Résumé
The large number of sorghum (Sorghum bicolor (L.) Moench) landraces maintained at ICRISAT represents a challenge for conservation and utilization of the germplasm collection. Different sampling strategies, either random or non-random, are being tested to obtain subsets of reduced size (core collection). Three subsets have thus been established from the entire landrace collection (ELC). One derives from a random sampling applied on the ELC stratified for photoperiod reaction (Logarithmic strategy: L). A second one takes into account the morpho-agronomic diversity (Principal Component Score strategy: PCS). The third subset is based upon classification, distribution and uses of sorghum (Taxonomic strategy: T). These three subsets do not differ as far as the overall phenotypic diversity is concerned (Shannon Indices = 0.85, 0.85, 0.82, 0.83 for the ELC, the subsets PCS, L and T, respectively). However, they may differ for the distribution of various traits: the PCS subset looks representative of the ELC; the L subset shows a shift for traits related to photoperiod reaction; the T subset shows skewed or flattened distributions, as it samples the most diverse landraces for eco-geographical adaptation, morpho-agronomic characters and uses selected by farmers. Genetic diversity retained by each subset has been assessed using the polymorphism at 15 SSR loci. The average allelic richness is equivalent for all subsets (16.1, 16.3, and 15.4 alleles per locus for PCS, L and T, respectively), as well as the average genetic diversity (0.81, 0.77 and 0.80). A high percentage of rare alleles (71 to 76% of a total of 206) is maintained in each of the three subsets. Thus, the molecular diversity retained in the subset is not affected by the sampling procedure based upon phenotypic characters. (Texte intégral)