Weed evolution: Genetic differentiation among wild, weedy, and crop radish

Abstract Approximately 200 weed species are responsible for more than 90% of crop losses and these comprise less than one percent of all named plant species, suggesting that there are only a few evolutionary routes that lead to weediness. Agricultural weeds can evolve along three main paths: they can be escaped crops, wild species, or crop‐wild hybrids. We tested these three hypotheses in weedy radish, a weed of small grains and an emerging model for investigating the evolution of agricultural weeds, using 21 CAPS and SSR markers scored on 338 individuals from 34 populations representing all major species and sub‐species in the radish genus Raphanus. To test for adaptation of the weeds to the agricultural environment, we estimated genetic differentiation in flowering time in a series of common garden experiments with over 2,400 individuals from 43 populations (all but one of the genotyped populations plus 10 additional populations). Our findings suggest that the agricultural weed radish R. r. raphanistrum is most genetically similar to native populations of R. r. raphanistrum and is likely not a feral crop or crop hybrid. We also show that weedy radish flowers more rapidly than any other Raphanus population or cultivar, which is consistent with rapid adaptation to the frequent and severe disturbance that characterizes agricultural fields.

There are three potential origin routes for agricultural weeds: crops going feral, wild populations invading fields, or crop-wild hybridization (de Wet, 1966;de Wet & Harlan, 1975;Vigueira et al., 2012); and these differing histories may have detectably different phenotypic and genetic effects. Escaped crops, for example, may already be resistant to herbicides or able to survive in disturbed habitats (Vigueira et al., 2012). These different origins would also leave a distinct genetic signature at neutral markers. Weeds could have a wild origin, with either a wild population that was preadapted, or one that rapidly evolved to weediness. Here we would expect strong genetic resemblance between the weed and one or more native populations (Vigueira et al., 2012). Alternatively, the weeds could be crop-wild hybrids, in which case we would expect the weeds to be a genetic mixture of cultivars and native plants with relatively high genetic diversity compared to the parental crop. Finally, if the weeds are most genetically similar to the cultivars at neutral markers, this would suggest that they descend from escaped crops. As feral crops would suffer from the founder effect of escaping cultivation after having undergone artificial selection, we might also expect weeds derived from crops to be less genetically variable than either wild invaders or hybrids (Vigueira et al., 2012).
Phenotypic adaptations of agricultural weeds have also been relatively neglected, with the exception of adaptations to human control, particularly the evolution of crop mimicry and herbicide resistance (reviewed in Barrett, 1983;Neve et al., 2009;Vigueira et al., 2012). However, weeds must already be well-adapted to agricultural habitats before they are problematic enough for humans to implement control practices. Some attention has been paid to the key adaptations of seed dormancy and shattering in the evolution of agricultural weeds from crop ancestors (Ellstrand et al., 2010;Vigueira et al., 2012), but there has been little work on the evolution of a shortened life-cycle. In an agricultural setting, frequent and regular disturbances from plowing and harvesting likely exert a strong selection on weeds for rapid flowering and seed set (Barrett, 1983;Warwick & Stewart, 2005).
Weedy radish, Raphanus raphanistrum ssp. raphanistrum, is an emerging model for studying both rapid adaptation and weed evolution (Campbell, Snow, & Ridley, 2006;Campbell, Snow, Sweeney, & Ketner, 2009;Conner et al., 2011;Conner, Mills, Koelling, & Karoly 2014;Klinger, Elam, & Ellstrand, 1991;Sahli, Conner, Shaw, Howe, & Lale, 2008;Snow & Campbell, 2005) and is a member of one of the four major weed and crop families (Brassicaceae). Determining the origins of weedy radish is tractable because the genus Raphanus includes only three named species, all of which are self-incompatible. The genus likely originated in the Mediterranean, as native populations exist only there (I. Al-Shehbaz, pers. comm.). R. raphanistrum is divided into two subspecies, with R. r. raphanistrum including native Mediterranean populations as well as the globally distributed weed, and R. r. landra existing only as native Mediterranean populations. The crop R. sativus is divided into four major types-two root crops (European radish and Asian daikon) and two fruit crops (Oilseed and ediblepod Rattail). R. pugioniformis is a little-studied endemic of the eastern Mediterranean (Ziffer-Berger, Hanin, Fogel, Mummenhoff, & Barazani, 2014). The relationships among these species are not well-resolved (Ziffer-Berger et al., 2014), but an analysis of cDNA sequence from one population of each of eight Raphanus taxa (not including R. pugioniformis) provided strong support for the monophyly of the crop cultivars, as well as monophyly of native and weedy R. r. raphanistrum (Shen et al., 2013).
Previous work has shown that eight populations of weedy radish flowered much more quickly than one native R. r. raphanistrum population in a greenhouse common garden (Sahli et al., 2008).
Similarly, five weedy radish populations flowered faster than five root-crop radish cultivars in greenhouse (Hegde, Nason, Clegg, & Ellstrand, 2006) and field  common gardens. However, since flowering time has only been reported for one native population, and the phylogeny in Shen et al. (2013) was based on only a single population of each Raphanus taxon and did not include R. pugioniformis, neither the origin of weedy radish nor whether the weeds have evolved earlier flowering is clear. We estimated genetic differentiation for flowering time and molecular markers for all named species and subspecies in the genus Raphanus, including a total of 15 populations from all three wild taxa from the native range, eight weedy R. r. raphanistrum populations from outside the native range, and 21 crop cultivars from all four groups (Supporting Information Table S1), to address two questions. First, did weedy radish originate from native R. r. raphanistrum as previous work suggests, or instead as an escaped crop or a crop-wild hybrid?
Second, is there evidence that the weeds have evolved more rapid flowering relative to the rest of the genus, again as hypothesized based on previous work?

