Intraspecific Classification of Cucumis melo: How is the Morphological and Biochemical Variation of Melons Reflected at the DNA Level?

Rafael Perl-Treves1 and Asya Stepansky

Department of Life Sciences, Bar-Ilan University, Ramat-Gan 52900, Israel

Arthur A. Schaffer

Department of Vegetable Crops, The Volcani Center, ARO, P.O. Box 6, Bet-Dagan, 50250, Israel

Irina Kovalski

Department of Life Sciences, Bar-Ilan University, Ramat-Gan 52900, Israel

Additional index words. Cucumis melo, germplasm, diversity, sugars, DNA fingerprints, sucrose, glucose, fructose

Abstract. Intraspecific classification of Cucumis melo L. has been quite difficult, and most taxonomists still rely on Naudin's work from 1859. In this study, morphological and biochemical variation among 54 Cucumis melo accessions was compared to DNA level variation. Several morphological and biochemical traits were scored including fruit sugar composition. Large variability was observed in sucrose levels, but hexose sugars, glucose and fructose, varied as well. Among genotypes that accumulate sucrose, the levels of this sugar, and not of the hexoses, were correlated with the total sugar concentration. Hexose levels were correlated with total sugar levels only among those low-sugar genotypes that did not accumulate sucrose. Sucrose accumulation was observed not only among dessert melons of the inodorus and cantalupensis types, but also in representatives of other subgroups of C. melo, including some of the accessions from agrestis and conomon groups. DNA variation was assessed using the inter-SSR-PCR and RAPD techniques, and both fingerprinting techniques detected abundant polymorphism among melon genotypes. Cluster analysis indicated largest divergence between North American and European cantalupensis and inodorus cultivars as one group, and the more exotic varieties conomon, chito, dudaim, agrestis and momordica from Africa and Asia. Var. flexuosus, an ancient crop in the Middle East, may occupy an intermediate position. The molecular phylogeny agreed, broadly, with the classification of melon into two subspecies, and did not contradict the division into convarieties. It was however apparent the infraspecific division within the species is rather loose, the variation being distributed evenly between and within cultivar groups. Despite the great morphological diversity, it appears that the history of melon cultivation is too short, and the separation between varietal groups may be based on only a few genes, to enable an unambiguous classification based on DNA diversity.


We express their gratitude to J. H. Kirkbride for sending valuable reprints on melon taxonomy. We thank J. Felsenstein for providing his versatile PHYLIP package and providing continuous, patient instructions for their use. We are grateful to H.M. Munger from Cornell University, K. Reitsma of the USDA­ARS, Regional Plant Introduction Station, Ames, Iowa, K. Hammer of the I.P.K. at Gatersleben, M. Gomez-Guillamon of Experimental Station La Majora, Y. Cohen from Bar-Ilan University, and S. Niego and R. Herman from Zeraim Gedera Ltd. for providing seed samples for this study. We thank E. Galun and K. Nerson-Kanter for translation from German. This work was supported by grant no. IS-2129-92 from BARD, the United States-Israel Binational Agricultural Research and Development Fund., and grant no. 5036-1-96 of the Israeli GenBank, the Israeli Ministry of Science and Technology.

1Corresponding author.


Cucumis melo is considered the most diverse species of the genus Cucumis (Kirkbride, 1993; Jeffrey, 1980; Bates and Robinson, 1995). The species comprises feral, wild and cultivated varieties, including sweet dessert melons, as well as nonsweet forms that are consumed as vegetables. The most ancient records on cultivated Cucumis melo (reviewed by Pangalo, 1929) appear in Egyptian mural paintings.

Among the vegetables listed in the bible as being eaten by the Hebrews in Egypt (Deuteronomium 11,5) are the qishuim, likely identified as nonsweet C. melo varieties (similar to var. flexuosus or adzhur; M. Kislev, Identification of cucurbit species in Jewish sources, in press). Extensive records are also found in ancient Chinese writings from 2000 BC


Cucurbitaceae '98

(Walters, 1989) and Greek and Roman documents from the first century BC. Pangalo (1929) sustained that sweet melon forms were not known in the roman period, and were imported from Persia or Caucasus by travelers, making their appearance in Europe only around the 13th century.

