In this paper, I will sketch some features of weed ecology and evolutionary biology that, in my view, are relevant to assessing the prospects of pest resistance transgene escape into populations of agricultural weeds. I will focus on weeds of field crop agroecosystems, rather than addressing the broader category of invasive plants in general.

This discussion is organized around a model for transgene escape that distinguishes three phases leading to the establishment of widely distributed populations of weeds carrying a transgene. In this scheme, the first event is hybridization between a weed and a transgenic crop. Second, a process of introgression and adaptation occurs in which evolutionary mechanisms improve maladaptive features of the early-generation products of hybridization, resulting in a weed bearing a pest resistance transgene and having a reasonably high level of adaptation to certain agroecosystems. Finally, a process of dispersal distributes this 'neo-weed' over the landscape accompanied by local adaptation to variable conditions encountered when dispersal is over a sufficiently broad area.

I will survey aspects of weed ecology and evolutionary biology that appear important to the operation of each of these three phases. Frequently, I will be in the uneasy position of suggesting plausible implications of suspected features of weed ecology. Unfortunately, in many instances neither these features nor their implications have received more than fragmentary documentation. There are enormous gaps in our knowledge of weed ecology. Many aspects that would likely be widely agreed-upon by weed scientists simply have not been described by published observations and experiments. This paucity of data reflects the prevailing focus of weed science in recent decades on herbicidal weed control to the neglect of ecological inquiries and, especially, of theoretical frameworks.

Clearly, hybridization between transgenic or conventional crops and sexually compatible relatives (Snow and Palma 1997) occurs in many crops and has produced new forms of weed behavior in resulting populations (Barrett 1983). Recent work has documented such hybridization in detail and makes clear that transgenes can be expected to escape even across large spatial barriers and significant barriers of genetic incompatibility (Snow and Palma 1997). Transgene escape by hybridization appears inevitable in some systems. However, in other cases it is unclear whether hybridization is a rate-limiting phase in the escape of transgenes. My premise is that hybridization may indeed be rate-limiting in some circumstances, for example when hybridization is occurring across a substantial incompatibility barrier. Aspects of weed ecology that may affect hybridization rates in these situations include weed breeding systems and effects of spatial and temporal distributions of weeds at several scales.

The most common breeding system among weeds of field crop agroecosystems is a mixed mating system in which both self-fertilization and cross-fertilization occur, although other reproductive systems are also known (Barrett 1992). Therefore, the most prevalent weed breeding system permits hybridization, but such crosses must occur in the face of a substantial rate of self-fertilization.

Breeding systems and other aspects of genetic systems and reproductive ecology that affect hybridization rates are known to vary within and among weed populations. For example, jimsonweed (Datura stramonium) populations in North Carolina have flowers that open to pollinators and show approximately 10% outcrossing rates (Motten and Antonovics 1992). In contrast, certain populations in Indiana are exclusively self-pollinating, with flowers that do not open to pollinators (Weller and Jordan, unpubl. data). In some cases, this variation may reflect adaptation of the breeding system after range expansion (Barrett 1992); pollinator behavior may also vary geographically as well. These aspects of reproduction, therefore, should not be regarded as fixed characteristics within weed species (Barrett 1992).

The spatial distribution of weeds may strongly affect weed-crop hybridization. First, many weeds have a highly patchy distribution within fields, and recent work suggests that patches in some species have some degree of temporal stability (Walter 1996). Patches may result from edaphic factors or from persistent effects of high seed production. Within fields, patchy weed distribution may mitigate against weed-crop hybridization if weeds occur in patches of sufficient density that the proportion of weed individuals on the periphery of patches is small, thus limiting the population rate of hybridization. More homogenous and sparser spatial distributions may favor considerably higher rates of outcrossing, because isolated individuals may experience much higher local abundance of crop pollen or because of changes in pollinator movement as a function of local density. Alternatively, weed density may have the opposite effect on hybridization rates when the crop serves as female parent. In this case, high-density patches may promote hybridization by virtue of attaining high local densities of weed pollen, and homogenous weed density may reduce hybridization.

