Adaptation to new habitats might facilitate species’ range shifts in response to climate change. In 2005, we transplanted experimental populations of coastal dune plant Camissoniopsis cheiranthifolia into four sites within and one site beyond its poleward range limit. Beyond-range transplants had high fitness and often delayed reproduction. To test for adaptation associated with experimental range expansion, we transplanted descendants from beyond and within-range populations after 10 generations in situ into two sites within the range, one at the range edge, and two sites beyond the range. We expected to detect adaptation to beyond-range conditions due to substantial genetic variation within experimental populations and environmental variation among sites. However, individuals from beyond-range experimental populations were not fitter than those from within the range when planted at either beyond-range site, indicating no adaptation to the beyond-range site or beyond-range environments in general. Beyond-range descendants also did not suffer lower fitness within the range. Although reproduction was again delayed beyond the range, late reproduction was not favored more strongly beyond than within the range, and beyond-range descendants did not delay reproduction more than within-range descendants. Persistence in beyond-range environments may not require adaptation, which could allow a rapid response to climate change.

Study species

Camissoniopsis cheiranthifolia (Hornem. Ex Spreng.) W.L. Wagner & Hoch (Onagraceae) is a short-lived herbaceous perennial that inhabits open sandy patches of coastal dunes from San Quintin, Baja California, Mexico to Coos Bay, Oregon, USA. The species is common throughout its range, occurring in 87% of 124 dune sites surveyed (Samis & Eckert, 2007). The northern range limit in Oregon has been stable as long as herbarium and natural history records have been collected in this region (i.e., 150 yrs; Samis & Eckert, 2007), despite nearly continuous dune habitat extending along the coast into southern Canada. Individuals usually live 1–3 years, but ~ 80% of individuals reproduce and die in their first year (Samis & Eckert, 2009). The species lacks obvious mechanisms for dispersal, with tiny (~ 0.15 mg) seeds dropping from low-lying fruits and passively dispersing with blowing sand, or often germinating within or directly below a mature fruit ([authors], personal observation). Populations in the northern part of the range where this study occurred are highly self-fertilizing (~ 86% seeds self-fertilized; Dart & Eckert, 2013) and all belong to the same genetic subdivision within the species (López-Villalobos & Eckert, 2018).

Transplant experiment

To test for adaptation to beyond-range conditions, we executed a transplant experiment including descendant individuals of the original genetically mixed experimental populations from all five planting sites. The experimental populations differed in genotypic composition, with those at sites W133 and E0 and beyond-range site B60 including genotypes from all eight source populations, and those at sites W36 and W102 including genotypes from only the four northernmost source populations (Table S1).

In 2018, after each set of experimental populations had persisted for about 10–12 generations, we planted seed from descendants from each of the five sites (total n = 2951 seeds) in the glasshouse at Queen’s University, ON, Canada (Supp. Mat. S2). We then transported seedlings to Oregon and planted 7-wk old seedlings (n = 1081 seedlings) into five sites; two within the range (W133 & W36), one at the range-edge (E0), the original beyond-range site (B60), and another site 220 km beyond the northern range limit (B220; Fig. 1C; Table S4). We planted 3 – 19 individuals into each of 19 1.8 m² blocks of suitable habitat per site (suitable habitat defined as in Samis & Eckert, 2009). This experiment was run in parallel with an experiment using seedlings from natural source populations, leading to unequal numbers of transplants from experimentally mixed source populations per block. Both experiments were randomized together for a total of 31 or 32 transplants per block, with adjacent plants 30 cm apart (Cross, 2022). Surveying the original beyond-range site in 2019 revealed that most experimental populations had spread beyond transplant blocks into similar habitats (i.e., open sandy areas), further justifying our definition of suitable habitat. Blocks were not weeded, watered, protected from herbivores, or manipulated in any way. Naturally occurring C. cheiranthifolia were not removed, and 31.6% of within-range blocks contained at least one natural individual.

We monitored survival and reproduction during 20 visits to each planting site over 1321 days after transplanting, including visits at the end of each of four growing seasons (late November – mid-December, 2018 – 2021). During years 1 and 2, we visited blocks throughout the growing season and at season end, whereas in years 3 and 4, we could only sample at season end due to COVID-19 travel restrictions. All transplants had died by year 4, and we consider total fruit production across three seasons to be lifetime reproductive success (Fig. 2). Because the species is highly selfing in this part of its range, fruit production represents both female and male fitness. Due to the large scale of this experiment, we used open-pollinated seed, but non-genetic maternal effects (such as seed provisioning; Roach & Wulff, 1987) are unlikely to contribute to fitness variation among source populations of C. cheiranthifolia as variation in seed mass among maternal families did not correlate with progeny fitness in the original experiment (Samis, 2007) or this experiment (Cross, 2022).

Statistical analyses

All analyses used R v. 4.1.3 (R Core Team, 2022) through RStudio v. 2022.02.1.461 (RStudio Team, 2022), using the tidyverse package v. 1.3.1 (Wickham et al., 2019).

