Adaptation to drought is coupled with slow growth in marginal silver fir (Abies alba Mill.) populations

Drought is increasingly considered as the most important selection pressure for forest trees in the context of climate change. We studied adaptation to drought in marginal populations of silver fir (Abies alba Mill.) from the French Mediterranean Alps. Drought tolerance was assessed using proxies both from seedlings and adult trees. We measured water stress response, growth and bud break of seedlings originating from 16 populations in a greenhouse common garden experiment (N=8199) and water use efficiency via δ13C of adult trees of the source populations in-situ (N=315). Further, 357 single nucleotide polymorphisms (SNPs) were used to uncover the demographic history of the populations. Demographic distances between populations were used to generate a null expectation for trait divergence, thereby detect the signature of natural selection. We found evidence for adaptive population divergence in drought tolerance across life stages. Seedlings originating from source populations with low soil water capacity resisted better to water stress in the greenhouse, and additionally, adult trees from these populations had a higher water use efficiency. Seedling growth showed an evolutionary trade-off with drought tolerance: seedlings with fast growth and high stature came from populations that had lower drought tolerance. In contrast, population divergence in bud break showed only a weak signal of adaptation, which was independent of that in drought tolerance. Variation in phenology between populations was associated with variance in temperature and drought frequency and severity at the source populations. Our results highlight the adaptive value of marginal populations, advance our understanding of the different processes that have allowed silver fir to cope with drought stress under a warming climate, and contribute to our knowledge to advise assisted migration programs.

Introduction grouped in plastic crates by sets of 32. Each mother tree (subsequently family) was represented by 16-20 seedlings. The experimental design consisted of two greenhouses each divided into 162 two complete blocks. Populations, families and seedlings were randomized across greenhouses 163 and blocks. Greenhouse 2 was exposed to a water stress treatment that involved a complete lack 164 of watering starting from 6 May 1998. Greenhouse 1 received watering for the whole duration 165 of the experiment. 166 Traits were recorded starting from the 2 nd growing season (i.e. from 1997) to the 4 th (i.e. 167 until 1999) ( Table 1). We analyzed a total of 8199 observations, of which 3931 were in green- 168 house 1 and 4267 in greenhouse 2. Growth Increment and Height were used as raw measure-169 ments (trait names are capitalized hereafter; see Table 2 for trait definitions). Spring bud break 170 phenology was scored at two dates each year from which we calculated a Bud Break Score (see 171   Table 2). Bud Break Score ranged between two and 10, where higher numbers indicate earlier 172 bud break. Response to water stress was scored at five consecutive dates in 1998 (Table 1). We 173 used a derived trait, Water Stress Score, to characterize response to water stress, which ranged 174 between zero and 10 (see Table 2). Low values indicate drought-hardy seedlings and high val-175 ues indicate drought sensitive seedlings. The distribution of the Water Stress Score was highly 176 skewed and zero inflated (see Water Stress Score (raw sum) on Fig. S1). Thus, we calculated 177 an integrative measure of water stress by weighting the stress scores with the log Julian dates 178 of the observations. This weighted Water Stress Score had a close to Normal distribution (see Table 1: The time-line of the common garden experiment performed in the experimental forest nursery located in Milles (see Fig. 1). See Table 2   Top of the needles broke through the bud 5 Needles are completely free and grow Growth Increment Height difference between spring and fall in mm Height From soil to highest point in mm Water Stress Score Sum of scores recorded at five consecutive dates: 0 No visible sign of stress 1 The side shoots point downwards 2 The terminal shoot is not upright 3 Needles are yellow across the whole plant values of Tajima's D between 2 and -2 and dN/dS between 0.9 and 1.1, and with low LD with 189 the existing SNPs (r 2 < 0.1 and p-value > 0.05). Further, we selected 149 SNPs from the control   greenhouses were considered), and seed weight, population, and seedling as random effects, 256 and a random residual error for each seedling. V A was estimated from the covariance between 257 relatives, for which, we reconstructed a pedigree assuming seedlings from the same mother tree 258 to be half-sibs. We implemented the model using the R package ASreml-R 3.0, and constructed 259 the inverse kinship matrix using the function asreml.Ainverse. 260 We performed several tests to check the validity of the above full model. We tested if 261 environmental heterogeneity generated by the experimental designed contributed to the trait 262 variance by comparing models with and without block effect using a Wald-test (wald.asreml). 263 We also tested for the significance of the random effects using a likelihood ratio test between 264 models with and without seed weight and population, and also tested significance of including 265 the pedigree itself (referred to "family" effect). Finally, the above model assumes that the 16 266 populations have a common V A . This assumption may not hold because the populations were 267 selected from an environmentally and perhaps demographically heterogeneous region. In order 268 to test the hypothesis of heterogeneity in V A across the study region, we repeated the above 269 analysis for the three main genetic clusters that the above demographic analysis identified. The . TAR and PUN were excluded from this regional 273 analysis because they did not belong to any of the three clusters ( Fig. 1).