The distribution of weeds in the broader landscape around field crop agroecosystems also has potential importance in modulating rates of weed-crop hybridization. If conditions permit weed establishment in non-cropped areas in this landscape, then many small isolated populations may exist (Wilkinson et al. 1995). Weed-crop hybridization may occur at higher rates in these populations than in field populations for a variety of possible reasons. For example, due to the plasticity of reproduction in many weed species, flowering can occur over a broad time period during the growth season for a given species. Commercial-scale fields of both wind and insect pollinated crops have been shown to disperse pollen for more than 1 km beyond field boundaries (Wilkinson et al. 1995). Therefore, crop pollen can be expected to reach non-field weed populations in agricultural landscapes within this distance. Thus, when weed populations are considered on a landscape scale, extensive plasticity of flowering time and availability of crop pollen across the landscape may markedly extend the range of opportunities for hybridization in many weed-crop systems.

Finally, weed abundance is highly variable. In certain years, weather factors can lead to weed control failures over extensive regions, producing very high weed densities in some fields. Weed density also varies on a regional basis due to interactions between weed biology and regionally variable weed management practices and other cropping system factors. Both forms of variation may result in substantial increases in the absolute number of hybridization events. Increased rates of hybridization may result in cases in which the rate is affected by density-dependent variation in pollinator behavior or spatial distribution. For example, increased local abundance of a species may allow it to colonize marginal habitats in agricultural landscapes that are not occupied at lower densities, perhaps increasing probability of hybridization. Thus hybridization rates may fluctuate considerably over years and over portions of a weed species range.

The few-but-strong selective agents seemingly result from the biological simplification that appears typical of contemporary high-input field crop agroecosystems. Apparently, although data are lacking, weed populations in these ecosystems are often limited by only a few management practices or natural enemies. The implication is that introgression of genes that improve weed adaptation to these predominant selective agents can dramatically increase the average fitness of a weed population. Moreover, tradeoffs among adaptations to different limiting factors (e.g., competitors vs. herbivores) resulting from introgression of a single gene may also be minimal. A prime example is the evolution of herbicide resistance in weeds. The advent of herbicide resistance often dramatically increases average fitness and growth rate of weed populations. Moreover, herbicide resistant mutations can have high absolute fitness despite major functional impairments that result from pleiotropic effects of resistance mutations. This example illustrates how selection can favor mutants that overcome key limiting factors despite performance tradeoffs. Another line of evidence for this notion comes from the multiple examples of increased distribution and abundance of weeds resulting from acquisition of a crop trait via hybridization (Barrett 1983). Finally, the many cases of major increases in distribution and abundance of certain weeds following modest changes in cropping systems (e.g., herbicide or fertility regimes) provide additional evidence that many weed populations are regulated by a few powerful factors (Froud-Williams 1998).

If accurate, this conjecture suggests that post-hybridization adaptation of weeds bearing escaped transgenes is greatly facilitated by the biological uniformity of current field crop ecosystems. Weeds may require relatively little evolutionary ‘refinement,’ such as breaking of linkages to disadvantageous crop traits, in order to acquire adaptation to large areas (Adam and Kohler 1996). One suggested criterion for assessing spread of transgenes into weed populations is that the fitness of the weed-crop hybrid bearing a transgene should be greater than the fitness of non-hybrid weeds. This criterion may more easily be met in contemporary field crop ecosystems than in most other ecosystems. Thus, transgene escape may be a rapid process compared to what it would be if cropping systems were, in effect, less forgiving of maladaptive features. Even quite poorly fit hybrids and early backcrosses may persist in agroecosystems at sufficient densities to allow opportunities for introgression and adaptive refinement, provided that these forms have a key adaptation that facilitates their persistence. These considerations may apply most strongly to escape of transgenes affecting tolerance to abiotic factors (e.g., herbicides or drought). However, it is possible that over extensive areas weed populations are limited by a single biotic factor to which adaptation would confer major increases in fitness. For field crop weeds, virtually nothing is known about this possibility.

