Plant materials were selected from a collection of Lagenaria siceraria maintained at Nangui Abrogoua University (Abidjan, Côte d’Ivoire). The seed samples of thirty accessions were collected mainly in five geographical zones (South, East, Northeast, North, and Centre) of Côte d’Ivoire. The selected accessions were representative of two cultivars. The first one, with round fruit is characterized by the presence of a cap on the distal side of seeds (C) and the second cultivar (SC), with elongated fruits and characterized by seeds without a cap (Figure 1).

The cultivar “C” contained 22 accessions and the cultivar “SC”, 8 accessions, according to seeds availability in each cultivar. A complete description of these accessions is given in Table 1. Ten seeds per accession, in total 300 seeds were analyzed.

C presence of a cap; LS, Lagenaria siceraria; SC without a cap.

The young leaves of each seedling were collected and stored at −80˚C until use. These samples were used for DNA isolation and PCR analysis. DNA isolation was carried out according to procedure described by Levi and Thomas [17] with a few modifications. Fresh leaf (0.1 g) tissue was finely ground in 1.5 ml microtubes (eppendorfs) in liquid nitrogen and resuspended in 700 µl CTAB extraction buffer [0.1 M Tris-Base, 1.4 M NaCl, 2.5% cetyltrimethylammonium bromide (CTAB), 20 mM EDTA-dissodium, 0.2% sodium dodecyl sulfate (SDS), 0.5% Sarkosyl, 250 mg polyvinylpyrrolidone (MW 40 (PVP-40) and 250 mg polyvinylpyrrolidone (PVP)]. Each tube was mixed by gentle agitation and then incubated for 30 min at 65˚C. The supernatant was taken and 350 µl of isopropanol were added to precipitate the DNA. The DNA pellet was washed in absolute ethanol and dried. Then the pellet was resuspended in TE to a final concentration of ca 100 ng∙µl⁻¹ containing 10 g∙ml⁻¹ RNAse. The DNA solution was stored at −20˚C until use. DNA concentration was measured by a Nanodrop ND-1000 spectrophotometer (NanoDrop Technologies, Inc).

The microsatellite markers had been set up for Lagenaria siceraria previously [16] . The PCR reaction condition used was as follows: genomic DNA samples (15 ng) were amplified in a 15 µl reaction volume containing 1× ThermoPol Reaction buffer (20 mMTris HCl, 10 mM (NH₄)₂SO₄, 10 mM KCl, 0.1% Triton X-100 pH 8.8 @ 25˚C), 0.2 mM each of the four dNTPs, 2 mM MgCl₂, 0.5 mM of each forward and reverse primer, and 0.5 U of Taq polymerase (BioLabsInc, NEW ENGLAND). The amplifications were performed in a thermocycler (Biometra) programmed as follows: an initial cycle at 94˚C for 3 min, followed by 40 cycles at 94˚C for 30 s, 52˚C - 55˚C for 30 s and 72˚C for 1 min. Cycling was followed by a final extension at 72˚C for 8 min, and a soak at 4˚C.

PCR products were separated in denaturing 6% polyacrylamide gels prepared using an acrylamide/bisacryla- mide ratio of 19:1, 0.53 TBE (Tris boric acid ethylenediaminetetraacetic acid) buffer, 0.1% ammonium persulfate (APS), and 8.33% tetramethylethylenediamine (TEMED). Polyacrylamide gels were cast in a vertical gel casting plate. Immediately after addition of APS, 70 ml of the gel solution was poured directly into the gel casting plate. The plate with gel solution was then kept at room temperature for approximately 1.5 hours to allow polymerization. The amplified DNAs were mixed with 20 µl of formamide dye (98% formamide, 10 mM EDTA pH 8.0, 1% xylene cyanol and 1% bromophenol blue) before denaturation by heating for 3 min at 90˚C. Three microliters of each denaturated DNA mixture were loaded onto a pre-warmed polyacrylamide gel. Electrophoresis was performed at 55 W for 2 hours. The separated DNA bands were revealed using a silver staining method as described by Creste et al. [18] which was slightly modified.

