There can be no clearer demonstration of the evolutionary potential of weeds than the rapid and widespread evolution of resistance to herbicides (Powles & Shaner, 2001). The propensity for evolution of resistance varies, with some species and herbicides being more prone to resistance than others (Heap & LeBaron, 2001). In the most extreme cases, resistance has evolved following exposure of no more than three or four generations of a weed population to a herbicide (Powles & Holtum, 1994). Herbicide resistance is arguably the single largest global weed management issue and studies concerned with herbicide resistance are at the forefront of current weed science research. Given this, it seems logical that evolutionary biology should play a central role in informing solutions to this escalating problem, yet, conversely, it is our view that herbicide resistance research most starkly highlights the lack of evolutionary thinking in weed science.
The majority of herbicide resistance research is conducted retrospectively. A suspected resistant population is reported, seed is collected from surviving plants in the field and the dose–response curve of the suspected resistant and a known susceptible population are compared under controlled glasshouse or field conditions. Following confirmation of resistance, further physiological, genetic and molecular characterization is conducted to diagnose the resistance mechanism. These studies are important for characterizing new mechanisms of resistance, but endless descriptions of the same mechanism in a different species or from a different cropping system provide rapidly diminishing returns in terms of their ability to better inform resistance management (Cousens, 1999; Neve, 2007). Indeed, it seems that weed researchers have become overly concerned with describing the outcome of resistance evolution to the detriment of studies that seek to better understand the process of selection for resistance. We believe this is a reflection of the alignment of weed science with crop science and physiology rather than the disciplines of plant ecology and evolution. It also represents a missed opportunity for herbicide resistance research to combine applied management advice with fundamental insight into evolutionary ecology as has been the case in insecticide resistance studies (Lenormand et al., 1999; Tabashnik et al., 2004).
The evolutionary dynamics of selection for herbicide resistance
Studies that focus solely on characterizing the outcome of resistance evolution may prejudice assumptions about the process of selection. For example, the ultimate fixation of a single major resistance allele with no fitness cost (Coustau et al., 2000), does not preclude the possibility that many other minor alleles were also initially selected or that an initial cost of resistance was compensated during the course of selection (Andersson, 2003; Wijngaarden et al., 2005). Evolution of herbicide resistance is a stochastic process and resistance management strategies attempt to ‘load the dice’ in favour of herbicide susceptibility. It is likely that the key steps towards evolution of resistance occur during the early stages of selection, long before field resistance is apparent, and that following this initial selection, resistance becomes an inevitable or deterministic consequence of further exposure to herbicides. Greater knowledge and understanding of genetic variation for herbicide susceptibility in weed populations, of fitness costs and trade-offs associated with this variation and of population genetic processes during the early stages of selection for resistance should be incorporated into simulation models, and will, we argue, greatly improve resistance management. Key to this understanding will be a greater appreciation of the relative contributions of spontaneous mutation and standing genetic variation to the evolution of resistance (Lande, 1983; Orr, 1998; Hermisson & Pennings, 2005). In the following text, we consider this question in relation to the impact of herbicide dose on potential for evolution of resistance.
The potential for reduced herbicide application rates to accelerate evolution of resistance has been keenly debated (Gressel, 2002; Beckie & Kirkland, 2003; Neve, 2007) and has practical significance given economic and environmental incentives to reduce herbicide application rates. Low doses of the acetyl-coenzyme A carboxylase (ACCase) inhibiting herbicide diclofop-methyl have been shown to rapidly select for resistance to very much higher doses via the selection and reassortment of minor genes in L. rigidum, an outcrossing species (Neve & Powles, 2005a). This phenomenon has also been demonstrated for low-dose selection with glyphosate in L. rigidum, though the response to selection was less marked (Busi & Powles, 2009). These results suggest a high degree of additive genetic variation for herbicide susceptibility in a weed population never previously exposed to herbicides. High herbicide doses during the initial stages of selection would have prevented selection and reassortment of minor genes into highly resistant phenotypes. Even accepting that the majority of field-evolved herbicide resistance is endowed by single major genes, it is possible that initial selection at low doses is for putative minor genes, resulting in reduced herbicide efficacy, larger population sizes and an ultimately higher probability of subsequent selection for major gene resistance. The ‘low dose’ question also highlights the importance of understanding the process, rather than simply the outcome of selection for resistance.
Evolutionary biology, population genetics and physiology all suggest that evolved resistance to novel pesticides will be associated with a fitness cost (Coustau et al., 2000). These costs may be environment-specific (Plowman et al., 1999; Salzmann et al., 2008) and they may only be manifest at certain life-history stages (Roux et al., 2005; Vila-Aiub et al., 2005). Knowledge of the extent of these costs and of their environment-specific and life-history-specific attributes may be crucial for designing ‘biorational management tactics’ which could turn the costs and idiosyncrasies associated with resistance into valuable tools in resistance management (Jordan et al., 1999). There have been some excellent studies of herbicide resistance fitness costs. However, in many other cases, the concept of fitness as it relates to herbicide resistance has been poorly understood and many published studies have used wholly inappropriate methods to quantify fitness costs. Many studies have compared resistant (R) and susceptible (S) populations with completely different genetic backgrounds. Numerous studies have also mistakenly made the assumption that comparative growth rate alone is a proxy for fitness. Perhaps more than in any other case, these widespread and repeated faults in fitness studies highlight the application in weed science of methods from crop breeding and physiology rather than from ecology and evolution.
