As described in Sects. 2 and 3, glyphosate resistance can be mediated by a variety of mechanisms. These resistance mechanisms arise as a result of changes to one or more locations in the genome, resulting in structural or regulatory changes to gene products. Genetic changes that are beneficial (e.g., confer reduced sensitivity to glyphosate) are selected and increase in frequency in the selected populations. The source of the genetic differences that can be selected include standing genetic variation (i.e., they already exist in the population before the onset of selection), immigration from a different population or species, or new mutations. As is the case with resistance to other herbicides, the relative contribution of these sources for glyphosate-resistance evolution are largely unknown (Casale et al. 2019). Ultimately, a better understanding of the evolution of herbicide resistance could lead to novel strategies to mitigate it (Neve et al. 2009).

Naturally occurring plant tolerance cases to a given chemistry may provide insight into what mechanisms may evolve in the future. Several plant species have exhibited a natural tolerance to glyphosate, although the underlying mechanisms have been investigated in few cases. In both Convolvulus arvensis and lilyturf species, gene copy number has been attributed to at least part of the observed tolerance phenotype (Mao et al. 2016; Huang et al. 2019). However, these tolerance cases are frequently due to a combination of mechanisms. In lilyturf species, for example, EPSPS structural differences were also noted, relative to other plant EPSPS enzymes, due to multiple amino acid substitutions and deletions Both modeling and in vitro enzyme assays indicated that these structural differences resulted in reduced glyphosate sensitivity (Mao et al. 2016). In C. arvensis, a promoter-mediated overexpression, associated with glyphosate application, was also observed in addition to increased EPSPS copy number (Huang et al. 2019). Reduced glyphosate translocation was associated with increased tolerance in Ipomoea lacunosa (Ribeiro et al. 2015), whereas increased glyphosate metabolism is hypothesized to confer tolerance in I. purpurea (Van Etten et al. 2020). While there does not appear to be an overarching trend in tolerance mechanisms among species, similar mechanisms are observed across plant tolerance and resistance. The structure-based tolerance of lilyturf could be considered analogous to target-site resistance. Promoter-mediated overexpression, resulting in an increase in EPSPS protein abundance, also has been occasionally associated with evolved resistance (Baerson et al. 2002a). Gene copy number increase and reduced translocation both have been implicated in both tolerance and evolved resistance. The investigation of tolerance mechanisms to a given chemistry, even beyond the scope of glyphosate, can provide insight into what mechanisms might evolve in response to selection.

There have been a couple of cases in which a weed evolved glyphosate resistance via gene flow from a related species. In one case, the weed Brassica rapa acquired the transgene (CP4 EPSPS) conferring glyphosate resistance from cultivated rape (Warwick et al. 2003). This evolutionary path to glyphosate resistance in B. rapa subsequently has been shown to be a common event and has occurred in multiple countries (Simard et al. 2006; Pandolfo et al. 2018). Another case involves weed-to-weed gene flow, in which A. spinosus acquired EPSPS gene duplication that had evolved in A. palmeri (Nandula et al. 2014). These cases are the exception to the norm, however, and most weed species have evolved glyphosate resistance from either standing genetic variation or new mutations.

In general, nontarget-site glyphosate resistance mechanisms are still poorly understood, and even less is known about their evolutionary origins. In regard to enhanced detoxification, because glyphosate is metabolized readily through multiple pathways in bacteria, horizontal gene transfer could certainly be a source of resistance, though no evidence exists for this having occurred. As discussed above, AKR likely plays a role in glyphosate resistance in E. colona, perhaps via enhanced expression, and remains the only plant protein proven to directly metabolize glyphosate (Duke 2019). The evolutionary origin of enhanced expression of AKR in E. colona, or of any other herbicide-metabolizing enzyme selected in weed populations, remains unknown. Now that AKR has been identified to metabolize glyphosate, evaluation of homologous genes in other weed species likely will follow and should reveal the potential of AKR to confer glyphosate resistance in other species.

