Four segregating populations of rice (Oryza sativa L.) were used in this study. One was a primary mapping population consisting of 247 RILs developed from a cross between indica rice cultivars ZS97 and MY46. The other three were NIL populations segregating a region involving the RFT1 gene in an isogenic background, all of which have been reported by Zhu et al. (2017). Two of the NIL populations, namely TF6-15 and R1, were derived from a residual heterozygote (RH) of ZS97/MY46. An F₁₀ plant was selected, which was heterozygous in a 29.9-kb region covering the RFT1 and Hd3a loci and homozygous in other regions. The S₁ plants were assayed with DNA markers located in the segregating region. Homozygous plants were selfed to produce NILs. The TF6-15 population was established, consisting of 10 lines of ZS97 homozygotes and 10 lines of MY46 homozygotes differing in the 29.9 kb region. New RHs were identified from heterozygous progeny of the F₁₀ plant. An F₁₄ plant that was heterozygous at the RFT1 locus only was selected. The R1 population was constructed, comprising 20 lines of ZS97 homozygotes and 20 lines of MY46 homozygotes differing for the RFT1 gene only. The remaining NIL population consisted of 28 homozygous transgenic lines in the genetic background of japonica rice cultivar Zhonghua 11 (ZH11), of which 14 lines carried the RFT1 ^ZS97 transgene and the others carried no transgene.

The four populations were planted in the single-cropping rice season at the China National Rice Research Institute in Hangzhou, Zhejiang, China. The RIL population was raised in 2015, 2016, and 2017, and the other three populations were grown in 2017 only. Each line was represented by 12 plants per row, with an inter-plant spacing of 16.7 cm and an inter-row spacing of 26.7 cm. The RILs were grown without replication, while the other populations were represented by two replicates. Plants were managed using standard agricultural practice.

Grain was bulked from the middle ten plants of each row and dried to a moisture content of ~12%. De-hulling was achieved using THU-35A testing husker (Satake Engineering Co. Ltd., Hiroshima, Japan). The dehusked grain was ground using a Cyclotec 1093 Sample Mill (Tecator, Hoganas, Sweden) and the resulting flour was passed through a 0.42 mm sieve. A 250 mg batch of each flour sample was sealed in a vial containing 10 ml 6.0 M HCl and held at 110°C for 24 h. The resulting hydrolysate was diluted to 50 ml with deionized water and filtered. A 0.2 ml aliquot of the filtrate was transferred to a 2 ml tube and evaporated down to ~0.1 ml by bubbling nitrogen gas. After the addition of 2 ml 20 mM HCl, the solution was passed through a 0.2 μm Acrodisc membrane (Pall Corp., Port Washington, NY, USA). The amino acid content of each sample was acquired using a L8900 amino acid auto analyzer (Hitachi, High-Technology Corporation, Tokyo, Japan). Percentage contributions of each of the 17 amino acids were obtained using Ezchrom Elite software (High-Technology Corporation, Tokyo, Japan).

The amino acids studied were aspartic acid (Asp), threonine (Thr), serine (Ser), glutamic acid (Glu), glycine (Gly), alanine (Ala), cystine (Cys), valine (Val), methionine (Met), isoleucine (Ile), leucine (Leu), tyrosine (Tyr), phenylalanine (Phe), lysine (Lys), histidine (His), arginine (Arg), and proline (Pro). Each sample was analyzed in triplicate. Every analytical batch included a blank (20 mM HCl only) and a GBW (E) 100010 reference sample (Chinese National Research Center for Certified Reference Materials, Beijing, China), except that the reference samples for Asp and Cys were not available. Concentrations in the standard samples of Ile, Leu, Tyr, Phe, Lys, His, Arg, Thr, Ser, Glu, Pro, Gly, Ala, Met, and Val measured based on six replicates were 0.38 ± 0.060, 0.74 ± 0.040, 0.30 ± 0.020, 0.52 ± 0.068, 0.28 ± 0.012, 0.22 ± 0.012, 0.49 ± 0.048, 0.30 ± 0.014, 0.49 ± 0.014, 3.28 ± 0.18, 0.95 ± 0.010, 0.40 ± 0.048, 0.503 ± 0.026, 0.165 ± 0.046, and 0.50 ± 0.024%, respectively, matching well with the Certified Reference Material's values. The day-to-day reproducibility of the assay checked over three days was satisfactory, having the relative standard deviation lower than 5.0% for all the fifteen amino acids.

