Ecological experiments usually infer long-term processes from short-term data, and the analysis of geographic range limits is a good example. Species’ geographic ranges may be limited by low fitness due to niche constraints, a hypothesis most directly tested by comparing the fitness of populations transplanted within and beyond the range. Such studies often fail to find beyond-range fitness declines strong enough to conclude that geographic range limits are solely imposed by niche limits. However, almost all studies only follow transplants for a single generation, which will underestimate the importance of niche limitation because critical but infrequent range-limiting events may be missed and methodological issues may artificially boost the fitness of beyond-range transplants. Here, we present the first multi-generation beyond-range transplant experiment that involves adequate replication and proper experimental controls. In 2005, experimental populations of the coastal dune plant Camissoniopsis cheiranthifolia were planted at four sites within and one site beyond the northern limit. Fitness of initial transplants was high beyond the limit, suggesting that the range was limited by dispersal and not niche constraints. To better address the niche-limitation hypothesis, we quantified density and fitness of descendant C. cheiranthifolia populations 12–14 years (~10 generations) after transplant. Average annual fruit production and density of reproductive individuals were as high beyond the range as at four comparable experimental populations and eight natural populations within the range, and the beyond-range population had more than tripled in size since it was planted. This provides unprecedented support for the conclusion that northern range limit of C. cheiranthifolia results from something other than niche limitation, likely involving constraints on local dispersal.

Study species and original experimental design

Camissoniopsis cheiranthifolia is endemic to open sand patches on Pacific coastal dunes from Coos Bay, Oregon, USA, south to San Quintin, Baja California, Mexico. Herbarium records suggest that the northern range limit has been stable for as long as plants have been collected in this region (> 100 years; Samis and Eckert 2007).

In 2005, greenhouse-started seedlings were transplanted into replicated blocks at three sites within the range (W133, W102, W36), one site at the northern limit (E0), and one site 60 km beyond the range (B60). Seedlings were planted into 16 blocks at W133, E0, and B60 and were sourced from eight populations across the northern ~ 600 km of the species’ range, whereas seedlings at W102 and W36 were planted into eight blocks and sourced only from the four northernmost populations. The extensive planting at within-range sites provides an appropriate control for the beyond-range experimental population. There is a continuous dune complex from the northern range limit of C. cheiranthifolia past the selected beyond-range site, and the plant community composition at B60 was not substantially different from any of the within-range sites (Fig. 3A in Samis and Eckert 2009). Yet, there are no records of the species in the northern 60 km of these dunes, including at B60, over the last 150 years (Samis and Eckert 2007). B60 was chosen as representative of habitat beyond the northern limit, but close enough to the range limit to reasonably simulate a beyond-range colonization scenario. At each transplant site, 57–60 individuals were planted into 1.5 x 3.0 m blocks of seemingly-suitable habitat. Suitable habitat was defined according to Samis and Eckert (2007, appendix B) as stabilized to semi-stabilized dune behind the foredune but ahead of the deflation plain, supporting a community of other common dune plants, elevation < 50 m, ≥ 0.5 m from footpaths, and ≤ 75% plant cover. Neighbouring blocks at a site were separated by ≥ 10 m (range 10–160 m; Samis 2007) but were not self-contained, so seeds were free to disperse outside the blocks. Thus, for analyses we consider all blocks at a site to be one experimental population. At within-range sites, there were also naturally occurring C. cheiranthifolia, and individuals occurring within blocks were counted but not removed (range 0–58, median = 4 natural plants / block; Samis 2007). At within-range transplant sites, we measured 30 randomly chosen reproductive focal individuals (or 40 random individuals in 2018) within a 5-m radius of original blocks. At source sites, 30 (or 40 in 2018) individuals were randomly sampled at least 5 m apart. At the beyond-range site, we randomly sampled 30 (or 40) individuals from any persisting plants as all were descended from original transplants.

Sampling

At five intervals throughout summer 2017 (June 11– Aug 8) and at the end of the growing season in 2017, 2018, and 2019, we quantified size and reproduction of plants persisting at all five transplant sites and all eight natural source sites. We measured plants throughout summer 2017 to see if the timing of growth or reproduction varied among sites (which may have demographic consequences), but we did not detect any trends different from season-end sampling, and thus only sampled at season-end in 2018 and 2019. Transplant and source populations at a site were separated by 1–2 km. Individuals were sampled by walking 10 steps in a random direction, pointing in a random direction, and choosing among the plants in that vicinity with a random number. At within-range transplant sites, we measured randomly chosen reproductive focal individuals within a 5-m radius of original blocks. At source sites, individuals were randomly sampled at least 5 m apart (following Samis and Eckert 2009). At the beyond-range site, we randomly sampled individuals from among any persisting plants as all were descended from original transplants. There were naturally occurring C. cheiranthifolia at within-range transplant sites, so individuals we sampled likely included descendants from both natural and experimental individuals with possible crossing between them.

Density and reproductive fitness

At the five within-season intervals throughout summer 2017 (June 11– August 8), we counted vegetative and reproductive individuals within a 1-m radius (3.14 m²) of each of 30 randomly chosen focal plants. A new random sample of plants was measured each round. This local density estimate correlates positively with site-level measures of C. cheiranthifolia density (Samis and Eckert 2007, Fig. 2). In July 2019, we counted total vegetative and reproductive individuals at the beyond-range site to estimate population growth from initial transplanting.