Because inheritance studies have not yet been published regarding the rapid-response glyphosate-resistance mechanism, its genetic complexity is not known. Additionally, though similarities exist with a recently identified resistance mechanism to 2,4-D (Queiroz et al. 2020), it is unclear if these rapid response mechanisms have any evolutionary relatedness. The similarities with plant pathogen response (e.g., hypersensitivity and rapid cell death) suggest this mechanism evolved by somehow co-opting a pathway for plant defense against abiotic attack (Roden and Ingle 2009).

As discussed above, glyphosate resistance due to vacuolar sequestration might be mediated by an ABC transporter and has been most studied in C. canadensis. Just as enhanced herbicide metabolism can evolve through increased expression of a herbicide-metabolizing enzyme, sequestration could evolve through increased expression of an ABC transporter. A previous study found glyphosate resistance in C. canadensis to be mediated by a single gene (Zelaya et al. 2004), although the identity of that gene is unknown. Increased expression of both EPSPS and ABC transporters in a glyphosate-resistant C. canadensis biotype prompted Margaritopoulou et al. (2018) to investigate methylation of the EPSPS gene. Their finding of differential EPSPS methylation between resistant and sensitive biotypes suggests epigenetic changes could be playing an evolutionary role. The contribution of epigenetic changes to herbicide-resistance evolution in general, not just specifically to glyphosate, remains an unanswered question (Markus et al. 2018).

Nontarget-site herbicide resistance offers the field of weed science many novel research questions to be answered through a variety of omics-based approaches (Maroli et al. 2018; Patterson et al. 2019a). The recent establishment of an International Weed Genomics Consortium promises the development of reference genome assemblies for many of the world’s most problematic weeds (Ravet et al. 2018). This effort will supplement other recent but less coordinated efforts to produce genomic resources for driver weed species, including L. multiflorum (Copetti et al. 2019), A. tuberculatus (Kreiner et al. 2019), B. scoparia (Patterson et al. 2019b), and C. canadensis (Laforest et al. 2020). The availability of these genomic resources enables genetic mapping of traits such as glyphosate resistance (Korte and Farlow 2013; Van Etten et al. 2020) and will complement previous transcriptomic studies designed to identify candidate genes that may be involved in herbicide resistance (Piasecki et al. 2019). The identification of genomic regions associated with the trait of interest, via a genetic mapping experiment, allows for the filtering of candidate genes identified via expression- or variant-based transcriptomic analyses and hedges against the possibility that the trait is ultimately controlled by some regulatory element located far from the genes that would be identified through expression-based transcriptomic approaches. These filtered candidates should be judged, based on physiological characteristics of the trait, and functionally validated via loss- or gain-of-function experiments (Sauka-Spengler and Barembaum 2008; Housden et al. 2017). Pan et al. (2019) provide a good model for functional validation of a glyphosate-resistance gene (AKR), but additional genetic study may have identified the second locus (EPSPS, see Sect. 2.3) contributing to glyphosate resistance. With the identification of the genes involved in nontarget-site glyphosate resistance, researchers will be able to better understand the evolutionary origins of such resistance and predict how likely it is that other species will evolve similar resistance mechanisms in the future.

Because of the novelty and importance of EPSPS gene duplication as a resistance mechanism, its evolutionary origin is of great interest and has been addressed in several studies (Patterson et al. 2018). Except for the case of A. spinosus, wherein the EPSPS amplicon from A. palmeri introgressed into the A. spinosus population after a hybridization event (Nandula et al. 2014), EPSPS gene duplication evolved independently in each of these species. Accordingly, the mechanism of duplication and the length and content of the EPSPS amplicon varies across the different species. For two species with relatively low EPSPS copy numbers, A. tuberculatus and B. scoparia, cytogenomic analysis using fluorescent in situ hybridization (FISH) has shown the duplicated EPSPS genes are arranged as tandem repeats along one chromosome pair. In B. scoparia, these tandem repeats of EPSPS occurred at the distal end of one pair of homologous chromosomes, with approximately 40–70 kb between EPSPS genes and one copy inverted compared to the rest (Jugulam et al. 2014). The tandem arrangement of the EPSPS genes and their location in the telomeres suggests an unequal recombination-based mechanism of gene duplication since unequal crossing over occurs most frequently in telomeric regions of the chromosome and leads to tandem duplications. Similarly, in A. tuberculatus, the EPSPS repeats were found to occur at a single locus in one set of homologous chromosomes, but unlike in B. scoparia, these repeats were in the pericentromeric region of the chromosome, where recombination is less likely to occur (Dillon et al. 2017). Whether the mechanism of gene duplication in this species is also unequal recombination or some other form of chromosomal rearrangement or segmental duplication is unknown.