A distinctly different mechanism by which escaped transgenes can affect weed adaptation is via increased fitness in weed populations in non-cropped parts of agricultural landscapes. Weed populations in areas such as field margins or roadsides may be subject to a different range of selective pressures than weeds in cropped fields. For example, seed predation rates may be markedly higher in field-margin habitats, while selective factors affecting fitness in cultivated fields may be absent. Therefore, these non-cropland populations may offer refugia from certain selective factors such as seed germination behavior during early generations after hybridization. Also, introgression of transgenes into non-field populations may allow adaptation to unrecognized biotic population-regulating factors, such as herbivores, pathogens, and seed predators that are not active in field populations. This mechanism is speculative, since the role of non-field populations in the dynamics and evolution of weed populations in agricultural landscapes is currently unknown. Recent simulations (Blumenthal and Jordan, unpubl.) suggest that populations of perennial weeds along field margins can sometimes be important to maintaining field populations.

Another weed ecology feature likely to affect adaptation of crop-weed hybrids is the frequency of episodes of low effective population size due to small census sizes and high levels of selfing (Barrett 1992), particularly in the process of colonization. Low population sizes cause random changes in genetic composition through genetic drift and founder effects. These mechanisms can act on the genetic novelty produced by hybridization, producing a range of genetically differentiated small populations from a genetically diverse early-backcross weed population.

The implication here is that adaptation in weed populations containing escaped transgenes is likely to be affected by both selection and random genetic change. When both factors are present, evolutionary events can occur that would not occur when selection is the dominant evolutionary mechanism. Specifically, the adaptive effects of combinations of transgenes, other crop genes, and weed genes can be much more thoroughly "explored" by the joint action of random genetic change and selection than by selection acting alone (Wade and Goodnight 1998). This mechanism can be particularly forceful when weed populations experience high levels of extinction and recolonization, thus forming ecological and genetic metapopulations. Although it is not yet empirically clear whether agricultural weeds have a metapopulation structure, the occurrence of such structure, in combination with small population size and varied selection pressures, creates favorable conditions for the plausible operation of the shifting balance process. However, the action of this process may itself be unpredictable due to geographical variation in population structure in some weed species due to breeding system, local adaptation after colonization, time since colonization, and hybridization with related taxa (Barrett 1992).

A final dimension of weed ecology relevant to adaptation after hybridization is seed ecology. Ability to maintain persistent seed or propagule populations in soil, along with efficient dispersal and ability to rapidly and efficiently use available resources for reproduction, are the apparent hallmarks of successful weeds of field crop agroecosystems (Ghersa and Roush 1993). Many of the most intractable weed species are so because of the ecology of their seeds. Simulations of weed population dynamics show that seed demography (e.g., seed survival and germination rates) strongly affects weed population growth rates (Colbach and Debaeke 1998). Weed seeds vary substantially among species in longevity. Many soil management factors affect seedbank demography by preventing germination, evoking fatal germination, or by otherwise increasing seed mortality rates. These factors include use of cover crops, conservation tillage, and residue burning; they may have direct effects on seeds, or indirect effects via effects on seed predators and pathogens. Thus, the germination/dormancy behavior of weed seeds is a critical determinant of their survival rates in a given cropping system. The importance of seed ecology to weediness suggests that if transgenes affect seed ecology, these effects are likely to be a primary determinant of their fitness (Landbo and Jorgensen 1997). Similarly, maladaptive effects on seed ecology may be a major mechanism by which non-transgenes from crops hinder adaptation after hybridization.

There are several other aspects of weed seed ecology that appear relevant to the adaptation stage. First is the well-known effect of dormancy whereby weed genotypes, produced by plants growing under past environmental circumstances, can again be selected for despite intervening periods of unsuitable conditions. Thus, seed populations augment the genetic variability of weed populations. Also, seed populations, as a form of temporal dispersal, allow weed genotypes to be tested over a wider range of conditions than would otherwise be possible. This effect may significantly increase the opportunity for a weed carrying an escaped transgene to encounter conditions to which it is adapted.