Genetic map of the ZS97/MY46 RIL population was previously constructed, consisting of 256 markers and spanning 1,814.7 cM (Wang et al., 2017b). This map was applied for QTL analysis using composite interval mapping (CIM) and multiple interval mapping (MIM) in Windows QTL Cartographer v2.5 (Wang et al., 2011). A candidate QTL was identified with CIM using a threshold of logarithm of odds (LOD) > 2.0 and then evaluated with MIM using the Bayesian Information Criterion c (n) = ln (n). A putative QTL was claimed if it satisfied both criteria. QTLs were designated as proposed by McCouch and CGSNL (2008). Two-way analysis of variance (ANOVA) was conducted to test the differences between the two genotypic groups in each of the three NIL populations, using a general linear model (GLM) of the SAS Program as described by Dai et al. (2008).

Variation of the 17 components of AAC in unmilled rice of the ZS97/MY46 RIL population is summarized in Table 1 . There was a strong evidence for transgressive segregation in both directions for all the amino acids except His of which the contents of the RILs were all lower than the high-parental value in 2016. It was also shown that differences between the two parental lines varied greatly across the 3 years. Among the contents of the 17 amino acids, ZS97 was found to be the high-value parent for two amino acids only (Phe and His) in 2015, but ZS97 was the high-value parent for seven amino acids (Ser, Glu, Leu, Tyr, Phe, Lys, and His) in 2016 and for 14 amino acids (Asp, Thr, Ser, Glu, Gly, Ala, Val, Ile, Leu, Tyr, Phe, Lys, His, and Arg) in 2017.

Pearson correlation coefficients between the 17 components of AAC were calculated using mean values over 3 years and data of each year. Family error rates were controlled by dividing the P value of 0.05 by 17, thus a threshold of P < 0.003 was used for declaring a significant correlation. It was found that a large majority of the correlations were positively significant. Of the 136 estimates produced from the mean values, 114 were positively significant, 4 were negatively significant, and 18 were non-significant ( Table 2 ). The four negative correlations occurred between Ala and Gly, Cys, Met, and Pro. The 18 non-significant correlations included nine between Ala and others (Ser, Glu, Val, Ile, Leu, Tyr, Phe, Lys, and Arg), 8 between Cys and others (Asp, Thr, Gly, Val, Met, Lys, His, and Pro), and 1 between Tyr and Met. Common occurrence of significantly positive correlations between different AAC components were also observed when data of each year were used ( Supplementary Table S1 ). In 2015, 101 correlations were positively significant and the other 35 were non-significant. In 2016, 122 correlations were positively significant and the other 14 were non-significant. In 2017, 118 correlations were positively significant, three were negatively significant, and 15 were non-significant.

A total of 73 QTLs were detected based on 3-year's data of the 17 components of AAC in the ZS97/MY46 RIL population ( Supplementary Table S2 ). Of these QTLs, seven were identified in all the 3 years, eight were found in 2 years, and the others were detected in 1 year only. The number of QTLs detected for each amino acid ranged from two to eight, with the proportion of the variance explained (R² ) by a single QTL ranging from 2.2 to 35.9%. These QTL were distributed over all the 12 rice chromosomes except chromosome 5 ( Figure 1 ). Most of them were located in cluster, with chromosomes 1, 6, and 7 harboring the highest number of loci. Of the 15 QTLs detected in 2 or 3 years, 12, 2, and 1 were located in chromosomes 6, 7, and 11, respectively. Except qThr11, allelic directions of these QTLs all remained consistent across different years.

Fourteen QTLs were located in the RM190–RM6917 region on the short arm of chromosome 6, forming the largest cluster in terms of QTL number. Included were seven QTLs detected in three years (qAsp6, qSer6, qGly6, qLeu6, qPhe6, qHis6, and qArg6), five QTLs detected in 2 years (qThr6, qGlu6, qVal6, qMet6, and qTyr6), and two QTLs detected in 1 year (qLys6 and qPro6.1). Enhancing alleles of these QTLs were all derived from the male parent MY46, with qGly6 having the highest R² of 33.9%. Two other QTLs (qAla6 and qPro6.2) were detected in nearby intervals RM253–RM276 and RZ667–RM19784, respectively, of which the enhancing alleles were both derived from the female parent ZS97. Altogether, 16 QTLs were detected on chromosome 6.