274
Since the variance components are largely unaffected by the inclusion of model terms that 275 do not contribute to the trait variance, we used the full model to estimate the V A for each trait. Finally, note that driftsel uses a Bayesian approach, so we will use "unusual" and 296 not "significant" to indicate results with strong evidence of adaptive divergence. 297 We ran three independent chains for each trait with a burn-in of 100,000 iterations fol-

Results
Population structure and demographic history isolation-by-distance both from east to west and from south to north (Fig. 1a). Interestingly, the two populations that were the geographically closest (TAR and ISS) and situated in the middle 327 of the sampling range, were one of the most strongly differentiated population pairs.

328
Convergence was good using RAFM, the mean potential scale reduction factor (Gelman

329
& Rubin, 1992) was one across the ten chains and ranged between 0.9 and 1.1. Using the  (Table 4).

340
For three traits, Bud Break Score 1997 and 1998, and Growth Increment 1997, we were able to 341 compare the effect of the greenhouse. We found that variance components were similar when 342 taking all data or just one of the greenhouses, with slightly higher estimates for one greenhouse 343 only (Table 4, Fig. 2). The effect of year on the traits was addressed indirectly using Bud Break 344 Score and Height that were measured across three years of the experiment in greenhouse 1. 345 We found that seedlings most likely acclimated to the experimental conditions: the block effect 346 disappeared for Bud Break Score 1998 and 1999 and for Height 1998. Further, we observed that 347 the heritability decreased with years, most likely also due to acclimation, which made families 348 more similar to each other (Fig. 2). However, it is also possible that maternal effects, other than 349 seed weight, have contributed to an inflated heritability in the first year of the experiment, i.e.

351
Family (i.e. the pedigree) always explained a large percentage of the trait variation and a 352 model with pedigree was always better than a model without it (Table 4). All traits expressed 353 Table 4: Model comparison of 11 silver fir (Abies alba) seedling traits measured in a common garden using an animal model implemented in ASreml-R. Seedlings were grown in two greenhouses each comprising two blocks. Populations and families were randomized across greenhouses and blocks. N indicates the number the observations. The role of Block (fixed effect) was tested using a Wald-test, while the role of Family (i.e. the pedigree), and Seed weight and Population (random effects) were tested using a likelihood ratio test by excluding each of these variables one-by-one. Height (Fig. 2). We tested for the effect of seed weight to assess if genetic or non-genetic 358 maternal effects explain trait variation. Seed weight explained a negligible (at the maximum 359 0.027%, Fig. S3), yet significant part of trait variation in growth traits, i.e. Growth Increment 360 and Height, but not in phenology (Table 4). Population of origin also explained a significant 361 part of trait variation (Table 4). The proportion of variance explained by population varied 362 between 9.8% (Height 1998) and 0.7% (Bud Break Score 1998) with a mean of 5% across traits 363 (Fig. S3). Finally, the regional analysis, where we considered populations only from one of 364 the demographically most uniform clusters, revealed the same relative differences between the 365 traits in h 2 and CV A , however, estimates were often higher (i.e. inflated) most likely owing to a 366 smaller sample size (Fig. S4).   Table 3.
Next, we asked the question that in which directions populations evolved from a hypothet-  (Fig. 4a-d). We detected unusually early bud break for TAR, 385 BAY, BOS and VTX in 1997 (Fig. 4a). Typical of a continental-type climate, these popula-386 tions experienced the highest variance in temperature (PC1) at their environment of origin, the 387 longest drought periods and winter precipitation in the form of snow (Table 3, Fig. 3). PUN, BRG, PES had the latest bud break, but their additive genetic trait values were within a range 389 that could be expected based on neutral demographic processes alone (Fig. 4a). Nevertheless, 390 their climate was the opposite of the populations with the earliest bud break, and characterized 391 by low temperature variance and lack of prolonged drought periods (Table 3).

