Scientific and Technological Park “Moliseinnovazione S.C.p.A.”, Via F. De Sanctis snc, 86100, Campobasso, ItalyLaboratory of Analytical Technique and Proteomics, Research Laboratories, Catholic University, Largo A. Gemelli 1, 86100, Campobasso, ItalyMuseo Erbario del Molise, Department of Science and Technology for the Environment and Territory, University of Molise, Contrada Fonte Lappone, I-86090, Pesche, ItalyDepartment of Animal, Plant and Environmental Sciences, University of Molise, Via F. De Sanctis snc, 86100, Campobasso, ItalyInstitute of Plant Genetics (Research Division of Portici), National Council of Research, Via Università 133, Parco Gussone, 80055, Portici (NA), Italy
r t i c l e i n f o
rticle history:eceived 10 January 2011eceived in revised form 3 June 2011ccepted 30 June 2011
Three olive cultivars (Oliva Nera di Colletorto, Noccioluta, and a probably a new local genotype) from twostrictly related areas of Molise region (south-centre of Italy) were characterized by combining moleculardata (eight SSRs analyzed on leaves) and morphological features (thirty-one parameters from leaves,drupes and pits). Both molecular and morphological analyses have shown a very good separation of thethree endemic cultivars. A high correlation between morphological and molecular data was found usingMantel’s test. The morphological traits of pits were less influenced by environmental pressure than theleaves and drupes; therefore, the pits are more affected by genetic control and might be considered a
SRtatistical analysistatistical model
helpful tool for cultivar characterization and identification. Potential and limitations of three statisti-cal models computed to perform cultivar identification by morphological measures is also discussed.We demonstrated that molecular and morphological analyses are useful for distinguishing new acces-sions and studying local varieties to preserve genetic diversity, even at small geographical scale in suchan unequivocal way; hence the methodology could be proposed as a tool to discriminate widespreadcultivars, with long genetic distances.
. Introduction
The genetic diversity could be an important resource for theevelopment of modern olivicolture towards typical olive oil pro-uctions. From here, the study of less common cultivars representsn important tool to preserve this genetic diversity in respect toenetic erosion due to the introduction of few commercial culti-ars in the modern orchards. In fact, the modern olive oil industryequires new and more productive cultivars to sustain the newrends in olive growing. This phenomenon implies that only a fewommercial varieties are cultivated in the main production areas,hereas minor varieties are located in restricted areas and are
ometimes threatened. Hence, the importance of these less com-on cultivars is in the conservation of several adaptative traits that
ould support olive growing, especially in relation to the effects of
∗ Corresponding author at: Scientific and Technological Park “Moliseinnovazione.C.p.A.”, Via F. De Sanctis snc, 86100, Campobasso, Italy. Tel.: +39 0874 312 471;ax: +39 0874 312 710.
E-mail address: [email protected] (M. D’Imperio).1 These two authors have contributed equally to this article.
global change. Conservation programs could be useful tools for themanagement of this local genetic diversity. In this way, all acces-sions should be characterized to eliminate cases of mislabelling andredundancies (synonymy), identify the presence of different cloneswithin the same cultivar (multi polyclonal populations) (Alba et al.,2009; Muzzalupo et al., 2010) and safeguard all cultivars, in partic-ular the minor ones, to avoid a loss in genetic diversity.
In the last years, biochemical and molecular markers, such asisozymes (Belaj et al., 2008; Trujillo et al., 1995), AFLPs (Angiolilloet al., 2006), RAPDs (Besnard et al., 2001; Ganino et al., 2007),ISSRs (Gomes et al., 2009) and SNPs (Reale et al., 2006), have beenused to characterize olive germplasm. Recently, several SSRs havebeen isolated from olives (Breton et al., 2008; Carriero et al., 2002;Cipriani et al., 2002; Díaz et al., 2006; Muzzalupo et al., 2009; Sefcet al., 2000), and these represent the favorite markers for varietalidentification because they are transferable, hypervariable, highlypolymorphic, multiallelic polymerase chain reaction (PCR)-basedco-dominant markers, relatively simple to interpret and show a
high information content (Belaj et al., 2003; Ganino et al., 2006;Khadari et al., 2003).
Morphological characteristics have been widely used to describeolive cultivars (Cantini et al., 1999; Ganino et al., 2006; Morales-
GeneScan-350, Foster City, CA, USA), according to manufacturer’s
30 M. D’Imperio et al. / Scientia
illero et al., 2008; Ozkaya et al., 2006, 2008; Poljuha et al., 2008a;otondi et al., 2003; Taamalli et al., 2006; Terral et al., 2004). Thesearameters were usually used to redact catalogues describing, alsoy pictures, the morphology of the trees, leaves and fruits of eachenomination in different countries. For instance, the Internationallant Genetic Resources Institute (IPGRI, Rome, Italy) and Interna-ional Olive Oil Council (IOOC, Madrid, Spain) have catalogued andescribed the most well-known world cultivars (Barranco et al.,000). However, morphological descriptors proposed in world cat-logue principally are qualitative variables, and the total variabilityf a singular morphological trait is described by a limited number oflasses, which are drawn from well known cultivars. For these rea-ons, it is very difficult to use these qualitative variables to realizetatistical models. The use of morphological characteristics to dis-riminate olive cultivars is controversial compared with the use ofolecular markers. Indeed, morphological parameters are clearly
ffected by environmental and agronomical factors (Besnard et al.,001) but it is nevertheless important to provide an easy identifi-ation tool for farmers or to exploit in breeding programs (Cantinit al., 1999).
To our knowledge, no work has reported a high correlationetween morphological and molecular data. Taamalli et al. (2006),n Tunisian cultivars, reported only a significant but low correlationetween some morphological traits, such as the weight of fresh andry drupes and pits, and genetic distance matrices obtained withSR (r = 0.185) and AFLP markers (r = 0.156).
However, several studies on olive germplasm from minor areasave been performed to clarify the status of local endemic cultivarsBracci et al., 2009; Ganino et al., 2007; Omrani-Sabbaghi et al.,007; Rotondi et al., 2003; Taamalli et al., 2008). These reported
ittle about the olive cultivars from the Molise region (Angiolillot al., 2006; Muzzalupo et al., 2008b) and, principally, Oliva nerai Colletorto and Noccioluta were only cited as synonyms (Cicoriat al., 2000) or strictly related cultivars (Muzzalupo et al., 2008a).n any case, the identification of genetically close cultivars remainsne of the biggest weaknesses of the olive sector and in general forll crops.
In this paper, three minor olive cultivars were investigatednd characterized by combining quantitative morphological andolecular data with several statistical techniques. These three
ultivars were compared with eight Italian cultivars (national dif-usion) to evaluate their genetic similarity. Eight SSRs were used forenetic characterization and their discrimination power was eval-ated. Additionally, more morphological characteristics on leaves,rupes and pits were investigated, and these parameters were usedo build three models for cultivar identification.
