Plant hopscotch: The leaps and bounds of invasive species

Arabidopsis by PennState

Editor's Introduction

Rapid evolution accelerates plant population spread in fragmented experimental landscapes

Evolution, the study of how organisms change over time, is a core concept of biology that shapes our understanding of many organisms and species. Many people have the misconception that evolution happened only once a long time ago, but in fact, evolution is an ongoing process affecting populations of all living things, including humans, as groups of organisms become better suited to changing environments. This paper investigates a species of rapidly growing plants called Arabidopsis thaliana under a simulation of evolution. The purpose of the study was to determine the evolutionary effects on how far and how fast the plants were capable of spreading from their origins compared to plants that were not undergoing evolution.

NOTE: This article is part of a Collection of student-annotated papers that are the product of the SitC team’s research into best practices for using primary literature to support STEM education. For this reason, these papers have undergone an alternate review process and may lack educator guides. To learn more, visit the main Collection page: SitC Lab.

Paper Details

Original title
Rapid evolution accelerates plant population spread in fragmented experimental landscapes
Original publication date
Vol. 353, Issue 6298, pp. 482-485
Issue name


Predicting the speed of biological invasions and native species migrations requires an understanding of the ecological and evolutionary dynamics of spreading populations. Theory predicts that evolution can accelerate species’ spread velocity, but how landscape patchiness—an important control over traits under selection—influences this process is unknown. We manipulated the response to selection in populations of a model plant species spreading through replicated experimental landscapes of varying patchiness. After six generations of change, evolving populations spread 11% farther than nonevolving populations in continuously favorable landscapes and 200% farther in the most fragmented landscapes. The greater effect of evolution on spread in patchier landscapes was consistent with the evolution of dispersal and competitive ability. Accounting for evolutionary change may be critical when predicting the velocity of range expansions.


In an era of global environmental change, biological invasions and the movement of species ranges with climate change present two of the greatest disruptions to natural and managed ecosystems (12). At the core of each dynamic is the spread of populations across landscapes fragmented by natural and anthropogenic barriers to movement. It has long been appreciated that habitat fragmentation slows the velocity of spread (34), but its influence on the potential for evolution to increase population expansion is unknown (5). Theory shows that natural selection at the low-density front of populations expanding through continuously favorable landscapes, coupled with the spatial sorting of offspring, favors traits contributing to fecundity and dispersal, both of which accelerate the invasion velocity (610). Whether this eco-evolutionary process operates similarly in systems fragmented by unsuitable habitat is uncertain because spread in these systems depends on the buildup of high-density populations capable of dispersing over gaps (511). Although any factor that alters selection on an expanding population can influence spread, whether evolution operating through selection or genetic drift predictably affects spread velocity on the rapid time scale of ecological dynamics remains to be determined. Answering questions about how evolution affects population expansion has important implications for predicting the future spread of biological invasions and climate change migrants, based on currently measured rates.

Empirical progress toward understanding evolution in populations spreading through fragmented landscapes is limited, largely because the process occurs over many generations and at geographic spatial scales. Due to these constraints, nearly all empirical evidence for evolution affecting spread comes from a few retrospective, observational analyses (1216). The spread velocity of cane toads, for example, increased by a factor of 5 after the species was introduced to Australia, consistent with evolved changes in dispersal (141718). Nonetheless, with stochastic events contributing to the ecological and evolutionary trajectories of spreading populations (51921), replicated, controlled studies are necessary for understanding the predictability of this eco-evolutionary dynamic (15). Given the challenges of replicating invasions in the field and doing so in landscapes of varying fragmentation, model laboratory systems present an excellent opportunity to evaluate how evolution affects the speed at which populations expand through habitats of varying patchiness.

