The evolution of crop-related weeds may be constrained by recurrent gene flow from the crop. However, flowering time variation within weedy populations may open the way for weed adaptation by allowing some weeds to escape from this constraint. We investigated this link between phenology, gene flow and adaptation in weedy sunflower populations that have recently emerged in Europe from crop–wild hybridization.
We studied jointly flowering phenology and genetic diversity for 15 microsatellite loci in six cultivated sunflower (Helianthus annuus) fields infested by weedy sunflower populations.
The flowering overlap of cultivated and weedy sunflowers varied between and within populations: some weedy individuals were found to be completely isolated from the crop, the frequency of these plants being higher in populations from highly infested fields. Within weedy populations, we detected a pattern of isolation-by-time: the genetic divergence between individuals was positively correlated with their divergence in flowering period. In addition, earlier weeds, which flowered synchronously with the crop, were genetically more similar than late-flowering weeds to the cultivated varieties.
Overall, our results suggest that crop-to-weed gene flow occurred, but was limited by divergent phenologies. We discuss the roles of weed adaptation and population history in the generation of this partial reproductive isolation.
Plant domestication is a very recent process on the evolutionary scale: most crops and their progenitors diverged < 13 000 yr ago (Balter, 2007), which has generally not been sufficient to allow for the establishment of strong reproductive barriers (Hancock, 2005). Crop-to-wild gene flow has contributed to the constitution of a continuum between typical ancestral wild plants and current elite crop varieties (i.e. crop–wild complexes; see, for instance, Reagon & Snow, 2006; Barnaud et al., 2009). In these complexes, some forms, described as weeds, grow within cultivated fields and can strongly depress the growth and yield of crop plants. Notable examples include radish, rice, rye, sorghum, sugarbeet and wheat (for a review, see Ellstrand et al., 2010). Crop-related weeds share traits with other agricultural weeds that make them competitive with regard to the crop (rapid seedling growth, high seed output, strong seed shattering, dormancy and sometimes herbicide resistance; Basu et al., 2004; Ellstrand et al., 2010). Weedy lineages have been shown to result from the colonization of cultivated fields by plants from wild populations (e.g. wild rice taxa in Asia, Ellstrand, 2003; weed sunflower in North America, Kane & Rieseberg, 2008), crop–wild hybridization (e.g. weed beets; Boudry et al., 1993; Arnold, 2004) or the reversion of cultivated plants to a wild-type habit (suggested for some weedy rice; Vaughan et al., 2005; Reagon et al., 2010).
Over the last decade, crop-to-weed gene flow has been increasingly examined, but mainly to investigate the possibility of whether genes or transgenes can ‘escape’ from a crop into a weedy relative and equip it with novel traits to increase its invasiveness (Ellstrand, 2003; Claessen et al., 2005; Snow et al., 2005; Andow & Zwahlen, 2006). The actual consequences of crop-to-weed gene flow on the evolution of weedy populations have been studied more rarely. They are dependent on the hybridization rate between crops and weeds, the fitness of the resulting hybrids and the various successive backcrossing events, and the effects of specific crop genes (Zhivotovsky & Christiansen, 1995; Ellstrand, 2003). Indeed, although cultivated plants and their weedy relatives grow in the same environment, they experience different selective pressures. Traits selected during the domestication process are generally assumed to reduce fitness in weedy populations (e.g. the absence of seed dispersal and dormancy; Stewart et al., 2003). Strong and recurrent gene flow from the crop can thus place a major constraint on the adaptive evolution of weedy populations by increasing the frequency of maladapted genes. Reciprocally, the establishment in weedy populations of a domesticated gene improving invasiveness (e.g. resistance to diseases, rapid growth, self-compatibility) requires backcrossing between first-generation hybrids and weedy plants to break the linkage to deleterious domesticated genes (Jenczewski et al., 2003). An understanding of how crop-related weedy populations can extend and adapt to the agro-ecosystem constraints despite ‘a rain of cross-compatible pollen’ from the crop (Ellstrand et al., 2010) is a fundamental topic of research in evolutionary biology, well beyond the particular case of crop–wild complexes (Lenormand, 2002).
Plant populations most often contain individuals with different flowering schedules (onset of flowering, flowering duration), thus reproducing at different times. Whenever there is variation in reproductive time between individuals, and as long as this variation has a genetic component, it should cause some deviation from random mating – phenological assortative mating – and lead to a temporal genetic structure within populations (i.e. isolation-by-time; Fox, 2003; Weis & Kossler, 2004; Hendry & Day, 2005). By applying this conceptual framework to the case of crop-related weedy populations, Ellstrand et al. (2010) have suggested that a rapid evolution of flowering time should prevent weedy populations from being swamped by gene flow from the crop. In accordance with this hypothesis, these authors noticed that reports of phenological divergence existed for weeds descended from an outcrossing crop, but not for weeds descended from a highly selfing crop, for which crop-to-weed gene flow is already limited. Notwithstanding the potential significance of phenological divergence on gene flow, little is known about its actual importance in natural populations (Weis & Kossler, 2004). To our knowledge, the flowering overlap between a crop and its sympatric weedy relatives has rarely been quantified precisely (but see Burke et al., 2002; Barnaud et al., 2009), and even more rarely related to the intensity of crop-to-weed gene flow.