. Materials and methods
.1. Sampling (plant material)
A total sample of 57 olive trees was collected in an area of thedriatic district in the Molise region (centre-south Italy). The firstnit is located near the municipality of Colletorto (an hilly zone);he second unit is located near the municipality of Larino (a planeone) (for pedo-climatic characteristics see Reale et al., 2002).
Altogether, 55 olive trees were sampled in first unit, whereashe remaining two specimens, ortet of the cultivar Oliva nera diolletorto, were sampled in the Molise collection of olive treesCOTEB-Larino) located in second unit. A total of 23 samples weret first assigned to Noccioluta and 34 to Oliva nera di Colletorto,
wo local strictly related olive cultivars. Other eight samples,ach representing Italian commercial cultivars, were only usedor molecular comparison. These samples were generously pro-ided by Dr. L. Baldoni (CNR-Istituto di Genetica Vegetale, Perugia,
ulturae 130 (2011) 229–240
Italy) and were collected from the olive germplasm collectionof Cosenza-Italy (Centro di Ricerca per l’Olivicoltura e l’IndustriaOlearia, CRA-OLI).
The materials for the molecular and morphological analyseswere sampled in 30–50 years old plants, in the best stage, with-out parasites, on the branches placed at 1.5 m above the ground,and during the harvest period. The samplings on the brancheswere random, also to allow easy procedure by farmers. This studywas carried out in two consecutive crop years (2007/2008 and2008/2009).
2.2. Molecular analysis (DNA extraction and SSR-PCR analysis)
Molecular analysis was carried out on fresh leaves, since theyare always present on the trees and DNA isolation procedure worksvery well on them. Genomic DNA was isolated from 6 g of leavesaccording to Doyle and Doyle method (Doyle and Doyle, 1990)with few modifications. Leaves were ground in a mortar to a pow-der in liquid nitrogen. The powdered material was incubated for1 h at 60 ◦C with 20 mL of 2× cetyl trimethylammonium bromide(CTAB) buffer (2% CTAB, 0.1 M Tris–HCl pH 8, 1.4 M NaCl, 20 mMethylene diamine tetraacetic acid (EDTA) pH 8) added with 2%polyvinylpyrrolidone (PVP, 40.000) and 2% �-mercaptoethanol;3.25 M potassium acetate was added and incubated on ice for30 min. Centrifugation at 12,000 × g for 15 min at 4 ◦C was thencarried out. The supernatant was recovered and purified twicewith chloroform/isoamyl alcohol (24:1 v/v), and the DNA wasprecipitated with 2-isopropanol. The DNA obtained was resus-pended in 1× TE buffer (10 mM Tris–HCl pH 8, 0.5 M EDTA pH8). RNA was removed by incubation with RNase (10 �g mL−1) for30 min at 37 ◦C. The DNA fraction was precipitated with a solu-tion of 1/10× of 3 M sodium acetate and 2× of cold absoluteethanol; after incubation at –20 ◦C for 30 min, this was then cen-trifuged at 13,000 × g for 30 min at 4 ◦C. DNA was washed oncewith 70% cold ethanol and finally resuspended in 500 �L of 1×TE.
SSR analysis was performed using eight primers: ssrOeUA-DCA3, ssrOeUA-DCA7, ssrOeUA-DCA9, ssrOeUA-DCA16, ssrOeUA-DCA17 and ssrOeUA-DCA18 from the 15 sets described by Sefc et al.(2000), and GAPU103 and GAPU101 from the 20 sets described byCarriero et al. (2002). Optimization of annealing temperatures andMgCl2 concentration for each primer pair was accomplished by gra-dient PCR. PCR reaction in 10 �L final volume consisted of 1× PCRreaction buffer, 0.2 mM of each dNTP, 2 mM MgCl2, 0.3 �M of eachprimer, 0.5 U of GoTaq polymerase (Promega, Madison, WI, USA)and 50 ng of DNA template.
Amplifications were performed in a 96-well GeneAmp PCRSystem 9700 (Applied Biosystems, Foster City, CA, USA) underthe following conditions: 5 min at 95 ◦C, followed by five touch-down cycles and 30 cycles at 95 ◦C for 20 s, “X” ◦C for 30 s, 72 ◦Cfor 30 s and a final extension at 72 ◦C for 7 min. Four touch-down PCRs were set up: 60–55 ◦C for the ssrOeUA-DCA7 andGAPU103 loci; 63–58 ◦C for the ssrOeUA-DCA3 locus, 62–57 ◦C forssrOeUA-DCA4 and GAPU101 loci and 57–52 ◦C for the ssrOeUA-DCA9, ssrOeUA-DCA16, ssrOeUA-DCA17 and ssrOeUA-DCA18 loci.Capillary electrophoresis fluorescence-based SSR analyses wereconducted on an ABI PRISM 310 (Applied Biosystems, Foster City,CA, USA). Forward primers were labelled with either 6-FAM (blue),HEX (yellow) and TET (green) fluorescent dyes and TAMRA (red-labelled) was used as internal size standard (Applied Biosystems
instructions; co-loading was then performed when size or colourdid not overlap. Sequencing raw data were analyzed with Gen-eMapper software (Applied Biosystems, version 4.0, Foster City, CA,USA) to estimate the variant sizes.
Hortic
2
wcsww(
leto(tluttgrdDorfpmdotolwloAs
2
s(a(ittga((Pqio
atpvwa
M. D’Imperio et al. / Scientia
.3. Morphological analysis
For each of the 57 olive trees, 10 fresh leaves, drupes and pitsere sampled and acquired at 300 dpi, by a scanner and digital
amera. Digital images were used to record morphological mea-urements. Traditional linear variables and several shape indexere measured by the UTHSCSA ImageTool program (Dove, 2000)hereas the shape of objects was reduced to principal components
PCs) by Shape 1.3 (Iwata and Ukai, 2002).Eleven morphological variables were measured on the 10 fresh
eaves, drupes and pits sampled, then a mean was computed forach tree: ARea (AR) = the area of the object (cm2), measured ashe number of pixels in the polygon; PeRimeter (PR) = the lengthf the outside boundary of the object (cm); Major Axis LengthMaAL) = the length of the longest line (cm) that can be drawnhrough the object; Minor Axis Length (MiAL) = the length of theongest line (cm) that can be drawn though the object perpendic-lar to the major axis; ELongation (EL) = the ratio of the length ofhe major axis to the length of the minor axis (if the value is 1,he object is roughly circular or square, whereas it is more elon-ated when the ratio decreases from 1); ROundness (RO) = if theatio is equal to 1, the object is a perfect circle, when the ratioecreases from 1, the object departs from a circular shape; Feretiameter (FD) = the diameter of a circle having the same area as thebject (cm); COmpactness (CO) = provides a measure of the object’soundness: at 1 the object is roughly circular, when it decreasesrom 1, the object results less circular. WEight (WE) = drupes andits were measured by technical balance (g); VOlume (VO) = waseasured by graduated cylinder (the value was expressed as the
ifference in volume of water between the presence and absencef pits and drupes (cm3)); Shape (PC1Sh) = was measured by Ellip-ic Fourier Analysis (Kuhl and Giardina, 1982), the mean contourf leaves, drupes and pits was computed for each individual, fol-owing the procedure described in (Viscosi et al., 2009): outlines
ere analyzed by means of 30 harmonics and standardized by theongest radius method, a variance–covariance matrix of coefficientsf harmonics was computed and subjected to Principal Componentnalysis (PCA); the shape variation was then reconstructed with PCcores by means of the inverse Fourier’s transform.