What is the center of origin of Cucumis melo, and its primary site of domestication? A species center of origin should harbor all its present-day botanical variants; another criterion is the distribution of the wild species and relatives of the given crop. Twenty-eight wild Cucumis species are mostly found in Central and Eastern Africa (Kirkbride, 1993). Asia, on the other hand, appears to be an important center of intraspecific variation: wild (or feral) populations of Cucumis melo var. agrestis were described in Asia Minor, Afghanistan and India (rev. Pangalo, 1929; Jeffrey, 1980; Bates and Robinson, 1995), as well as in Africa. Both Asia and Africa have thus been proposed as melon's two possible sites of origin, although the Asiatic centers of variation may be secondary ones (Yang and Walters, 1992). Accordingly, the initial domestication may have occurred in Africa, Asia or at both sites in parallel (Mallick and Masui, 1986).

The extensive variation found in C. melo has led botanists to propose infraspecific classification schemes. Kirkbride (1993) emphasizes that such horticultural types should be treated under the rules of cultivated plants nomenclatures, and not as true botanical taxa. C. melo was divided into two subspecies, ssp. melo and ssp. agrestis (Grebenscikov, 1953; Jeffrey, 1980). The morphological character proposed as a key for the two subspecies involves the hairs that cover the female hypanthium: ssp. melo has pilose or lanate ovaries, while ssp. agrestis has sericeous ovaries.

In 1753 Linné described five species of cultivated melons. These were later united into a single species, Cucumis melo, by Naudin (1859), who developed a classification scheme based on a live collection of 2000 varieties (see Pangalo, 1929). Naudin's work remained the basis of all subsequent studies. Pangalo studied a live collection of 3000 specimen at the Russian Vavilov Institute, and proposed a more sophisticated multilevel taxonomyspecioids, convarieties, varieties

based on the idea of homologous series. Hammer et al. (1986) inherited a three-level classification system by Grebenscikov (1953) and tried to simplify it. He grouped under subspecies agrestis two convarieties, the eastern Asian conomon, and wild-growing agrestis. The second subspecies, melo, was divided into 10 convarieties. Munger and Robinson (1991) proposed a further-simplified version of Naudin's taxonomy, dividing C. melo into a single wild variety and six cultivated onesvar. agrestis,var. cantalupensis (Includes also former var. reticulatus), var. inodorus, var. flexuosus, var. conomon, vars. chito and dudaim (grouped together), and var. momordica.

In the last years, DNA fingerprinting techniques are used to resolve taxonomic and evolutionary relationships, including at the intraspecific level. In this study we tried to combine morphological scoring and DNA fingerprinting to address several open questions regarding melon classification. We asked whether the above mentioned classifications reflect a biologically significant demarcation between genotypes. Can we use morphological variation in certain key characters to sort melons into discrete groups, that would also be defined by molecular fingerprinting? Alternatively, do melons vary in a more continuous pattern, making varietal grouping more arbitrary?

A selection of 54 melon accessions from 23 different countries was described at the botanical­horticultural level, and DNA fingerprinting was applied to analyze the relationships among them. Two PCR-based methods were used, to exclude possible bias generated by a single technique. Random amplified polymorphic DNA (RAPD) profiles are obtained using decamer primers of arbitrary sequence (Williams et al., 1993). Intersimple sequence repeat (ISSR) PCR involves longer (16 to 18 nucleotides) primers encoding microsatellite elements that amplify DNA segments between microsatellite repeats (Gupta et al., 1994; Zietkiewicz et al., 1994). The data were subjected to cluster analysis, and the implications to C. melo germplasm classification are discussed. The full data and the detailed dendrograms will be published elsewhere (Stepansky, Kovalski and Perl-Treves, in press), while a survey of the fruit biochemical traits (sugars, pH invertase activity)

Cucurbitaceae '98

Table 1. Melon accessions used in this study. Accessions are ordered alphabetically. Data on country of origin and a tentative assignment to melon varieties are presented


Code Origin Name Accession no. Donor Group (tentative)

ACK Turkey Acuk PI 167057 2 flexuosus

AFG Afghanistan PI 125951 2 cantalupensis?