We measured reproductive output as the sum of buds, flowers, and developing or mature fruit for 30 reproductive individuals at each site (all 8 source and 5 transplant sites) five times throughout summer 2017, as well as in October 2017 and November 2019 (year-end sampling). In November 2018, we sampled 10–33 reproductive individuals (mean = 19.9) per site. We refer to within-season reproductive output as “estimated fruits / plants” for simplicity, but note that this measure is a more accurate representation of final fruit output as the season progresses.

 Late season reproductive output approximates lifetime reproductive fitness as 80% of individuals at these sites are semelparous (Samis and Eckert 2009) and 88% of individuals were finished flowering at year-end sampling (2017–2019). Fruits persist on plants after dehiscing, so are accurately counted. Remaining buds and flowers are presumed to become fruits, as fruit-set is near 100% in these highly-selfing populations (Dart and Eckert 2015). We collected the 1–3 most recently-matured fruits per measured plant (30–92 fruits / population) in 2017 and counted seeds / fruit to test for variation in seed production. This sampling approach does not include individuals that died before flowering or failed to become reproductive and thus had zero fitness (i.e. the “invisible fraction”; Grafen 1988). However, the prevalence of nonreproductive individuals is readily evaluated from our density measures (above). Given the large number of individuals at each site (~1000s) and high per-capita fruit production (mean = 19.7 fruits / plant), it is unlikely that sampling negatively affected the demography of study populations.

Statistical analysis

All analyses used R version 3.6.1 (R Core Team 2019) with the tidyverse package (version 1.2.1; Wickham 2017). For generalized linear model (GLM) analyses of nonbinomial response variables, we selected the error distribution (Poisson, quasi-Poisson, or negative binomial) that best fit the data based on producing a dispersion statistic closest to one, which was always the negative binomial (glm.nb function in MASS R package; version 7.3-51.4; Venables and Ripley 2002). When responses varied significantly among sites, we contrasted the beyond-range site (B60) with each within-range transplant site using Dunnett’s test, which is commonly used to contrast a single group (B60) to other groups (within-range transplant sites; “trt.vs.ctrl” contrast in emmeans package version 1.4.4; Lenth 2020).

To control for the potential effects of mixing genetically differentiated source populations on the fitness of plants descended from experimental transplants, we compared B60 to the four within-range transplant sites. This formal comparison is justified by a comparison of paired transplant and source populations at the four within-range sites, which showed that plants at transplant sites often produced fewer fruit and seed than those at nearby unmanipulated source sites, but are usually more abundant (Appendix S2). To put the fitness and density of plants at B60 in a broader context, we also compared B60 to the eight within-range source sites.

We fit variation in 2017 within-season reproductive output to a negative binomial GLM with transplant site, sampling round, and their interaction as fixed effects. We also compared plant density and proportion of individuals in the 1-m radius plots that were reproductive between B60 and the four within-range transplant sites by fitting mean density in 2017 to a negative binomial GLM and mean proportion reproductive to a binomial GLM. We used the same analysis to compare B60 to the eight natural within-range source sites (Appendix S3).

We fit variation in season-end fruit production to a negative binomial GLM with transplant site, year, and the interaction as fixed effects. If the interaction was significant, we analyzed the effect of site for each year. We used the same analysis to compare B60 to the eight natural within-range source sites. Fruit production captured most of the variation in total seed production, as estimated seeds / plant correlated strongly with fruits / plant in 2017 (Pearson r = +0.87, n = 145,  P < 0.0001; Appendix S4).

Dart, S., and C. G. Eckert. 2015. Variation in pollen limitation and floral parasitism across a mating system transition in a Pacific coastal dune plant: Evolutionary causes or ecological consequences? Annals of Botany 115:315–326.

Grafen, A. 1988. On the uses of data on lifetime reproductive success. Pages 454–471 in T. H. Clutton-Brock, editor. Reproductive success. University of Chicago Press. Chicago, IL, USA.

Lenth, R. 2020. emmeans: Estimated marginal means, aka Least-squares means. R package version 1.4.4. https://CRAN.R-project.org/package=emmeans

R Core Team. 2019. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/

Samis, K. E., and C. G. Eckert. 2007. Testing the abundant center model using range-wide demographic surveys of two coastal dune plants. Ecology 88:1747–1758.

Samis, K. E. 2007. The evolution of species’ geographical range limits: an empirical evaluation using two coastal dune plants, Camissonia cheiranthifolia (Onagraceae) and Abronia umbellata (Nyctaginaceae) [Doctoral dissertation, Queen’s University, Kingston]. QSpace: Queen’s Scholarship & Digital Collections.

Samis, K. E., and C. G. Eckert. 2009. Ecological correlates of fitness across the northern geographic range limit of a pacific coast dune plant. Ecology 90:3051–3061.

Venables, W. N., and B. D. Ripley. 2002. Modern applied statistics with S. Fourth Edition. Springer, New York.

Wickham, H. 2017. tidyverse: Easily install and load the 'Tidyverse'. R package version 1.2.1. https://CRAN.R-project.org/package=tidyverse

This dataset follows up on an experiment done in 2005 and previously published (see Samis and Eckert 2009). 

Samis, K. E., and C. G. Eckert. 2009. Ecological correlates of fitness across the northern geographic range limit of a pacific coast dune plant. Ecology 90:3051–3061.