Evolutionary Applications will be publishing a special issue on evolution and agriculture. Some of the papers are already available online. One that caught my eye is "Crops gone wild: evolution of weeds and invasives from domesticated ancestors" by Norman Ellstrand and others. (Among Ellstrand's past contributions is work on gene flow from transgenic crops to their wild relatives.)
They looked for cases where long-domesticated crops (>1000 years) evolved into weeds. Of 13 cases, seven involved hybridization, mostly with a wild relative. Many cases involved reversal of evolutionary changes that had occurred during domestication. As I point out in my forthcoming book on Darwinian Agriculture, much of crop genetic improvement has involved reversing effects of past natural selection that are undesirable in agriculture.
For example, nine species re-evolved seed shattering (scattering seeds as soon as they're mature), a trait seen in many weeds and in the wild ancestors of crops, but often lost during domestication. With harvesting by humans, seeds scattered on the ground apparently have worse survival prospects than those that get harvested. Even though the majority of harvested seeds get eaten, some are saved, protected from animals, and carefully planted the next year.
Similarly, three species re-evolved seed dormancy. This trait of wild plants and weeds delays growth of some seeds for a year or more, decreasing the risk that all of a plant's offspring will be killed in a bad year. Dormancy in a crop, however, results in (for example) soybeans coming up in the corn and being killed by herbicides to which corn is resistant.
One trait that was rarely reversed was looking like the crop. A weed that resembles the crop is less likely to be killed by hand weeding. I was familiar with an earlier study by Barrett (1983)showing that the rice-field weed, watergrass, evolved to resemble rice more than its own recent ancestor, barnyard grass, presumably because weeds that look more like rice are less likely to be removed by human weeders (Ehara and Abe 1950).
Weedy rice appears to have evolved from cultivated rice several times, with at least one case involving hybridization between two different kinds of cultivated rice. Weedy rye, on the other hand, appears to have a single origin. Like many weeds, it has smaller seeds than its crop ancestor, so it can make more of them with a given amount of carbon and nitrogen. This smaller seed size (and other traits, such as shattering and smaller leaves) evolved in less than 60 generations. (For comparison, 60 human generations would be about 1500 years, allowing significant evolutionary change within recorded history.)
Among species recommended for further study is strawberry guava. Although it's a serious invader of natural areas, apparently nobody knows how much, if it all, the invasive version has evolved from its cultivated ancestor.
Finally, they note that weedy versions of crops typically occur in the same areas as the crops themselves. They may grow at the edge of the field or in the wrong year of a crop rotation, but they could still be inundated by pollen from nearby crop plants, which could tend to slow evolutionary change. This would only apply to cross-pollinated species, of course. And all of those species evolved an earlier or later flowering date than the crop, limiting gene flow.
Barrett S. C. H. 1983. Crop mimicry in weeds. Economic Botany 37:255-282.
Ehara G., S. Abe. 1950. Classification of the forms of Japanese barnyard millet. Proceedings of the Crop Science Society of Japan 20:245-246.