AGA Afghanistan CuM 146 1 agrestis

AGN Nigeria CuM 287 1 agrestis

AGR Africa 3 agrestis

BAK South Balkan CuM 53 1

BLA Spain Blanco C-199 5 inodorus

BNG USA Burpee's Netted Gem 4 cantalupensis

CAS Spain Rochet Pamal 4 inodorus, Cassaba type

CCA Spain cc26 (Amarillo orange flesh) C-446 5 inodorus, Cassaba type

CHA France Charentais 6 cantalupensis, Charentais type

CHI unclear Chito PI 140471 7 chito/dudaim

CHT India Chito PI 164320 2 chito

CON Far East line 85-893 7 conomon

COC China CuM 206 1 conomon

COV Vietnam Kairyo Ogata Kogane Seumari CuM 246 1 conomon

DHA Israel Dvash Haogen2 4 cantalupensis, Haogen type

DUA Afghanistan CuM 254 1 dudaim

DUD unclear Dudaim line 85-895 7 dudaim

DUG Georgia CuM 296 1 dudaim

END Israel Ein Dor 4 cantalupensis, Ananas type

FLI India CuM 227(=VIR K2511) 1 flexuosus

FLN India CuM 225 1 flexuosus

FLR Iraq CuM 349 1 flexuosus

FLX Lebanon Facus 4 flexuosus

GIN Japan Ginsen Makuwa (Silver Spring) PI 420176 2 conomon

GUO China Gou Gua PI 532829 2

HCR Spain Hilo Carrete C-198 5 inodorus?

HON USA Honeydew line 89A-15 7 inodorus

IML Kazakhstan Imljskaja PI 476342 2 cantalupensis

INB India PI 124112 2

IRN Iran PI 140632 2 cantalupensis

ITA Italy CuM 298 1 cantalupensis

JPN Japan PI 266947 2 cantalupensis

KAK India Kakri PI 164493 2 agrestis

KRK Turkey Kirkagac PI 169305 2 inodorus

KUV URSS Kuvsinka PI 506460 2 cantalupensis

LYB Libya CuM 294 1 cantalupensis

MEA Spain Melona Amarilla C-193 5 inodorus

MEC China CuM 255 1 cantalupensis?

MOM India PI 414723 7 momordica

OGO Japan Ogon No.9 PI 266933 2 conomon

PDS Spain Pinonet Piel de Sapo C-207 5 inodorus, Cassaba type

SAF Afghanistan Safed Sard PI 116915 2

SAL Ukraine Salgirskaja PI 506459 2 cantalupensis

SEN Senegal G 22841 PI 436532 2 agrestis

SNG Senegal G-22841 PI 436534 2 agrestis

SON South Korea Songwhan Charmi PI 161375 2 conomon

TM USA Top Mark 7 cantalupensis

USM USA PI 371795 2 momordica

VEL India Velleri PI 164323 2

WNT Turkey Winter Type PI 169329 2 inodorus

ZA1 Zambia ZM/A 5317 PI 505599 2 agrestis?

ZM1 Zimbabwe TGR 1843 PI 482429 2 agrestis?

ZM3 Zimbabwe TGR 228 PI 482399 2 agrestis?

Cucurbitaceae '98

and the physiological correlations between them is in press (Stepansky et al.).

Materials and methods

Growth and morphological description of plant material. Table 1 lists the plant material used and the seed sources. Three plants of each accession were grown in the spring a field plot near Rehovot according to standard agronomic practice; backup plants were grown at Bar-Ilan University as well. Plants were described morphologically at the flowering stage, fruit were described and photographed when mature. Botanical description included 26 traits, including: stem thickness, density of stem hairs, leaf color, male flower diameter, ovary shape, ovary pubescence, sex type; traits examined in the mature fruit included primary and secondary rind colors, rind pattern and textures, fruit size and shape, flesh color, presence of aroma, taste, abscission, seed weight. At least three fruit from each accession were harvested when mature and taken to the laboratory for biochemical analysis. Biochemical methods were reported elsewhere (Miron and Schaffer, 1991; Stepansky et al., in press).

