PARENT SESSION
Contributed Oral Session 165: Agroecology: Diversity and Modeling
Friday, August 12, 8:00 AM - 11:30 AM, Meeting Room 521 A, Level 5, Palais des congrès de Montréal

Modelling pollen dispersal of genetically modified oilseed rape within the field.

Demsar, Damjan¹, Debeljak, Marko¹, Lavigne , Claire², Dzeroski, Saso¹, ¹ Department of Intelligent Systems, Ljubljana, Slovenia² Lab.Ecologie, Systematique et Evolution (ESE), Paris, FRance

ABSTRACT- Large-scale commercial release of transgenic modified crops raises many agronomic and ecological concerns. The primary concern is to limit the uncontrolled escape of the introduced genes from the genetically modified crops. A key process to assessing the risks is the study of dispersal capacity of pollen grains. Our research has conducted a field experiment (Rennes, France) to study the pollen dispersal from the herbicide-resistant transgenic variety of oilseed rape. The male-fertile (MF) transgenic winter oilseed rape line "BOO4.oxy" was used as the pollen source. The area (90 by 90m) surrounding transgenic donor plot (10 by 10m) was sown with the near-isogenic MF line "BOO4" susceptible to Oxynil herbicides. In order to maximize the detection of pollen dispersal and to determine the selfing rate of the MF plants, ten seeds of the non-transgenic male-sterile (MS) line "Fu58.BOO4" were sowed on every node of a 29 by 29 grid throughout the entire field. Seeds of the MS and MF plants were collected separately on each of the 841 nodes of the grid. Seeds were sown and seedlings from all plots were tested on resistance to herbicide Oxytril. We modeled the pollen-dispersal (using rate of herbicide resistance of MF and MS seedlings per every plot) in dependence of the position, cardinal direction and distance from the center of donor field, the visual angle between the sampling plot and the donor field, and the shortest distance between the plot and the nearest edge of the donor field. The dispersal of pollen from the donor transgenic plants was modeled with the Clus data mining algorithm for learning predictive clustering trees, that predicts the resistance of MF and MS simultaneously. The classification model shows that the visual angle and the distance from the donor field are the most important attributes for predictions of out-crossing rate of MF and MS plant. If the visual angle is very broad (more the 2.17rad) then the distance between the donor field and the receptors plants is not very important and the out-crossing rate is 6 to 10 time higher as compared to the situation where the visual angle is narrow. In the latter case, the distance becomes important as well and the out-crossing rate decreases with the distance from the donor field. The ratio between the contaminated MS and MF plants also decreases with the distance from the donor field.

Key words: genetically modified organisms, oilseed rape, pollen dispersal, data mining