Sunflower (Helianthus annuus) is an outcrossing, bee-pollinated species, native to North America and domesticated c. 4000 bp (Harter et al., 2004). Noxious weedy forms have repeatedly evolved from wild H. annuus populations in the USA (Kane & Rieseberg, 2008). In Europe, where H. annuus is absent in its wild form, weedy sunflower populations have been reported recently within sunflower fields (Faure et al., 2002; Holec et al., 2005; Vischi et al., 2006). In Spain and France, these populations show a wide diversity in phenotype, forming a continuum between typical wild plants (notably characterized by the presence of anthocyanin pigmentation in the stem and head, strong seed dormancy and high seed shattering, self-incompatibility, branching) and cultivated morphotypes (Muller et al., 2009). In the Lauragais region (south-east of Toulouse, France), c. 20% of all sunflower fields are affected, with weed density varying from only a few plants per field up to 15 plants per square metre. For such high levels of infestation, high yield losses are observed and weed eradication appears to be impossible. Molecular analysis of French and Spanish weedy populations has shown that these weeds originated from the unintentional introduction of crop–wild hybrids into farmers' fields through contaminated seed lots (Muller et al., 2011).
Weedy sunflowers have appealing features for the investigation of the impact of gene flow and phenological divergence on weed evolution. First, in North America, gene flow from crop to adjacent wild sunflower populations has been shown to be substantial (Arias & Rieseberg, 1994). Here, weedy populations are sympatric to their related crop, start flowering during the flowering period of the cultivated field and, despite its self-compatibility, cultivated sunflower still shows outcrossing rates of c. 60% (Ostrowski et al., 2010; Roumet et al., 2012). It thus appears highly likely that crop-to-weed gene flow occurs in sunflower infested fields. Second, wild H. annuus is generally late flowering relatively to the cultivated form (Blackman et al., 2011). Crop–wild hybrids and their descendants usually exhibit a wide diversity in flowering time, intermediate between the wild and cultivar-like phenotype (Reagon & Snow, 2006). This diversity provides the substrate for the evolution of flowering time in weedy populations. Finally, the different weedy populations have been shown to result from independent and recent introductions (on the order of 10–20 generations for the oldest; Muller et al., 2011). Fields with different levels of infestation may therefore be considered as more or less successful repetitions of the invasion process and compared to investigate factors that affect invasion success.
Following Ellstrand's hypothesis, we first considered whether there was any partial temporal reproductive isolation which might prevent weedy sunflower populations from being ‘swamped’ by gene flow from the crop. Second, by comparing fields with different levels of infestation, we considered whether such a partial reproductive isolation was stronger in highly infested fields, as expected if it was linked to invasion success (i.e. either as a favourable condition or as an outcome of adaptive evolution).
To address these issues, we characterized phenologically and genetically six weedy populations in six sunflower fields with different levels of infestation. The data collected were used to quantify the flowering period overlap between weedy populations and the sunflower varieties grown in the same fields, to determine whether variation in flowering time among weedy individuals was associated with variation of the genetic proximity with the crop, and especially with variation of inferred levels of gene flow from previously cultivated varieties, and to compare patterns detected in highly and weakly infested fields.
Materials and Methods
Choice of populations and sampling method
In the Lauragais area, the level of infestation by weedy sunflowers is recorded each year in c. 100 fields (Muller et al., 2009), six of which were surveyed by us in 2009. Three, namely Odars, Verfeil and Villefranche, were classified as weakly infested (less than one weed per square metre on average, which corresponded to only a few plants or clusters of plants scattered in the field), and the other three, namely Balma, Escalquens and Fourquevaux, were classified as highly infested (large and continuous infested zones in the field, with local densities reaching over 15 weeds per square metre).
In 2009, sunflower (Helianthus annuus L.) was cultivated in these fields according to the farmers' usual agricultural practices with no interference on our part. To ensure a representative sampling of weedy populations, we delimited, in each field, 6–14 quadrats in which all weeds (between nine and 44) were surveyed. Quadrats were spread over the surface of the field and their size was dependent on the local density of weeds (between 0.5 and 32 weeds per square metre). Quadrats were set up in June, c. 1 month after sowing, when cultivated sunflower plants had reached between 10 and 50 cm. All sunflowers growing between cultivated sunflower rows were considered as weeds and individually identified by a numbered plastic ring. The number of weeds per square metre was recorded in each quadrat and denoted as the local density.