RAPD and ISSR analyses. RAPD reactions were performed using random decamers (Operon Technologies, Alameda; primers OPA7,10,16,18; OPB6; OPC8; OPD7,8,11,13,20; OPL7; OPR2,10), according to Williams et al. (1993). The reaction products were subjected to electrophoresis on 1.5% TBE-agarose gels, stained with ethidium bromide and visualized under UV light. DNA fingerprinting by ISSR were according to the PCR protocol by Gupta et al. (1994). The primers used were from Primer set #9, University of British Columbia: (TC)8C, (AG)8T, (GGGTG)3, (ATG)6, (AC)8YC, (GA)8YG, (TG)8G, (AC)8G, (AC)8T. Annealing temperature was 5 oC lower than the estimated melting temperature. PCR products were separated through 1.8% agarose gels, stained and photographed as above.

Band scoring and cluster analysis. DNA fragments were scored as present (state 1) or absent (0). Morphological and biochemical data were coded as discrete characters for cluster analysis. Two softwares employing different algorithms were used: cluster analysis using parsimony meth

ods was performed using the PAUP program (Swofford, 1993). Shortest trees were also computed by distance-based methods using the program RESTDIST (distributed by J. Felsenstein, University of Washington, Seattle), that belongs to the PHYLIP package (Felsenstein, 1993), and cluster analysis was carried out using the NEIGHBOR, KITSCH, BOOTSTRAP and CONSENSE programs.


Phenotypic variation of Cucumis melo accessions. In order to study the phylogenetic relationships within Cucumis melo, an assembly of 55 accessions, including cultivars, landraces and wild or feral types from 23 countries was grown and described (Table 1). We have included representatives of all the horticultural types described by Munger and Robinson (1991) and sampled a substantial number of genotypes from Africa, Southern and Western Asia, and the Far East, i.e., the primary and secondary centers of diversity of the species. Twenty six morphological and biochemical traits were scored, and we tried to assign the accessions to horticultural groups (Table 1).

Sugar composition of the varieties is represented in Figure 1. Sucrose content ranged from 1 to 91 mg·g­1 fresh weight in individual fruit. Large, less expected variation was also observed in the reducing sugars, glucose and fructose. A strong, linear correlation (r2 = 0.8) can be computed between sucrose content and the total sugars content in the different varieties (Figure 2A). The correlation holds for those varieties that accumulate large amounts (>25 mg·g­1 fresh weight) of sucrose, indicating that in these fruit sucrose is the most significant component that contributes to the variation in total sugars. In most sweet varieties, sucrose constitutes 1/2 to 2/3 of the total sugars. In contrast, varieties with low total sugars (<30 mg·g­1 fresh weight) do not accumulate sucrose, and reducing sugars contribute most of their fruit sugars, and most variation in sugar content. This is illustrated by plots of glucose versus total sugars (Figure 2B; a similar plot was obtained for fructose), whose patterns are complementary to those of Figure 2A. A linear correlation is observed between reducing sugars and total sugars in the low-sugar varieties,

Cucurbitaceae '98

whereas sucrose accumulating fruit have variable contents of fructose and glucose, that are not significantly correlated with sucrose content.

Since each phenotypic trait may only provide partial taxonomic information, we tried to combine all the morphological and biochemical data and subject them to cluster analysis by the PAUP program. When two traits appeared to be interdependent, e.g., TSS and sucrose content, or ovary shape and fruit shape, only one was included. The seven quantitative traits in the set were coded as discrete-character states as in Hosoki et al. (1990)

by dividing them into three to five discrete classes. The main topological features of the phenotypic trees (not shown) broadly reflect the horticultural classification into two subspecies and five to six varietal types. The tree could be dissected into two large clusters, one that includes all the large-fruited, sweet accessions of vars. cantalupensis and inodorus, and the other comprising all the nonsweet, more exotic types of African and Asiatic origin. Within the sweet cluster, accessions are parted into subclusters, but no consistent dichotomic separation between cantalupensis and inodorus was

Figure 1. Histograms depicting average sucrose, glucose and fructose contents (expressed in mg·g­1 fresh weight) in 55 melon accessions.

Cucurbitaceae '98

observed. The exotic subtree included a few well-differentiated clusters: one included var. flexuosus, a second included small-fruited agrestis genotypes, and a third cluster comprised larger, nonsweet fruit (larger agrestis ZM1, ZM3 and ZA1, and var. momordica). Genotypes belonging to the chito, dudaim and conomon oriental varieties formed additional, intermixed subclusters. The tree topology reflected traits that were used as taxonomic keys: sweetness, aroma, fruit size and shape, sex type, and ovary pubescence.