.4. Statistical analysis
The PCR fragments were scored as present (1) or absent (0). Fortudying the informative potential of the SSRs data, the expectedHe) heterozygosities was defined as He = 1 − ∑
p2i, where pi is the
llele frequency for the ith allele. The Power Discrimination (PD)Tessier et al., 1999) of each SSR locus was also calculated accord-ng to the formula as above, where pi represents the frequency ofhe ith genotype. Fst provides a measure of the genetic differen-iation among populations, which is the proportion of the totalenetic diversity that separates the populations. Fst was defineds Fst = (Ht − mean He)/Ht where Ht is the total heterozygositiesHartl and Clark, 1997). The Polymorphic Information Content (PIC)Botstein et al., 1980) was calculated using the following formula:IC = 1 − ∑
p2i
−∑
(i)
∑(j=I+1)2p2
ip2
jwhere pi and pj are the fre-
uencies of the ith and jth alleles respectively. The PIC value gives andea of level of polymorphism of each locus and is another estimatef discriminatory power.
The Analysis of Molecular Variance (AMOVA) allowed the hier-rchical partitioning of genetic variation among populations andhe estimation of the widely used PhiPT; it is calculated as the
roportion of the variance among populations relative to the totalariance, and was defined as PhiPT = VAP/(VWP + VAP) = VAP/VTOThere VAP is the variance among populations and VWP the vari-
nce within populations (Excoffier et al., 1992). Genetic similarity
ulturae 130 (2011) 229–240 231
among groups and among individuals was calculated using Nei’sgenetic distance. The Unweighted Pair Group Method using Arith-metic Averages (UPGMA) procedure was used for cluster analysis.
Mantel’s test was also computed. First, three PCAs were com-puted on morphological traits of the leaves, drupes and pits, andthen another Principal Component Analysis (PCA) was computedon total phenotype. Then, a simple Mantel’s test was computed totest the correlations between the morphological and genetic dis-tance matrices. A three-way Mantel’s test was computed to verifythe correlation among the three morphological data sets (leaves,drupes and pits). Significance level (p) of matrix correlation (r) wastested by means of 999 random permutations. For each morpholog-ical data set the Euclidean distance matrix was computed on scoresof the first two PCs, while Nei’s genetic distance matrix was usedfor genetic data.
The Analysis of Variance (ANOVA) was applied on the morpho-logical data to detect discriminant variables among genotypes, andmultiple comparisons (Bonferroni’s post-hoc test) were computedto identify the difference between each pair of groups (alpha levelwas 0.05).
The PCA provides a global overview of the compositional vari-ability in the samples through the projection of the morphologicaldata into hyperspaces defined by linear combinations, i.e. the PCsof morphological variables.
Linear Regression Model (LRM) was built on the morpho-logical data. The reliability of models is given as R, R-squaredand Durbin–Watson. The Durbin–Watson statistic is useful forevaluating the presence or absence of a serial correlation ofresiduals and, therefore, estimating the model’s reliability. Theresidual represents the difference between predicts and real val-ues. If the residuals turn out to be independent according tothe Durbin–Watson table (Savin and White, 1977), the systemis extremely reliable with a good foretelling capacity. Thesemodels were validated by calculating the Mean Absolute Error(MAE = the average of the absolute value of differences betweenthe predicted and observed values, expressed in %) (Orlandi et al.,2010).
Finally, three Linear Discriminant Analysis (LDA) were per-formed only on three model combinations individuated by LRM.The discriminant functions deduced with an a priori hypothesiswere calculated. The relative contribution of the variables to thediscrimination can be explained by the coefficients of each variable.The results are the linear combinations of variables that predict themembership of each sample to the corresponding group. Wilks’lambda is a measure of how well each function separates casesinto groups. Smaller values of Wilks’ lambda indicate greater dis-criminatory ability of the function. The models obtained by LDAwere validated by cross-validation. In cross-validation, the predic-tion ability of the model is determined by developing a model withpart of the data set (training set) and using another part of data (testset) for testing the model. The 57 samples are divided in training(70% of all samples) and test set (30%); each set containing samplesrepresentative of three cultivars. The percentage of samples cor-rectly identified was evaluated by the parameters of recognition(the percentage of samples in the training set correctly classified)and prediction (the percentage of samples in the test set correctlyclassified) (Berrueta et al., 2007).
He and PIC were calculated using the Cervus 3.0 software(Kalinowski et al., 2007). Fst and PhiPT were calculated using theGenAlEx 6.2 software (Peakall and Smouse, 2006). Population 1.2software (Langella, 1999) was used to calculate the Nei geneticdistances and construct the clusters. To generate a dendrogram
we used TreeViewX 0.5 software (Page, 1996). NTSsys-pc softwarepackage for Windows (version 2.2, 2005) was used to compute theMantel test. Statistical software package for Windows (version 6.0,1997) was used to compute PCA, whereas SPSS software package for
232 M. D’Imperio et al. / Scientia Horticulturae 130 (2011) 229–240
Fig. 1. (A) Cluster analysis performed on 57 samples of the three endemic olive genotypes (Gen) from Italy; in brackets the sample numbering is reported; * identifies theo e locac ter an
Wd
3
3
iat
TGg
rtet; (B) cluster analysis of the eight Italian cultivars (national diffusion) and threultivars was used). Nei’s distance and UPGMA of eight SSR data were used for clus
indows (version 15.0; 2006) was used to compute the Euclideanistance, ANOVA, LRM and LDA on morphological data.
. Results
.1. Molecular characterization
Cluster analysis was computed among the 57 samples groupednto three separate genotypes (Fig. 1A). The AMOVA was computedmong the three detected genetic groups, which resulted statis-ically different (PhiPT = 0.969; p < 0.001). Moreover, the greater
able 1enetic parameters for the eight SSRs markers obtained on three Italian olive cultivars:enotype 3.
l cultivars of the Molise region (genotypes 1–3) (one sample from each of the 11alysis. Bootstraps are reported on horizontal connecting segments.
part of the genetic variability explained the differences among thegroups (97% of the total variance was distributed among the groupswith only 3% distributed within the groups).
The eight primer pairs selected for polymorphism and clearerbands revealed 31 alleles, ranging from two at the ssrOeUA-DCA17locus to six at the ssrOeUA-DCA7 and GAPU103 loci, with a meanvalue of 3.8 alleles per locus (Table 1). The alleles ranged from
109 bp for ssrOeUA-DCA17 to 253 bp for ssrOeUA-DCA3 (Table 2).The genetic differentiation among the genotypes was also eval-uated by Fst, with ssrOeUA-DCA7 the most significant primer(Table 1).
