Young, healthy, and fully-expanded leaves from the 627 plants representing 131 Spanish accessions as well as leaves belonging to three foreign accessions, were analyzed (Table S1 [suppl.]). Four or five plants per accession were used. This number was considered enough as populations in cucumber (and other cucurbits) are generally small, not only at farms but also in wild populations, favoring endogamy. The Spanish accessions included in the study were all landraces, collected from local farms. Thus, these accessions derive from a low number of plants, since local farmers usually grow small populations for self-consumption.

One hundred and twenty-seven Spanish accessions come from the COMAV and the other four accessions were provided by the BGHZ. Some of the accessions conserved at the COMAV genebank were collected by the holder institution and some others were collected by the BGHZ and maintained as duplicates in the COMAV. All these landraces are routinely reproduced using 10 plants per accession. Considering the origin of the Spanish accessions, 31 accessions came from Andalusia, 26 from Valencian Community, 21 from Castilla-La Mancha, 16 from Aragon, nine from Castilla-Leon, seven from Extremadura, six from Murcia, four from Canary Islands and three from the Basque Country, two accessions from Cantabria, Catalonia and Navarra, and one accession from Galicia and Asturias. Besides, three accessions provided by the Center for Genetic Resources (CGN, The Netherlands) and by the Chinese Academy of Agricultural Science (CAAS, China) were used as outgroups in the analysis. The phenotypic characterization of the complete collection of cucumber hold at the genebank of the COMAV has been recently conducted (Valcárcel et al., 2018). They established five groups (‘White’, ‘French’, ‘Short’, ‘Long’ and ‘Very long’) after a visual inspection, according to their similarity to commercial types or the phenotypic traits of the fruits, mainly skin colour and fruit length. Using this classification, 131 accessions were selected and used in the present study to perform the molecular characterization. Seventy of the accessions belonged to the ‘Short’ type, 48 to the ‘Long’ type, 11 to the ‘French’ type, one to the ‘Very long’ type, and the other to the ‘White’ type. The selection of the 131 accessions was based on the criteria of excluding accessions with very similar fruit characteristics and maximizing the diversity of origins. Still, the proportion of selected accessions corresponding to each group was similar to those in the complete collection.

DNA was extracted using the protocol proposed by Doyle & Doyle (1987) with some modifications and DNA quality and quantity was assessed using gel electrophoresis and spectrophotometry. A total of 23 SSR previously described by Lv et al. (2012) were analyzed in three sets of multiplex polymerase chain reactions (PCR) reactions (Table S2 [suppl.]). Each multiplex was carefully assembled according to the compatibility of the SSRs during PCR and the molecular size of their amplicons. The forward primer of the SSR markers was labelled with one of the three fluorescent dyes, carboxy fluorescein (FAM), carboxytetramethylrhodamine (TAMRA) and hexachloro-6-carboxyfluorescein (HEX). Multiplex PCR was carried out in a 11 μL volume using 5 μL of commercial Master Mix PCR Multiplex (Takara Multiplex Hot Short PCR, Takara), 20-40 ng of genomic DNA and labeled multiplexed SSR primers (0.4 pmol). The amplification was performed in an Mastercycler ^® personal thermocycler, and the amplification conditions were 95 °C for 5 min followed by 35 cycles of 95 °C for 20 s, specific annealing temperature for 90 s, and 72 °C for 30 s, and a final extension of 72 °C for 8 min. Multiplex PCR product was visualized using gel electrophoresis and then PCR fragment sizes were determined using capillary electrophoresis. The capillary electrophoresis was carried out by the Sequencing Service at the Institute for Plant Molecular and Cell Biology (IBMCP, Valencia, Spain), on an ABI 3100 ^® platform (Appl Biosyst, Foster City, CA, USA), and the allele sizes were calculated using GeneScan 3.7 (Appl Biosyst). For PCR fragment size determinations, 0.05 μL of an internal size standard (Rox-500, ROX) was mixed with 1 μL of diluted PCR product (1/100) and 6 μL formamide. The mixture was heated at 94 °C for 3 min and then cooled within icy water. The size of the SSR fragments was determined with the software packages GeneScan 3.7 (Appl Biosyst).