Cluster analysis of melon varieties based on dna fingerprints. DNA fingerprints were generated using 14 RAPD primers. Each primer amplified an average of 6.9 bands. Following agarose-gel electrophoresis, 97 bands were scored, ranging in size between 0.3 and 2 kbp; 70% of these were polymorphic among the 54 accessions, and the brightest, clearly scorable bands were selected for cluster analysis. A second data set was produced by running ISSR reactions with 9 different microsatellite primers, producing 5 to 16 bands per primer (average 9 bands/reaction). Following agarose gel


electrophoresis, 116 bands of 0.3 to 2 kbp were scored among the 54 genotypes. About 90% of these bands were polymorphic among the genotypes, and the brightest bands were selected for cluster analysis. Although technically similar to RAPD, ISSR amplify a different genomic compartment, i.e., regions that are enriched with specific microsatellite repeats. Each of the two sets was analyzed using two different computational methods. The first was parsimony analysis, and the second involved distance-matrix programs. Cluster analysis of each data set separately produced rather similar trees, so we combined them into a single set of 42 ISSR and 44 RAPD bands. In this report we present, as an example, a single dendrogram produced by PAUP (Figure 3).

In this dendrogram, Western-world cultivars of the inodorus and cantalupensis types may be separated (by the 14-step long node) from the nonsweet types of theagrestis, conomon, momordica, dudaim, and chito varieties. It is also apparent that, despite their morphological distinctiveness, flexuosus accessions (FLN, FLI, FLR, FLX, ACK) did not cluster together, but are probably closer to the Middle-Eastern and European sweet types, than to the


Cucurbitaceae '98

Figure 2. Scatter plots of sucrose and glucose contents versus total sugar content in 55 C. melo accessions. Sugar contents, averaged from 3 HPLC measurments, expressed in mg·g­1 fresh weight. Total sugars are the sum of the sucrose, glucose and fructose measurements. (A) Sucrose versus total sugars. (B) Glucose versus total sugars.


nonsweet types of the second group. Within the first group, cantalupensis cultivars tended to cluster together (e.g., ITA, CHA, LYB, TM; SAL, KUV, IML), as did some of the inodorus genotypes (CCA, PDS, CAS; BLA and MEA), but such subclusters were intermixed, not allowing a clear-cut separation between the two types. Only two Indian nonsweet variety (VEL, INB) clustered amongst the dessert-flexuosus accessions and they in fact

seem to occupy an intermediate position between the two groups.

Within the exotic group, a few subclusters can be identified. A rather tight one (as indicated by the long internode, 11 steps, separating it from the rest of the tree) comprises the conomon cultivars (CON, COV, COC, OGO, GIN) and two dudaim lines (DUD and DUG). Chito, momordica and agrestis accessions are dispersed among a few subclusters.

Figure 3. Cluster analysis of molecular fingerprint data from 54 melon accessions. Data-base included 44 RAPD bands and 42 ISSR bands. A heuristic search was conducted by the PAUP software using TBR optimization options, resulting in 50 shortest trees of 766 steps. One of these trees is depicted. Numbers indicate the length (no. of steps) of each branch.

Cucurbitaceae '98

The African accessions with medium-size fruit, ZA1, ZM1 and ZM3 were set aside from the weedy, tiny-fruited agrestis accessions from Africa and Asia (AGR, AGA, AGN, SEN, SNG, KAK). No tight clustering was observed between the two momordica, the two chito and the three dudaim accessions, respectively. The same data were analyzed with the RESTDIST and NEIGHBOR programs, and rather similar results were obtained (not shown).


Variation in sugars. Our survey shows that sucrose can vary over a 90-fold range among melon varieties, but fructose and glucose content vary as well, over a 30-fold range. In most melon varieties, fructose and glucose attain similar levels; CHT was an interesting accession in this respect, since it had about twice the level of fructose as compared to glucose. Moreover, variation in hexoses is found in varieties both with high sucrose and lower sucrose, suggesting that one may, in principle, select for higher reducing sugars either instead, or without a parallel decrease, in sucrose level. An interesting conclusion of this study is that sucrose storing ability is not restricted to the dessert melon types, and varieties that accumulate significant amounts of sugars are found among conomon types (GIN, COC), chito (CHT), and even agrestis types (SEN, KAK). We still do not know whether or not the diverse genetic backgrounds are indicative of different sweetness loci that are expressed in each group. It would be interesting to see whether bringing together different sugar genes from diverse backgrounds will result in additive effects, or rather a physiological barrier will prevent additional gain in sweetness.