| Populations
Eleven R. r. raphanistrum populations from the native Mediterranean range were included, with six from the western part of the range (collected in Spain or France) and five eastern (collected in Israel). Only one of these populations (AFFR from southern France) was collected in an agricultural field; all but one of the rest were from human-disturbed nonagricultural habitats (Supporting Information Table S1). We used eight populations from outside the native range; all but one were collected as weeds of agricultural fields in the USA, Europe, and Australia, and the one exception (MAFI from Finland) was collected in an agricultural landscape. Thus, R. r. raphanistrum exists primarily as an agricultural weed outside its native Mediterranean range, and our conclusions about the evolution of the weeds are based on comparing native to non-native populations. Three populations of R. r. landra from Spain and one R. pugioniformis population from Israel were also included, as were twenty-one crop varieties purchased from seed companies (Supporting Information Table S1). The natural populations were collected by a variety of individuals using a variety of methods, including collecting from all fruiting individuals in small populations and collecting from transects or grids in large populations; in all cases, the goal was to sample the genetic variation of each population in an unbiased manner.

| Genotyping
To assay patterns of neutral genetic differentiation among these populations, we used a panel of 13 CAPS (cleaved amplified polymorphic sequences) in addition to the 8 SSR (microsatellite) markers from Sahli et al. (2008). To create the CAPS markers, cDNA sequencing of seven lines of Raphanus populations and cultivars (Moghe et al., 2014) was used to assemble and align line-specific contigs against each other (Supporting Information Table S2). None of the markers are closely linked, with the closest pair being 6 cM apart (Supporting Information Table S2). Genotyping was completed on 10 randomly sampled plants from each of 34 populations for a total of 338 individuals (Supporting Information

| Visualization of marker variation
To complement the AMOVA and F ST analyses, we also performed a principal components analysis to visualize the patterns of marker variation in the genus. SmartPCA (v.13050-from the program Eigensoft 6.0.1) (Patterson, Price, & Reich, 2006) was used to perform an eigen decomposition optimized for genomic data to rotate the data onto a set of orthogonal axes defined by the amount of variation explained.  Figure S3).
As the SmartPCA algorithm is designed for biallelic markers, each SSR marker was expanded into several biallelic markers as described in Patterson et al. (2006), prior to analysis using a custom script.
Experimentally determined linkage groups (http://radish.plantbiology.msu.edu/) were used as a proxy for chromosomes. Markers that could not be assigned to linkage groups were given unique chromosome numbers. The total number of linkage groups (7)

| Flowering time common gardens
To test for genetic differentiation in flowering time, we combined data from eight common garden experiments performed over a period of 11 years and including a total of 2,441 plants (Table 1) Figure S7). Still, we accounted for these aspects of the dataset by analyzing the flowering time data using two mixed models in a Bayesian framework, similar to a modern meta-analysis. shown. This method resulted in two reasonably balanced datasets and provides two estimates for native R. r. raphanistrum. We also provide un-modeled population level estimates of both flowering time and vernalization requirement.
We modeled days to flower as a function of geographic origin to verify the differences in flowering time between native and non-native R. r. raphanistrum populations reported by Sahli et al. (2008), but with multiple native populations. We also tested for differences between native populations from eastern and western Mediterranean. For this analysis, we used a subset of the data that only included R. r. raphanistrum populations (Supporting Information