Polymorphic SSR markers were used to analyze the genetic diversity of 131 Spanish cultivars and the three foreign accessions considered as outgroups. The number of alleles (Na), the number of genotypes (Ge) and the effective number of alleles (Ne) was determined for each SSR locus in the Spanish collection using the PowerMaker software (Liu & Muse, 2005). To estimate the discriminatory power of the microsatellite loci, polymorphic information content (PIC) value for each locus was estimated by where pi and pj are the frequency of the ith and jth alleles, and n is the number of alleles (Botstein et al., 1980). The observed heterozygosity (H ₀) and expected heterozygosity (H _e) were computed for each SSR locus using GenAlEx version 6.501 (Peakall & Smouse, 2012).

Nei (1978) genetic similarities were calculated and an unweighted pair group method with arithmetic mean (UPGMA) phenogram was performed using genetic distances with the PowerMaker software (Liu & Muse, 2005) and plotted using TreeView software (Page, 1996). The reliability and robustness of the dendrogram was tested by bootstrap analysis with 1,000 replicates to assess branch support. Genetic distances were also used to graphically represent genetic relationships among accessions by principal coordinates analysis (PCoA) using GenAlEx 6.5 software (Peakall & Smouse, 2012).

Pearson’s correlation coefficients were computed between the number of alleles per locus found by Lv et al. (2012) and in our study using Statgraphics Centurion XVI software (Statistical Graphics, Rockville, MD, USA).

A total of 627 plants belonging to 131 accessions were analyzed using 23 SSRs. Two of them, SSR02895 and SSR07543, yielded amplification in less than 50% of the plants. SSR20852 and SSR23370 were monomorphic in all samples analyzed, while SSR16068 was monomorphic except for two plants of outgroup R41, homozygotes for the alternative allele. These five markers were not included in subsequent analysis.

The remaining 18 polymorphic SSR markers generated a total of 58 alleles in the Spanish collection (Table 1). The mean number of alleles per locus was 3.2, ranging from two to six. Significant correlation was found between the number of alleles per locus found by Lv et al. (2012) and in our study for the 18 polymorphic SSRs ( r=0.512, p=0.03). Lv et al. (2012) reported a number of alleles between six and 16 for these 18 SSRs when analyzing the mega-collection of 3,342 accessions with representation from the five continents. The lower number of alleles per locus obtained in our study may be due not only to the lower number of accessions analyzed but also to their limited origin. In fact, other authors found ranges for the number of alleles per locus similar to our study, using different SSRs, when analyzing smaller collections of cucumber accessions from specific origins (Kong et al., 2006; Watcharawongpaiboon & Chunwongse, 2008; Hu et al., 2010a, 2010b, 2011; Pandey et al., 2013; Yang et al., 2015).

Polymorphism levels and mutation rate were correlated with the number of repeat units, with different results depending on the species. A positive correlation was found in grapevine (Thomas & Scott, 1993), tomato (Areshchenkova & Ganal, 1999) and watermelon (Zhu et al., 2016), meanwhile negative correlation was found in rice (Panaud et al., 1996), brassica (Szewc-McFadden et al., 1996) and barley (Struss & Plieske, 1998). No correlation was found in our study between the nucleotide repeats and the level of polymorphism ( p›0.05). However, Watcharawongpaiboon & Chunwongse (2008) found a positive correlation between these traits. This different result could be explained because different SSRs were used, and therefore different regions of the genome were analyzed.

