Evaluation of characters
The distribution frequency of the examined characters was normal or only slightly left- or right-biased (data not shown), which enabled further statistical analyses.
The coefficients of variation (CV) of the characters at the population level varied between 1.5 and 33 % (Table ). The average values of cone length (CL) and cone diameter (CD) were ≈9 mm. These two characters varied from 6.3 to 13.5 mm (CV = 9 %) and from 6.1 to 14.4 mm (CV = 10 %). The average value of the ratio CL/CD was almost equal to one and varied between 0.72 and 1.28 (CV = 6 %). The largest cones were found in the populations from southern Lebanon and Crimea (samples LB5, LB6 and CR2, respectively), while the smallest ones were from the high-altitude population in Lebanon (sample LB3) (Table ). The majority of the cones measured had four scale rows (CSR) and only a few cones (20 % of the trees from TU2) had six scale rows. The cone scale number (CSN) varied between four and 10 (CV = 10 %), and averaged 6.00 ± 0.57. The mean number of seeds per cone (SN) was close to 6, with cones containing 1–13 seeds (CV = 19 %) (Table ). As for the cones, the highest value of SN was found in the sample from the eastern side of Mount Lebanon (sample LB5) and the lowest value was found in the sample from the high mountains of this country (sample LB3). The seeds had a mean length (SL) and width (SW) of 4.75 ± 0.35 and 2.89 ± 0.28 mm, respectively. Seed length varied between 3.08 and 6.5 mm (CV = 7 %), and SW between 1.8 and 4.87 mm (CV = 10 %). The mean ratio SL/SW was 1.66 ± 0.14 with CV = 8 %. The seed dimensions did not show the same geographical trend as for the cone dimensions: the smallest seeds were found in CR1 and LB1 (Table ). The ratios CD/SN and SL/SW had a high level of variation (CV = 22 and 26 %, respectively). The ratios of the cone dimension over the seed dimension averaged 3.32 ± 0.39 (CD/SW, CV = 12 %).
Descriptive statistics of the measured morphological traits at the population level.
The average number of leaves per 5-mm apical section of the ultimate lateral branchlet (LN) was 22.73 ± 2.91, and varied between 12 and 36 (CV = 13 %) (Table ). The average value of the thickness of the last ramification shoot with leaves (ST) was 0.72 ± 0.07 mm and varied between 0.3 and 1.05 mm (CV = 6 %). All the populations had almost the same mean ST except for LB3 and LB4, which had a higher mean of 0.8 mm (Table ). The ratios ST/LN, SW/SN and CD/SN had the highest levels of variation.
Cone length and cone diameter (CL and CD) are highly and significantly positively correlated (R2 = 0.98, P< 0.01), as well as seed length (SL) and seed width (SW) (R2 = 0.74, P< 0.01) (Table ). The two features of the cone dimensions (CL and CD) significantly affected the number of seeds per cone (SN) (R2 = 0.77, P< 0.01 and R2 = 0.8, P< 0.01, respectively). Likewise, CL and CD were positively correlated with the number of cone scales (CSN) (R2 = 0.83, P< 0.01 and R2= 0.86, P< 0.01, respectively). We obtained a high, but less significant, correlation coefficient (P< 0.05) between the number of seeds per cone (SN) and the number of cone scales (CSN). This correlation was most likely derived from the higher correlations noted earlier between these two characters and the cone dimension features.
Correlation coefficients between nine characters of J. excelsa subsp. excelsa from all populations sampled; character acronyms as in Table .
Tukey's post-hoc test [Additional Information—File 1] showed that the highest number of statistically significant differences between the populations was found for the ratio of cone diameter/seed width (CD/SW), the ratio of cone diameter/number of cone scale rows (CD/CSR) and cone diameter (CD) and cone length (CL). The ratio of cone length/cone diameter (CL/CD), cone diameter/number of seeds per cone (CD/SN) and the seed length (SL) were significantly different between only a few populations, but mainly between the Turkish population (sample TU2) and some of the others. On the other hand, seed width (SW) showed significant differences only between Crimean and Turkish populations (CR2 and TU2). Moreover, the thickness of the last ramification shoot with leaves (ST), the number of leaves (LN) and the ratio of cone diameter/number of cone scale rows (CD/CSR) only differed significantly (P< 0.01) between two or three of the sampled populations.
