographic purposes. Nevertheless, we consider that it could bring potentially important information at the lineage level, particularly to solve the origin of haplotype E1.2 in the cultivated gene pool (7% of cultivars).

Chloroplast genome sequencing

In order to maximize polymorphism detection, the analysis focused on seven individuals of O. europaea L. (subgenus Olea sect. Olea, or olive tree complex), which were chosen to represent one haplotype of each previously described lineage [15]. The following genotypes were thus investigated: 'Manzanilla de Sevilla' (Spanish cultivar; lineage E1), oleaster "Haut Atlas 1" (Morocco; lineage E2), oleaster "Gué de Constantine 20" (Algeria; lineage E3), subsp. maroccana "Imouzzer S1" (Morocco; lineage M), subsp. cuspidata "Maui 1" (Hawaii; lineage A), subsp. cuspidata "Guangzhou CH1" (China; lineage C1), and subsp. cuspidata "Almhiwit C5.1" (Yemen; lineage C2). In addition, we characterised one outgroup species [O. woodiana Knobl. subsp. woodiana (South Africa); sect. Ligustroides Benth. & Hook.], which belongs to the sister group of O. europaea [16, 29]. Appropriate PCR primers were designed to amplify 105 overlapping cpDNA fragments (Additional file 4). Each PCR reaction (25 μl) contained 10 ng DNA template, 1× reaction buffer, 2 mM MgCl₂, 0.2 mM dNTPs, 0.2 μmol of each primer, and 0.75 U of Taq DNA polymerase (Promega, Madison, WI, USA). The reaction mixtures were incubated in a thermocycler (T1; Biometra, Göttingen, Germany) for 2 min at 95°C, followed by 36 cycles of 30 s at 95°C (denaturing), 30 s at the annealing temperature (Additional file 4), and 2 min at 72°C (extension). The last cycle was followed by a 10-min extension at 72°C. Direct sequencing of PCR amplicons was performed with an ABI Prism 3100xl Genetic Analyzer, using the Big Dye v3.1 Terminator cycle-sequencing kit, according to the manufacturer's instructions (Applied Biosystems, Foster City, CA, USA). Additionally, nested (internal) primers were also designed to complete the sequencing of each fragment (Additional file 4). The eight Olea genomes were thus reconstructed using a similar approach to the one used by Mariotti et al. [18].

Characterisation of cpDNA polymorphisms in the Mediterranean olive tree

Based on the seven O. europaea sequences, length polymorphism was detected in 62 regions. These polymorphisms were either due to a variable number of repeats in a microsatellite motif or another type of indel (Additional file 2). The PCR primers were designed in flanking regions to specifically amplify short segments (generally inferior to 240 bp). For locus multiplexing, the annealing temperature of all these primers needed to be similar, while the size of PCR products of each locus should be as different as possible. Finally, these primers were also designed to allow amplification of short DNA segments for characterization of poorly preserved material and highly degraded DNAs from herbarium samples. Additionally, the 5' end of the reverse primer of locus 19 was tagged with the sequence GTGTCTT to minimize band stuttering. All primer pairs and specific characteristics of generated fragments are given in Additional file 2. To reduce the cost of the PCR characterization (i.e., time and costs), we used the method described by Schuelke [46]. For each locus (except loci 8, 10, and 61), an 18-bp tail of M13 was added on the forward primer (Additional file 2). When each locus was amplified separately, each PCR reaction (25 μl) contained 10 ng DNA template, 1× reaction buffer, 2.5 mM MgCl₂, 0.2 mM dNTPs, 0.2 μmol of one universal fluorescent-labelled M13(-21) primer (5'-TGTAAAACGACGGCCAGT-3'; labelled with one of the three following fluorochromes: HEX, 6-FAM or NED), 0.2 μmol of the reverse primer, 0.05 μmol of the forward primer, and 0.5 U of Taq DNA polymerase (Promega). The reaction mixtures were incubated in a T1 thermocycler for 2 min at 95°C, followed by 28 cycles of 30 s at 95°C, 30 s at 57°C, and 1 min at 72°C, and then by 8 cycles of 30 s at 95°C, 30 s at 51.5°C, and 1 min at 72°C. The last cycle was followed by a 20-min extension at 72°C. Usually, we amplified five or six loci in the same reaction, but in this case, the MgCl₂ concentration was increased to 5 mM, and the concentration of primers (except the labelled M13 primer) was decreased by five or six. Loci 8, 10, and 61 (without the M13 tail) were amplified separately with the following conditions: each PCR reaction (25 μl) contained 10 ng DNA template, 1× reaction buffer, 2 mM MgCl₂, 0.2 mM dNTPs, 0.2 μmol of each primer, and 0.75 U of Taq DNA polymerase. The reaction mixtures were incubated in a T1 thermocycler for 2 min at 95°C, followed by 36 cycles of 30 s at 95°C, 30 s at 53°C, and 2 min at 72°C. The last cycle was followed by a 10-min extension at 72°C.

The PCR products labelled with a fluorochrome were mixed together with GeneScan-500 ROX as internal standard to run the maximum of loci at the same time (considering the colour and the expected allele size range). They were separated on an ABI Prism 3100xl Genetic Analyzer and the fragment size was determined with GeneMapper version 4.0. For the two non-labelled loci 8 and 61, indels of 342 and 225 bp were revealed under UV after migration on a 2.5% agarose gel electrophoresis stained with GelRed (Biotium, Hayward, CA, USA).