At this stage, portions of leaves were taken from all marked weeds in Verfeil, Villefranche and Escalquens, from five random weeds per quadrat in Balma, Odars and Fourquevaux and from at least three cultivated sunflowers (in theory genetically identical, see later) in each field studied and in each adjacent sunflower field. A summary of the sampling design is given in Table 1.
Table 1. Summary of the sampling design
Number of quadrats
Total sample size
Sample size for genetic analysis
Values in parentheses correspond to the number of sampled plants for which full phenological data were available. The difference from the initial sample size is a result of mortality during the season.
From the end of June until the second week of August, we visited the fields twice a week. After this period, all varieties had finished flowering and over 70% of the weeds had started to flower. For practical reasons, data were then recorded once a week until harvest at the beginning of September.
Within each field, cultivated sunflowers flowered highly synchronously, except for a few early and late outliers, which we excluded to determine a reference period for crop pollen emission (the ‘crop flowering period’). Namely, for a given field F, the crop flowering period was defined as the time interval [Tc1, Tc2]F, where Tc1 was the date on which c. 2–3% of cultivated plants had started to flower and Tc2 was the date on which c. 97–98% of the cultivated plants had finished flowering.
Weedy individuals and populations
Of the 1350 marked weedy plants, only the 1024 surviving individuals (i.e. which set seeds) were included in our dataset (Table 1).
We recorded the onset of flowering (T0) for each weed as the date on which the primary head started to flower. At each visit and for each plant, we recorded the number of heads flowering on the first (I), second (II), third (III) and fourth (IV) level branches. At the end of the flowering of head I, we measured its diameter. We estimated the mean diameters of heads II, III and IV by measuring the diameters of a representative sample of these heads (1–10 depending on the number of heads available) at the end of flowering.
We used these data to estimate, for each date of visit t, the flowering area for each plant and the total flowering area of our sample of the weedy population in field F (AF(t); for details, see Supporting information Methods S1).
For each weed, we determined the end of flowering (Te) as the date from which no further heads were in flower, the ‘individual flowering duration’ as Te – T0, and the date on which the flowering area was at its maximum (Tm). For the weedy population in field F, we defined the onset of flowering (T0F) as the onset of flowering of the first flowering weed, and the end of flowering (TeF) as the end of flowering of the last flowering weed. The ‘population flowering duration’ was defined as TeF – T0F, and TmF was the date on which AF(t) was at a maximum.
All dates, T0, Te, Tm, T0F, TeF and TmF (Table 2), were given and analysed as the number of days after the Tc1 date in the corresponding field.
Table 2. Parameter definitions
Crop flowering period in field F
Flowering period of weedy population in field F
Date of the onset of flowering of a weedy individual
Date of the end of flowering of a weedy individual
Date at which the flowering area of the weedy population of field F is at a maximum
Date at which the flowering area of a weedy individual is at a maximum
Overlap in flowering period of the weedy population of field F with the cultivated variety
Overlap in flowering period of weedy individual i with the variety cultivated in field F
P07, P05, P03
Possibility (0 or 1) of a given weed being an offspring of the variety cultivated in its field in 2007, 2005 or 2003
F07, F05, F03
Proportion of weeds which could be offspring of the variety cultivated in the same field in 2007, 2005 or 2003
In total, 766 plants were genotyped at 15 independent microsatellite loci: ORS297, ORS337, ORS342, ORS344, ORS371, ORS380, ORS432, ORS610, ORS620, ORS656, ORS674, ORS735, ORS788, ORS887 and ORS925 (Tang et al., 2002). Molecular methods for DNA isolation, genotyping and allele scoring were as described in Muller et al. (2011).
Almost all sunflower varieties cultivated in France are genetically homogeneous F1 hybrids. A database of 19 genotypes of F1 hybrids genotyped at the same 15 microsatellite loci has been reported by Muller et al. (2011), covering a representative part of the varietal diversity grown in the Lauragais region in the previous 20 yr. The data obtained from cultivated plants in the present study were integrated into this database: we included five ‘new’ genotypes, corresponding to varieties PR41H32 (Pioneer), NK Sinfoni (Syngenta), Vellox (RAGT) and two unknown varieties.
Overlap in flowering periods of crop and weeds
In each field F, we computed the flowering period overlap between any marked weed i and the variety cultivated in the field (OVFi, (Eqn 1)) and the flowering period overlap of the weedy population of field F with the cultivated variety (OVF, (Eqn 2)).
(AFi[Tc1, Tc2]F and AF[Tc1, Tc2]F, areas of weed i and of the weedy population of field F, respectively, in bloom during the crop flowering period; AFi[T0Fi, TeFi] and AF[T0F, TeF], total flower areas of weed i and population F, respectively, over their flowering period). The flower areas of each weed and each population during the different time intervals were computed as described in Methods S1.