Melon infraspecific classification in view of phenotypic and molecular data. Using the assignment of gene-bank curators (mainly from IPK, Gatersleben), and our own morphological description, we have tried to classify 55 melon accessions to the horticultural types described by Munger and Robinson (1991). Such assignment relies on a small set of key traits that are of horticultural significance. When we performed cluster analysis based on a larger set of morphological

and biochemical traits, including ones that were not used for classification, the subdivision into most of the varietal groups persisted. This indicates that the botanical­horticultural traits were well chosen: the other traits either covary with the former, or do not suggest any consistent, alternative grouping of the germplasm.

The dendrograms based on molecular datawhich should represent neutral traits of simple inheritancecan provide a quantitative and more objective measure of variation, and of the relationships between specific taxa. The dendrograms obtained by using two independently derived molecular data sets, and two clustering algorithms, were similar in their general topology. Interestingly, they did not contradict substantially the topology of the phenotypic-traits tree. Thus, a node separating the varieties cantalupensis + inodorus from the other types is consistent with both the phenotypic and molecular data. The separation between the two sweet typescantalupensis, with climacteric, aromatic fruit, and inodoruswith long-keeping, nonclimacteric ones, is only partially substantiated by both data sets. Some cantalupensis-like genotypes cluster together, as do some of the inodorus lines, but the two groups are rather close. This was also observed in the smaller-scale surveys by Silberstein et al. (in press) and Staub et al. (1997), and may indicate that the maturation-related differences may be controlled by a small number of key genes and do not represent extensive genetic divergence between the two types.

The varieties agrestis, conomon, chito, dudaim and momordica appear related to each other, and more distant from the two dessert melon groups and from var. flexuosus. This feature of the tree may be interpreted in favor of the proposed subdivision into two subspecies, ssp. melo and ssp. agrestis. However, the data would support the grouping, within ssp. agrestis, not only of var. conomon and var. agrestis, as was proposed by Grebencikov (1953), but also of vars. chito, dudaim and momordica. The trait of ovary pubescence, suggested as a key for the definition of subspecies due to its neutrality, is only in partial agreement with the molecular phylogeny, since we encountered a few accessions with pilose ovaries in the

Cucurbitaceae '98

agrestis group, and sericeous ovaries in the inodorus group. While the trait apparently reflects phylogenetic reality, its use as a single dichotomic key may be problematic. According to the molecular data (but not the phenotypic data), accessions belonging to var. flexuosus were dispersed among the branches, closer to the inodorus and cantalupensis types, placing it in ssp. melo, which fits with the near-eastern origin of this ancient vegetable.

Another trait that divides rather naturally the melon germplasm into groups is sex type, encoded by the single locus, A (Pitrat, 1991). In our survey, all the accessions designated as flexuosus, momordica, agrestis and chito were monoecious. Accessions of var. dudaim (DUA and DUG) were andromonoecious, while DUD, that is more similar to chito, was monoecious. Three groupsconomon, inodorus and cantalupensishad andromonoecious flowering, although a few accessions had at least some monoecious individuals (AFG, CCA, IRN). It appears therefore that andromonoecy may have appeared at least twice, in the conomon-dudaim, and in the dessert-melon lineages.

The molecular phylogenyeven more than phenotypic clusteringidentifies var. conomon as a well distinct group. This probably reflects an ancient history of independent melon domestication in the Far East, separate from the domestication lineage leading to cantalupensis and inodorus types. The occurrence of sweet fruited genotypes in all three groupsagrestis, conomon, and dessert melonsprobably suggests a scenario of multiple domestication events that occurred in parallel in different places.