Effective spatial dispersal of seeds is considered a primary attribute of a successful weed (Ghersa and Roush 1993), and the dispersal ecology of weeds is expected to affect the fate of escaped transgenes in a number of ways. On a field scale, simulation modeling indicates that high rates of weed seed dispersal generally greatly increase weed populations (Perry and Gonzalez-Andujar 1993).

For most weeds of field crops, dispersal is determined by the interaction of weed attributes and human activities, such as contaminated crop seeds (Ghersa and Roush 1993), equipment such as combine harvesters, irrigation water, livestock, and trucking of grain. Management actions in agroecosystems can affect weed dispersal, perhaps regulating weed populations in some cases. When human activities serve as principal weed seed dispersal vectors, dispersal distances are difficult to characterize. They are strongly affected by the particular dispersal vector and geographically variable due to variations in cropping system factors. As a result, the extremes of the dispersal distance distribution are poorly known in most cases.

On a broader scale, many cases of rapid, sub-continental scale dispersal of weed species are known. Weed species have been observed to disperse and become abundant over large regions of the western US (Mack 1986). Due to cropping system changes that promote its abundance, jointed goatgrass (Aegilops cylindrica), a sexually compatible weed of wheat, substantially expanded its range in Utah during a period of eight years. Expansion over hundreds of kilometers of roadsides in less than a decade has been observed in herbicide resistant weeds. These observations suggest that roadside and other non-field weed populations may be important to weed range expansion, again suggesting the importance of weed ecology across agricultural landscapes in the escape and dispersal of transgenes.

Finally, because of the apparent biological simplification of current field crop agroecosystems, a weed may gain markedly higher fitness across a large spatial domain from an escaped transgene. In theory, the resulting spatial homogeneity of favorable habitat (Tomiuk and Loeschcke 1993) and the absence of a need for local adaptation of colonizing populations promote rapid range expansion by colonizing organisms. Therefore, the ecology of weed dispersal and population regulation in current agroecosystems and agricultural landscapes appear to permit large and rapid range expansions of adapted weeds.

Weed attributes affecting dispersal (e.g., seed size, shape, resemblance to crop seed, etc.) should be regarded as adaptive traits that are probably subject to strong selection. As for seed ecology, any effects of transgenes on dispersal ecology are likely to be primary determinants of the fitness effects of those transgenes, and effects of crop genes on dispersal ecology may cause major fitness costs in hybrids. Weed dispersal may also have an evolutionary role, as mentioned above. Dispersal of small founding populations can trigger adaptive processes in these small populations that would not occur in larger populations. Also, in the shifting balance process, dispersal has a critical role in distributing evolutionary products of events in small populations across the landscape and in triggering change in other populations.

There are several summary points to emphasize. First, the ecology of weeds in contemporary cropping systems may facilitate transgene escape by permitting survival of weed-crop hybrids that are maladapted, relative to "wild-type" weeds, in a variety of fitness components. This most likely occurs when the hybrids and subsequent backcross progeny carry a transgene of sufficient adaptive value. Second, seed and dispersal ecology are major determinants of weed fitness and population growth rate, although this is not widely appreciated as such among non-weed scientists. Effects of crop-derived transgenes and all other crop genes on these traits will strongly affect the adaptation of weed-crop hybrids and backcross progeny. Third, most major weed species show extensive spatial and temporal variation in reproductive, seed, and dispersal ecology on several scales. This variation has both genetic and environmental causes and may strongly affect processes involved in all phases of transgene escape. Finally, a landscape perspective may be important for proper assessment of prospects for transgene escape. Populations of agricultural weeds are distributed across agricultural landscapes, including many populations that occur outside of cropped fields. Particularities of the ecology of these populations may affect all three phases of transgene escape.

¹ Paper presented at the "Workshop on Ecological Effects of Pest Resistance Genes in Managed Ecosystems," in Bethesda, MD, January 31 - February 3, 1999. Sponsored by Information Systems for Biotechnology.