We manipulated evolution in populations of the model plant Arabidopsis thaliana spreading through continuous and fragmented landscapes, each consisting of a linear array of rectangular pots (Fig. 1A) (22). We initiated each replicate invasion in the leftmost pot of the array by sowing equal fractions of 14 genotypes (recombinant inbred lines), which varied in spread-relevant traits. Due to nearly complete self-pollination of A. thaliana (23), the 14 genotypes can be treated as clones (24), facilitating our measurements of evolutionary change. In evolving populations, the resulting plants produced seeds, which dispersed across the array (assisted via a simulated rain event), constituting the next generation of the population (Fig. 1B). In nonevolving treatments, germinants emerging in the next generation were replaced with individuals randomly drawn from the initial seed pool, thus maintaining population dynamics while eliminating any change in the frequency or spatial sorting of genotypes. We manipulated habitat patchiness by separating individual pots of suitable habitat by gaps that were 0 (continuous landscapes), 4, 8, or 12 times the mean dispersal distance. This protocol was repeated over six generations of spread, at which point individuals at the leading edge and back of the invasions were genotyped, and traits of all 14 genotypes were measured.

figure 1

Fig. 1. Spread of A. thaliana in experimental greenhouse arrays. (A) Leading edge of an invasion of a continuous landscape. (B) Spread in a continuous landscape for one replicate in the evolving treatment. Each colored line represents a successive generation (pink, founding population; red to purple from left to right, first to sixth generation of spread). Points show abundance in the individual pots that make up the arrays.

Experimental landscape

This picture shows how the experimental landscape was set up. The plants shown are modeling an invasive spread, meaning they are edging further and further from the original population.

Distance grown

The lines of this graph indicate the number of plants in each generation, and the distance they grew from the original population. The blues and purple lines are the latest generation. You can see that every color moves a little bit further along the x axis (distance from parents). The number of these plants growing far away from the original location also increases with every generation (y-axis counts the number of plants).

We found that after six generations of spread in continuous landscapes, evolving populations spread a modest 11% farther than nonevolving populations (Fig. 2A), a difference that was only marginally significant (t13.5 = –2.05, P = 0.060). By contrast, in experimental landscapes with gaps 12 times the mean dispersal distance, evolving populations spread three times as far as their nonevolving counterparts (Fig. 2D) (t10.4 = –3.36, P = 0.007), leading to a significant gap size by evolution interaction (F1,72 = 10.77, P = 0.002). The effects of evolutionary change were so strong in patchy landscapes that evolving populations showed no significant reduction in velocity as the size of gaps increased from 4 to 8 to 12 times the mean dispersal distance (generation-six location of dark green line in Fig. 2, B to D) (F1,25 = 0.014, P = 0.908), even as velocity slowed in the nonevolving populations (F1,28 = 8.52, P = 0.007). Patchiness and evolutionary change also influenced the among-replicate variability in expansion velocity (Fig. 2). The coefficient of variation for spread was four times greater in the patchiest landscapes than in the continuous ones (fig. S1), consistent with a spread process driven by infrequent long-distance dispersal events in fragmented systems. We also found that evolving populations showed significantly less among-replicate variation in spread than nonevolving populations (fig. S1). Thus, despite the theoretical expectation for greater genetic drift at the leading edge of spreading populations (25), invasion speed was more predictable in evolving populations.

figure 2

Fig. 2. Farthest distance colonized in each generation. Distance moved by evolving (thin green solid lines) and nonevolving (thin gray dashed lines) replicate invasions and their mean values (thick green and black lines, respectively) in landscapes that are (A) continuous or separated by gaps that are (B) 4, (C) 8, and (D) 12 times the mean dispersal distance. Lines in the three patchy landscapes are jittered for visibility.

Continuous Setting

After six generations, evolving populations only spread 11% farther than nonevolving populations which was only considered marginally significant.

Gap Setting

In experimental landscapes, the evolving populations spread three times as far as their nonevolving counterparts when gaps = 12 times. The effects of evolutionary change were so strong in patchy landscapes that they did not slow down as the size of gaps increased from 4 to 8 to 12 times.