Fourteen primer pairs used to evaluate 44 entries of Chinese bottle gourd were tested to estimate the genetic diversity among Lagenaria siceraria accessions in the present study.

The genetic diversity was evaluated based on genotype and allele frequencies, using the level of polymorphism 0.95 criterion. There should be at least two alleles each with a frequency of at least 0.05. Only one allele has a frequency of 0.95 and the rest of the alleles have less than 0.05. Hence, the locus cannot be considered polymorphic. To evaluate the informativeness of each marker, polymorphic information content (PIC) of an SSR locus was calculated, based on the allele frequencies [19] . The number of alleles per locus, estimates of observed and expected heterozygosity, and Shannon’s Information Index were calculated for each population an each locus using GenAlEx v. 6.1 [20] . Comparison between observed and expected heterozygosities were examined according to Mann-Whitney U test using software STATISTICA version 7.1 [21] . The fixation indices were estimated at each polymorphic locus and tested for significant deviation using an exact test performed by the software Genepop [22] . Within each accession, null allele frequencies were estimated using the maximum likelihood estimator based on the EM algorithm of Dempster et al. [23] and implemented in Genepop 4.0 [22] .

Analysis of molecular variance (AMOVA) was calculated for the sampled accessions to estimate the partitioning of genetic variation at different levels and then to investigate the hierarchical level upon which genetic variation can be attributed. Significance of AMOVA was tested using a nonparametric permutation approach with 999 permutations [24] .

A model-based on clustering algorithm in order to search for the most likely number of accessions sampled was used. This algorithm assigns individuals to accessions and also assesses accessions heterogeneity as implemented in the STRUCTURE program [25] . The STRUCTURE analysis was conducted at five replications of K (assumed number of subpopulations), ranging from 1 to 10, with 100,000 repetitions of Markov Chain Monte Carlo (MCMC) and a burn-in period of 50,000, using the admixture model. Each assessment of K was repeated five times to check the repeatability of the results.

An UPGMA tree based on the Nei’s genetic distances matrix was constructed in PHYLIP package version 3.6 [26] . Cluster analysis was used to describe the relationships among and within different L. siceraria accessions. First, 1000 times bootstrapping was performed on SEQBOOT program to generate confidence in the dataset. Then, biased genetic distance from gene frequencies on GENDIST program was computed [27] . The cluster analysis tree was produced with the NEIGHBOUR program which use a matrix of pairwise distances (based on gene frequency genetic distances) between all pairs of accessions and CONSENSUS program.

Confidence in tree topology was assessed by bootstrapping over loci (1000 iterations) and the phylogenetic tree was visualized in TREEVIEW 1.6.6 [28] .

Among fourteen primer pairs tested, only three showed polymorphism in this studied. The allelic composition of each marker in each genotype was determined to calculate a PIC value. The average PIC value was 0.61 with a maximum of 0.65 observed with LSR030 and a minimum of 0.55 observed with LSR020 (Table 2).

The population statistics generated by the three microsatellites has been summarized in Table 3. A total of 116 alleles were detected across the three loci. The mean effective number of alleles per locus (A), varied respectively from 1 (Ls020) to 5.56 (Ls166) with a mean of 2.88 (Table 3). The average observed heterozygosity (Ho) was 0.631, ranging from 0 (Ls020) to 1 (LS207) and the average expected heterozygosity (He) was 0.645, ranging from 0 (Ls020) to 0.863 (LS166) (Table 3). Mann-Whitney U test indicated that there is no significant difference (p > 0.05) between observed heterozygosity and expected heterozygosity. Indeed, of the 89 inbreeding coefficients calculated, only 36 (40.5%) were significantly different from zero (p < 0.05). Eighteen inbreeding coefficients showed negative indices. The average inbreeding coefficients (F_IS = 0.040) was significantly different from zero (p < 0.05) for the analyzed accessions. Null alleles frequencies estimates ranged from 0 (Ls005) to 34.15% (Ls147), and were consistent with the F_IS estimates. Overall, the three markers investigated appeared to be affected by at least one null allele. The mean accession diversity using the Shannon Information Index (I) was 1.113. Accession Ls166 was the most diverse (I = 1.805) and the least diverse accession was Ls020 (I = 0).