How distinct are melon varietal groups? distribution of molecular variation within and among varietal groups. The optimal trees derived from our data did not contradict the botanical classification into varietal groupsat least in their main features. On the other hand, the distinction between varietal groups as infraspecific taxa is not as strong or clearcut, as compared to the distinction between different species or genera (e.g., Perl-Treves et al., 1985). This is reflected in the tree topology (Figure 3): terminal branches, leading to individual accessions are long, due to the sensitivity of the molecular techniques that were applied, while those internodes that divide

taxa into clusters that may be taxonomically informative, are short. Subjecting the data to the Bootstrap test (not shown) proved the same point: The most significant, or consistent nodes are the terminal ones, that group two to three accessions together, and not the more internal ones. This indicates that the data points (i.e., polymorphic bands) that served to group together larger clusters of accessions are few, as compared to the numerous bands that specify terminal branches.

What could be the possible reasons for the relatively poor resolution among groups of accessions? Technical aspects of the fingerprinting techniques may be one source of non phylogenetic noise to our data (Schierwater, 1995). Unclear PCR bands would represent such a source; we tried to avoid it by scoring only the brightest bands, and carefully assessing the reproducibility of our PCR patterns. In fact, reducing the data set from an initial set of 200 bands to 86 best-looking bands resulted in an improvement in data consistency. PCR bands that are not genetically inherited are also a source of bias/noise, and have been widely documented (Schierwater, 1995; Staub et al., 1996). Checking the polymorphism for Mendelian inheritance and using for cluster analysis only bands exhibiting the expected segregation would solve the problem, but require considerable additional labor. In a preliminary screen of RAPD and ISSR markers we have ascertained the Mendelian inheritance of many of the bands that we scored, by analyzing a sample of 17 F2 progeny individuals of a cross between 'Top Mark' (TM) and PI 414723 (MOM; Stepansky, Kovalski and Perl-Treves, unpublished results). For example, 68% of the ISSR bands that were polymorphic between these parents segregated at a 3:1 ratio.

Other reasons that can account for a less sharp, more continuous distribution of genetic variation in the germplasm would involve biological, rather than technical, considerations. Short, less-significant internodes between groups of accessions may simply indicate that the history of melon evolution and domestication is too short to be resolved by DNA fingerprinting: the varietal groups formed over a relatively very short time span, which can be probed by only a small proportion of the traits that are scored. An examination of

Cucurbitaceae '98

the pairwise distances between the accessions indicates that the range of distance values between accessions of the same horticultural group are somewhat smallerbut not very much sofrom the values between pairs that belong to more distant groups, again suggesting a more gradual, or continuous, distribution of variation in the germplasm, rather than a well defined varietal subdivision. For example, Nei's genetic distance values among accessions of var. cantalupensis range between 0.2 to 0.4 units, and within inodorus0.15 to 0.38, while typical distances between a cantalupensis and a conomon, or an agrestis accession are 0.6 and 0.54 units, respectively.

Another, complementary explanation for the patterns of variation observed in this study, relates to the fact that melon varieties did not evolve any reproductive barriers between them. Wild and feral genotypes continue to grow, in many countries, in proximity of sweet or vegetable landraces, with whom they may freely hybridize. The occasional occurrence of sweet agrestis fruit may have resulted from such exchange. Selection by breeders who have combined genetic material from different groups (modern breeders have done so for inodorus and cantalupensis lines) may also result in poor resolution of the molecular phylogenies.

In conclusion, the molecular data generally seems to support the taxonomic classifications by Naudin, Pangalo and Jeffrey, albeit not in all its details. The distinction between infraspecific groups in the melon germplasm is, however, not a sharp one and should be more safely considered as a horticultural, rather than botanical, classification.

Literature cited

Bates, D.M. and R.W. Robinson. 1995. Cucumbers, melons and watermelons, p. 89­96. In: J. Smartt and N.W. Simmonds (eds.). Evolution of crop plants. 2nd ed. Longman Scientific, Essex.

Felsenstein, J. 1993. PHYLIP (Phylogeny Inference Package), version 3.5c. Distributed by the author, Department of Genetics, University of Washington, Seattle.

Grebenscikov, I. 1953. Die entwicklung der melonsystematik. Kulturpflanze 1:121­138.

Gupta, M., Y-S. Chyi., J. Romero-Severson, and J.L. Owen. 1994. Amplification of DNA markers from evolutionarily diverse genomes using single primers of simple-sequence repeats. Theor. Appl. Genet. 89:998­1006.

