Meta-QTL and ortho-MQTL analyses identified genomic regions controlling rice yield, yield-related traits and root architecture under water deficit conditions

Meta-QTL (MQTL) analysis is a robust approach for genetic dissection of complex quantitative traits. Rice varieties adapted to non-flooded cultivation are highly desirable in breeding programs due to the water deficit global problem. In order to identify stable QTLs for major agronomic traits under water deficit conditions, we performed a comprehensive MQTL analysis on 563 QTLs from 67 rice populations published from 2001 to 2019. Yield and yield-related traits including grain weight, heading date, plant height, tiller number as well as root architecture-related traits including root dry weight, root length, root number, root thickness, the ratio of deep rooting and plant water content under water deficit condition were investigated. A total of 61 stable MQTLs over different genetic backgrounds and environments were identified. The average confidence interval of MQTLs was considerably refined compared to the initial QTLs, resulted in the identification of some well-known functionally characterized genes and several putative novel CGs for investigated traits. Ortho-MQTL mining based on genomic collinearity between rice and maize allowed identification of five ortho-MQTLs between these two cereals. The results can help breeders to improve yield under water deficit conditions.

. To the best of our knowledge this is the first MQTL analysis for GW, TN and DT in rice.
A total of 10 MQTLs were detected in the same chromosomal regions with similar yield and yield-related traits under well-water condition in rice 19 . This indicates the same loci might control aforementioned traits under both water deficit and well-water conditions (Additional file 2). They include five MQTLs for PH (MQTL-PH2, www.nature.com/scientificreports/ The MQTL analysis considerably narrowed the CI allowing for exploration of a reduced number of candidate genes (CGs) for the investigated traits. The average CI was reduced from 15.57 cM in the initial QTLs to 5.48 cM in the MQTL with 65% of MQTLs having CI < 5 cM (Table 3). In 10 MQTLs, MQTL-GW4, HD4, HD7, PH3, PH5, YLD1, YLD7, DT1, DT3, RL4, RDW1 and RDW7, the CI was reduced to < 1 cM (Table 3). Therefore, MQTL analysis can significantly raise the accuracy of identification of CGs. All the annotated genes located within the CI of each MQTL and the most promising CGs based on their reported function in previous studies are reported in Additional file 3. Some functionally characterized genes such as GRAIN SIZE 2 (GS2), GRAIN WEIGHT 7 (GW7), Early heading date 1 (Ehd1), DWARF 10 (d10) and Grain number, plant height, heading date7 (Ghd7), OsPIN3t, OsSAUR45, and WEG1 were located within MQTL-GW1, GW3, HD6, PH2, PH8, RT1, RL4, RDW5, respectively, and OsAIR1 located at MQTL-RN2 and RDW2, and OsMGT1 located at MQTL-RT2 and RDW1, that are assumed to control the aforementioned traits. Putative novels CGs for each trait are discussed below. In addition, the positions of MQTLs on the rice genome were compared with the gene density, and densities of SNPs, structural variants (SV), recombination and functional variants, and the reported selective sweep regions 25 (Fig. 3). Most of the detected MQTLs were located in sub-telomeric regions where generally the gene, SNP, SV and recombination densities are higher (Figs. 2,3). This is consistent with previous results in barley, maize and rice 15,19,26 . The regions with high SV frequency could play an effective role in stress response 27 . A total of 13 MQTLs (MQTL-YLD3, YLD6, GW2, GW5, HD5, TN2, DT2, RT3, RT4, RL3, RL4, RDR3 and RDW6) were co-located with selective sweep regions reported by Huang et al. These MQTLs are likely effective for selection towards drought adaptation during rice breeding and domestication processes 25 . Five of these MQTLs including MQTL-YLD3, TN2, GW5, RT4 and RDW6 were also co-located with the position of reported functional variants 25 .
The investigation of collinear regions within the rice genome resulted in identification of five duplicated regions containing MQTLs for the same traits. MQTL-YLD2 and MQTL-YLD4 on chromosomes 2 and 4, and MQTL-YLD8 and MQTL-YLD9 on chromosomes 11 and 12 for yield, MQTL-RWD1 and MQTL-RWD2 on chromosomes 1 and 5, MQTL-RWD4 and MQTL-RWD5 on chromosomes 8 and 9, and MQTL-RN1 and MQTL-RN2 on chromosomes 1 and 5 for root-related traits are co-located at rice genome duplicated regions (Fig. 3). Duplicated genomic regions derived from common ancestors might contain paralogous genes with similar functions that can be considered as promising CGs controlling the trait 28 . Consequently, we carefully surveyed these regions for detecting possible paralogous CGs in the duplicated regions. In MQTL-RN2, we note the OsABIL3 or PP2C50 gene which has a key role in root architecture and response to drought stress by affecting ABA signaling: overexpression of this gene was reported to lead to the ABA insensitivity along with stomatal density and root architecture 29 . The paralogous gene Os01g0618200 encoding PP2C07 is also present at the duplicated region on chromosome 1 with MQTL-RN1 for the same root number trait. Moreover, at MQTL-YLD4 interval on chromosome 4, we detected GRAS23 that contributes to drought response in rice 30 , with paralogues HAM1 and HAM2 colocalizing with the duplicated genome on chromosome 2 with MQTL-YLD2 for the yield under drought stress.