To further complicate the story, some A. tuberculatus individuals with higher EPSPS copy numbers (>15 copies) showed multiple EPSPS signals on an additional small chromosome (Dillon et al. 2017). Further cytogenomic work found this extra chromosome to be a ring chromosome that was derived from the pericentromeric region of the chromosome with multiple EPSPS gene duplications (Koo et al. 2018a). FISH assays of F1 progeny showed variation in the size and EPSPS copy number of these ring chromosomes across different individuals and, surprisingly, additional EPSPS gene copies on other pairs of chromosomes, indicating reintegration of the ring chromosomes into the linear chromosomes through ectopic recombination (Koo et al. 2018a). The hypothesized model of ring chromosome formation includes breakage of the linear chromosome at two spots flanking the original EPSPS gene duplicates (perhaps via aneuploidy-triggered destabilization), followed by fusion of the broken chromosome ends into a shortened linear chromosome. The excised middle region containing one or more EPSPS genes then undergoes fusion of its proximal ends to form a ring chromosome, that may then form varying sizes of ring chromosomes via a breakage-fusion-bridge cycle model (Koo et al. 2018a). Work looking into the EPSPS gene duplication mechanism in A. palmeri has found similar results, with the additional EPSPS gene copies occurring on extrachromosomal DNA. In the initial report of gene duplication in this species, a FISH image showed EPSPS gene signals distributed across all 34 chromosomes of A. palmeri (Gaines et al. 2010), but a later study (Koo et al. 2018b) showed these gene signals were not actually on the linear chromosomes but were located on extrachromosomal circular DNA (eccDNA) tethered to the main chromosomes. Inheritance of these eccDNA molecules was highly variable and displayed unequal mitotic segregation, illustrating the need for glyphosate selection for retention of glyphosate-resistant plants with high numbers of EPSPS copies. Further work has highlighted that these eccDNA molecules are highly structured with 59 genes, 41 of which are expressed under glyphosate application, and a complex array of mobile genetic elements, repeat sequences, and clustered palindromes (Molin et al. 2017, 2020). The contribution of these additional genes/sequences to the overall resistance phenotype is unknown. Syntenic analysis using genomic assembly of closely related species (Amaranthus hypochondriacus and A. tuberculutus) suggested that the eccDNA was built from several regions across the genome, rather than derived from a single locus (Molin et al. 2020). Consequently, some of the genes (in addition to EPSPS) within the eccDNA may have been selected by glyphosate. An alternative hypothesis is that one or more genes in the eccDNA were selected in the evolutionary past by some other plant stress, and EPSPS happened to get captured within the amplicon, priming the species for the later evolution of glyphosate resistance.

In grass species with the EPSPS gene duplication mechanism, some recent publications have begun to shed light on the arrangement and origin of the EPSPS gene copies. In L. perenne ssp. multiflorum, FISH mapping of the EPSPS gene on somatic metaphase chromosomes revealed a similar pattern as that observed in A. palmeri, with EPSPS signals distributed across all chromosomes in plants with high EPSPS gene copy number (Putta 2017). As with A. palmeri, the signals appeared to be on the outer edges of the chromosomes, perhaps indicating a similar mechanism of gene duplication involving circular extrachromosomal DNA tethered to the main chromosomes, but conclusive evidence of this does not yet exist. Conversely, in E. indica, EPSPS gene copies in a resistant individual appeared to be restricted to two pairs of homologous chromosomes, as indicated by FISH work in this species (Chen et al. 2019). In B. diandrus, no FISH assays have yet been published, but inheritance work has shown F2 offspring to have a range (3–30) of EPSPS gene copies, with all F2 offspring showing an increase in the baseline copy number (Malone et al. 2016). If the EPSPS gene copies were inherited as a single locus, as would be expected in a tandem repeat model, 25% of the F2s should have a single EPSPS copy, and the fact that this is not observed indicates these EPSPS gene copies likely occur on multiple chromosomes. For the other three grass species (C. truncate, H. glaucum, and P. annua), no cytogenetic or inheritance work has yet been completed and the mechanism of EPSPS gene duplication is unknown.