8 samples of Noccioluta, 38 samples of Oliva Nera di Colletorto, and 11 samples of
b In brackets, the number of data for each group.c The number represents the genotype.
The mean expected He was 0.645, ranging from 0.565 (ssrOeUA-CA17) to 0.767 (ssrOeUA-DCA7), whereas PD ranged from 0.241
ssrOeUA-DCA17) to 0.521 (ssrOeUA-DCA7) with a mean value of.442 (Table 1). He and PD values revealed, from each primer, aigher genetic variability and discriminant efficiency among geno-ypes, respectively (Besnard et al., 2001).
These results allow us to associate genotypes to cultivar names.enotype 2 was assigned to Oliva nera di Colletorto; in fact, the twortets (COTEB) were grouped in this cluster (Fig. 1A). Genotype 1as assigned to Noccioluta without an ortet as reference, but withstrong correspondence between several morphological featuresf the drupes and pits (see results of morphological data) and ety-ology of the name Noccioluta, as indicated by local farmers. No
ssignment was performed for genotype 3, and this was considereddifferent cultivar, probably a new local genotype not investigatedntil now.
The genetic profile of the eight primers for each recognised cul-ivar is reported in Table 2 and the genetic relationship betweenhese three cultivars and the other eight Italian cultivars was inves-igated by cluster analysis (Fig. 1B).
.2. Morphological characterization
The simple Mantel’s test was computed between phenotypicmean Euclidean distance matrix) and genotypic distances (Nei’sistance matrix) (Table 3), detecting a high correlation between
hese two data sets (r = 0.999, p < 0.01). This indicated that for sig-ificant variations in morphological traits there were significantifferences between genotypes, and vice versa. The ANOVA wasomputed using morphological parameters and, similar to AMOVA,
showed that the three groups were statistically different (Table 4).Only the roundness of the drupes was not significant in cultivar dis-crimination, whereas all other parameters discriminated the threegroups, as shown by multiple comparisons (Bonferroni’s post-hoctest).
The three-way Mantel test was computed among the threeEuclidean distance matrices of morphological traits measuredon the leaves, drupes and pits that resulted highly correlated(r = 0.92072, p < 0.01). This explained that for each significant mor-phological variation in leaf traits, analogous significant variationswere shown in the morphological traits of drupes and pits, andvice versa. Hence, we compared the morphological groups eitheraltogether or separately in the following analyses.
Then, multivariate analysis was performed. The first PCA wascomputed for all morphological parameters of the leaves, drupesand pits (except the roundness of drupes) (Fig. 2A). The greaterpart of differentiation was explained along PC1 (57.4% of the totalvariance). The variables that principally weighed along this com-ponent were PI MiAL (loading −0.940), DR MaAL (loading −0.918),DR WE (loading −0.917), PI AR (loading −0.905) and PI FD (loading−0.899). Along the PC2, which explained 17.2% of the total vari-ance, the variables that principally weighed were LE AR (loading−0.950), LE PE (loading −0.911), LE MaAL (loading −0.877) andLE MiAL (loading −0.854). Along this component, the two ortets ofOliva nera di Colletorto (genotype 2) sampled in the COTEB-Larino(olive cultivars from the Molise collection) were separate from allothers.
The second PCA was only computed on leaf traits (Fig. 2B).Along PC1, which explained the 54.2% of the total variance, sev-eral variables were highly correlated: LE MiAL (loading −0.953),
M. D’Imperio et al. / Scientia Horticulturae 130 (2011) 229–240 235
F t for t( ) PCA
Lwl−
vtiDv−b
PoaPiWo
at
ig. 2. (A) PCA applied to all parameters of the olive leaves, drupes and pits (excepC) PCA applied to all parameters of the olive drupes (except for the roundness); (D
E AR (loading −0.822) and LE FD (loading −0.810). Along the PC2,hich explained the 44.5% of the total variance, the more corre-
ated variables were LE MaAL (loading −0.928) and LE PE (loading0.895).
The third PCA was only computed on drupes traits (Fig. 2C). Theariables that had the highest weight for PC1, explaining 81.7% ofhe total variance, were DR MiAL (loading −0.995), DR WE (load-ng −0.978), DR AR (loading −0.973), DR VO (loading −0.954) andR PE (loading −0.942). Along PC2, explaining 16.1% of the totalariance, the variable with the highest weight was DR EL (loading0.605). The two ortets were separate from other samples as well,ut less than in the PCA on leaf traits.
Finally, morphological measures of pits were used for the fourthCA (Fig. 2D). A clear separation of the three genotypes wasbserved especially along PC1 explaining 74.3% of the total vari-nce. The variables that had the highest weight for PC1 wereI MiAL (loading −0.990), PI WE (loading −0.969), PI VO (load-ng −0.968), PI AR (loading −0.963) and PI PE (loading −0.913).
hereas, the variable that had the highest weight for PC2 (20.7%
f the total variance) was PI RO (loading −0.706).
The mean shapes of the leaves, drupes and pits of each cultivarre reported in Fig. 3. Genotype 3 showed a larger leaf than the otherwo, whereas the Noccioluta had a strictly elongated leaf. Moreover,
he roundness of the drupes); (B) PCA applied to all parameters of the olive leaves;applied to all parameters of the olive pits.
Noccioluta presented drupes and pits with a higher roundness thanthe other two cultivars, whereas the presence of a distinct mucronon the drupes was characteristic of genotype 3. Finally, Oliva nera diColletorto could be distinguished by the reduced and sharp mucronof pits, which is pungent to the touch.
Multiple regression models (Table 5) allowed therefore toassign a sample to a cultivar by measuring several morphologicalparameters: if the equation gave a value of 1, the cultivar underinvestigation corresponded to Noccioluta (genotype 1); if the valuewas 2, the cultivar under investigation corresponded to Oliva neradi Colletorto (genotype 2); if the value was 3, the cultivar underinvestigation corresponded to genotype 3.
Model A used four variables to assign a specimen to a cultivar.Two of these were relative to pit parameters; roundness of pit andof leaf had a high weight in the model. The high values of R, R-squared and the Durbin–Watson test suggested an extremely goodreliability of the model. The value of Durbin–Watson also ensuredthe independence of the residues. The validation by MAE showsan extremely good result for the genotypes 2 and 3 (3.9 and 5.6%
respectively) and a good result for genotype 1 (16.6%) (Table 5). Thesame variables have been used for the construction of the model byLDA (Table 6). The first function has a smaller value of Wilks’ lambdaand so it has a greater discriminatory ability. The coefficients of
236 M. D’Imperio et al. / Scientia Hortic
Tab
le5
Mu
ltip
lere
gres
sion
mod
els.
Mod
eln
ame
Dep
end
ent
vari
able
(cu
ltiv
ar)a
Ind
epen
den
tva
riab
les
(equ
atio
ns)
bR
R2
Du
rbin
-Wat
son
Cor
rela
tion
ofre
sid
ual
sV
alid
atio
nA
E(M
ean
Abs
olu
teEr
rors
)(%
)
Gen
ot.1
(Noc
ciol
uta
)G
enot
.2O
l.n
era
di
Col
let.
orto
)
Gen
ot.3
Gen
ot.1
(Noc
ciol
uta
)G
enot
.2O
l.n
era
di
Col
let.
orto
)
Gen
ot.3
A1
23
−3.8
5[P
IR
O]+
3.05
[LE
RO
]+0.