All the calculated characters used in this study significantly discriminated between the samples at the level of P< 0.01. The discriminatory powers of the characters were very close, with values of partial Wilks' λ varying between 0.8 and 0.9 (Table ).
Discriminant power testing for the calculated characters of J. excelsa subsp. excelsa.
Sampling within each population was relatively uniform, with comparative numbers of individuals (Table ). The within-population differentiation using Ward’s agglomeration method was generally found at similar levels, with the first split into two groups of individuals at a distance between 11 and 17 [Additional Information—File 2]. We note that the smallest distances for this split are observed in the most fragmented and marginal populations, such as LB3, CY, CR1 and GR. The maximum separation distances of individuals within populations varied between 6 and 9, and were observed for individual 22 from TU3, 13 from CY and 3 from GR.
In the discrimination analyses using the calculated characters (without CD/CSR), all of the individuals were grouped together, except for the high-mountain Lebanese population (LB3) and the Cypriote population (CY) (data not shown). The discrimination analysis at the population level shown in Fig. reveals a clustering of the populations. According to the variable U1, which explained ~53 % of the total variation and mostly depended on the ratio of cone diameter/seed width (Table ), the high-altitude Lebanese population of Wadi El Njass (LB3) was separated from the other Lebanese populations (LB1, LB2, LB5 and LB6). The second high-altitude Lebanese population from Jbab el Homr (LB4) was dislocated at a middle distance in between. The populations most closely grouped were from Qammouaa, Barqa and Afqa in Lebanon (LB1, LB5 and LB6, respectively). All the other compared populations formed one group, although more dispersed.
Discrimination analysis results for J. excelsa. The results obtained by the two main discriminant variables (U1, U2) based on seven ratios are shown (acronyms as in Table ).
Coefficients of determination between discrimination variables and analysed characters of J. excelsa subsp. excelsa (character acronyms as in Table ). Bold values correspond to the highest coefficient values.
According to the variable U2, which was responsible for ~20 % of the total variation and mostly depended on the ratio between seed length and width (SL/SW), all the populations except for three of the Turkish populations and the Greek one (TU1, TU2, TU3 and GR, respectively) formed one group. However, the separation of the high-altitude Lebanese population (LB3), and the populations from Cyprus and one from Turkey, can be recognized in the space between the second and third discrimination variables U2 and U3. These two variables also differentiated between all Lebanese and Turkish, and Greek and Crimean populations [Additional Information—File 3] (Table ).
According to the cluster analysis by the Ward method, the sampled populations could be divided into three main sub-clusters (Fig. ). The first included the high-altitude population from Lebanon (LB3) with the coastal population from Crimea (CR1) and the population from Cyprus (CY). The second cluster grouped the southern Turkish populations (TU2, TU3 and TU4) with the population from Greece (GR). The third cluster included all the other populations from Lebanon (LB1, LB2, LB4, LB5 and LB6) with the northern Turkish population (TU1) and the mountain population from Crimea (CR2).
Dendrogram constructed by the Ward method of cluster analysis on the Euclidean distances between samples of J. excelsa (acronyms as in
The K-grouping method also revealed the most probable split of the populations into three groups (Fig. ). The classification matrix (Table ) showed that the level of congruence was the highest for the Lebanese high-altitude, the Cypriote and the coastal Crimean populations (LB3, CY and CR1, respectively), where 67, 63 and 57 % of the individuals, respectively, were included in the correct population group (Table ).
Classification matrix for individuals of J. excelsa subsp. excelsa as a result of a K-means cluster analysis for calculated characters; acronyms as in Table .
The first three barriers revealed by Monmonier's maximum difference algorithm applied by using BARRIER 2.2 software confirmed the separation of the high-mountain Lebanese population from all of the others and in the second level the differentiation of the Cypriote population (Fig. ).