We also focused on the characterisation of two substitutions, which were detected by Mariotti et al. [18] in lineage E1 (the most frequent one in cultivated olive trees; see [13, 17]) and may be potentially useful for forensic analyses and the study of olive tree domestication. We chose to develop two Cleaved Amplified Polymorphism Site (CAPS) loci as in Besnard et al. [47], in order to rapidly characterise a high number of individuals. The PCR primers are given in Additional file 2. The two loci were amplified following the same PCR conditions as for microsatellites. The PCR products were digested with a restriction enzyme (EcoRI or XapI) according to the manufacturer recommendations. The restricted fragments of the two loci were then mixed (with the internal standard ROX 500) and separated on an ABI Prism 3100 xl Genetic Analyzer. The polymorphism for the presence/absence of a restriction site was scored for each genotype. The possibility of multiplexing three different colours (e.g., NED, FAM and HEX) allows the characterisation of 288 (96 × 3) samples per run.

We then characterised 186 cultivated olive tree accessions from different areas with the 64 loci (Table 2, Figure 3a and Additional file 5). The analyzed germplasm includes 106 cultivars from the OWGB Córdoba [48]. These cultivars represent major cultivars from all Mediterranean countries. A few local cultivars from different places were also included in our study for a better representativeness of the cultivated gene pool. First, we characterized 55 cultivated local forms from Morocco (41) and Corsica-Sardinia (14) previously genotyped with nuclear markers [49, 50]. In addition, cultivated trees with or without known denominations from Algeria-Tunisia (6), Italy (6), France (2), Greece-Turkey (3), the Levantine region (5), Libya-Egypt-Sudan (2) and South Africa (1) were added to this study. Beforehand, we tested with nuclear microsatellites that these latter accessions were genetically different (G. Besnard, unpubl. data), except for one herbarium leaf sample from Kufra, Libya (Newberry, sn; 1933 - Kew Herbarium). In addition, to assess the cpDNA variation in the wild Mediterranean olive trees, 129 individuals from five distant populations (Figure 3b) were also characterized: Rajo (Syria; 36°43'50''N, 36°40'00''E), Gialova (Greece; 36°55'12''N, 21°42'42''E), Pugnochiuso (Italy; 41°47'46''N, 16°10'05''E), Minorca (Spain; 39°56'52''N, 04°14'42''E) and Bin El Ouidane (Morocco; 32°03'00''N, 06°35'00''W). To test the reproducibility of the method, the characterisation of ten accessions (i.e., 'Picholine Marocaine', 'Manzanilla de Sevilla', 'Frantoio', 'Moraiolo', 'Ciarasina', 'Confetto', 'Itrana', 'Giaraffa', 'Kalamon' and 'Souri') were repeated three times at random.

Based on this analysis of wild and cultivated accessions, 40 polymorphic loci were detected in the Mediterranean olive trees (Additional file 3). We first proposed to combine 36 of these loci for a rapid characterisation of Mediterranean olive tree germplasm. The multiplex PCRs of five or six loci are proposed in Table 3, but this can be easily modified. The PCR conditions are those previously reported (with the M13 primer). After PCR, these products are mixed together (with no overlap for allele size between loci in a given colour). The locus 10, which needs to be amplified separately, is combined with these multiplex PCRs. Second, when amplified in a multiplex PCR, we encountered some difficulties with locus 19 (not reported in Table 3), and we thus recommend to use it separately and to combine it with the two CAPS (CAPS-XapI and CAPS-EcoRI) for a second combination of three loci. Lastly, the locus 61 is independently characterised on 2.5% agarose gel electrophoresis.

Data analysis

A phylogenetic tree based on the complete plastid genomes was constructed. A partial cpDNA sequence of Forsythia (DQ673256; [51]) was used as an outgroup to root the tree. Sequences were aligned with the application MEGA v4.1 [52]. The alignment was manually refined. Firstly, a maximum parsimony analysis was performed. All characters were equally weighted. The gaps were treated as missing data. A heuristic search was used to find the most parsimonious trees. The close-neighbor-interchange algorithm was used with a search level of 3, as recommended and implemented in the software [52]. The searches included 100 replications of random addition sequences. All the best trees were retained. A strict consensus tree was generated from the equally most-parsimonious trees. The bootstrap values were computed using 10,000 replicates. Secondly, the tree inference was made under a maximum likelihood criterion, using the application PHYML v3.0 [53]. The best-fit substitution model, determined through hierarchical likelihood ratio tests, was the GTR model, with invariable sites and a gamma shape parameter estimated from the data. Support values were obtained by 1,000 bootstrap replicates. Based on fragment genotyping (i.e., microsatellites and indels), the relationships among cpDNA haplotypes were visualized by constructing a reduced median network implemented in the application NETWORK v4.112 [54]. Multi-state microsatellites were treated as ordered alleles and coded by the number of repeated motifs for each allele (e.g., number of T or A; see also [15]) whereas the presence or absence of other indels was coded as 1 and 0, respectively. Basically, this coding strategy assumes that variation at cpDNA microsatellites is mainly due to single-step mutations (e.g., [15, 18]), while allowing consideration of length polymorphisms (microsatellites or indels) with similar weight. However, whether we used different weights or not for indels versus microsatellites did not affect the topology. In addition, for loci combining indels and microsatellite motifs (loci 10, 11, 54 and 57), we separately coded the two types of characters based on available sequences for these loci. The matrix used for the analysis is given in Additional file 6.

The probability that two individuals taken at random display a different haplotype was computed as D = 1 - Σ p _i ², where p _i is the frequency of the haplotype i. This parameter was calculated separately on cultivated and wild olive trees, but also on sub-samples or populations. The groups of cultivated olive trees were defined according to their geographic origin.