Development of a core collection for the Australian annual medic collection
D.Z. Skinner,¹ G.R. Bauchan ², G. Auricht ³ and S. Hughes ^3
1USDA-ARS, Manhattan, KS and ²Beltsville, MD; and ³South Australian Research & Development Institute, Adelaide, South Australia.
A collection of 20,997 annual Medicago accessions, representing 34 species and 42 subspecies, was evaluated for 27 agronomic characteristics from 1968 to 1994. We have developed a method to assemble a core collection from these accessions by maximizing the selected diversity from within groups of accessions defined by species, subspecies, and geographic origin.
The mean Euclidian distance of each accession to all other accessions was calculated; passport data was used to group the accessions by species, subspecies, and geographical origin yielding 793 groups. Each group was represented in the core collection by the extremes of average Euclidian distance found within the group. A small average Euclidian distance indicated "typical" accessions, while a large average Euclidian distance indicated the more unusual types. Groups comprised of one or two accessions were included as a unit.

The mean Euclidian distances calculated ranged from 5.4 to 48.7. Retaining 0.5% of the typical end and 3.0% of the diverse end of each of the 793 groups resulted in the retention of 1705 accessions, or 10.4% of the entire collection. To evaluate how well this 1705 accession core collection captured the variation in the original collection, the ranges of each character in the core collection were compared to the corresponding ranges in the entire collection. Both extremes were captured in 20 of 27 characters. An additional accession to represent each of the extremes not included, easily could be added to the core collection.

Associations of trait expression were evaluated by calculating correlation coefficients of all combinations of the evaluated traits. Because the proportion of variance in one trait that can be attributed to its linear relationship to a second trait is indicated by the square of the correlation coefficient, we suggest that meaningful correlation coefficients in this case are those with absolute values greater than 0.71, i.e.more than 50% of the variance of one trait is predicted by the other. By this criterion, the relationships of interest are: seedling vigor and average winter yield (r=0.74), average spring yield and average winter yield (r=0.78), grams of seed per plant and grams of pods per plant (r=0.91), number of seed per plant and number of pods per plant (r=0.74), 1000 seed weight and 1000 pod weight (r=0.75), number of seeds per plant and grams of seeds per plant (r=0.77), florets per peduncle and pods per peduncle (r=0.86). None of these relationships were surprising, but they do suggest that it may not be necessary to measure all of the traits involved in future germplasm evaluations.
The core collection assembled here very likely represents the great majority of the genotypes present in the annual medic collection, and provides a resource for efficient further development of medics as part of an agricultural system.