We then defined the frequency of isolated plants in field F as the proportion of weeds for which OVFi = 0 (i.e. weeds which had finished flowering before Tc1 or started flowering after Tc2).
The effect of the field infestation level on the weedy population phenology, namely on variables T0F, TmF, population flowering duration and OVF, was tested using a one-way analysis of variance (ANOVA, performed in R-2.5.1; R Development Core Team, 2008; package lmtest). Differences in the frequencies of isolated plants between population pairs were subjected to a chi-squared test.
Within each weedy population, a Spearman's correlation test was used to analyse the effect of local density (i.e. the number of weeds per square metre in the quadrat in which a weed was located) on the variables Tm, T0 and individual flowering duration, and on the frequency of isolated plants per quadrat.
Standard diversity statistics were computed in the database of varieties and in weedy populations using GENETIX (Belkhir et al., 2001) and FSTAT (Goudet, 2001): Nei's expected heterozygosity (He; Nei, 1987), the number of different alleles and the allelic richness standardized for similar sample sizes (Ra; Petit et al., 1998). We denoted as ‘cultivated’ (C) alleles, alleles that were observed at least once in the 24 F1 hybrids contained in our database. All other alleles were considered as specific to the weedy populations and hereafter called ‘original’ (O) alleles. As shown by Muller et al. (2011), these O alleles are the footprints of the wild origin of weedy populations. The number and frequency of O alleles per population were computed separately for each locus and averaged over all loci. The frequency of O alleles was also computed at the individual level. In each population, the number of O alleles was standardized for samples of similar size, by analogy with allelic richness. Namely, the standardized number of O alleles per population () was computed as the mean number of O alleles in 1000 random subsamples of 30 individuals from the population. The significance of He, Ra and differences between pairs of populations was tested in R using Wilcoxon signed-rank tests comparing values for the same loci in different populations. Genetic differentiation between each weedy population and the database of varieties was quantified through FST estimation by the method of Weir & Cockerham (1984). The significance of FST values was tested by 10 000 random permutations of multilocus genotypes among populations, using the G log-likelihood statistic (Goudet et al., 1996).
Temporal genetic structure of weedy populations
Temporal genetic structure within weedy populations was assessed by adapting a method classically used for the investigation of isolation-by-distance, replacing spatial coordinates with reproductive time. As individual flowering durations could be very long (15 d on average, with a maximum of 58 d), we used Tm rather than T0 as a surrogate of reproductive time. Within each population, the temporal distance between two individuals (Dij) was computed as the absolute value of the difference between their Tm values. Genetic relatedness between individuals was computed as the multilocus kinship coefficient (Fij) (Loiselle et al., 1995). The association between Dij and Fij was tested by computing the average Fij value for 10 temporal distance classes and comparing these values with those expected under the null hypothesis of no temporal genetic structure. Distance classes were defined in such a way that the number of pairwise comparisons within each distance interval was approximately constant. The 95% confidence interval associated with the null hypothesis was assessed by 10 000 random permutations of temporal distances among pairs of individuals. To quantify the extent of the temporal genetic structure, the Sp statistic (Vekemans & Hardy, 2004) was calculated as −b/(1 – F(1)), where b is the regression slope of Fij against the temporal distance between individuals and F(1) is the mean kinship coefficient in the first distance class. All computations were performed using SPAGeDi v1.2 software (Hardy & Vekemans, 2002).
Temporal variation of crop–weed genetic differentiation
For each weedy population, the relationship of the reproductive time of the weed (Tm) with crop–weed genetic differentiation was investigated using two methods. First, we used the FST between weedy individuals and the database of 24 varieties. The effect of Tm on FST was tested using a sliding-window analysis. A window of a constant length of 14 d was moved along the season in 2-d steps, successively including all weeds displaying their Tm within the window. Each window created a subset of weeds for which we computed the mean Tm and FST values with the database of varieties. The 95% confidence interval associated with the null hypothesis of temporal randomness was built using the same procedure for 1000 random permutations of Tm values over the weedy individuals.
In a second analysis, we quantified the genetic differentiation between each weedy individual and the cultivated varieties by the frequency of O alleles in its genotype. The effect of Tm on this frequency was tested within each weedy population using a Spearman's correlation test.
Weed crop ancestry
To investigate more specifically the extent of gene flow between sunflower varieties and weedy populations, we assessed the possibility that a given weed could or could not be an offspring of a variety previously cultivated in the same field. This possibility is denoted as ‘weed crop ancestry’ in the following, and can be considered as an indicator of the intensity of gene flow from the cultivated variety in a given year.