392
Five to eight populations, depending on the greenhouse, showed unusually low Growth BRG, and SET (Fig.4c). The fact that only one population showed unusual trait divergence for 396 Height in 1998 could be due to the lower sample size in this year (see Table 4). we performed a climate led to enhanced growth (Fig. 3, Table 3). were more resistant to water stress, while populations that evolved towards a large stature were 409 the least resistant to water stress (Fig. 4e). Additional evidence for a correlated evolution is that 410 drought tolerance was influenced by a similar set of environmental variables as the growth traits 411 (Fig. 3). Finally, the PCA analysis also suggested that spring phenology evolved independently 412 from the growth-drought tolerance trait complex (Fig. 4e), which is also supported by the fact 413 the two groups of traits seem to have responded to different environmental cues (Fig. 3).  Finally, we were able to corroborate the evidence for adaptive divergence in drought tol-415 erance by comparing observations on seedlings and adults. We estimated drought tolerance 416 in adults using δ 13 C, a proxy for water use efficiency. We found that the populations' me-417 dian δ 13 C was strongly correlated with seedling's mean additive trait value for the Water Stress conditions. The main environmental drivers of drought tolerance in seedlings were soil water 422 capacity (PC3) and, to a lesser extent, mean temperature (PC2) (Fig. 3, Table 3). Fig. 5 also 423 illustrates, Available Water Capacity at 30m, which had the highest loading on the synthetic 424 environmental variable, soil water capacity (PC3) (Table 3). Indeed, even a simple visual in-

Discussion
Drought tolerance at the southern range margin 430 We found evidence for adaptive divergence between silver fir populations from the southern 431 margin of the species distribution range for drought tolerance and growth traits. Further, we 432 identified a trade-off between growth and drought tolerance: seedlings originating from popu-433 lations with relatively slow growth and the small stature were the most drought tolerant both in 434 terms of adult and seedling traits (Fig. 4, Fig. 5

455
The main environmental driver of drought tolerance and growth were soil water capacity 456 and mean temperature ( Fig. 3 and Fig. 5) phenology. Indeed, we found weak evidence for adaptive divergence in bud break approxi-470 mated using a Bud Break Score (Fig. 4), which suggests the role of phenotypic plasticity. The  Adaptive divergence in bud break was independent of that in the growth-drought tolerance 480 trait complex (Fig. 4e). As we argued above, many aspects of drought tolerance can be limited was explained by family than this study (Latreille & Pichot, 2017). Heterogeneity was also 502 introduced via the orientation of the greenhouse: greenhouse 1 was exposed to more wind,  Silver fir is a predominantly outcrossing species, thus we assumed that all seedlings from 511 the same mother tree are half-sibs. However, the mating system in silver fir is likely more com- that facing drought Scots pine is taken over by pubescent oak.

543
Our results suggest that rapid adaptation to drought conditions could be possible suggested 544 by the high heritability and evolutionary potential for water stress response (Fig. 2). However, 545 we also found that the evolution of drought tolerance is linked to growth (Fig. 5e)       . Two genetically isolated populations, TAR and PUN, were excluded from this analysis. Not all traits were scored in both greenhouses. Greenhouse 2 received a water stress treatment in 1998 after which only Water Stress Score was recorded. Parameters were estimated using an animal model implemented in ASreml-R including block as a fixed effect, and seed weight and population as random effects.  Fig. S5: The lower bound on heritability (h 2 ) and additive genetic coefficient of variation (CV A ) estimated assuming that all seedlings were issued from selfing. 11 traits measured on silver fir (Abies alba) seedling in a common garden. The experiment consisted of two greenhouses and counted 8199 observations. Not all traits were scored in both greenhouses. Greenhouse 2 received a water stress treatment in 1998 after which only Water Stress Score was recorded. Parameters were estimated using an animal model implemented in ASreml-R assuming including block as a fixed effect, and seed weight and population as random effects.