In the summer of 1996, we planted 144 A. thaliana rosettes at random locations in a grid at our field site in central Illinois, at densities reflecting those of nearby A. thaliana populations. One-quarter of our plants were homozygous for the dominant and mutant allele of acetolactate synthase, Csr1-1; this allele confers resistance to the herbicide chlorsulphuron. Mutants were originally isolated through mutation of the wild-type A. thaliana ecotype, Columbia, by ethyl methanesulphonate, and had been backcrossed for six generations3. One-quarter of the plants in the field were wild-type Columbia, and the remaining half were Columbia-strain plants transformed to express Csr1-1 in a pBin vector4,5. The latter possessed insertions at a single site, were homozygous for this insertion, and were equally divided between two independently transformed lines.

Plants were grown in the absence of herbicides, and were visited frequently by syrphid flies which consume pollen and nectar. At the season's end, we identified progeny produced through outcrossing by germinating seeds produced by each wild-type Columbia plant on plates containing 100 nM chlorsulphuron. Chlorsulphuron-resistant seedlings were transplanted to the greenhouse, where outcrossing was prevented with pollination bags, and their selfed seeds were collected. We then identified progeny fathered by transgenic A. thaliana by germinating selfed seeds on plates containing 50 mg l−1 kanamycin (only transgenic fathers were resistant to both chlorsulphuron and kanamycin).

A survey of approximately 100,000 seeds showed that the per-plant outcrossing rate was 0.30% for mutant fathers and 5.98% for transgenic fathers. Transgenic A. thaliana were roughly 20 times more likely to outcross than ordinary mutants. We screened a subsample of 281 transgenic progeny with primers specific to each insertion site in order to identify which of the transgenic lines had fathered them. We calculated the outcrossing rates for these two lines as 1.2% and 10.8%. Both transgenic lines showed increased outcrossing relative to the mutant (χ2 tests, P<0.003 in each case), but the two transgenic lines differed in their propensity to outcross (P<0.001).

To be certain that seed contamination did not contribute to estimates of outcrossing, we re-plated the seeds from all selfed, resistant plants on chlorsulphuron plates. Contaminants would have been homozygous for resistance whereas progeny resulting from outcrossing would have been heterozygous. Over 99% of our resistant progeny were heterozygous, and only heterozygous individuals were included in our calculations. We also placed known resistant and susceptible seeds on all selection plates to confirm our ability to determine the phenotypes of progeny.

Our results show that wild-type A. thaliana are more likely to be fertilized by the pollen of transgenic rather than mutant A. thaliana when each expresses the mutant allele Csr1-1. Although A. thaliana is unlikely to become a pernicious weed, these results show that genetic engineering can substantially increase the probability of transgene escape, even in a species considered to be almost completely selfing.

Our results do not prove that enhanced outcrossing is due to the transgene itself, but rather that a difference in outcrossing between transgenic and mutant plants exists. Because we do not know the underlying genetic mechanism, the generality of our result is unclear at present. Even if enhanced outcrossing is restricted to Csr1-1, however, our results are of broad relevance because this transgene has been introduced into dozens of agricultural crops, and is advocated as a selectable marker for plant transformation vectors6.