To test for adaptation to beyond-range conditions and potential tradeoffs (predictions 1–3), we fit variation in estimated lifetime fitness of individual plants using unconditional aster models (Shaw et al., 2008; aster pkg v. 1.1-2; Geyer, 2021). The models incorporated, for each year of life, whether an individual survived to reproduce or to season-end (Bernoulli; S = 0 or 1), whether it flowered (Bernoulli; F = 0 or 1), and how many fruit were set if it did flower (R = zero-truncated Poisson). Because plants were regularly surveyed throughout the first and second growing seasons, S1 and S2 indicates whether a plant survived to reproduce or to season end (see Fig. 2 for aster predecessor structure) and we used maximum fruit production observed during the growing season on each plant as R. However, we could only survey plants once in December 2020 due to COVID-19 travel restrictions, so S3 indicates survival to season-end, and it is possible that some plants survived to flower and perhaps even produce fruit but died before season-end. Moreover, season-end fruit number may not represent total fruit production if some plants lost reproductive branches due to damage. However, in years 1 and 2 season-end fruit number correlated strongly with maximum fruit reproduction (Pearson’s r = +0.85 in yr 1 and +0.96 in yr 2), so season-end fruit production is a reliable measure of total season fruit production in year 3.

To test for fitness variation of experimental populations across planting sites, we fit variation in fitness components (Fig. 2) to a mixed-effects aster model (using the reaster function; aster pkg v. 1.1-2; Geyer, 2021) with experimental source population, transplant site, and their interaction as fixed effects and transplant block as a random effect. The significance of each fixed effect was evaluated with a likelihood ratio test performed using the anova function in base R. A significant interaction indicated the potential for local adaptation (Blanquart et al., 2013) and motivated a comparison of source populations at each site separately (predictions 1–3). Means and standard errors for lifetime fruit production were estimated for each source population at each site using a fixed-effects aster model (using the aster function and not including the effect of block; C. Geyer, personal communication). When source populations varied at a site, we compared the B60 experimental descendants to those from sources within the range using pairwise comparisons with fixed-effects aster models (mixed effects models generated inappropriate Fisher matrices). Pairwise comparisons involved contrasting models with and without source population (df = 1) as a fixed effect using likelihood ratio tests as above. To account for the multiple comparisons, we held the Type I error rate at 5% using sequential Bonferroni. Higher fitness of B60 descendants compared to within-range descendants when planted at B60 would be consistent with adaptation to beyond-range conditions (prediction 1), and when planted at B220 with adaptation to beyond-range environments more broadly (prediction 2). At within-range sites, lower performance of B60 descendants compared to those from within-range experimental populations would be consistent with a cost to adaptation to beyond-range environments (prediction 3).

To test whether variation in reproductive timing impacted lifetime fruit production, we reanalyzed data from the 2005 experiment to investigate whether delayed reproduction was associated with higher fitness and whether any such association was stronger beyond than within the range (the first part of prediction 4). In 2005, most transplanted individuals that successfully reproduced were semelparous (80.7%), so most of the variation in timing of reproduction involved whether semelparous individuals flowered in their first or second year. In contrast, 2018 was very dry, which doubled pre-reproductive mortality (66.4% of 1077 plants in 2018 vs. 35.1% of 3782 plants in 2005) and delayed reproduction of those that survived. In 2005, 74.6% of reproductive plants flowered in their first year compared to only 5.8% in 2018. To quantify reproductive timing in a way that captured variation in 2018 and could be compared between experiments, we calculated, for each reproductive plant, the weighted year of fruit production Y_w = (F₁*1 + F₂*2 + F₃*3)/(F₁ + F₂ + F₃), where F_i is the number of fruits produced by an individual in year i. Y_w averaged (± 1 SE)  2.15 ± 0.017 in the 2018 experiment (range = 1–3, median = 2.00) and 1.35 ± 0.009 in the 2005 experiment (range = 1–2.99, median = 1.00; Table S3).

The effect of delaying reproduction on lifetime fruit production in 2005 was evaluated using two complementary analyses. First, we compared lifetime fruit production between the semelparous individuals that reproduced in their first vs. second year. At site E0, only five plants flowered in their second year, so this site was excluded from analysis. Individuals from the four Oregon source populations were planted at all five sites in 2005, whereas individuals from the four California source populations were only planted at three sites in 2005 (see Fig. 1), so we analyzed these two groups separately. Variation in lifetime fruit number was fit to a generalized linear model with negative binomial errors and reproductive timing, site, and their interaction as potential predictors using the glm.nb function in the MASS package (v. 7.3-60, Venables and Ripley, 2002). Second, we regressed lifetime fruit production over weighted year of fruit production (Y_w) using a generalized linear model with negative binomial errors (glm.nb) and evaluated whether this regression varied among planting sites by including site and the interaction between site and Y_w as potential predictors. For both analyses, the significance of predictors was evaluated with likelihood ratio tests using the Anova function in the car package (v. 3.1-2, Fox and Weisberg, 2019). For both analyses, the interaction term was significant, so we performed the analyses for each site separately.

We tested whether plants tended to delay reproduction when planted beyond the range in 2018 and evaluated the second part of prediction 4 that the descendants from B60 would exhibit delayed reproduction across all planting sites by fitting variation in Y_w to linear models with either site or experimental source population, respectively, as a potential predictor and significance evaluated as above.