Differences were also found depending on the study when comparing the relative frequencies of the alleles. In our study, among the 58 alleles found in the Spanish collection, 34 (58%) were ‘common’ alleles, as their frequency was higher than 5% (Table 2). In fact, frequency of the most frequent allele for the 18 polymorphic SSR markers ranged between 50 and 98%, being for 12 of them higher that 75%. Out of the total 58 alleles, 15 (26%) were denoted as ‘less common’ alleles, with frequencies between 1 and 5%, while eight (14%) were ‘rare’ alleles (frequency between 0.1 and 1%). One of the alleles of SSR31399 was classified as ‘very rare’ allele (frequency lower than 0.1%) as it was only present in one plant, in heterozygous state. Among the 316 SSR alleles found by Lv et al. (2012) when analyzing the collection of 3,342 accessions, only 20% were classified as ‘common’ alleles. Again, differences may be due in part to the different size of the collections, but mainly to the most diverse origins of the accessions analyzed by Lv et al. (2012). The fact that the accessions included in their analysis represented different origins explains the higher variability, which conditions the lower frequency of ‘common’ alleles. Moreover, 59% of the alleles identified by Lv et al. (2012) showed a frequency below 1%, suggesting a broad representation of the diversity of cucumber. The Spanish collection merely represents part of the diversity of the species, thus the lower percentage of ‘rare’ and ‘very rare’ alleles is justified.

Accession-specific alleles were found for SSR19998, SSR22653 and SSR31399. SSR19998 allele 339 bp was present in two of the plants of accession Sty17, in homozygosis in one of them and in heterozygote state for the other one. Allele 401 bp for SSR22653 was specific for three plants of accession Lti113, in homozygosis in one of them and combined with allele 411 bp in the other two. SSR31399 allele 202 bp was the only ‘very rare’ allele of the collection. As previously stated, it was only present in heterozygote state in one plant, which belonged to accession Sty157. Moreover, some alleles appeared either exclusively or almost exclusively in some cucumber types. SSR16695 allele 172 bp and SSR10018 allele 137 bp were only present in accessions belonging to the short type. SSR10018 allele 147 bp, SSR19998 allele 344 bp and SSR22653 allele 409 bp were mainly present in short type accessions. On the other hand, some of the ‘rare’ alleles and the ‘very rare’ alleles were in many cases specific of concrete geographical origins. As an example, allele 143 bp of SSR16056 was only identified in accessions from Andalusia or SSR14861 allele 382 bp was exclusive of accessions from Andalusia or Castilla-La Mancha. All these results must be considered when using this information to rationalize management of the collection.

Most of the possible genotypes for each of the SSRs were identified in the Spanish collection, concretely, 105 out of the 133 genotypes (Table 1). As previously stated, the very rare allele identified in the Spanish collection (allele 202 bp for SSR31399) was only present in heterozygote state. The remaining homozygote combinations were present, except for homozygotes for allele 202 bp for SSR20218. This allele is ‘rare’ in the population (frequency=0.0088), thus the lack of homozygotes could be expected. The rest of genotypes absent in the Spanish population were heterozygotes, in no case formed by two ‘common alleles’.

The values of H ₀ ranged from 0.00 (SSR05723) to 0.22 (SSR16056), with a mean of 0.07. All individuals analyzed in the present study presented the SSR05723 in homozygosis, corresponding to the lack of heterozygosity observed for this marker. The highest value corresponded to SSR16056, the SSR with the highest number of alleles identified in our collection, two of them ‘common’ (with similar frequency), other two ‘less common’ and only the other two with frequencies lower than 1%. Results for this SSR were similar for the three groups that include more than one accession, i.e., ‘Short’, ‘French’ and ‘Long’ type. Using the same SSRs, Lv et al. (2012) obtained the maximum value of H ₀ (0.36) for SSR05012. This is one of the SSRs with higher H ₀ in the present study (0.14 as an average) although in this case differences were observed among groups, given that all ‘French’ type plants were homozygote for this locus. Two alleles were identified for this SSR in our collection (with frequencies 75 and 25%, respectively), while six different alleles were found in the samples analyzed by Lv et al. (2012). In both assays other SSRs presented a higher number of alternative alleles, so the high values for heterozygosity observed for this marker must correspond to the fact that none of the alleles appear at very low frequencies. The minimum value obtained by Lv et al. (2012) was 0.07, for SSR23220, also corresponding to low heterozygosity in our collection. In fact, only plants of four accessions of the ‘Short’ type were heterozygous for this marker. However, Kong et al. (2006) obtained higher values of the H ₀ (0.41), similar to the results obtained by Watcharawongpaiboon & Chunwongse (2008). Their results could be explained by the fact that some of the materials included were hybrids.