Gene duplication as a herbicide-resistance mechanism thus far has been reported in only one other case, resistance to acetyl-CoA-carboxylase inhibitors (Laforest et al. 2017). Why, then, has it repeatedly evolved for glyphosate resistance? As can be seen in Fig. 1, besides multiple amino acid substitutions in EPSPS, gene duplication confers the highest magnitude of resistance among the known resistance mechanisms evolved to date. Perhaps EPSPS duplication is the evolutionary “path of least resistance” for robust glyphosate resistance (Tranel 2017).

Recent population genetics analysis of glyphosate-resistance evolution in A. tuberculatus indicated that EPSPS duplication in this species – which appears to be due primarily to tandem duplications – independently occurred multiple times (Kreiner et al. 2019). In contrast, the EPSPS-containing eccDNA in A. palmeri was nearly identical among geographically dispersed populations, suggesting a single evolutionary origin (Molin et al. 2018). Conservation of the eccDNA among these populations suggests a relatively recent evolutionary event, arguing against the hypothesis mentioned above, that the amplicon was selected by some plant stress prior to glyphosate selection. Kreiner et al. (2019) presented evidence suggesting EPSPS duplication preexisted as standing genetic variation in A. tuberculatus, in contrast to the eccDNA in A. palmeri being a relatively recent event. Certainly, more work is needed, but comparison of these two species suggests that tandem duplication is a higher probability event than the eccDNA-based duplication. Why these two related species used different evolutionary paths to EPSPS duplication is unknown. One possibility is that tandem duplication may not have evolved as a glyphosate-resistance mechanism in A. palmeri because this species is inherently more sensitive than A. tuberculatus to glyphosate. Therefore, A. palmeri needed tens of copies of EPSPS for resistance, which was enabled only after evolution of the EPSPS-containing eccDNA. In fact, if the linear correlation between EPSPS copy number in A. palmeri and resistance magnitude shown in Fig. 2 is extrapolated, resistance would not be observed below 10 copies. As mentioned above, it is also possible that other genes within the eccDNA augment the glyphosate resistance conferred by EPSPS duplication.

The relative contributions of standing genetic variation versus new mutations for target-site resistance likely vary among herbicides. In the case of target-site resistance to glyphosate, repeated occurrence of double mutations and the occurrence of a triple mutation (discussed in Sect. 3.1) present additional evolutionary questions. These multiple-mutation alleles could preexist in a population as part of the standing genetic variation, or the multiple mutations could arise sequentially during the course of herbicide selection. In addition, the spontaneous occurrence of a double or triple-mutation allele (e.g., both or all three of the mutations occurring in a single generation) is formally possible, but the probability is so low that this route probably can be considered inconsequential (Ossowski et al. 2010). Sequential evolution could occur by a second mutation occurring in an allele that already has one mutation, or via recombination between two alleles each carrying one of the two mutations. Given the close proximity of the double and triple mutation sites in the gene, however, recombination between them will be exceedingly rare. Therefore, the two most likely evolutionary paths to the multiple-mutation alleles are either they existed prior to selection or a single-mutation allele increased in frequency as a result of herbicide selection, and then acquired one or more additional mutations.

If a multiple-mutation allele preexisted in the population, then one would expect it to have a limited fitness cost, because a large fitness cost would result in it having been purged from the population. From limited studies to date on fitness costs of multiple-mutation EPSPS alleles, however, at least some seem to have significant fitness costs (see Sect. 5.4). Additionally, if a multiple-mutation allele preexisted, one would expect to find this allele in essentially all resistant plants, i.e., occurrence of alleles containing only one of the mutations would be rare (since they would only come about via recombination or a mutation back to wild type). In an E. indica population with the Thr-102-Ile + Pro-106-Ser double mutation, both the double mutant and the single mutant Pro-106-Ser, but not the single mutant Thr-102-Ile, allele were found at high frequencies, leading the authors to conclude that the two mutations evolved sequentially (Yu et al. 2015).