Analysis of molecular variance (AMOVA) within and among 30 accessions of Lagenaria siceraria revealed that 39% of the total variation resides among accessions and 61% within accessions (Table 4). Calculations carried out separately for differentiation among cultivars, exhibited similar trends of AMOVA taking into account no prior grouping of accessions. We found more genetic diversity within cultivars (90%) than among cultivars (10%).

PIC, polymorphism information content.

^ap-value of the score test for heterozygote deficiency, with: *1%\p-value \5%; **0.1% \p-value \1%; ***p-value \0.1%; ^b95% confidence intervals (CI) for null allele frequencies provided by Genepop. A effective number of alleles per polymorphic locus; Ho observed heterozygosity; He expected heterozygosity under Hardy-Weinberg equilibrium; FIS Fixation index; Fnullnull alleles frequencies per locus and accession; I Shannon’s Information Index and SE standard error of sample mean.

PA, partitioning all accessions; PCu, partitioning per cultivar; df, degrees of freedom; SS, sum of square; MS, mean square; Est. Var. estimated variance and %D distribution of total variance. The probability was estimated computing 999 permutation.

The Bayesian analysis using the software STRUCTURE indicated the presence of two main clusters in the entire set of accessions. The highest value for ΔK, the rate of change in the log probability of the data between successive potential numbers of clusters, was obtained for K = 2. Estimated log probability of the data was higher under K = 2 (−3576.96) than under K = 1 (−4170.1). High proportions of admixed individuals were observed in the region of six accessions with assigned membership different from one to another individual. The results were plotted to evaluate the geographical relationships of the accessions and the cultivars in different genetic clusters (Figure 2). The first cluster was composed of ten accessions. All of them are characterized by the presence of a cap on the distal side of seeds. The second was composed of 20 accessions. Twelve of these accessions are characterized by the presence of a cap on the distal side of seeds, eight showed the absence of a cap on the distal side of seeds. Accessions from southern part of Côte d’Ivoire are exclusively member of the first cluster and accessions from northeastern, northern and center part of Côte d’Ivoire are exclusively member of the second cluster. Only accessions from the eastern part of Côte d’Ivoire are member of the two clusters.

The UPGMA phylogenetic tree based on the Nei’s genetic distances matrix from SSR data is shown in Figure 3. The dendrogram consisted of two major clusters. The branch separating these two clusters was well supported (bootstrap = 1000). The first cluster contained a group of twenty-seven accessions (I), the second one (II), consisted of three accessions from eastern part of Côte d’Ivoire (one from Gontougo and the two others from Moronou).

The PIC value is a measure of the polymorphism level detected by a particular marker and is dependent on the number of alleles detected and their distribution in the population. The number of alleles (1 to 7) and PIC mean value (0.61) observed in this study suggested that the SSR markers selected were highly informative with sufficient discriminatory power.

Indeed, based on the PIC values, all the markers were highly informative (PIC > 0.50). Such a result indicated the high utility of used set of markers for genetic diversity analysis. A similar PIC value was reported in eight C. lanatus (another cucurbit species) accessions collected in Zimbabwe [29] using nine SSR primers. A similar trend was observed in investigations carried out by Ram et al. [30] in five germplasm lines of bottle gourd (Lagenaria siceraria) using six RAPD primers. These results also indicated profound polymorphism in bottle gourd landraces.