| Marker data are consistent with the hypothesis that weeds evolved from a native R.r. raphanistrum ancestor
The F ST analyses show that the lowest levels of differentiation across the genus Raphanus are between the non-native and native R. r. raphanistrum (Figure 1). Similarly, the AMOVA hierarchical ϕ ST values were lower when the non-natives were tested against native R. r. raphanistrum compared to the tests against crops or R. r. landra (Table 2)

| Non-native range R. r. raphanistrum flower more rapidly
We found that non-native radish consistently flowers more quickly than any other Raphanus. In model 1, non-natives flowered about 58 days earlier than the Western range R. r. raphanistrum, and about 24 days earlier than the Eastern populations (Figure 3 and (1) (2) Supporting Information Figure S4). The native R. r. raphanistrum population with the fastest flowering time (AFFR) was the only one collected from an active agricultural field, while the slowest flowering (DEES) was the only one collected from an undisturbed habitat (Supporting Information Table S1). R. r. landra populations took even longer to flower. In model 2, they required an additional 76-139 days to flower over the average native R. r. raphanistrum plant, depending on whether vernalization time was included as a fixed effect (Supporting Information Figure S5). Not surprisingly, on average the root crops (daikon and European) flowered more slowly than the crops that are used for their fruits (oilseed and rattail), as flowering causes resources stored in the roots to be reallocated to the fruits.

| D ISCUSS I ON
We assayed a diverse set of populations across the genus Raphanus, to address two interconnected questions concerning the evolution of weedy radish. First, what is the likely origin of the weeds; and second, has weedy radish evolved rapid flowering in comparison with its progenitors?
F I G U R E 1 Pairwise F ST calculated for all 21 markers, and clustered by Euclidean distance. Populations are colored along the axes to match the putative groups from the SmartPCA (Figure 2) analyses; for population codes on the other axes, see Supporting Information Table S1 TA B L E 2 Results of three AMOVAs In all three models, populations were nested in groups, and % variance indicates the amount of variation accounted for by each hierarchical level of that model. Note that there is evidence for population structure at within groups in all three models.

| Origins of radish as an agricultural weed
There are three main pathways to agricultural weediness; feral crops, wild invaders, and hybridization, either wild-wild or wild-crop. All our results are most consistent with the hypothesis that the weeds evolved from a native R. r. raphanistrum ancestor, in agreement with taxonomic designation of the weeds as members of this subspecies.
F ST analyses clearly cluster weedy and native members of this subspecies together, and while the AMOVA found that weedy and native R. r. raphanistrum are significantly differentiated, they are less differentiated than the weeds are from either the crop R. sativus or the other native subspecies R. r. landra (Table 2). This conclusion is in agreement with a previous phylogenetic analysis based on eight transcriptomes, which found that weedy R. r. raphanistrum and native R. r. raphanistrum are sister taxa, and that both are more closely related to the other native Raphanus than to any cultivar (Shen et al., 2013).
While our results are all consistent with native R. r. raphanistrum as the ancestor of the weeds, we cannot rule out introgression from other Raphanus taxa. Introgression is suggested both by the central position of the non-native populations relative to all the other Raphanus groups in the PCA (Figure 2), as well as by the significant genome-wide differentiation between the non-native and native R. r. raphanistrum in the AMOVA ( Table 2). The differentiation between non-native and native could also have been caused by drift early in weed evolution before they spread across the globe; however, the weeds have the highest expected heterozygosities of all the groups tested (Supporting Information Table S8), consistent with introgression and inconsistent with strong effects of drift. Introgression with native R. r. landra could have occurred in the Mediterranean early in the evolution of the weeds, and there are ample current opportunities for weeds and crop radish to hybridize. The California invasive wild radish is the product of weedy and crop hybridization (Hegde et al., 2006;Panetsos & Baker, 1967), and weed-crop hybridization has occurred elsewhere (Snow & Campbell, 2005). As the marker density of our data is relatively low, we would be unable to resolve small scale introgression followed by strong selection for adaptive alleles.
Resolving patterns of past introgression is difficult in general, and strong evidence would require a much larger set of genomic markers.
In cases where the origin of agricultural weeds are known, researchers have frequently found them to be escaped crops (Ellstrand et al., 2010;Vigueira et al., 2012); however, our results do not support this.
The crops are significantly different from the weeds in the AMOVA with a higher ϕ-ST value than the weed-native R. r. raphanistrum comparison (Table 2), and the latter two groups form a distinct cluster separate from the crops. These results are inconsistent with the crop origin theory but are consistent with previous work that found no shared chloroplast haplotypes between crop and weed populations (Ridley, Kim, & Ellstrand, 2008). Additionally, weeds resulting from de-domestication are expected to have very low genetic diversity (Vigueira et al., 2012), but non-natives in our study have the highest expected heterozygosity of any group (Supporting Information Table S8). This does not seem to be due to pooling genetically differentiated populations, as three of the five weed populations we genotyped also have the highest expected heterozygosites that we measured, and the weedy populations clustered together in our genetic analyses. Empirical work also suggests that newly escaped radish cultivars would make poor weeds;  found no evidence that R. sativus could establish feral populations without introgression from weedy radish and were unable to artificially select for greatly reduced flowering time in the Red Silk cultivar. While there are some reports of feral crop radish in the literature, it seems likely that these are actually hybrids (Snow & Campbell, 2005 crop" or "crop hybrid" origin for weedy radish, but introgression of crop genes into weedy radish is a possibility.