In the cases of multiple-mutation EPSPS alleles in Bidens subalternans (double mutant) and A. hybridus (triple mutant), however, only the multiple-mutant alleles were observed (Perotti et al. 2019; Takano et al. 2020), which is consistent with the alleles preexisting in the population. Furthermore, because B. subalternans is tetraploid, it was suggested that fitness cost of the double-mutation allele could be masked by the second, wild type EPSPS gene (Takano et al. 2020), which could explain how such an allele persisted in the population prior to glyphosate selection. Because the multiple-mutation alleles confer higher resistance than the single-mutation alleles, there are caveats with the expectation that lack of finding the single-mutation alleles is evidence of the multiple-mutation alleles preexisting in the population. For example, with repeated selection of glyphosate, especially with high doses, the multiple-mutation alleles will be favored over the single-mutation alleles and, therefore, the single-mutation alleles will be purged over time. Thus, one must consider the glyphosate selection timeframe. In addition, if the multiple-mutation allele arose sequentially in one population, but then migrated to a second population, analysis of the second population would incorrectly lead to support of the hypothesis that the multi-mutation allele preexisted.

In summary, there is good evidence that multiple-mutation EPSPS alleles evolved from sequential events in at least some cases. More evidence is needed, however, to conclude that glyphosate resistance also has evolved via selection of multiple-mutation EPSPS alleles that preexisted as part of the standing genetic variation of a population.

In many organisms, the evolutionary adaptation to a new environment or to a new selection pressure is often accompanied by tradeoffs that can affect the general fitness of the organism, commonly referred to as fitness cost (Purrington 2000; Strauss et al. 2002; Vila-Aiub 2019). The presence of resistance alleles in a biotype can cause pleiotropic effects that will enhance some negative phenotypes, such as lower number and viability of seeds, less biomass, and less attraction to pollinators. All of these effects can prevent the fixation of resistance alleles, making the adaptation process occur slower (Tian et al. 2003; Vila-Aiub 2019). On the other hand, studies have also shown that, in some cases, no fitness cost was observed due to the presence of herbicide-resistance alleles (Vila-Aiub 2019). Understanding fitness costs related to the presence of herbicide resistance traits is important to understand the evolution patterns that these traits will follow (Cousens and Fournier-Level 2018).

Studies to investigate fitness cost due to glyphosate resistance have shown different results according to the mechanism of resistance involved. In the case of target-site glyphosate resistance, there is generally a correlation between higher levels of resistance and greater fitness costs (Vila-Aiub et al. 2019). For example, substitution of two amino acids in EPSPS in E. indica was accompanied by a high fitness cost, whereas a single mutation in the same species—which provided lower resistance—conferred a negligible fitness cost (Yu et al. 2015; Han et al. 2017). Fitness studies of EPSPS gene duplication generally have identified little if any fitness costs, although costs may be higher in certain genetic backgrounds (Giacomini et al. 2014; Vila-Aiub et al. 2014; Martin et al. 2017; Osipitan and Dille 2019). That EPSPS duplication does not confer a large fitness penalty is particularly surprising in A. palmeri, given both the large number of copies in resistant plants and the size of the amplicon (Vila-Aiub et al. 2019). The EPSPS amplicon in A. tuberculatus also appears quite large (Kreiner et al. 2019) but, nevertheless, only modestly decreased in frequency in a multi-generational fitness study (Wu et al. 2017). Vila-Aiub (2019) provides a recent and more comprehensive review of fitness costs associated with glyphosate resistance. When considering fitness costs of herbicide-resistance mechanisms, it is important to keep in mind that those mechanisms that confer extremely high fitness penalties are unlikely to be selected. Consequently, our vantage point is skewed by studying only those mechanisms that have evolved in weed populations.