81[D
RM
aAL]
−0.
26[P
IW
E]+
4.09
0.96
0.93
2.05
no
16.6
3.9
5.6
B1
23
−6.4
1[P
IR
O]−
4.77
[PI
MiA
L]+
10.9
00.
940.
891.
53N
Ec16
.25.
97.
1
C1
23
−2.5
6[P
IM
iAL]
+2.
00[D
RM
aAL]
−0.
98[D
RM
iAL]
−0.
21[P
IW
E]+
3.05
0.95
0.91
1.80
no
17.3
3.8
7.4
aIf
the
equ
atio
ngi
ves
the
valu
e1,
the
oliv
ecu
ltiv
aru
nd
erin
vest
igat
ion
corr
esp
ond
sto
Noc
ciol
uta
;ift
he
equ
atio
ngi
ves
the
valu
e2,
the
oliv
ecu
ltiv
aru
nd
erin
vest
igat
ion
corr
esp
ond
sto
Oli
van
era
diC
olle
tort
o;if
the
equ
atio
ngi
ves
the
valu
e3,
the
oliv
ecu
ltiv
aru
nd
erin
vest
igat
ion
corr
esp
ond
sto
gen
otyp
e3.
bIn
brac
kets
are
the
vari
able
sre
por
ted
inTa
ble
4.c
Cor
rela
tion
ofre
sid
ual
sca
nn
otbe
eval
uat
edac
cord
ing
toth
eD
urb
in–W
atso
nta
ble.
ulturae 130 (2011) 229–240
variables in the function indicate that the roundness of pits andleaves have a big weight in the discrimination. The relative weightof each variable in the first function is the same of the model Aobtained by LRM. The cross-validation of the model gives a perfectvalue of recognition (100% of samples in the training set correctlyclassified) and prediction (100% of samples in the test set correctlyclassified). The graphical representation of model A built by LDA isreported in Fig. 4A.
Model B required a lower number of parameters, and therebya lower error of measurement; only two pit variables were used(Table 5). The R value was again high but the R-squared togetherwith the values of the Durbin–Watson test were slightly lowerthan other models. In this case, the value of the Durbin–Watsonalgorithm does not allow the independence of residues to be eval-uated. The validation by MAE shows an extremely good result forthe genotypes 2 and 3 (5.9 and 7.1% respectively) and a good resultfor genotype 1 (16.2%). In the LDA (Table 6), the first function has agreater discriminatory ability (Wilks’ lambda = 0.10) and the vari-able with biggest weigh is the PI RO. The cross-validation of themodel gives a perfect value of recognition (100% of samples inthe training set correctly classified) while the prediction is slightlylower than the other two models (94.4% of samples in the test setcorrectly classified). In this case, a sample of genotype 2 was mis-classified as genotype 3. The graphical representation of model B isreported in Fig. 4B.
Model C also required four traits, two of them, PI MiAL andDR MaAL, had the higher weights in this model (Table 5). Asmodel A, the parameters of reliability were extremely good. Thevalidation by MAE is in line with the results obtained with theother two models. In the LDA (Table 6), the first function has avery good discriminatory ability (Wilks’ lambda = 0.03) but alsothe second function has a sufficient discriminatory ability (Wilks’lambda = 0.36) (Fig. 4C). The variables with greatest weigh are thoserelated to the axis length of drupes and leaves. The cross-validationof the model gives a perfect value of recognition (100% of sam-ples in the training set correctly classified) and prediction (100% ofsamples in the test set correctly classified).
4. Discussion
4.1. Molecular characterization
The He and PD values of microsatellite loci, clearly indicated thatthese markers were valid tools to discriminate among the threeolive cultivars. In particular, we showed 17 private alleles by geno-types, with alleles 131 and 151 bp that are in all samples of thegenotype 1 and genotype 2, respectively, and never in the others(Table 2). Both these alleles were in primer ssrOeUA-DCA7, whichdiscriminated all three genotypes, and proved to be the most dis-criminating primer (PD = 0.521) (Table 1). The PIC values rangedfrom 0.726 (ssrOeUA-DCA7) to 0.464 (ssrOeUA-DCA17).
As shown in Fig. 1A, the low differentiation of sample 8 forgenotype 1, samples 2 and 3 for genotype 3 and samples 14 and37 for genotype 2 were imputable to the one or two allele differ-ences. This might have occurred because of mutations in the allelesequences of genotypes under investigation (polymorphisms at anintra-cultivar level, as also described in Muzzalupo et al. (2010)).Hence, we can consider these five samples assimilated to corre-sponding genotypes (cultivars) also because of the correspondingmorphological data; the slight genetic variability found is negli-gible, in fact, it does not have effects on phenotype. This finding
highlights the importance to pursue both genetical and morpho-logical analyses.
Fig. 1A also shows that genotype 1 has separated more clearlyfrom the other two, which share a certain genetic similarity.
M. D’Imperio et al. / Scientia Horticulturae 130 (2011) 229–240 237
F transfm e leavo
p(A(1zoiattcTwi
itNt
TL
ig. 3. Mean shape of the olive leaves, drupes and pits obtained by inverse Fourier’sean dimensions of Major Axis Length and Minor Axis Length from each of the oliv
n the olive drupes of genotype 3.
As for the genetic profile results (Table 2), these values are onlyartially comparable with the results reported in Muzzalupo et al.2008b). In fact, we only have a primer in common (GAPU103A).s for Oliva nera di Colletorto and Noccioluta, Muzzalupo et al.
2008a) gave 136–170 as a pair of alleles, whereas our results gave36–174 and 136–136. Any discrepancy between our and Muz-alupo’s data can be justified by intra-cultivar variation, as reportedwn by Muzzalupo et al. (2010) or Baldoni et al. (2009). Moreover,n another paper, Muzzalupo et al. (2008a) reported that Nocciolutand Oliva nera di Colletorto were two strictly related cultivars withhe greater genetic similarity in a population of 119 Italian olive cul-ivars. On the contrary, our results showed that between these twoultivars there is a significant genetic difference (Tables 2 and 3).hese discrepancies could be due to the use of different primers,ith the result that our eight microsatellite loci were more efficient
n distinguishing the three investigated cultivars.Cluster analysis, as shown in Fig. 1B, revealed four major groups:
n the first cluster, Pendolino, Leccino and Frantoio are grouped,he second comprises only the Tonda iblea, in the third Moraiolo,ociara, Oliva nera di Colletorto and Dritta are grouped with geno-
ype 3 and the fourth is constituted by Carolea and Noccioluta. It
a In brackets are the variables reported in Table 4.
orm (program Shape 1.3) and computed on the three analyzed Italian cultivars. Thees, drupes and pits were reported in cm. The arrow indicates the mucron observed
is important to note that Noccioluta is separate from the othertwo cultivars from Molise, whereas Oliva nera di Colletorto clus-ters with genotype 3, which is more related to Dritta and Nociara(Table 3), widespread cultivars from the neighbour Abruzzo andPuglia regions, respectively. These four cultivars could be relatedby their common geographical origin (Adriatic district). Allele pro-files of genotype 3 were investigated by the online Olea database(Bartolini and Cerreti, 2007), and this showed that it was strictlyrelated to Istarska Belica (synonymous: Istarska Bjelica (Poljuhaet al., 2008b)), a typical Croatian/Slovenian cultivar, with a 62.5%similarity, as confirmed by the allele profiles reported in Poljuhaet al. (2008a). This could be justified by the strong link betweenthe human populations of the Balkan Peninsula and Italian Adri-atic coast; in fact, there are many historically well-integrated Slaviccommunities in the Adriatic district of the Molise region.