Crop rotations bring sunflowers back into a field every second year. The identity of the sunflower varieties cultivated in 2007, 2005 and 2003 in the fields studied were provided by the growers. When the genotype of the variety was recorded in our database, we used an exclusion procedure to determine whether a given weedy individual sampled in 2009 could have that variety as a parent. Because genotyping errors, mutations and residual heterogeneity within the varieties could contribute to false exclusions, we allowed one locus mismatch between a parent and its potential offspring. Namely, the variety was excluded as a putative parent of a weed when there was more than one locus for which they did not share any allele. For each weed, the outcome of this procedure (exclusion or not) was stored in a variable, denoted as P07, P05 or P03, for the compatibility with the variety cultivated in the field in 2007, 2005 or 2003, respectively. For each population, the proportions of weeds that could be an offspring of the variety cultivated in the field in 2007, 2005 and 2003 were computed and named F07, F05 and F03, respectively (see Table 2 for parameter definitions).
Field and reproductive time effects on weed crop ancestry
The significance of the differences in F07, F05 and F03 between pairs of populations was tested using chi-squared tests.
Within populations, we investigated how weed crop ancestry was related to local density and weed phenology. As the proportions of weeds which could be an offspring of the variety were substantial only for the varieties cultivated in 2007 (see the 'Results' section), we focused on the ancestry with these varieties for the following analysis. The effects of P07 on local density, and on T0, Tm and individual flowering duration, were tested using a hierarchical ANOVA, assuming population and P07 (nested within population) as fixed factors.
The variation of weed crop ancestry across the season was investigated within each population using the sliding-window procedure already described: the mean value of Tm and the frequency F07 were computed within each window. The 95% confidence interval associated with the null hypothesis of temporal randomness was built on the basis of 1000 random permutations of Tm values over all individuals.
Crop and weed flowering phenology
In the six fields studied, the weedy population had a longer flowering duration than the cultivated variety: on average, the flowering periods of the crop and weedy populations extended over 19.8 and 62.6 d, respectively; in all fields, the first weeds (12% on average) started to flower before the crop, and some weeds (3% on average) were still in bloom at harvest time. This result reflected the fact that the flowering period duration for weedy individuals was quite long (14.6 d on average and up to 58 d) and that their flowering onset was highly variable (the first and last weed started to flower 16 d before and 59 d after the date Tc1, respectively). The maximum flowering area of the populations was reached between 14 and 24 d after Tc1. The flowering periods of weedy populations and cultivated fields largely overlapped: overall, c. 40% of the total weed flower area was in bloom during the crop flowering period [Tc1, Tc2] (Table 3, Fig. 1a). As shown in Fig. 1(b), the flowering overlap with the cultivated variety varied between 0% and 100% between individuals of the same population: each weedy population was composed of plants which entirely or partially flowered during the [Tc1, Tc2] period and of plants completely isolated from the crop; 99.7% of isolated plants corresponded to late-flowering weeds.
Table 3. Descriptors of the flowering schedules and crop–weed flowering overlap in each field
Frequency of isolated plants (%)
[Tc1, Tc2], crop flowering period; [T0F, TeF], weedy population flowering period; TmF, date at which the flowering area of a weedy population is at a maximum; T0F, TeF and TmF given as the number of days after the date Tc1 of the corresponding field; OVF, overlap in flowering period of the weedy population in field F with the cultivated variety.
Frequencies of isolated plants with the same letters were not significantly different (P < 0.05, chi-squared test).
[16 July, 3 August]
[10 July, 1 August]
[11 July, 2 August]
[25 July, 11 August]
[11 July, 3 August
[16 July, 2 August]
The overlap in flowering periods of the weedy population and crop differed significantly between infestation levels: it was c. 50% in weakly infested fields and only 33% in highly infested fields (P = 0.042, Table 3). This result reflected differences in the frequency of isolated plants, rather than in population phenology. Indeed, the frequency of isolated plants was higher in highly infested fields (43.9% vs 18.4%, Table 3), whereas T0F, TmF and the population flowering duration did not vary significantly as a function of the infestation level (P = 0.33, 0.64 and 0.11, respectively).
No strong effect of local density on weed phenology was detected at the intrapopulation level. However, there was a trend for plants to flower later in quadrats with higher densities. In the weedy populations of Verfeil, Odars, Fourquevaux and Villefranche, Tm values increased significantly with local density. Local density also affected T0 in Verfeil, Odars and Fourquevaux, flowering duration in Verfeil and the frequency of isolated plants (computed per quadrat) in Fourquevaux and Villefranche (i.e. T0 and the frequency of isolated plants increased significantly with local density, whereas the flowering duration decreased). By contrast, we found no significant impact of density on weed phenology in Balma and Escalquens.