| Evolution of faster flowering in weedy radish
Although non-native R. r. raphanistrum populations are genetically similar to native R. r. raphanistrum, they flower much faster (25-58 days earlier flowering on average, Model 1). Weedy radish flowers much more rapidly and uniformly than any of the other Raphanus taxa, and the weeds never require vernalization (Figure 3). This lack of variation and decreased mean suggests that the weeds have undergone strong directional selection for flowering time. This supports the hypothesis that the weedy radish most likely arose from native R. r. raphanistrum, and subsequently evolved a faster flowering time. This more rapid flowering is in agreement with previous work, which has shown that non-native R. r. raphanistrum flower faster than crop (Hegde et al., 2006; or native radish (Sahli et al., 2008).
This difference is especially striking in the raw data. We assayed days to flowering for non-native radish from three continents in common garden experiments across a span of eleven years and eight experiments across three locations; however, we find almost no variation in the raw flowering time among or within the weed populations ( Figure 3, Supporting Information Figure S7). This finding is somewhat surprising as phenotypic plasticity is expected to be a common feature of weeds (Baker, 1965) and invasives (Davidson, Jennions, & Nicotra, 2011;Richards, Bossdorf, Muth, Gurevitch, & Pigliucci, 2006) and flowering time has been found to be plastic in invasive plants, for example (e.g., Claridge & Franklin, 2002;Colautti & Barrett, 2010). As our common garden experiments were performed either in the summer or using summer conditions, it is possible that they were simply not different enough to trigger a plastic response.
In stark contrast to the phenotypic uniformity of the weeds, native range populations of both R. r. landra and R. r. raphanistrum varied in both days to flowering, and the need for vernalization.
F I G U R E 3 Raw Flowering times in Raphanus. Medians, quartiles, and outliers for raw days from germination to first flower for each population are shown, with shading to denote the proportion of plants that flowered without experiencing vernalization. Boxplot widths are a function of number of individuals per population, with wider plots indicating more individuals. Maximum: AUFI (N = 250); minimum: RBBC (N = 3), total = 2,054. Note that Y-axis scales are the same except for R. r. landra, which has much longer flowering times. None of R. r. raphanistrum populations from outside the native range required vernalization Interestingly, this also appears not to be a plastic response to our variable conditions. In native populations that were assayed in at least four common garden experiments, we find far more variation within experiments than between them, and that some native populations have reproducibly bimodal distributions of flowering times (Supporting Information Figure S7). Taken together, these data suggest that the native ancestors were likely more variable and slowerflowering, at least partly due to a vernalization requirement, and this is consistent with the hypothesis of recent and rapid directional selection for shortened weed flowering time.
In summary, we found no evidence to support a crop origin for weedy radish, either by hybridization or exoferality. Non-native R. r. raphanistrum likely descended from native R. r. raphanistrum, with possible introgression from other Raphanus taxa. All of these potential source populations have longer flowering times than we found among the weeds, which suggests that wild populations were not preadapted to field conditions and is evidence for rapid local adaptation to an agricultural habitat. Whether or not adaptation of weeds to agricultural conditions other than human control efforts is a more general phenomenon in requires additional studies comparing potential adaptive traits between agricultural weeds and their progenitors, as well as studies of present-day selection on weeds in agricultural habitats.

ACK N OWLED G EM ENTS
We

CO N FLI C T O F I NTE R E S T
None declared.

DATA ACCE SS I B I LIT Y
Seed stock information is in Supporting Information Table S1.