One explanation for the greater effects of evolutionary change on spread velocity in patchier landscapes might be faster evolution due to stronger selection in these systems. However, the extent of genotypic change did not differ significantly with gap size (fig. S2 and table S1; Fig. 3shows the initial and final genotypic compositions), and the extent of trait change increased only marginally with increasing gap size (Fig. 3, fig. S2, and table S1). In fact, trait and genotypic change occurred in populations spreading through all landscape types, irrespective of whether evolution enhanced the spread velocity (significant intercepts in the fitted models of table S1). These evolutionary changes reflect the combined effects of selection and drift. In the continuously favorable landscapes in particular, we found more among-replicate variation in the genotypic composition of leading individuals than expected by chance (fig. S3), consistent with spatial priority effects where genotypes that initially got ahead due to chance dispersal were able to stay ahead (525).

figure 3

Fig. 3. Genotypes and traits at the invasion fronts. The central pinwheel of each panel depicts the equal frequency of genotypes in the founding population and is located at the mean trait rank for three spread-relevant traits: competitive ability (dominance in nonspreading context), dispersal (average distance of farthest dispersed seed from a solitary individual), and plant height. Pies show the genotypic composition of the 10 leading individuals for each replicate invasion after six generations of spread through landscapes that are (A) continuous or separated by gaps that are (B) 4, (C) 8, and (D) 12 times the mean dispersal distance. The location of each replicate is given by the genotype-weighted trait rank mean (22). A fourth trait, seed mass, also evolved, but its evolution did not vary with landscape patchiness and is not shown here. The central panel shows trait ranks of the 14 genotypes; numbers indicate genotype identity.

Continuous setting

Plants in a continuous setting were seen to have a large variety in the traits expressed with height ranging between 2 and 12, dispersal ranging between 2 and 14, and the competitive ability was between 2 and 13.

Gap setting X4

Plants in a setting with gaps 4 times the size of mean dispersal distribution had a more condensed grouping, with height ranging between about 4 and 13, dispersal ranging between about 4 and 10, and the competitive ability was between about 4 and 13.

Gap setting X8

Plants in a setting with gaps 8 times the size of the mean dispersal distribution show a grouping rather similar to condition (B), but had higher height and dispersal values. Height values ranged from between about 9 and 14, dispersal ranged between about 7 and 12, and the competitive ability was between about 4 and 13.

Gap setting X12

Plants in a setting with gaps 12 times the mean dispersal distribution show a grouping similar to that of condition (C) except for a lower dispersal value. Height values ranged between about 8 and 14, dispersal ranged between about 4 and 12, and the competitive ability was between about 4 and 13.

Despite similarities in the extent of trait and genotypic change across gap sizes, landscape patchiness affected the direction of evolution. Height and the average distance of the farthest dispersed seed, traits correlated with one another (Spearman rank correlation coefficient rs = 0.55, P = 0.046), increased with landscape patchiness (backward and rightward shift of the replicates with increasing patchiness in Fig. 3P = 0.008 and 0.060, respectively, Table 1). These trait changes were associated with changes in the genotypic composition of the leading individuals with increasing patchiness (Fig. 3) (F1,34 = 2.54, P = 0.042). Considering theory showing that greater dispersal increases the invasion velocity (610), the evolution of greater height and dispersal in patchier systems is consistent with the greater effects of evolution on spread in these landscapes. Nevertheless, whether landscape patchiness selected directly for better dispersal or indirectly via unmeasured traits that are correlated with dispersal remains an open question.

table 1

Table 1. Evolution of spread-relevant traits as a function of landscape patchiness. Results of linear models examining the change in height, dispersal, competitive ability, and seed mass at the invasion front after six generations of evolution as a function of landscape patchiness (size of gaps between suitable habitat). Trait change was measured as the difference between the genotype-weighted trait rank for each replicate (N = 36) and 7.5 – the mean trait rank of 14 genotypes in the founding population. Significant slopes indicate that the amount of change in the trait increased with increasing gap size (units of mean dispersal distance). Significant intercepts indicate that the trait changed significantly from the founding population, even in continuous landscapes. For competitive ability and seed mass, two traits with nonsignificant slopes, zero-slope models yielded highly significant intercepts (P ≤ 0.001). Est., estimated value.

Change in height & dispersal

Height and the average distance of the farthest dispersed seed increased as patchiness increased.

Change in seed mass

Increasing competitive ability had a significant effect on seed mass, which can be seen on the table by the intercept values. In this table, significant intercepts indicate that the trait changed significantly from the founding population even in the continuous landscape.  

