EU Perspective - Sampling for Testing of GM-Impurities

The distribution of a contaminant in a bulk mass greatly influences the effectiveness of sampling procedures. Most currently used sampling guidelines for genetically modified organisms (GMO) testing have stringent distribution requirements. Specifically, all these protocols are based upon the assumption that GM material, if present, is randomly distributed. Yet, assuming randomness is risky because the applied sampling plan may not provide an accurate estimate of the content of the contaminant. Despite the general consensus that an evaluation of the possibility of non-random distribution of GM materials within lots is a pre-requisite for the definition of effective sampling protocols, no experimental data on the distribution of GMOs in kernel lots are available anywhere in the world. Additionally, to date there is no distribution-free statistical model to estimate sampling errors, thus making the theoretical definition of optimal sampling technique impossible when randomness cannot be assumed. Here we presents preliminary results of the KeLDA project, which is the first study investigating the distribution of GM materials in large soybean lots imported within the European Union. The extensive variability of GM content, observable both within and among soybean lots, indicates that heterogeneity issues cannot be overlooked when defining sampling strategies for large lots of particulate materials and that randomness cannot be assumed a priori. A new statistical model free from any distribution assumptions, developed in order to estimate the sampling error associated with different sampling protocols together with the risk of false-negative results, is also presented. The new model, which is applicable to any consignment of particulate material with respect to any kind of contamination, presents two novelties: first, freedom from any distribution assumption; second, the possibility of estimating the magnitude of the sampling error associated with different sampling protocols as a function of specific distributional ...