Hybridization and introgression between crops and wild relatives may have important evolutionary and ecological consequences such as gene swamping or increased invasiveness. In the present study, we investigated hybridization under field conditions between crop lettuce (Lactuca sativa) and its wild relative prickly lettuce (L. serriola), two cross-compatible, predominantly autogamous and insect pollinated species. In 2003 and 2004, we estimated the rates of hybridization between L. sativa and L. serriola in close-to-reality field experiments carried out in two locations of Northern Switzerland. Seeds set by the experimental wild plants were collected and sown (44 352 in 2003 and 252 345 in 2004). Progeny was screened morphologically for detecting natural hybrids. Prior to the experiment, specific RAPD markers were used to confirm that morphological characters were reliable for hybrid identification. Hybridization occurred up to the maximal distance tested (40 m), and hybridization rates varied between 0 to 26%, decreasing with distance. More than 80% of the wild plants produced at least one hybrid (incidence of hybridization, IH) at 0 m and 1 m. It equaled 4 to 5% at 40 m. In sympatric crop-wild populations, cross-pollination between cultivated lettuce and its wild relative has to be seen as the rule rather than the exception for short distances.

Many varieties of transgenic rice are under development in countries where wild and weedy relatives co-occur with the crop. To evaluate possible risks associated with pollen-mediated transgene dispersal, we conducted a two-year survey in Vietnam to examine overlapping flowering periods of rice (Oryza sativa L.), weedy rice (O. sativa), and wild rice (O. rufipogon Griff.), all of which are inter-fertile. We surveyed populations in two regions of the Mekong Delta, northern and southern, and at three sites in each of three habitats per region: fresh water, saline water, and acid sulfate soil. Weedy rice frequently flowered simultaneously with neighboring cultivated rice plants. Flowering was more seasonal in wild rice and often peaked in November and December. Peak flowering times of wild rice overlapped with adjacent rice fields at all of the saline sites and half of the acid sulfate sites. The longer flowering season of wild rice ensured that crop-to-wild gene flow was possible in fresh water habitats as well. Our second objective was to determine whether wild and weedy rice populations are exposed to pests that could be targeted by future transgenes, which may then provide fitness benefits. These populations shared many pathogen and insect herbivore species with cultivated rice (leaffolder, locust, cricket, planthoppers, rice bug, stem borer, sheath blight, blast, bacterial leaf blight, and brown spot). Damage by leaffolders and locusts was the most frequently observed insect feeding damage on all three rice types. Indicator species analysis revealed that most of the insect herbivores were associated with particular habitats, demonstrating the importance of broad geographic sampling for transgenic rice risk assessment. These survey data and the strong likelihood of gene flow from cultivated rice suggest that further studies are needed to examine the effects of transgenic traits such as resistance to pests on the abundance of wild and weedy rice.

Outcrossing of non-Bt cotton (Gossypium hirsutum (L.)) in refuges by transgenic Bt cultivars could reduce the efficacy of refuges for delaying resistance in seed-feeding pests. Based on reports that outcrossing decreased as distance from Bt cotton increased in small-scale studies, we hypothesized that increasing refuge width or distance from Bt fields would reduce outcrossing. In a large-scale study in Arizona, we quantified Bt seed in refuges of experimental and commercial fields, comparing outcrossing between in-field (narrow) and external (wide) refuges and among rows of refuges at various distances from Bt fields. Some refuges, including those in tightly controlled experimental plots, contained up to 8% adventitious Bt plants. Some, but not all, Bt plants likely resulted from Bt seed in the non-Bt seed bags. We did not detect a difference in outcrossing between in-field and external refuges. However, statistical power was low because outcrossing was low (< 0.4% of seeds) in both treatments. Higher outcrossing levels (≤ 4.6% of seeds) were observed in the studies measuring outcrossing at various distances from Bt fields, yet outcrossing did not decrease as the distance from Bt fields increased. We hypothesize that Bt plants in refuges cross-pollinated surrounding non-Bt plants, overshadowing the expected association between distance from Bt fields and outcrossing.

Spatially isolating genetically modified (GM) maize fields from non-GM maize fields is a robust on-farm measure to keep the adventitious presence of GM material in the harvest of neighboring fields due to cross-fertilizations below the European labeling threshold of 0.9%. However, the implementation of mandatory and rigid isolation perimeters can affect the farmers' freedom of choice to grow GM maize on their fields if neighboring farmers do not concur with their respective cropping intentions and crop plans. To minimize the presence of non-GM maize within isolation perimeters implemented around GM maize fields, a method was developed for optimally allocating GM maize to a particular set of fields. Using a Geographic Information System dataset and Monte Carlo analyses, three scenarios were tested in a maize cultivation area with a low maize share in Flanders (Belgium). It was assumed that some farmers would act in collaboration by sharing the allocation of all their arable land for the cultivation of GM maize. From the large number of possible allocations of GM maize to any field of the shared pool of arable land, the best field combinations were selected. Compared to a random allocation of GM maize, the best field combinations made it possible to reduce spatial co-existence problems, since at least two times less non-GM maize fields and their corresponding farmers occurred within the implemented isolation perimeters. In the selected field sets, the mean field size was always larger than the mean field size of the common pool of arable land. These preliminary data confirm that the optimal allocation of GM maize over the landscape might theoretically be a valuable option to facilitate the implementation of rigid isolation perimeters imposed by law.


We comment on Zaida Lentini's summary of Session V (titled “Estimating Likelihood and Exposure”) of the 9th International Symposium on the Biosafety of Genetically Modified Organisms. We provide an explanation of the drawbacks of using empirical pollen dispersion models, based largely on the general representativeness of the data used to generate the empirical models. We exemplify the drawbacks by highlighting the limited data used to develop the empirical model of Gustafson (presented in the same Symposium session). We provide a discussion of the meaning of “worst-case” assessments for pollen dispersion, how “worst-case” assumptions are commonly used in environmental impact assessments and how regulators will view worst-case impact assessments differently from the regulated (biotech) community. Finally, we clarify the advantages and disadvantages of mechanistic models and explain why they are often used in preference to empirical models in environmental impact assessments.

DiGiovanni and Kevan (2008, Environ. Biosafety Res.7: 105–108) commented extensively on the empirical approach that I and my co-authors took in our previous modeling of pollen-mediated gene flow in maize (Gustafson et al. (2006) Crop Sci.46: 2133–2140). As we detailed in that original paper, gene flow is a highly complex process that necessarily requires at least some level of empiricism in order to adequately quantify all of the biological, meteorological, and physical phenomena that are involved. DiGiovanni and Kevan favor a mechanistic modeling approach, and they proposed a number of potential advantages for such a method over our entirely empirical technique. However, the 20 m buffers we had proposed based on our empirical model continue to be supported by the rapidly growing body of experimental data on maize gene flow that has now been collected in Europe and elsewhere around the world. This does not mean there is no place for mechanistic modeling of gene flow, but it does suggest that properly implemented empirical approaches have a valid role to play. They offer a degree of simplicity and practical utility that is not available from more complicated approaches.