Increased competitive ability probably also contributed to the greater effects of evolutionary change on spread velocity in patchier systems. Although competitive ability evolved to the same extent regardless of gap size [upward shift of replicates (Fig. 3 and Table 1); a similar result was found for seed mass (Table 1)], theory (511) predicts that increasing competitive ability will have a greater effect on spread in fragmented versus continuously favorable landscapes (fig. S4 shows this result applied to our system). In fragmented habitats, individuals often compete at crowded invasion fronts, enabling genotypes that make more offspring at high density (i.e., better competitors) to spread faster (511) (fig. S4). Though weaker, this effect also emerges in models of finite populations in continuously favorable landscapes (fig. S4) (26), consistent with the evolving populations moving modestly farther than the nonevolving populations in continuous landscapes (Fig. 2A).

Extrapolating our results to wild populations requires care for several reasons. First, the focal populations were effectively asexual, meaning that trait variation was not continuous and traits were perfectly linked. Nonetheless, it is not clear how more continuous variation or less linkage between traits would influence the effect of evolutionary change on spread velocity. Second, although we manipulated genetic change in this experiment, we cannot rule out the influence of maternal and epigenetic effects on our results. Third, we explored the effects of fragmentation, assuming it has no influence on the initial pool of genetic variation. If fragmentation in the nonspreading portion of a species range was to select for reduced dispersal (1627), then populations that spread from such sources might have less genetic variation in dispersal-related traits, limiting the response to selection. Related to this point, the effects of evolution in our study arose through drift and selection on standing variation; our results do not bear on the rates of evolution resulting from the rise of novel mutations.

Our results demonstrate that evolution on ecological time scales can increase the speed of advance in spreading populations, and markedly so in the most patchy landscapes. However, further studies are needed to evaluate whether patchiness per se generally selects for traits that increase spread (24). Our results for less patchy landscapes show that large evolutionary changes in spreading populations can have little or no consequence for spread velocity. More generally, our findings add a more process-focused perspective to past work that has shown either accelerating invasion fronts consistent with evolution (131517) or trait differences between individuals at the front and back of spreading populations (182829). We conclude that accounting for evolutionary change on ecological time scales may be critical when predicting the rate at which biological invasions and climate change migrants reach new locations.

Supplementary Materials

Materials and Methods

Figs. S1 to S4

Tables S1 to S3

References (3043)