The second largest cluster consisting of nine QTLs was located in the RM3325–RM3859 region on the short arm of chromosome 7. Included were two QTLs detected in 2 years (qAsp7 and qArg7) and seven QTLs detected in 1 year (qThr7, qSer7, qGly7.1, qAla7, qVal7, qTyr7.1, and qPhe7). Enhancing alleles of these QTLs were all derived from ZS97, with qAsp7 having the highest R² of 19.8%. Five other QTLs were clustered in the RZ471–RZ395 region on the long arm of this chromosome. The enhancing alleles were derived from ZS97 at qTyr7.2 and qPro7.1, and from MY46 at qGly7.2, qCys7, and qPro7.2. The five QTLs had high R² ranging from 10.7 to 35.9%. Altogether, 14 QTLs were detected on chromosome 7.

The third largest cluster consisting eight QTLs (qSer1, qVal1, qIle1, qLeu1, qTyr1, qLys1.1, qHis1, and qPro1.1) was located in the RM283–RM3746 region on the short arm of chromosome 1. Enhancing alleles of these QTLs were all derived from ZS97, with qLeu1 having the highest R² of 15.0%. Three other QTLs (qAsp1, qGly1, and qPro1.2) were located in the pericentromeric region of chromosome 1. They had high R² ranging from 17.0 to 32.8%, and the enhancing alleles were all derived from ZS97. Two sparsely-distributed QTLs (qPro1.3 and qLys1.2) were located in lower regions of the long arm. Altogether, 13 QTLs were detected on chromosome 1.

Among the remaining 30 QTLs, two single QTL were located on chromosomes 8 and 10, respectively, and the others were distributed on chromosomes 2, 3, 4, 9, 11, and 12 with 2–6 QTLs per chromosome. The six QTLs on chromosome 2 were all located in the lower region of the long arm; the two QTLs on chromosome 3 were tightly linked; the six QTLs on chromosome 4 involved two pairs of tightly-linked QTLs with two nearby QTLs; the five QTLs on chromosome 9 may be viewed as one cluster and two single QTL; the five QTLs on chromosome 11 was separated into two clusters; and the four QTLs on chromosome 12 included one cluster and one single QTL.

As described above, the largest QTL cluster for the 17 components of AAC detected in the ZS97/MY46 RIL population was located in the RM190–RM6917 region on the short arm of chromosome 6. This region covered the two florigen genes of rice, RFT1 and Hd3a (Hori et al., 2016), suggesting a possible involvement of RFT1 and/or Hd3a in controlling AAC of rice grain. This assumption was firstly tested using the NIL population TF6-15 segregating a 29.9-kb interval covering both RFT1 and Hd3a. Significant differences (P < 0.05) between the 10 homozygous lines of NIL^ZS97 and 10 homozygous lines of NIL^MY46 were detected on 15 of the 17 components of AAC ( Table 3 ). The R² for individual components ranged from 16.7 to 61.2%. The enhancing alleles were all derived from MY46, which is in agreement with the effects detected in the ZS97/MY46 RIL population.

Then, QTL analysis was performed using the NIL population R1 that was homozygous at the Hd3a locus but segregated for the RFT1 gene. Significant differences (P < 0.05) between the 20 homozygous lines of NIL^ZS97 and 20 homozygous lines of NIL^MY46 were detected on 15 of the 17 components of AAC ( Table 4 ). The R² for individual components ranged from 9.5 to 63.2%. Again, the enhancing alleles were all derived from MY46. It is also noted that the two components showing no significant difference between NIL^ZS97 and NIL^MY46 were commonly found to be Met and Pro in the TF6-15 and R1 populations. These results indicate that the RFT1 gene has a strong and stable effect on most components of AAC in unmilled rice.

The effect of RFT1 on AAC in unmilled rice was further tested using a transgenic population segregating the RFT1 ^ZS97 transgene in the genetic background of japonica cultivar ZH11. Significant differences (P < 0.05) between the 14 lines of NIL^ZS97 carrying homozygous transgenes and 14 lines of NIL^ZH11 carrying no transgene were detected on 13 of the 17 components of AAC, with R² ranging from 8.1 to 38.0% ( Table 5 ). Integration of the RFT1 ^ZS97 transgene into the genome of ZH11 reduced the contents of the amino acids. In addition, the two components showing no significant difference between NIL^ZS97 and NIL^MY46 in the TF6-15 and R1 populations, Met and Pro, were included in the four components having no significant difference between NIL^ZS97 and NIL^ZH11 in the transgenic population. These results indicate that the effects of RFT1 on AAC of unmilled rice are consistent in the genetic background of different subspecies of Asian cultivated rice.