Long-distance dispersal of oilseed rape seeds: The role of grain trailers

In agroecosystems, anthropogenic activities can modify the natural dispersal capacity of crops and their capacity to establish feral populations. In the case of oilseed rape (OSR), seed spillage from grain trailers during harvest was first quantified by an in situ scientific study (Selommes, Loir-et-Cher, France). Demographic analysis of seeds collected from 85 traps set on road verges suggested that OSR dispersal distance due to seed spillage from grain trailers can be up to 400m. In the present study, we used SSR markers to genotype seeds collected from trap-sites and from surrounding OSR fields to precisely estimate the distances between traps and fields. Trailer directions on each road were also considered. Few seeds (5.8%) were not linked to a field in the studied area, while most of the seeds (59.2%) were linked to a field situated over 400 m away. The overall mean dispersal distance was 1250 m. It ranged from 308 m to 1392 m for one-lane roads, and from 1048 m to 1404 m for two-lane roads. Events of seed dispersal at greater distances (> 5 km) were rare but still possible. It thus follows that OSR seed dispersal due to spillage from grain trailers should be carefully considered in the context of genetically modified plant cultivation.


34
Seed dispersal is often a complex phenomenon that governs the dispersal of annual plants [1]. 35 It is crucial to quantify this dispersal in order to understand the population dynamics and thus 36 the spatial distribution of the species [2]. In areas where human activities are intense, human-37 mediated seed dispersal (i.e., anthropochory) considerably affects plant dispersal [3,4].   In the case of genetically modified (GM) cultivation, it has been shown that feral OSR 58 populations could arise from the spillage of imported seeds during transportation [16,[38][39][40][41]  The origin of feral OSR populations seems to be linked to anthropochory. They could originate 69 from feral seed banks [24], the harvesting of adjacent fields [24] found that the number of seeds in traps depended on the trap-field distance as well as the 85 distance to the main silo in interaction with the number of road lanes. However, this approach 86 did not permit an accurate conclusion to be drawn in terms of an estimate of the effective 87 dispersal distances of the seeds. For example, a seed found in a trap placed next to an OSR 88 field, that is to say, at a distance of 0 m, could come from this field or indeed another field 89 situated further away. Based on this scientific study and the analysis of the seeds collected in the traps and fields in 92 the area, we used microsatellite markers to estimate precise seed dispersal distance due to seed 93 spillage from grain trailers. We discuss these results in the framework of GM cultivation and 94 the coexistence of GM, non-GM, and organic crop cultivation.   1  31  0  39  R11  1  24  0  49  R12  1  12  0  50  R12  1  60  40  51  R12  1  13  400  26  R8  1  0  0  27  R8  1  0  40  73  R2  2  39  400  74  R2  2  65  0  75  R2  2  69  0  76  R2  2  54  40  77  R2  2  137  0  78  R2  2  6  40  79  R2  2  57  0  80  R2  2  100  40  81  R2  2  208  400  102  R2  2  235  0  103  R2  2  318  0  52  R3  2  36  400  53  R3  2  32  400  54  R3  2  21  0  55  R3  2  20  40  56  R3  2  199  0  57  R3  2  36  40  58  R3  2  20  400  82  R4  2  67  0  83  R4  2  106  40  84  R4  2  160  400  85  R4  2  13  0  86  R4  2  225  40  87  R4  2  35  400  88  R4  2  227  0  89  R4  2  We obtained certified seeds from 20 more recent cultivars (12 pure-line and 8 hybrid cultivars) 160 and completed them with seeds of the 50 most probable previous cultivars that were still grown. 161 We genotyped a total of 1976 cultivar seeds with most of the time 30 seeds per cultivar.

169
Thus, in total, 11 usable polymorphic and independent loci were amplified from these 9 SSR  Cross-recombination among cultivars was not allowed and only considered within cultivars; we 183 also did not consider inter-cultivar hybrids. If the findings were ambiguous, the plant was 184 assigned to the most consensual cultivar with the highest likelihood.

186
Field assignment 187 Due to sequencing problems (see Results section), cultivar assignments were not possible. We 188 decided to directly assign fields to each grain from the trap-sites without cultivar identification. 189 We kept only unique and entire field genotypes. Contrary to cultivar assignment (i.e. a field is  196 We computed the distances between each trap and each field in our area based on the road 197 network. We then coupled this geographic distance information to the field assignment The data analyses and figures generation were done with R software version 3.6 [55].

208
The datasets generated during the current study are available in the Dryad repository [will be 209 accessible on Dryad after journal acceptation].

213
Cultivar genotypes 214 Due to sequencing problems on cultivars, we did not obtain any alleles from Ol11B0 and,

219
Data collection genotypes 220 The two new markers (Na12C08 and Na12E01-A/Na12E01-B) returned 53% of missing data 221 for field and seed-trap genotypes (against 15.1% for the other markers).