References and Notes

  1. M. H. Williamson, Biological Invasions (Chapman & Hall, 1996).
  2. C. Parmesan, G. Yohe, Nature 421, 37–42 (2003).
  3. D. B. Lindenmayer, J. Fischer, Habitat Fragmentation and Landscape Change: An Ecological and Conservation Synthesis (Island Press, 2013).
  4. N. Shigesada, K. Kawasaki, E. Teramoto, Theor. Popul. Biol. 30, 143–160 (1986).
  5. J. L. Williams, R. E. Snyder, J. M. Levine, Am. Nat. 188, 15–26 (2016).
  6. O. J. Burton, B. L. Phillips, J. M. J. Travis, Ecol. Lett. 13, 1210–1220 (2010).
  7. R. Shine, G. P. Brown, B. L. Phillips, Proc. Natl. Acad. Sci. U.S.A. 108, 5708–5711 (2011).
  8. M. Kot, M. A. Lewis, P. van den Driessche, Ecology 77, 2027–2042 (1996).
  9. A. Hastings et al., Ecol. Lett. 8, 91–101 (2005).
  10. B. L. Phillips, G. P. Brown, J. M. J. Travis, R. Shine, Am. Nat. 172 (suppl. 1), S34–S48 (2008).
  11. E. Pachepsky, J. M. Levine, Am. Nat. 177, 18–28 (2011).
  12. A. D. Simmons, C. D. Thomas, Am. Nat. 164, 378–395 (2004).
  13. C. D. Thomas et al., Nature 411, 577–581 (2001).
  14. B. L. Phillips, G. P. Brown, J. K. Webb, R. Shine, Nature 439, 803 (2006).
  15. E. A. Fronhofer, F. Altermatt, Nat. Commun. 6, 6844 (2015).
  16. C. L. Hughes, C. Dytham, J. K. Hill, Ecol. Entomol. 32, 437–445 (2007).
  17. B. L. Phillips, G. P. Brown, R. Shine, J. Evol. Biol. 23, 2595–2601 (2010).
  18. T. A. Perkins, B. L. Phillips, M. L. Baskett, A. Hastings, Ecol. Lett. 16, 1079–1087 (2013).
  19. B. A. Melbourne, A. Hastings, Science 325, 1536–1539 (2009).
  20. T. E. X. Miller, B. D. Inouye, Ecol. Lett. 16, 354–361 (2013).
  21. B. L. Phillips, Biol. Invasions 17, 1949–1960 (2015).
  22. Materials and methods are available as supplementary materials on Science Online.
  23. R. J. Abbott, M. F. Gomes, Heredity 62, 411–418 (1989).
  24. S. Fakheran et al., Proc. Natl. Acad. Sci. U.S.A. 107, 19120–19125 (2010).
  25. L. Excoffier, M. Foll, R. J. Petit, Annu. Rev. Ecol. Evol. Syst. 40, 481–501 (2009).
  26. R. E. Snyder, Ecology 84, 1333–1339 (2003).
  27. J. M. J. Travis, H. S. Smith, S. M. W. Ranwala, Divers. Distrib. 16, 690–702 (2010).
  28. A. Chuang, C. R. Peterson, Global Change Biol. 22, 494–512 (2016).
  29. F. F. Kilkenny, L. F. Galloway, Evolution 67, 722–731 (2013).
  30. C. Alonso-Blanco, A. M. J. Peeters, M. Koornneef, C. Lister, C. Dean, N. van den Bosch, J. Pot, M. T. R. Kuiper, Plant J. 14, 259–271 (1998).
  31. C. Alonso-Blanco, H. Blankestijn-de Vries, C. J. Hanhart, M. Koornneef , Proc. Natl. Acad. Sci. U.S.A. 96, 4710–4717(1999). 
  32. N. Cappuccino, R. Mackay, C. Eisner, Am. Midl. Nat. 148, 263–270 (2002). 
  33. L. A. Turnbull, M. Rees, M. J. Crawley , J. Ecol. 87, 899–912 (1999). 
  34. M. L. Stanton, Ecology 65, 1105–1112 (1984). 
  35. P. G. Krannitz, L. W. Aarssen, J. M. Dow, Am. J. Bot. 78, 446–450 (1991). 
  36. J. S. Clark, M. Lewis, L. Horvath, Am. Nat. 157, 537–554 (2001). 
  37. J. J. Doyle, “DNA protocols for plants.” in Molecular Techniques in Taxonomy, G. M. Hewitt, A. W. B. Johnson, J. P. W. Young, Eds. 57, 283–293 (1991).
  38. M. Schuelke , An economic method for the fluorescent labeling of PCR fragments. Nat. Biotechnol. 18, 233–234 (2000). 
  39. M. Kearse, R. Moir, A. Wilson, S. Stones-Havas, M. Cheung, S. Sturrock, S. Buxton, A. Cooper, S. Markowitz,C. Duran, T. Thierer, B. Ashton, P. Meintjes, A. Drummond, Bioinformatics 28, 1647–1649 (2012). 
  40. A. C. Davison, D. V. Hinkley, Bootstrap Methods and Their Applications (Cambridge Univ. Press, 1997).
  41. A. Canty, B. Ripley, boot: Bootstrap R (S-Plus) Functions, R package version 1.3-18 (2016).
  42. J. Oksanen et al., vegan: Community Ecology Package (2016).
  43. R Core Development Team, R: A Language and Environment for Statistical Computing (R Foundation for Statistical Computing, 2016);

Acknowledgments: We thank S. Giovanettina, R. Guidon, A. Bieger, M. Clerc, and other members of the ETH Zurich Plant Ecology group for help with performing the experiments and F. Altermatt, A. Angert, T. Miller, and the Plant Ecology group for providing comments on the manuscript. J.M.L., J.L.W., and B.E.K. acknowledge support from the Swiss National Science Foundation (grants 31003A_141025 and IZK0Z3_163497), B.E.K. from the U.S. NSF (grant DEB-1120865), and J.L.W. from the Natural Sciences and Engineering Research Council of Canada. Data described in the paper are available from the Dryad Digital Repository: