The genetic architecture of local adaptation II: The QTL landscape of water-use efficiency for foxtail pine (Pinus balfouriana Grev. & Balf.)

Water availability is an important driver of the geographic distribution of many plant species, although its importance relative to other climatic variables varies across climate regimes and species. A common indirect measure of water-use efficiency (WUE) is the ratio of carbon isotopes (δ13C) fixed during photosynthesis, especially when analyzed in conjunction with a measure of leaf-level resource utilization (δ15N). Here, we test two hypotheses about the genetic architecture of WUE for foxtail pine (Pinus balfouriana Grev. & Balf.) using a novel mixture of double digest restriction site associated DNA sequencing, species distribution modeling, and quantitative genetics. First, we test the hypothesis that water availability is an important determinant of the geographical range of foxtail pine. Second, we test the hypothesis that variation in δ13C and δ15N is genetically based, differentiated between regional populations, and has genetic architectures that include loci of large effect. We show that precipitation-related variables structured the geographical range of foxtail pine, climate-based niches differed between regional populations, and δ13C and δ15N were heritable with moderate signals of differentiation between regional populations. A set of large-effect QTLs (n = 11 for δ13C; n = 10 for δ15N) underlying δ13C and δ15N variation, with little to no evidence of pleiotropy, was discovered using multiple-marker, half-sibling regression models. Our results represent a first approximation to the genetic architecture of these phenotypic traits, including documentation of several patterns consistent with δ13C being a fitness-related trait affected by natural selection.


52
Descriptions of the genetic components underlying fitness-related phenotypic variation 53 have been a focus of quantitative genetics for over a century (Shull 1908;Fisher 1918

107
Water is crucial to the survival of many plant species (e.g. Sorenson 1983), although its 108 importance relative to other environmental factors varies depending upon the environmental 109 factors that are most limiting within local environments (Dudley 1996). The intrinsic efficiency by 110 which plants use water (WUE) is defined as the ratio of net assimilation of carbon from CO 2 111 during photosynthesis to the loss of water during transpiration (Bacon 2004). Carbon isotopic 112 composition (δ 13 C) is an indirect measure of intrinsic WUE and is based upon the ratio of two 113 isotopes of carbon ( 13 C and 12 C) within plant tissue standardized to a reference. This ratio is 114 related to WUE because it has been demonstrated that the discrimination by C 3 plants of 13 CO 2 115 relative to 12 CO 2 is correlated to the ratio of carbon assimilation during photosynthesis to 116 stomatal conductance (Farquhar et al. 1982;Farquhar and Richards 1984; e.g. Zhang and 117 Marshall 1994). The physiological and environmental mechanisms, however, driving the linkage 118 between δ 13 C and intrinsic WUE at various levels of biological organization are numerous, so 119 that the expected linear relationship between δ 13 C and WUE may not always hold (Seibt et al.

132
Water availability is often an important driver of tree distributions (Stephenson 1990

171
(2015). Of these 141 maternal trees, offspring, assumed to be half-siblings, from five were 172 selected for analysis of water-use efficiency (see Phenotype determination, Table 1). The 173 megagametophyte associated with each germinated seed from these five maternal trees was 174 rescued and used to construct a high-density linkage map based on four of the five maternal 175 trees (Friedline et al. 2015). The seedlings from each maternal tree were allowed to grow for a 176 full year after which needles were sampled (n = 32 to 40/maternal tree) for determination of 177 phenotypes and genotypes. As done by Friedline et al. (2015), families were named using 178 colors (i.e. these were the colors of family identifier tags in the common garden), with families Genetic architecture of water-use efficiency 8 sampled from the Klamath Mountains being labeled as blue, yellow, and purple and families 180 sampled from the southern Sierra Nevada being labeled as red and green.

182
Two phenotypic traits were measured from needle tissue sampled from each growing 183 seedling -carbon isotope discrimination (δ 13 C) and foliar nitrogen content (δ 15 N). These were 184 chosen because (δ 13 C) is a proxy for intrinsic WUE (Farquhar et al. 1982; 185 Richards 1984), while δ 15 N is a proxy for plant growth and resource utilization during 186 photosynthesis (Prasolova et al. 2000). Tissue was sampled in year 1 of growth, which was also 187 prior to formation of randomized blocks in the common garden. Given the age of the seedlings,

215
Raw FASTQ sequences were quality-checked and filtered as in Friedline et al. (2015).

216
Briefly, reads must pass a three-stage filtering procedure to be retained for downstream 217 analysis. First, if the average quality for all bases in the read was below 30, the read was 218 discarded. Second, a five-base pair sliding window was evaluated along each raw sequence.

219
Consecutive windows were retained if their mean quality was greater-than or equal-to 30. If the 220 mean score of a window fell below this threshold, the read was trimmed at this point. If the 221 length after trimming was at least 50% of the original read length, the read was kept, otherwise

248
We used species distribution models (SDMs) to justify water-use efficiency as a fitness-249 related trait and to quantify niches of each regional population relative to one another. The

250
former provides an a priori justification for the measured traits as ecologically relevant, while the 251 latter provides an estimate of niche differentiation between regional populations comparable to 252 the effect of region on trait differentiation (see Quantitative Genetic Analysis).