222
We thus focused only on the former eight SSR loci for all the field and seed-trap genotypes.

223
On this basis, for seed-trap genotypes, we only considered genotypes with a maximum of 6 224 missing alleles (for a total of 16 alleles). We finally obtained 2923 proper genotypes.

225
For field genotypes, as they constituted our references, we only considered entire genotypes 226 with no missing data. We obtained 865 full genotypes from 98 fields (only 1 field was 227 discarded). We then constituted a field database reference of 444 genotypes with unique 228 genotypes for each field.

Field and seed-trap plant assignment 231
For trap-site seeds, only 131 seeds were not linked to a field genotype. This means that 2792 232 (over 2923, i.e. 95.5%) seeds had a potential field origin in our area. From the perspective of 233 the fields, every field was assigned to at least one trapped seed.

Minimum distance with the information on grain trailer trips
These field assignment results combined with the geographical distances between each trap and 237 field enabled us to extract the minimum distance between each trapped seed and corresponding 238 field. Taking into account the trailer trips from the fields to the main silo, we corrected the 239 initial distances (see Supplementary Information S1): each seed was linked to the closest field 240 according to the driving direction and farmer trailer trips (Fig 1 & 2). As some putative fields 241 were not genotyped for seeds from roads R2, R4, and R5, minimal distances were not 242 considered from these roads. The other roads were not impacted by the missing fields.

250
(116) to a field at 40 m, and 4.6% (58) to a field at 400 m. A huge number of seeds (688, 55.1%) 251 was linked to a field further than 400 m away. 40% of the seeds were linked to a field further 252 than 1165 m, the overall mean minimal dispersal distance estimated. A few seeds (12, less than 253 1%) were associated with very long dispersal distance (> 5 km). Additionally, 5.8% (72) were 254 not linked to a field in this area. Some seeds (5.5%, 69) were linked to fields at distances 255 between 0 m and 400 m.   The demographic analysis of the study of Bailleul et al. [18] suggests that OSR seed dispersal 284 distance due to spillage from grain trailers can be up to 400 m. In this article, 400m was 285 considered as long distance dispersal and the farther long distance dispersal ever quantified.
However, our present study shows for the first time that seed dispersal can go far beyond 400 287 m, as it was the case for more than two-thirds of the seeds trapped on road verges. The average 288 seed dispersal distance is 1164.7 m (or 847 m if we do not considered the information obtained 289 from local farmers, Supplementary Information S1) and 40% of the seeds collected were linked 290 to a field farther than this average distance. This information from grain trailer trips was thus 291 essential to properly assign a seed to its source field. Even events of seed dispersal at "very-292 long" distances (> 5 km) are rare (less than 1%) but still possible. As seeds trapped on road 293 verges were assigned to the closest and most likely field from where they could originate, we 294 should state that our seed dispersal distances are certainly underestimated. long distance from the source field seems more than likely.

301
The knowledge of grain trailer trips was crucial in this study. It was improbable that grain 302 trailers did not take the most direct road leading to the silo. However, in two cases, the large 303 dispersal distance we observed could be due either to the fact that the trailers take an alternative 304 route or that seeds experiencing a rare and intense dispersal event. As the number of collected 305 seeds is related to the distance between a trap and the road verge, we thought that a trap placed 306 on a side of the road could not collect seeds from a field located on the opposite side.

307
Nevertheless, we found one case (site 29) on the narrow one-lane road R7 where three seeds 308 seemed to originate from the field on the other side of the road.

309
From the genetic analysis of this study area, we showed that the diversity of the feral population 310 (i.e., successful seed dispersal and survival) in terms of cultivars was greater on road verges two-lane roads, and as a result, seed dispersal due to wind blowing would be greater.

317
These results should be interpreted with caution, as a number of choices limit their scope. In 318 terms of assignment, we retained the missing data for some seed genotypes, which limits the 319 reliability of the procedure. However, the genotypes with missing data are relatively few. Also, 320 only 10 leaves per field were genotyped, which is likely to limit the characterization of the 321 genetic diversity of the seed sources. However, we were able to show that 10 leaves allow a 322 non-negligible part of this diversity to be estimated for a reasonable sampling effort and 323 genotyping cost (in Supporting Information [54]). Finally, we chose to assign the seeds to the 324 nearest field whose likelihood was non-zero, which leads to an underestimation of the dispersal 325 distance.

326
This study faced some limitations. First, an inherent difficulty in this study relates to the very 327 nature of the organization of biodiversity in the agroecosystem under investigation. As the 328 diversity of OSR cultivars is relatively low, and these cultivars are relatively homogeneous in 329 term of genotypes, it is therefore difficult to discriminate them with only a few genetic markers.

330
Second, the fact that some fields could not be genotyped forced us to disregard three roads, 331 which would have had an impact on the proportion of seeds linked to another location in the 332 study area. Another potential bias could potentially originate from the fact that the frequency 333 estimation of seed origin could differ between trap were all the plants were analyzed (trap with 334 less than 50 seeds) and those were 50 seeds were randomly selected.