Population-level summary statistics are given in Table 4. Between six and 23 alleles per locus were scored over the whole sample. As shown previously by Muller et al. (2011), varieties had a lower allelic richness relative to weedy populations: 70 alleles were detected over all 15 loci in the database of varieties (i.e. C alleles), against 235 alleles in the weedy pool. All C alleles were present in the weedy pool and, despite their lower number, they were prevalent in all the weedy populations. On average, over loci, the frequency of O alleles was only 17%. When computed at the individual level, this frequency varied continuously between 0% and 50% between weedy individuals.
Table 4. Population genetics statistics in the database of varieties and in weedy populations
A, mean number of alleles per locus; Ra, mean allelic richness per locus, standardized for a sample size of 22 individuals; He, expected heterozygosity over loci; Korig, total number of original alleles (sum over loci); , number of original alleles over loci, standardized for a sample size of 30 individuals; forig, frequency of original alleles per population; FST, FST estimated between each weedy population and the database of varieties. Values of He, Ra and with the same letters are not significantly different (P < 0.05, Wilcoxon signed-rank tests). ***, P <0.0001.
Populations in highly infested fields presented a higher allelic richness and had more O alleles than those in weakly infested fields (Table 4). Populations in Odars and Balma represented extreme cases with 24 and 71 O alleles, respectively, for a standardized sample size of 30 individuals.
A significant genetic differentiation (P <0.05) was detected between each weedy population and the database of varieties. The estimated FST values did not vary significantly with the level of infestation (P =0.41) or with population flowering overlap (P =0.92).
Temporal genetic structure of the populations
Within the weakly sampled populations of Fourquevaux, Balma and Odars (< 55 individuals genotyped), we found no or only a weak relationship between pairwise genetic similarities and temporal distances. By contrast, within the populations of Villefranche, Escalquens and Verfeil, in which at least 107 individuals were genotyped, we found a significant decrease in genetic similarity with temporal distance between individuals, as expected under isolation-by-time (Fig. 2).
Temporal variation of crop–weed genetic differentiation
For each weed population, genetic differentiation with the varieties varied with individual reproductive time (Fig. 3a). Except for the Villefranche population, the FST value between windows of weedy individuals and the database of varieties increased with the average reproductive time (Tm) of the windows. In four populations, lower than expected FST values were detected for the early windows and, in three populations, higher than expected FST values were detected for the groups of late weeds. In Villefranche, this pattern was reversed: FST values with the varieties decreased across the season.
Except for the two weakly sampled populations of Balma and Fourquevaux, we found that the individual frequency of O alleles increased significantly with reproductive time (Tm) (P <5 × 10−3 for all populations).
Weed crop ancestry
In the fields in which historical data on the varieties were available, the proportion of weeds which could have descended from a cultivated variety was higher for recently cultivated varieties: on average, F07 (18.1%) was higher than F05 (5.2%) and F03 (1.3%) (Table 5).
Table 5. Varieties cultivated in 2007, 2005 and 2003 and frequencies (%) of the sampled weeds which could be an offspring of these varieties: F07, F05 and F03, respectively
Empty cells correspond to missing information.
Values of F07, F05 and F03 with the same letters were not significantly different (P < 0.05, chi-squared test).
We found no effect of infestation level on the proportion of weeds compatible with the variety cultivated in 2007 (F07, Table 5). Within weedy populations, we found a slight, marginally significant effect of local density (P = 0.05 for the effect of local density on the variable P07): when computed at the quadrat level, F07 tended to decrease when the local density increased.
Weed phenology also affected weed crop ancestry. Indeed, within a field, weeds which could not have been derived from the variety cultivated in the field in 2007 presented significantly higher T0 (delay of 4.8 d on average, P = 1.4 × 10−2) and Tm (delay of 10.4 d on average, P = 1.8 × 10−11) values, and a longer individual flowering duration (6.9 d longer on average, P = 7.2 × 10−9).
Temporal variation of weed crop ancestry
Within each weedy population, weed crop ancestry varied with individual reproductive time (Tm). Except for the Odars population, F07 decreased across the season: the frequency of weedy individuals which could have resulted from crop–weed hybridization was lower within the late-flowering than within the early-flowering group. This effect was significant in the two extensively sampled populations of Villefranche and Verfeil (Fig. 3b): for these populations, the F07 value observed was not included in the confidence interval for at least eight windows. Such a result strongly suggests that late-flowering weeds had fewer crop ancestors in their genealogy than early-flowering forms.
Flowering overlap and variable opportunities for crop-to-weed gene flow
In all studied fields, the flowering periods of the cultivated varieties and the weedy sunflower populations largely overlapped, providing the opportunity for hybridization, as shown previously in another case of crop–weed sympatry, that is, sorghum (Barnaud et al., 2009). In addition, we showed that weedy populations were composed of a mixture of individuals that reproduced at different times, with some plants, mainly late-flowering ones, totally isolated from crop pollen. This variable exposure to pollen flow from the crop makes prezygotic temporal reproductive isolation possible.