Hammer, K., P. Hanelt, and P. Perrino. 1986. Carosello and the taxonomy of Cucumis melo L. especially of its vegetable races. Kulturpflanzen 34:249­259

Hoey, B.K., K.R. Crowe, V.M. Jones, and N.O. Polans. 1996. A phylogenetic analysis of Pisum based on morphological characters, allozyme and RAPD markers. Theor. Appl. Genet. 92:92­100.

Hosoki, T., A. Ishibashi, H. Kitamura, N. Kai, M. Hamada, and T. Ohta. 1990. Classification of oriental melons based on morphological, ecological and physiological differences. J. Jpn. Soc. Hort. Sci. 58: 959­970.

Jeffrey C. 1980. A review of the Cucurbitaceae. Bot. J. Linnean Soc. 81:233­247.

Kirkbride, J.H. 1993. Biosystematic monograph of the genus Cucumis (Cucurbitaceae). Parkway Publishers, N.C. p. 159.

Mallick, M.F.R. and M. Masui. 1986. Origin, distribution and taxonomy of melons. Scientia Hort. 28:251­261.

Miron, D. and A.A. Schaffer. 1991. Sucrose phosphate synthase, sucrose synthase and acid invertase in developing fruit of Lycopersicon esculentum Mill. and the sucrose accumulating Lycopersicon hirsutum Himb. and Bonpl. Plant Physiol. 95:623­627.

Munger, H.M. and R.W. Robinson. 1991. Nomenclature of Cucumis melo L. Cucurbit Genet. Coop. Rpt. 14:43­44.

Naudin C.V. (1859) Essais d'une monographie des espèces et des variétés du genre Cucumis. Ann. Sci. Nat. Bot. sér. 4, 11:5­87

Pangalo, K.J. (1929. Critical review of the main literature on the taxonomy, geography and origin of cultivated and partially wild melons. Trudy Prikl. Bot. 23:397­442 [in Russian, translated into English].

Perl-Treves, R., D. Zamir, N. Navot, and E. Galun, 1985. Phylogeny of Cucumis based on isozyme variability and its comparison with plastome Phylogeny. Theor. Appl. Genet. 71:430­436.

Pitrat, M. 1991. Linkage groups in Cucumis melo L. J. Hered. 82:406­411

Schierwater , B. 1995. Arbitrary amplified DNA in systematics and phylogenetics. Electrophoresis 16:1643­1647.

Silberstein, L., I. Kovalski, R. Huang, K. Anagnostou, M.M. Kyle, and R. Perl-Treves. 1998. Molecular variation in Cucumis melo as revealed by RFLP and RAPD markers. Scientia Hort. (in press).

Staub, J., B. Jeffrey, and K. Poetter. 1996. Sources of potential errors in the application of random amplified polymorphic DNA in cucumber. HortScience 31:262­266.

Staub, J.E., J. Box, V. Meglic, T.F. Horejsi, and J.D. McCreight. 1997. Comparison of isozyme and random amplified polymorphic DNA data for determining intraspecific variation in Cucumis. Genet. Res. Crop Evol. 44:257­269.

Stepansky, A., I. Kovalski, A.A. Schaffer, and R. Perl-Treves. 1998. Variation in sugar levels and invertase activity in mature fruit representing a broad spectrum of Cucumis melo genotypes. Genet. Res. Crop Evol. (in press).

Swofford, D.L. 1993. PAUP: Phylogenetic analysis using parsimony. version 3.1. Computer program distributed by the Illinois Natural History Survey, Campaign.

Walters, T.W. 1989. Historical overview on domesticated plants in China with special emphasis on the Cucurbitaceae. Econ. Bot. 43:297­313.

Williams, J.G.K., M.K. Hanafey, J.A. Rafalsky, and S.V. Tingey. 1993. Genetic analysis using random amplified polymorphic DNA markers. Methods Enzymol. 218:704­741.

Yang S-L. and T.W. Walters . 1992. Ethnobotany and the economic role of the Cucurbitaceae in China. Econ. Bot. 46:349­367.

Zietkiewicz, E., A. Rafalski, and D. Labuda. 1994. Genome fingerprinting by simple sequence repeat (SSR)-anchore PCR amplification. Genomics 20:176­183.

Cucurbitaceae '98