253
Species distribution models were used to assess the relative importance of precipitation-

259
When the latitude and longitude of locations associated with these herbarium records were 260 missing, visual inspections of maps from Google Earth were used to find the best approximation 261 to the locality described on the herbarium sample. Climate data for each regional population 262 were obtained from WordClim (http://www.worldclim.org/) and are represented as 19 bioclimatic 263 variables, which are functions of temperature and precipitation variables (Table S1)

276
We used also used SDMs to quantify niche differentiation between regional populations 277 of foxtail pine (Warren et al. 2008). We tested two null hypotheses. First, we tested the null 278 hypothesis that the two SDMs were based on a single, underlying SDM common to each 279 regional population. Second, we tested the null hypothesis that the two SDMs are no more 280 differentiated than those randomly drawn from a common SDM with non-overlapping 281 geographical distributions for each regional population. Both tests are based on the D and I

293
The genetic basis for each measured trait was assessed using linear models. We fit

376
Bioclimatic variables used to predict occurrences of foxtail pine within each regional 377 population were highly correlated with one another ( Figure S7)

407
Predicted niches based on SDMs for each regional population were dissimilar, with 408 estimates of D (0.072) and I (0.258) being much closer to zero (dissimilar) than to 1 (similar)

409
( Figure S15). These differences were significant enough to reject a null model of a single shared 410 SDM common to both regional populations (P < 0.01 for D and I). Even if differences were 411 accounted for in the background environments of each regional population ( Figure S5)

448
statistic at a resolution of 6 cM or less for δ 13 C and 3 cM or less δ 15 N ( Figure S16), but there was 449 no correlation between F-statistics for each trait (Pearson's r: -0.014, P = 0.734; Figure S17). In 450 general, 95% confidence levels of positions for each QTL were large (Table 5).

451
For the 11 QTLs detected using one-locus models, 10 were consistent with multiple 452 QTLs using two-locus models (Table 6). In general, the QTLs from the one-locus models were

466
QTL effects from the one-locus QTL models were consistent with differentiation between 467 regional populations, with family effects opposite in sign more often than expected by chance for

525
this suggests that water may be more limited throughout the year (e.g. summer drought) than 526 expected based on annual precipitation totals. Additional work, however, would be needed to 527 quantify trait variation within each regional population and correlate it to both climate and soil 528 characteristics.

529
Genetic architecture of water-use efficiency

530
Both δ 13 C and δ 15 N were consistent with non-zero heritabilities. Families and regions 531 accounted for approximately 50% of the total phenotypic variance for δ 13 C and 30% for δ 15 N.

532
Models with effects due to families or families nested within regions were also strongly preferred 533 over models without these effects (Table 4). The effect of region, however, was highest in 534 magnitude for δ 13 C, with the variance component for region larger than that for family. This is consistent with previous estimates of quantitative genetic parameters for these phenotypic traits

587
Sampling more families, measurement of other traits (e.g. growth), and experimentation in 588 multiple environments, however, would be needed to test these ideas. Importantly, δ 13 C should 589 be measured within natural populations to assess correspondence between inferences from 590 common gardens and natural populations.

591
Using one-locus QTL models, the observed segregating genetic variance for δ 13 C was 592 dissected into two major QTLs and four suggestive QTLs (

603
water availability is an important driver of its current geographical distribution and genetic 604 structure is moderate to high between regional populations and among stands within regional QTLs were consistent with differentiation among regions, so it is plausible that the architecture 607 discovered here for δ 13 C largely represents genomic regions underlying trait divergence 608 between the regional populations. If this is the case, this architecture has evolved since the divergence of the regional populations from their common ancestor on the order of one million 610 years ago ).

611
Summaries of the results from two-locus QTL models were largely consistent with those 612 from the one-locus models. For the 11 QTLs reported in Table 5, 10 were consistent with at 613 least two segregating QTLs. This brings the total number of QTLs to four major and seven 614 suggestive QTLs for δ 13 C and two major, four minor, and four suggestive QTLs for δ 15 N.

615
Interestingly, the correlation of family-level effects for the two QTLs on the same linkage group 616 was negatively related to the distance between these QTLs, so that QTLs close together tended

636
We leveraged the annotations of contigs at or near (± 3 cM) the estimated QTL positions 637 to search for putatively functional genes as the drivers of the genotype-phenotype correlations 638 for each QTL (Table S3)

663
We have used a mixture of species distribution modeling and quantitative genetics to 664 test two hypotheses about WUE, as measured by δ 13 C, for foxtail pine. We showed that 665 precipitation-related variables structured the geographical range of foxtail pine, that climate-666 based niches differed between regional populations, and that similar patterns were apparent for 667 δ 13 C, which was also demonstrated to be heritable. We subsequently dissected this heritability 668 into a set of large-effect QTLs (n = 21 total, with 11 for δ 13 C and 10 for δ 15 N), which we interpret 669 in light of population genetic theory about local adaptation. While we cannot definitely say that 670 WUE, as measured by δ 13 C, contributes to local adaptation, we have described to a first 671 approximation its genetic architecture, while noting several patterns consistent with δ 13 C being a

692
• Genotype data are available as raw short read data as part of NCBI BioProject 693 PRJNA310118, processed short read data in VCF format (File S1), and imputed data in 694 a tab-delimited text file (File S2).

695
• Phenotypic trait data are available for all half-siblings within each of the five families 696 used for QTL mapping in a tab-delimited text file (File S3).

697
• Location data used for species distribution modeling are available in a tab-delimited text 698 file (File S4).