The values of H _e ranged from 0.04 (SSR19998) to 0.58 (SSR16056), with a mean of 0.261. These values were lower than those obtained by Lv et al (2012) analyzing the same SSRs in their mega-collection (0.58). Similar H _e values have been obtained by Kong et al. (2006) and by Watcharawongpaiboon & Chunwongse (2008). The H ₀ was relatively low if compared with the H _e value for all SSRs, suggesting certain degree of inbreeding of these materials. Besides, the relatively low number of alleles and gene diversity estimates for cucumber reflect its narrow genetic base (Dijkhuizen et al., 1996). This result was also obtained in other Cucurbitaceae like C. melo L. (Raghami et al., 2014). It should be stated that endogamy depression is not important in the cucurbit family (Fehet, 1992; McCreight et al., 1992; Tatlioglu, 1992). Besides, populations in cucumber, and other cucurbits, are generally derived from seeds obtained from one or few fruits in small populations, not only at farms but also in wild populations, favoring endogamy. The Spanish accessions included in the study were all landraces, collected from local farms. Thus, these accessions derived from a low number of plants, since local farmers usually grow small populations for self-consumption.

Taking into account the classification performed by Botstein et al. (1980), seven SSRs were reasonably informative (0.5›PIC›0.25), and the rest were only slightly informative (PIC›0.25). Similar to our results, Pandey et al. (2013) and Yang et al. (2015) analyzing around 40 cucumber cultivars each other from India and China obtained similar PIC values.

The AMOVA of the distance matrix for all the analyzed plants permitted a partitioning of the overall variation into two levels. The proportion of variation attributable to within-accessions differences was high, 45%, whereas 55% occurred among accessions. The intra and inter-accession variability’s distribution has implications for the management of conserved accessions when regeneration has to be done in the genebank. When similar intra and inter-accessions variability is found, the number of plants to be regenerated must be high, in order not to lose the variability among plants of the same accession and to avoid the influence of genetic drift and selection, resulting in changes in genetic composition. The effective population size and the methods of pollination and harvesting seeds have to be also taken into account. However, when variability is higher among accessions, the effort must be devoted to increase the number of regenerated accessions.

The mean similarity coefficient of the 131 cucumber accessions was 0.14, ranging for 0.01 (between Ltp8 and Ltp48) to 0.44 (between Shs101 and Fty183). Diversity in Spanish cucumber is lower than that found by Yang et al. (2015) analyzing 42 Chinese accessions (mean similarity coefficient=0.76). These results were expected and probably due to the fact that China is a secondary center of diversity.

None of the accessions showed the same genotype for all the analyzed plants. Sty122 accession showed the highest number of plants sharing the same genotype, four plants. The other plant showed three different alleles. Shs101 and Sty116 accessions had three plants with the same genotypes, and the other plants showed different genotype inter and intra-accessions. Lti47, Lhs121 and Stp144 accessions had two groups of two plants with the same genotype differing among genotypes in two or three SSRs, depending on the accessions. A total of 12 accessions (Ltp8, Stp89, Lti131, Ltp147, Sty165, Sty180, Sel79, Sty146, Stp56, Sty94, Stp59 and Stp62) showed two plants with the same genotype. The rest of the Spanish accessions showed plants with a unique genotype within accession.

Discrimination between closely related individuals has been reported using SSRs, even when few loci were employed (Powell et al., 1996). Therefore, SSR markers could be useful molecular markers in order to estimate the cucumber genetic diversity and also to perform marker assisted breeding programs. The dendrogram based on Nei coefficient revealed that the accessions were grouped into three unbalanced major clusters consisting of 1, 131 and 2 accessions, respectively (Fig. 1). As expected, accession Shs101 was separated from the rest of the accessions since it showed the lower similarity coefficient with most of the rest of accessions. Shs101 accession showed low variability since three plants had the same genotype with all the SSRs in homozygosity. For four of the SSRs, the plants of this accession were homozygote for alleles appearing at low frequency in the Spanish population. This accession was collected in Canary Islands. Cluster II included almost all accessions of the Spanish collection and one of the outgroups, and the subclusters obtained were not supported by high bootstrap values. Two of the outgroup accessions were grouped to the cluster III. Both accession showed different phenotype but they shared their Asiatic origin. Moreover, they both accessions showed specific alleles (alleles 202 bp and 292 bp for SSR31399 and SSR23220, respectively).