Some fitness studies have used isogenic (R) and (S) lines to demonstrate fitness costs associated with triazine resistance in standardized genetic backgrounds (Gressel & Bensinai, 1985; McCloskey & Holt, 1990; Arntz et al., 2000; Salzmann et al., 2008). While accepting that isogenic lines are the gold standard for unequivocally demonstrating fitness costs, we suggest that future research should also compare fitness between plants arising from controlled crosses of R and S plants (Menchari et al., 2008) or where plant cloning techniques have enabled the identification and propagation of discrete R and S phenotypes from single populations (Vila-Aiub et al., 2005; Pedersen et al., 2007). In this way, fitness of R alleles can be compared in a broader range of genetic backgrounds, reflecting more closely the situation in natural populations. Wherever possible, fitness studies that have proper control of genetic background should also report the molecular genetic basis of resistance, measure fitness and fitness components at a range of life-history stages, under competitive conditions and in a range of environments.
As fitness is directly related to the average contribution of an allele or genotype to future generations, the evolution of R allele frequency in pesticide-treated and untreated populations may provide a better estimate of fitness cost than those based on direct measures of fitness-related traits. Using migration-selection models developed to estimate migration rates and selection coefficients in clines, Lenormand et al. (1999) and Roux et al. (2006) empirically showed that studying R allele frequency along a transect of pesticide treated and untreated areas gave more precise, and sometimes contrasting estimates of fitness costs than estimates based solely on fitness-related traits. We argue that in future, the most accurate estimates of fitness costs will be obtained by measuring changes in R allele frequencies in studies such as those described earlier.
Models and model organisms in herbicide resistance research
It is inherently difficult to design and perform experiments that study the dynamics of herbicide resistance evolution in weed populations. To be informative, these experiments must select for resistance at realistic spatial and temporal scales, so that herbicides are applied to millions of individuals over multiple generations. Some studies have sought to explore the efficacy of weed and resistance management strategies on small field plots (Westra et al., 2008), but weed populations are too small to represent the full range of genetic variation on which selection acts at the agronomic scale. Other studies have attempted to overcome this constraint by sowing weed populations with a low frequency of herbicide resistance into small field plots (Beckie & Kirkland, 2003; Moss et al., 2007). However, this approach has limited application as it examines the effectiveness of proactive resistance management strategies against populations that are already resistant.
Model organisms and mathematical models that simulate evolution of resistance may each have features that overcome some of the difficulties described above, although for some purposes their relevance to the field may be questioned. Simulation models (Maxwell et al., 1990; Diggle et al., 2003; Jacquemin et al., 2008) may be relatively inexpensive to develop and enable rapid comparisons of resistance management strategies over many generations. These models may be used solely to explore the relative importance of parameters that underpin resistance evolution or to address very specific cropping system-related questions (Neve et al., 2003). However, in some cases, a lack of understanding of key model parameters such as the fitness costs associated with R alleles, the extent of standing genetic variation for herbicide resistance and gene flow between metapopulations is hampering further model development and application. As these parameters become available, new models incorporating quantitative genetics, demographics and metapopulation dynamics can begin to explore some of the important questions discussed in the preceding sections and relating to the direct or interacting effects of: the impact of fitness costs on initial R allele frequency before the first herbicide exposure and resistance trajectories; the evolution of fitness costs by compensatory evolution; the relative contribution of major gene and quantitative resistance, and the role of herbicide dose; and the impact of environmental heterogeneity, degree of connectedness among patches and cropping systems on the evolution of herbicide resistance.
Model organisms may be useful in their own right for developing experimental evolutionary approaches (Elena & Lenski, 2003) to study the dynamics of evolution of herbicide resistance. For example, the unicellular chlorophyte, Chlamydomonas reinhardtii reproduces rapidly, and millions of individuals can be cultured in a few millilitres of liquid medium. It is also susceptible to many herbicides (Reboud, 2002) and has been used as a model experimental organism in herbicide resistance research (Reboud et al., 2007). Model organisms, such as Arabidopsis thaliana may also provide valuable insight for important parameters that drive resistance evolution (Jander et al., 2003). A series of studies examining costs associated with herbicide resistance alleles in A. thaliana has provided valuable insights for models of herbicide resistance evolution as well as demonstrating the potential for herbicide resistance to provide fundamental insight into the evolutionary genetics of plant adaptation (Roux et al., 2004, 2005; Roux & Reboud, 2005).