Temporal structure of the populations: the result of variable crop-to-weed gene flow
In the weedy populations in which large samples were collected, we found that differences in reproductive time between individuals were correlated with divergence on neutral genetic markers. Such an isolation-by-time pattern can occur when variation in reproductive time has both a genetic basis and causes nonrandom mating among individuals (Hendry & Day, 2005). High heritability values for phenological traits are commonly found in plants (Hendry & Day, 2005) and the effects of temporal variations of flowering schedules on mating patterns have been shown previously (e.g. Gerard et al., 2006b; Oddou-Muratorio et al., 2006). However, to our knowledge, temporal population structure has been reported only twice in plants: first, in maize populations for markers mapped close to genes or quantitative trait loci involved in the flowering schedule (Pressoir & Berthaud, 2004) and, second, within a hybrid zone of trees for which the flowering period extended over 4 months (Gerard et al., 2006a). In fish species, several similar results have also been published, based either on markers partially linked to quantitative trait loci for breeding time, or on the assessment of differentiation between early and late breeders from different locations, which entailed confusion with the isolation-by-distance effect (Maes et al., 2006).
The detection of an isolation-by-time pattern within populations based exclusively on a neutral set of markers therefore appears to be quite remarkable. In addition to the scarcity of studies on this topic, the small number of isolation-by-time examples may result from the restrictive conditions required for assortative mating to induce such a pattern: high heritability, sufficient variance of individual reproductive time and short individual flowering period (Franks & Weis, 2009). These conditions are even more restrictive in the absence of disruptive selection on mating traits (Devaux & Lande, 2008). In the case of weedy sunflower populations, individual flowering duration can be quite long. In addition, cultural practices limit flowering periods and thus the variance of the individual reproductive time: seed germination must occur after sowing (date of the last herbicide treatment), and seeds must mature and disperse before harvesting time. Flowering overlap between weedy individuals is thus quite substantial and assortative mating is probably not sufficient to explain the isolation-by-time pattern observed. By contrast, flowering overlap between weedy individuals and the cultivated variety was highly variable, and weed genetic differentiation with the cultivated varieties increased with the reproductive time of weeds. This suggests that more or less intense gene flow from the crop may be the main factor explaining this isolation-by-time pattern.
As a whole, therefore, our results have three simultaneous implications: reproductive time is heritable, crop-to-weed gene flow occurs and reproductive time variation within weedy populations entails a partial reproductive isolation from the crop. Geographical distances and their effects on gene flow and population structure are regularly reported in the literature (e.g. Arias & Rieseberg, 1994; Chun et al., 2011). However, to our knowledge, there are, at present, no data showing that variation in reproductive time affects weedy population genetic structure, thus probably influencing their evolution.
It should be stressed that the present conclusions are not based on direct estimations of gene flow, but on three different ways to quantify genetic differentiation between weedy individuals and sunflower varieties: FST, forig (the frequency of O alleles) and F07. These statistics can be influenced by factors other than pollen-mediated gene flow. First, our database of varieties has been shown previously to catch almost exhaustively the alleles present in F1 hybrids cultivated in the Lauragais region (Muller et al. 2011); it might, however, not be representative of the allelic frequencies in the varieties cultivated in a given field. Strong discrepancies could notably lead to the variation in FST values independent of crop-to-weed gene flow variation. Such bias may explain why, in Villefranche, FST variations were incompatible with both our expectations and the variations in F07 and forig. Second, the O alleles track the footprint of the wild gene pool which gave rise to the population. Crop-to-weed gene flow will decrease the value of forig at the within-field level, but its raw value variation among fields is dependent on both the gene flow and the initial constitution of the population, that is, the number and genotypes of the first hybrids introduced. Finally, the statistic F07 measures the compatibility of the weeds with a variety. Compatible weeds could also arise from crosses between weeds, especially if the weeds have been produced by crop–weed hybridization in the previous generations. It is therefore an overestimation of single-generation crop-to-weed gene flow. As for forig, we can interpret its within-population variation, but we cannot compare its raw value among fields.
The best method would be to estimate the exact probability of a weed being derived from a weed–variety parental pair or a weed parental pair, as is the case for classical assignment procedures (Paetkau et al., 2004). This is impossible because of the unknown genetic composition of weedy populations in previous years and possible pollen-mediated gene flow from surrounding fields (sunflower is one of the main crops cultivated in the area). In future studies, it would be interesting to properly quantify crop-to-weed pollen flow and its relation to different flowering phenology descriptors (T0, Tm, flowering overlap, etc.). In particular, direct estimations could be obtained from the progenies collected on the plants surveyed (Jones & Ardren, 2003).