That genotype 3 is considered as a distinct cultivar from Olivanera di Colletorto and Noccioluta is significantly confirmed by the
genetic distance values (Table 3). The genetic distance betweenFrantoio and Leccino (0.640) is in fact smaller than the distancebetween Noccioluta and genotype 3, is similar to the distancebetween Noccioluta and Oliva nera di Colletorto and is slightly
Wilks’ lambda Cross-validation
Recognition(training set)
Prediction(test set)
0.04 100% 100%
0.72
0.10 100% 94.4%0.700.03 100% 100%
0.36
238 M. D’Imperio et al. / Scientia Hortic
Fig. 4. LDA applied to quantitative morphological parameters to discriminate threeendemic Italian olive cultivars; A: model built on [PI RO], [LE RO], [DR MaAL] and[PI WE] parameters of the olive leaves, drupes and pits (see Table 4); B: model builton [PI RO] and [PI MiAL] parameters of pits; C: model built on [PI MiAL], [DR MaAL],[DR MiAL] and [PI WE] parameters of pits and drupes.
ulturae 130 (2011) 229–240
higher than the distance between genotype 3 and Oliva nera diColletorto. However, to avoid another possible synonym, in an oliveworld germplasm already rich in several, we postponed the char-acterization of genotype 3 into future works. Hence, any doubtthat the synonymy between Oliva nera di Colletorto and Noccio-luta (Cicoria et al., 2000) was cleared, and these resulted in twowell-separated cultivars.
4.2. Morphological characterization
A high correlation between the morphological and moleculardata was found using Mantel’s test. In Olea europaea, this was thefirst time, to our knowledge, that a strictly significant relationshiphas been recorded between molecular and morphological differ-entiation. Previously, only Taamalli et al. (2006) have reported asignificant, but poor, correlation between genetic (AFLPs and SSRs)and agronomic data on Tunisian cultivars (r = 0.185, p = 0.05 andr = 0.156, p = 0.05, respectively) analyzing fourteen morphologicalcharacters measured on twenty-six cultivars.
The PCA computed for all morphological parameters of theleaves, drupes and pits (except the roundness of drupes) showed agood separation among the three genotypes in the scatter plot of thefirst two PCs (Fig. 2A). When morphological leaf traits were used(Fig. 2B), the separation of the three genotypes was less evidentthan for the other traits, probably because of the micro-variationof climatic conditions and agronomical factors that can influenceleaf morphology. In fact, the different locations of COTEB and thestudied area produced differences in climatic and edaphic condi-tions and for this reason the two ortets of Oliva nera di Colletortoare clearly separated from the other samples of same cultivar.
When morphological traits of drupes were analyzed, the threegenotypes were clearly separated (Fig. 2C). In particular, genotype1 (Noccioluta) was well separated from the other groups. This indi-cated that the drupes were useful traits to identify Noccioluta.
Finally, when morphological measures recorded on pits wereused (Fig. 2D), a clearer separation of the three genotypes wasobserved especially along PC1. These PCA showed that the twoortets of Oliva nera di Colletorto were well grouped with othersamples from the same cultivar. These results suggested that themorphological traits of pits were less influenced by environmentalpressure and were more affected by genetic control; this is proba-bly due to the following causes: (i) the wooden nature of the pits;(ii) the protection effect of pulp on the pit; (iii) the short expositionof environmental factor on the fruit (the fruits have one annualcycle whereas the leaves have two). These results were partiallyconfirmed by Hannachi et al. (2008).
The three models obtained by LRM have all an extremely goodreliability (Table 5). However, all three models have some trou-ble to identify the Noccioluta cultivar, perhaps because of greaterdiversity of the latter from the other two cultivars.
The procedure based on the first construction of the model usingthe LRM and the subsequent construction and validation of a similarmodel by LDA allows: (i) the reduction of number of attempts atblind to build a good model; (ii) the obtaining a model much morereliable and accurate because it is based on very useful variables;(iii) to make a much more rigorous validation. The good results ofthree models built by LDA are also shown in Fig. 4: a good separationof three groups can be observed.
In the model C is also evident a small separation of the geno-type 2 (Oliva nera di Colletorto) along the function 2. In this case,it was not necessary to use the software for morphological mea-surements; only a calibre and a balance were necessary. Hence, the
model C can be assumed as practical and reliable tool for cultivaridentification.
In any case, to allow a proper use of the experimental model,we must remember the limits: (i) it can only identify the three
Hortic
cabpmidec
5
cm
dsp
tmtttipfotato
tgrmi
pdicav
OibtAr
A
d1Cp(a
M. D’Imperio et al. / Scientia
ultivars investigated here; (ii) the applicability of these modelsre confined on restricted area of diffusion of three local cultivars,ecause there is an influence of environmental conditions on mor-hological characteristics. However, this does not mean that ourethodology cannot be adapted on other areas or other cultivars;
t is possible to construct new models from simple experimentalata. These models may contain the same variables but with differ-nt coefficients, or may contain other morphological variables thatould integrate or replace existing ones.
. Conclusions
In this paper, three olive cultivars have been investigated andharacterized by combining morphological and molecular data. Theost interesting results obtained in this work are as follows.A high correlation between the morphological and molecular
ata was found. Probably, this was the first time that a strictlyignificant relationship was recorded between molecular and mor-hological data.
There is an evidence of major genetic control on morphologicalraits of the pits, compared with the leaves and drupes, which were
ore affected by environmental pressure. The PCA demonstratedhat the morphology of the pits reflected a high correspondenceo the genetic assignment. Morphological parameters were usedo produce three statistical models (by LRM and LDA) for cultivardentification. For the limited environmental effect on their mor-hology, drupes and especially pits could be considered good toolsor cultivar characterization and identification. This method is verybjective because it is based on the automated sampling of quan-itative data, limiting the error related to the observation of traitsnd their assignment to qualitative classes, which may be not ableo describe the variability of traits in relation to a natural gradientf variation.
