*4.4. Analysis of Reads and Variants*

The raw reads were cleaned and trimmed using BBDuk program of BBTools (http://jgi.doe.gov/ data-and-tools/bbtools/). The paired reads were mapped to the IRGSP-1.0 reference rice genome (http://rapdb.dna.affrc.go.jp) by using Burrows-Wheeler Aligner based on the Maximal Exact Matches algorithm (BWA MEM) and the alignments were processed by SAMTools [38–40]. Freebayes was used to call SNPs and InDels, with default parameters [41]. To obtain reliable polymorphic markers, variant filtering was performed by custom perl scripts: firstly, SNPs or short InDels exhibiting polymorphism between the two parents were screened; secondly, to further avoid severe segregation distortion, SNPs or short InDels with AF values from 0.3 to 0.7 were retained. These markers were annotated by snpEff [42].

### *4.5. QTL Analysis*

The marker set was employed to map QTLs. The replicated sub-pools ER-1, ER-2, and ES-1, ES-2 were firstly incorporated into one ER pool and one ES pool, respectively. AFD value between the ER and ES pools was calculated and then smoothed by block regression, following the Block Regression Mapping methodology [43]. The block size used for the regression was set to be 20 kb. The AFD curve threshold at the overall significance level of 0.05 was estimated under the assumption of theoretical allele frequency (= 0.5) in the F2 population. For each significant AFD peak (candidate QTL), the 95% confidence interval was estimated. In addition, unpaired t-tests based on the two biological replicates of the ER and ES pools were performed to validate the candidate QTLs following the X-QTL-seq method [44]. According to the peak AFD value, the heritability of each QTL was estimated using the method of Pooled QTL Heritability Estimator (PQHE) [45].