Stronger temporal reproductive barrier in highly infested fields: adaptation or population history?
The flowering periods of weedy populations and cultivated varieties were less synchronous in highly infested fields. Indeed, the flowering overlap and frequency of isolated individuals decreased and increased, respectively, with field infestation. Individual differences in the timing of flowering may be caused by several environmental factors, including plant density (Kelly & Levin, 2000). However, in the present study, we detected no strong effect of intrapopulation density variation (i.e. local density) on the frequency of isolated plants or on the flowering schedule. It thus appears that the increase in the prezygotic temporal reproductive barrier with infestation levels is attributable, at least in part, to genetic effects.
Two types of hypothesis can be proposed to explain this result. First, the different levels of infestation may reflect different stages of weedy population evolution, and the reduced flowering overlap in highly infested populations would then be the product of adaptation. This hypothesis appears to be plausible considering that the flowering time can evolve on ecological timescales as short as five to seven generations when selective pressures are strong (Franks et al., 2007; Franks & Weis, 2009). Moreover, it has been demonstrated that selection against hybrids between two populations locally adapted to different environments could strengthen prezygotic reproductive barriers (Freeland et al., 2010). Crop–wild hybrids of H. annuus have been shown to be less fecund than their wild parents, although with a wide variance among hybrids (Snow et al., 1998; Cummings et al., 2002; Mercer et al., 2006), and their fitness could also be negatively affected by domesticated maladapted traits, such as the absence of seed dormancy and dispersal (Snow et al., 1998). In weedy populations, selection against hybrids may thus act to reinforce the temporal reproductive barrier between weedy and cultivated sunflower. This hypothesis is in line with the expectation of Ellstrand et al. (2010) that reproductive isolation should evolve rapidly in crop-related, outcrossing weeds subjected to intense crop pollen flow.
However, the differences observed could also reflect differences in population history. These may result from different phenological characteristics of the introduced weeds, with earlier weeds introduced in the presently weakly infested fields. Alternatively, the initial proportion of late individuals may have been similar in all fields, but, in the case of a small introduction, the late individuals would have had reduced mating opportunities and the weedy population would have been more strongly exposed overall to gene flow from the crop; this would have pushed it towards earlier flowering. By contrast, highly infested fields may have benefited from higher initial seed lot contamination, recurrent weed introductions or pollen-mediated gene flow from surrounding infested fields. For both scenarios, earlier flowering time would have made the weedy populations more sensitive to genetic swamping by the crop, and hence have prevented an increase in infestation. In accordance with these ‘population history’ alternatives, the allelic richness and the number of O alleles were highest in populations of highly infested fields (Table 4). This increase in diversity together with infestation level cannot be explained by weedy population adaptation by itself. By contrast, it is expected as a consequence of both less intense crop-to-weed gene flow and recurrent or larger introductions in the highly infested fields. Such effects of population history on the success of an invasion have been demonstrated, for instance, via the role of gene flow in increasing genetic diversity (Lachmuth et al., 2010).
To sum up, two nonexclusive hypotheses could explain the differences observed in flowering overlap among populations: adaptation and population history. It is a hard task to demonstrate that adaptation has occurred when the initial steps are not available (as is the case in Franks & Weis, 2009; Freeland et al., 2010). Progress could be made through a molecular population genetics approach looking for the footprints of selection in candidate genes (i.e. Kane & Rieseberg, 2008; for American weedy populations of sunflower). Here, promising candidates could be found among genes involved in flowering time evolution during sunflower domestication and breeding (Blackman et al., 2011). Another approach would be to identify the phenologic and genetic characteristics that are more often associated with a successful weedy population by gathering information on additional populations (i.e. Gomez-Gonzalez et al., 2011). Finally, we could also investigate how reproductive isolation could evolve by quantifying the respective strengths of evolutionary forces at work (i.e. gene flow and selective pressures on phenotypic traits), especially by testing the presence of adaptive divergence through the season (i.e. adaptation by time; Hendry & Day, 2005).
We are very grateful to Muriel Tavaud and the Experimental Unit of the Centre Technique des Oléagineux et du Chanvre Industriel (CETIOM) in En Crambade for technical assistance. We thank Mathieu Siol and anonymous reviewers for their valuable comments on previous versions, and Philippe Chatelet for help in text revision. This work was funded by CETIOM. Marie Roumet was supported by a fellowship from the French Ministry of Research. Authors' contributions: M.R., J.D. and M.-H.M. conceived and supervised the study; M.R., C.N. and M.-H.M. collected the field data; M.R. and M.L. carried out the molecular analyses; M.R. and C.N. performed the statistical analyses; M.R. and M.-H.M. wrote the manuscript. Data archiving: genotype and phenotype data have been submitted to Dryad: http://dx.doi.org/10.5061/dryad.17qs8.