In spite of the low costs related to the analysis of morphologicalraits, molecular analysis remains an essential tool for investi-ating variability within and between genotypes and detectingelationships related to geographical and environmental effects onorphological traits. Furthermore, molecular data are useful to val-
date morphological models.However, the molecular and morphological analyses were com-
lementary tools for olive cultivar characterization and are valid foristinguishing new accessions. Separately, the two techniques are
ncomplete because molecular data are useless for cultivar identifi-ation by farmers or agronomists. By contrast, morphological datare useless without a molecular-based assignment because of theariability from environmental pressures.
Finally, the controversy about the possible synonymy betweenliva nera di Colletorto and Noccioluta was finally resolved, show-
ng significant differences between these cultivars, as supportedy molecular and morphological data. A possible new local geno-ype was identified. This was strictly related to two cultivars frombruzzo and Croatia/Slovenian called Dritta and Istarska belica,espectively.
cknowledgements
This project was funded by the Molise Region (art. 15ell’O.P.C.M. n. 3268/2003; decreto del Commissario Delegato n.92 del 9 ottobre 2006). We thank the San Giuliano di Puglia Localouncil and colleagues from PST “Moliseinnovazione” S.C.p.A., inarticular the Local Unit. We are grateful to Dr. Maurizio CorboARSIAM), Dr. Alessandro Patuto (Cooperativa Olearia larinese s.r.l.)nd the farmers for providing samples of olive trees.
ulturae 130 (2011) 229–240 239
References
Alba, V., Sabetta, W., Blanco, A., Pasqualone, A., Montemurro, C., 2009. Microsatellitemarkers to identify specific alleles in DNA extracted from monovarietal virginolive oils. Eur. Food Res. Technol. 229, 375–382.
Angiolillo, A., Reale, S., Pilla, F., Baldoni, L., 2006. Molecular analysis of olive cultivarsin the Molise region of Italy. Genet. Resour. Crop Evol. 53, 289–295.
Baldoni, L., Cultrera, N.G., Mariotti, R., Ricciolini, C., Arcioni, S., Vendramin, G.G.,Buonamici, A., Porceddu, A., Sarri, V., Ojeda, M.A., Trujillo, I., Rallo, L., Belaj, A.,Perri, E., Salimonti, A., Muzzalupo, I., Casagrande, A., Lain, O., Messina, R., Testolin,R., 2009. A consensus list of microsatellite markers for olive genotyping. Mol.Breed. 24, 213–231.
Barranco, D., Cimato, A., Fiorino, P., Rallo, L., Touzani, A., Castaneda, C., Serafín, F., Tru-jillo, I., 2000. World Catalogue of Olive Varieties. Internacional Olive Oil Council,Madrid, Espana.
Bartolini, G., Cerreti, S., 2007. World DataBase of Olive Tree.Belaj, A., Ojeda, A., Munoz, C., Trujillo, I., Barranco, D., Rallo, L., de la Rosa, R., Rallo, P.,
Díaz, A., Martín, A., Baldoni, L., 2008. Molecular markers in olive: an integratedapproach. Acta Hort. 791, 45–48.
Belaj, A., Satovic, Z., Cipriani, G., Baldoni, L., Testolin, R., Rallo, L., Trujillo, I., 2003.Comparative study of the discriminating capacity of RAPDm, AFLP and SSR mark-ers and of their effectiveness in establishing genetic relationships in olive. Theor.Appl. Genet. 107, 736–744.
Berrueta, L.A., Alonso-Salces, R.M., Heberger, K., 2007. Supervised pattern recogni-tion in food analysis. J. Chromatogr. A 1158, 196–214.
Besnard, G., Breton, C., Baradat, P., Khadari, B., Bervillé, A., 2001. Cultivar identifica-tion in olive based on RAPD markers. J. Am. Soc. Hort. Sci. 126, 668–675.
Botstein, D., Raymond, L., Skolnick, M., Davis, W.R., 1980. Construction of a geneticlinkage map in man using restriction fragment lenght polymorphisms. Am. J.Hum. Genet. 32, 314–331.
Bracci, T., Sebastiani, L., Busconi, M., Fogher, C., Belaj, A., Trujillo, I., 2009. SSR markersreveal the uniqueness of olive cultivars from the Italian region of Liguria. Sci.Hort. 122, 209–215.
Breton, C., Pinatel, C., Mèdail, F., Bonhommea, F., Berville, A., 2008. Comparisonbetween classical and Bayesian methods to investigate the history of olive cul-tivars using SSR-polymorphisms. Plant Sci. 175, 524–532.
Cantini, C., Cimato, A., Sani, G., 1999. Morphological evaluation of olive germplasmpresent in Tuscany region. Euphytica 109, 173–181.
Cicoria, M., Corbo, M., D’Uva, T., Ruggiero, A., 2000. Il germoplasma dell’olivo nelMolise, Quaderno divulgativo dell’E.R.S.A. Molise n.5/2000. Ente Regionale diSviluppo Agricolo per Il Molise “Giacomo Sedati”, pp. 1–63.
Cipriani, G., Marrazzo, M.T., Marconi, R., Cimato, A., Testolin, R., 2002. Microsatellitemarkers isolated in olive (Olea europaea L.) are suitable for individual finger-printing and reveal polymorphism within ancient cultivars. Theor. Appl. Genet.104, 223–228.
Díaz, A., De la Rosa, R., Martín, A., Rallo, P., 2006. Development, characterization andinheritance of new microsatellites in olive (Olea europaea L.) and evaluation oftheir usefulness in cultivar identification and genetic relationship studies. TreeGenet. Genom. 2, 165–175.
Dove, S.B., 2000. UTHSCSA ImageTool program 3.0.http://ddsdx.uthscsa.edu/dig/itdesc.html (accessed 2011 July 18).
Doyle, J.J., Doyle, J.L., 1990. Isolation of plant DNA from fresh tissue. Focus 12, 13–15.Excoffier, L., Smouse, P.E., Quattro, J.M., 1992. Analysis of molecular variance inferred
from metric distances among DNA haplotypes: application to human mitochon-drial DNA restriction data. Genetics 131, 479–491.
Ganino, T., Bartolini, G., Fabbri, A., 2006. The classification of olive germplasm – areview. J. Hort. Sci. Biotechnol. 81, 319–334.
Ganino, T., Beghè, D., Valenti, S., Nisi, R., Fabbri, A., 2007. RAPD and SSR markers forcharacterization and identification of ancient cultivars of Olea europaea L. in theEmilia region, Northern Italy. Genet. Resour. Crop Evol. 54, 1531–1540.
Gomes, S., Martins-Lopes, P., Lopes, J., Guedes-Pinto, H., 2009. Assessing geneticdiversity in Olea europaea L. using ISSR and SSR markers. Plant Mol. Biol. Rep.27, 365–373.
Hannachi, H., Breton, C., Msallem, M., El Hadj, S.B., El Gazzah, M., Berville, A., 2008.Differences between native and introduced olive cultivars as revealed by mor-phology of drupes, oil composition and SSR polymorphisms: a case study inTunisia. Sci. Hort. 116, 280–290.
Hartl, D.L., Clark, A.G., 1997. Principles of Population Genetics, 3rd ed. Sinauer Asso-ciates, Sunderland, MA.
