Climate sensitivity of Cryptomeria japonica in two contrasting environments: Perspectives from QTL mapping

each month (S2 Fig). The magnitude of drought stress on sampled populations can change with the length of the dry period [i.e., short dry periods (<1 day) may have no effect on tree growth]; thus, dry periods were accumulated for each month with varying thresholds, ranging from 1 to 7 days. As setting a higher threshold of dry days produces a larger ratio of zero data (i.e., raining days) per month, the largest threshold was set as 7 days to make the ratio of zero data less than 50% (S3 Fig) to avoid inaccurate estimation of climate sensitivity (described below). Climate data described above were linearly detrended prior to the analysis. 2.4. Climate sensitivity traits Based on the methods of [17], climate sensitivity traits were obtained by calculating correlation coefficients between climate data (i.e., temperature, precipitation, and dry period) and relative BAI scores. Monthly temperature, precipitation, and dry period during the experimental years were used for the climate sensitivity traits. Correlation coefficients were calculated for the months starting from June of the year prior to tree-ring formation until September of the year of tree-ring formation. 2.5. QTL analysis For each study site, QTL analyses of climate sensitivity traits in the sampled population were conducted using the Bayesian Lasso linear regression model: y ¼ μI þ Xβ þ e where y is climate sensitivity traits, μ is the overall mean, I is the identity matrix, X is the geno- type of each marker, β is the effect of the genetic marker on climate sensitivity traits, and e is

the error. Heterozygous and homozygous marker genotypes were converted to 0 and 1, respec- tively. The Bayesian Lasso model computation was performed using the Empirical Bayesian Lasso algorithm (EBlasso; [43]) with three-level hierarchical prior distribution (normal, expo- nential, and gamma prior; [44]): bi � Nð0; siÞ; ði ¼ 1; 2; . . . ; kÞ s2 i � expðlÞ; l � gammaða; bÞ where k is the number of genetic markers and a and b are regularization parameters that control the level of shrinkage of the marker effects (β) toward zero. The regularization parameters were determined using five-fold cross validation. The EBlasso was conducted using R package “EBglmnet” [44]. Genetic markers with non-zero effects and P < 0.01 were considered as significant QTLs. The phenotypic variance explained (PVE) by i-th QTL was estimated as PVEi ¼ ^b2 i sxi =sy, where ^b is the estimated value of parameter β, σx is the vari- ance of the marker genotype, and σy is the variance of climate sensitivity traits. PVE by all QTLs was estimated as PVEall ¼ s^y =sy, where s^y is the variance of the estimated value of the climate sensitivity traits, which is obtained from a multiple linear regression using significant QTLs as explanatory variables [45]. The prediction accuracy of significant QTLs was tested by correlating the observed and predicted values using five-fold cross validation. The results of the QTL analysis were validated with composite interval mapping (CIM) using R package “qtl” [46]. Climate sensitivity of Cryptomeria japonica

PLOS ONE | https://doi.org/10.1371/journal.pone.0228278 January 28, 2020 4 / 14