Model Application
The validated model was used to predict broadscale patterns of climate
response in A. thaliana . This requires inferring the spatial
distribution of genetic variation and germination timing on a
continental scale. Both components are crucial because they dictate the
distribution of plant genotypes and the temperatures they experience,
respectively.
Inferring the spatial distribution of
Genetic Variation and Germination Date
We inferred the distribution of A. thaliana genetic variation
using kriging (Oliver & Webster, 1990), a method of interpolation used
in geostatistics for spatially autocorrelated data (Appendix S5).
Kriging was considered suitable because spatial autocorrelation inA. thaliana’s genetic variation was observed in our dataset
(average Moran’s I =0.146, P=0) and is consistent with isolation
by distance previously reported in the species (Platt et al., 2010;
Sharbel et al., 2000). We produced a kriged genetic landscape at 1°x1°
resolution across Europe by kriging each column of the GSM using the
autoKrige function from the automap package (Hiemstra, 2013) in R.
Across its European range, A. thaliana germinates at different
times of the year (Donohue, 2002). In order to determine the most likely
growing season of different sites, we used data from Exposito-Alonso
(2020) which identified k= 4 germination strategies defining
coherent climate regions. We smoothed boundaries by replacing the value
of outlier cells (those assigned to a different cluster from all its
neighbours) with the most common value in the 8 neighbouring cells. The
four regions (Central Europe CEUR, South Mediterranean SMER, North
Mediterranean NMER, Scandinavia SCAN) corresponded to three germination
seasons (spring, summer, fall). We assumed all plants germinated on a
single date for each season. These dates were February 27 for spring, 25
May for summer, and 3 October for fall and were chosen based on the
transplant dates of our plantings. These three dates were used to
predict trait changes across years within regions. For landscape
predictions in a given year, we allowed germination dates to differ
across individual cells (1 minute spatial resolution). We predicted
individual dates using the climate region, longitude, and latitude as
predictors in a linear model and transplant dates as the response
variable.
Projected Climate
Response
We first predicted climate responses across Europe to identify sites
that are susceptible to future decline under the RCP2.6 and RCP8.5
climate change scenarios (van Vuuren et al., 2011), with RCP8.5 being a
worse scenario. We obtained daily minimum/maximum temperature
projections for RCP2.6 from CCSM4 ensemble r1i1p1 (Gent et al., 2011)
and for RCP8.5 from CMCC-CM ensemble r1i1p1 (Scoccimarro et al., 2011).
Temperature rasters were resampled to 1°x1° to match the resolution of
the kriged genetic landscape using the R\raster package
(Hijmans et al., 2020).
We predicted DTB and SP from 2041 to 2099 using the RCP projections and
in 2006 using temperature records from E-OBS v19.0eHOM (Cornes et al.,
2018). We assumed a single genotype present in each cell (inferred
through kriging) and a constant germination date across years. Negative
values of SP were set to zero.
Finally, we emulate revegetation trials by using our model to predict
the performance of specific genotypes across Europe under climate
change. This allowed us to determine whether known genotypes could be
used as a source of climate-resilient genetic variation at sites where
the fitness of local populations was predicted to decline. As a
proof-of-concept, we focused on predicting the fecundity of the Eden-2
and Ll-2 genotypes in 2006 and 2099 under RCP8.5. Eden-2 is a Swedish
genotype that must be exposed to prolonged chill before flowering
(‘vernalization’,https://www.arabidopsis.org),
while Ll-2 originates from Spain and shows a low expression of the key
flowering repressor FLC (Rosloski et al., 2013); the two
genotypes were predicted to be the latest- and earliest-bolting of the
2029 genotypes, respectively.
Testing additional scenarios
The high levels of genetic diversity observed in A. thalianameans Ll-2 and Eden-2 are unlikely to represent the diversity of climate
responses in the species. The performance of other genotypes may be of
interest, and land managers may want to predict the effect of different
germination dates on trait values. In order to aid in the visualisation
of these multiple alternative scenarios, we have developed AraCast
(https://adaptive-evolution.biosciences.unimelb.edu.au/shiny/AraCast2/),
a shiny app that generates trait predictions for different genotypes,
germination dates, and climate change scenarios.
Results