Iwata, H., Ukai, Y., 2002. SHAPE: a computer program package for quantitative eval-uation of biological shapes based on elliptic Fourier descriptors. J. Hered. 93,384–385.
Kalinowski, S.T., Taper, M.L., Marshall, T.C., 2007. Revising how the computer pro-gram CERVUS accommodates genotyping error increases success in paternityassignment. Mol. Ecol. 16, 1006–1099.
Khadari, B., Breton, C., Moutier, N., Roger, J.P., Besnard, G., Berville, A., Dosba, F.,2003. The use of molecular markers for germplasm management in a Frencholive collection. Theor. Appl. Genet. 106, 521–529.
Kuhl, F.P., Giardina, C.R., 1982. Elliptic Fourier analysis of a closed contour. Comp.Grap. Im. Proc. 18, 259–278.
Langella, O., 1999. Population, 1.2.29 ed. CNRS, UPR9034.Morales-Sillero, A., Rapoport, H., Fernández, J.E., Troncoso, A., 2008. Olive fruit pulp
and pit growth under differing nutrient supply. Sci. Hort. 117, 182–184.
uzzalupo, I., Chiappetta, A., Benincasa, C., Perri, E., 2010. Intra-cultivarvariability of three major olive cultivars grown in different areas of central-southern Italy and studied using microsatellite markers. Sci. Hort. 126,324–329.
uzzalupo, I., Lombardo, N., Salimonti, A., Perri, E., 2008a. Molecular character-ization of Italian olive cultivars by microsatellite markers. Adv. Hort. Sci. 22,142–148.
uzzalupo, I., Salimonti, A., Caravita, M.A., Pellegrino, M., Perri, E., 2008b. SSRmarkers for characterization and identification of cultivars of Olea europaeaL. in the Abruzzo and Molise regions in south-central Italy. Adv. Hort. Sci. 22,129–135.
uzzalupo, I., Stefanizzi, F., Perri, E., 2009. Evaluation of olives cultivated in southernItaly by simple sequence repeat markers. Hort. Sci. 44, 582–588.
mrani-Sabbaghi, A., Shahriari, M., Falahati-Anbaran, M., Mohammadi, S.A., Nankali,A., Mardi, M., Ghareyazie, B., 2007. Microsatellite markers based assessment ofgenetic diversity in Iranian olive (Olea europaea L.) collections. Sci. Hort. 112,439–447.
rlandi, F., Sgromo, C., Bonofiglio, T., Ruga, L., Romano, B., Fornaciari, M., 2010.Yield modelling in a Mediterranean species utilizing cause–effect relation-ships between temperature forcing and biological processes. Sci. Hort. 123,412–417.
zkaya, M.T., Cakir, E., Gokbayrak, Z., Ercan, H., Taskin, N., 2006. Morphological andmolecular characterization of Derik Halhali olive (Olea europaea L.) accessionsgrown in Derik–Mardin province of Turkey. Sci. Hort. 108, 205–209.
zkaya, M.T., Ergulen, E., Ulger, S., Ozilbey, N., 2008. Molecular, morphological and oilcomposition variability within olive (Olea europaea L.) at semi-arid conditions.Biotechnol. Biotechnol. Equip. 22, 699–704.
age, R.D.M., 1996. TREEVIEW: an application to display phylogenetic trees on per-sonal computers. Comput. Appl. Biosci. 12, 357–358.
eakall, R., Smouse, P.E., 2006. GenAlEx 6: genetic analysis in Excel. Population
genetic software for teaching and research. Mol. Ecol. Not. 6, 288–295.
oljuha, D., Sladonja, B., Brkic Bubola, K., Radulovic, M., Brscic, K., Setic, E., Krapac,M., Milotic, A., 2008a. A multidisciplinary approach to the characterisation ofautochthonous Istrian olive (Olea europaea L.) varieties. Food Technol. Biotech-nol. 46, 347–354.
ulturae 130 (2011) 229–240
Poljuha, D., Sladonja, B., Setic, E., Milotic, A., Bandelj, D., Jakse, J., Javornik, B., 2008b.DNA fingerprinting of olive varieties in Istria (Croatia) by microsatellite markers.Sci. Hort. 115, 223–230.
Reale, T., Di Lisa, G.A., Rasio, R., 2002. Pedopaesaggi del Molise. Quaderno divulgativodell’E.R.S.A. Molise n.1/2002. Ente Regionale di Sviluppo Agricolo per Il Molise“Giacomo Sedati”, pp. 1–103.
Reale, S., Doveri, S., Díaz, A., Angiolillo, A., Lucentini, L., Pilla, F., Martín, A., Donini,P., Lee, D., 2006. SNP-based markers for discriminating olive (Olea europaea L.)cultivars. Genome 49, 1193–1205.
Rotondi, A., Magli, M., Ricciolini, C., Baldoni, L., 2003. Morphological and molecularanalyses for the characterization of a group of Italian olive cultivars. Euphytica132, 129–137.
Savin, N.E., White, K.J., 1977. The Durbin–Watson test for serial correlation withextreme sample sizes or many regressors. Econometrica 45, 1989–1996.
Sefc, K.M., Lopes, M.S., Mendonca, D., Rodrigues Dos Santos, M., Laimer Da CamaraMachado, M., Da Camara Machado, A., 2000. Identification of microsatellite lociin olive (Olea europaea) and their characterization in Italian and Iberian olivetrees. Mol. Ecol. 9, 1171–1173.
Taamalli, W., Geuna, F., Banfi, R., Bassi, D., Daoud, D., Zarrouk, M., 2006. Agronomicand molecular analyses for the characterisation of accessions in Tunisian olivegermplasm collections. Electr. J. Biotechnol. 9, 467–481.
Taamalli, W., Geuna, F., Bassi, D., Daoud, D., Zarrouk, M., 2008. SSR marker-based DNAfingerprinting of tunisian olive (Olea europaea L.) varieties. J. Agron. 7, 118–176.
Terral, J.F., Alonso, N., Capdevila, R.B., Chatti, N., Fabre, L., Fiorentino, G., Marinval, P.,Pèrez Jordà, G., Pradat, B., Rovira, N., Alibert, P., 2004. Historical biogeography ofolive domestication (Olea europaea L.) as revealed by geometrical morphometryapplied to biological and archaeological material. J. Biogeogr. 31, 63–77.
Tessier, C., David, J., This, P., Boursiquot, J.M., Charrier, A., 1999. Optimization of thechoice of molecular markers for varietal identification in Vitis vinifera L. Theor.Appl. Genet. 98, 171–177.
Trujillo, I., Rallo, L., Arús, P., 1995. Identifying olive cultivars by isozyme analysis. J.Am. Soc. Hort. Sci. 120, 318–324.
Viscosi, V., Fortini, P., Slice, D.E., Loy, A., Blasi, C., 2009. Geometric morphometricanalyses of leaf variation in four oak species of the subgenus Quercus (Fagaceae).Plant Biosyst. 143, 575–587.