The clustering pattern of cucumber genotypes based on SSR markers was not in consonance with the groupings based on morphological traits or geographical area either including (Fig. 1) or not the outgroup (data not shown). Similar results were obtained when a PCoA (Principal Coordinates Analysis) based on SSR markers was done (Fig. 2); the first three principal coordinates accounted for 40.8% of the variation. These results seem to be in agreement with the absence of relationship between morphological traits and molecular markers found in cucumber (Yang et al., 2015) and other plant species like perennial ryegrass ( Lolium perenne), durum wheat ( Triticum durum), sorghum ( Sorghum bicolor), potato ( Solanum tuberosum), garlic ( Allium sativum) or tall fescue ( Festuca arundinacea) (Baker et al., 1998; Roldan-Ruiz et al., 2001; Geleta et al., 2006; Elameen et al., 2011; Chen et al., 2014; Sun et al., 2015). A higher variability in morphological traits was usually found. Neutral molecular markers such as SSRs are commonly used in molecular diversity studies and they may sample diversity in non-coding regions of the genome. Therefore, they have limited use in predicting the phenotypic diversity of individuals, especially in complex traits, which usually have a polygenic control. Consequently, low molecular variability may not be accompanied by low phenotypic variability for important traits (Collard et al., 2005). Null or low correlation between phenotypical and molecular diversity could also be due to low marker saturation (Sun et al., 2015). Therefore, more SSRs could improve the accuracy of cucumber genetic diversity estimation.

Dendrogram for the specific main groups (‘Short’, ‘Long’ and ‘French’) were performed, and a lack of agreement among subgroups were also found. Similar results were found when the geographical origin was taken into account (data not shown). Although environmental conditions and market orientation in a specific geographical area may select a specific morphological appearance, which can explain the high similarity among accessions from the same area, several accessions collected in different geographical areas could have at least some common genetic bases as a consequence of gene flow.

Given the low molecular diversity among the Spanish cucumber accessions found in the present study, a combined strategy was followed to perform the rationalization of the collection, including phenotypic traits, origin and molecular data. According to the groups established using morphological traits (Valcárcel et al., 2018), accessions of all of them were selected trying to conserve the phenotypic variation, to include accessions of most of the Spanish Autonomous Communities (Table S3 [suppl.]). Moreover, the genetic distances calculated from molecular data among accessions inside each group were taken into account. Priority was given to accessions carrying rare alleles, and phenotypic groups and Autonomous Communities with only one accession. The final set of selected accessions (Table S1 [suppl.]) included 47 accessions, six belonging to ‘French’ type, 15 to ‘Long’ type, and 24 to ‘Short’ type. The unique accessions molecularly analyzed belonging to ‘Very long’ and ‘White’ types were also selected. These accessions came from 14 Autonomous Communities. The phenotypic variation of the selected group was similar to the one of the complete collection, including mean, range and coefficient of variation for the studies quantitative traits and also for the qualitative ones (data not shown). The parameters used to explore the molecular variability (Na, Ne, H ₀ and H _e) were comparable for each marker between the complete and the selected collection, validating the selection made. However, further validation of the selection made should be done with data not previously used for the selection of the accessions, such as other relevant phenotypic traits and or a different type of molecular markers, as suggested by van Hintum et al. (2000), Parra-Quijano et al. (2011) and Odong et al. (2013).

According to the results obtained, the use of the stratified method, first constructing groups based on phenotypic traits, maximizing the origin and uniqueness of the accessions, and checking then the genetic distances among the groups and subgroups established, seem to be useful for rationalizing the collections of this crop. The rationalization of the collection done will be useful for optimizing the management of the cucumber collection conserved at the COMAV genebank.