The used microsatellites with wide range of heterozygosity reduced the risk of overestimating genetic variability, which might occur with microsatellites exhibiting only high heterozygosity. Although varying across the loci, the mean values of observed heterozygosity were lower than the expected mean heterozygosity values. However, failure of significant differences between observed and expected heterozygosities according to Mann-Whitney U test (p > 0.05) suggested random mating in Lagenaria siceraria.

The number of alleles at different marker loci serves as measure of genetic variability having a direct impact on the differentiation of accessions. The allelic variation in this study was lower than those obtained by Gürcan et al. [31] in 60 Turkish bottle gourd accessions. This discrepancy could be attributed to the methods used in

sample genotyping. Indeed, Gürcan et al. [31] used the capillary electrophoresis method for allele determination, whereas we used a direct reading on polyacrylamide gels. The capillary electrophoresis method is known to be the most efficient approach in sample genotyping [32] . The gene diversity indices obtained from the present study were higher than those (He = 0.073; Ho = 0.053) reported for 30 L. siceraria accessions from Nangui Abrogoua University germplasm using allozyme markers [15] . This discrepancy could be due to the fact that molecular markers are more polymorphic than isozymes. Significant deviations from Hardy-Weinberg expectations due to heterozygotes deficits were observed in 11 accessions, confirming random mating in the plant material studied. The same results had been reported by Koffi et al. [15] . A relative high prevalence of null alleles has been observed in the concerned accessions. Null alleles affect population parameter estimates. The observed heterozygosity would be largely underestimated [33] .

We found a significantly low level (39%) of genetic differentiation among accessions and within them the estimated genetic variation was 61% (AMOVA, P = 0.001). The results of this study were different from the findings of Minsart et al. [34] in Citrulluslanatus with, 88% variation among accessions and 12% within them. These results show that the structuring of the genetic diversity could be depend on the mode of sampling. For the present study, we selected ten seeds and a relatively higher number of accessions (30) while in the study conducted by Minsart et al. [34] , 20 seeds were randomly chosen per accession and only three accessions were selected. Such contrasted sampling schemes should have resulted in inverted trends in accession’s genetic structure as revealed by AMOVAs. Accession genetic structure observed in this study was similar to that reported in the previous study [29] . Performing AMOVA within and among seven accessions of watermelons divided into two major groups (cow-melons and sweet watermelons). Authors demonstrated that only 0.8% of the total variation resided between the two groups, 10% between accessions within groups and 89.2% within accessions.

Overall, results from accession genetic structure showed that L. siceraria maintained a high level of variability within cultivars in accordance with its mating system, coupled with farmer’s seed management approaches. Indeed, at the collecting sites a few seeds are usually saved from the previous season’s harvest, or obtained from neighboring farmers or local markets, resulting in the gradual depletion of genetic variability. Similar results have been reported in Cucumeropsismannii, another oleaginous cucurbits cultivated in Côte d’Ivoire [35] .

Accession structuring pattern derived from Bayesian clustering analysis revealed two clusters.

Data collected show a clustering according to geographical location. This result indicated that besides the forces such as exchange of genetic stock, genetic drift, spontaneous variation, natural and artificial selection, geographical origin also is responsible for genetic diversity.

A dendrogram established based on SSR genotyping of 30 accessions, also detect clustering by geographical location, which is not in agreement with Yetişir et al. [36] in which clustering of bottle gourd accessions from Turkey was based around fruit morphology much more than on geographical origin. No significant grouping pattern based on cultivar was observed in group I, on other hand, in group II, all the accessions are characterized by the presence of a cap on the distal side of seeds. However, according to the findings of Koffi et al. [15] , the UPGMA cluster analysis of morphological differentiation among cultivars of Lagenaria siceraria showed that the used two cultivars were well separated. Consistent with the result from Uluturk et al. [37] , morphological and molecular genetic diversity are distinct factors and must be considered separately in germplasm characterization. This is especially important for crops like cucurbits which have limited molecular genetic diversity.