MQTLs and CGs for heading date.
It is well known that HD is highly correlated with YLD 38 and drought adaptation 39,40 . We detected seven MQTLs for HD under water deficit conditions including two MQTLs on chromosomes 3 and 9 and one MQTL each on chromosomes 5, 10 and 12 (Table 3). MQTL-HD1 on chromosome 3 had the highest number of supporting QTLs with five QTLs from two independent studies (Table 3).
Among annotated genes within MQTL-HD1, MQTL-HD3 and MQTL-HD6 intervals, OsCCT11 41 , HBF1 42 and Ehd1 38,43,44 , respectively, were identified as potential candidates for HD under water deficit conditions. OsCCT11 is considered as a positive regulator of heading date since RNAi-mediated downregulation of this gene delays HD 41 , while HBF1 is considered as a negative regulator of HD since mutation promotes flowering 42 . Among genes within the MQTL-HD3 interval, basic region/leucine zipper motif (bZIP), FT-like and circadian clock genes are promising candidates 43,45,46 . Another CG at this MQTL is OsHAPL1, known to prevent flowering under long-day conditions 47

MQTLs and CGs for plant height.
Since the Green Revolution, PH has been considered as a major target for YLD improvement 50 and it also contributes to drought tolerance 40   www.nature.com/scientificreports/ chromosome 4 had the largest number of initial QTLs with six QTLs from four independent studies followed by MQTL-PH10 on chromosome 12 with three QTLs reported from three independent studies. These MQTLs are the most stable QTLs for PH under water deficit conditions. The d10 51 and Ghd7 52 genes are reported to regulate plant height in rice, and they are positioned within MQTL-PH2 and MQTL-PH8 genomic regions, respectively. MQTL-PH4 contains EP3/LP gene, whose mutant shows increased panicle size and PH in rice 53 . Conversely, mutations in OsKS2 and NAL1 54,55 at MQTL-PH7 and OsSEC3A 56 at MQTL-PH6 decrease PH in rice. In MQTL-PH10, we detected Os12g0271600 that encodes BRI1 and the mutant alleles could act as a dwarfism gene 50 .

MQTLs and CGs for yield. The maintenance of YLD under drought condition is the ultimate goal in cereal
breeding 57,58 . We identified 10 MQTLs for YLD consisting of two MQTLs on chromosome 12 and one MQTL each on chromosomes 1, 2, 3, 4, 6, 8, 10 and 11 (Table 3). Among them MQTL-YLD3 on chromosome 3 overlapped with a YLD MQTL identified under well-water conditions 19 . Therefore, the same genes might control YLD under both mentioned conditions at this position.
We detected some genes which affect the photosynthetic rate including Roc5 at MQTL-YLD2 59 , UCL8 at MQTL-YLD3 60 and OsPTR6 at MQTL-YLD4 61 that might indirectly contribute to the final YLD. OsALMT7 is located at the MQTL-YLD2 interval with pleiotropic effects on YLD and panicle size 62 . TAC3 might indirectly regulate YLD through changing HD and tiller angle in rice at MQTL-YLD3 63 . The most likely CGs at MQTLs and CGs for number of tillers. The number of fertile tillers is a major contributor to YLD and its alteration during drought stress can result in drought adaptation 4,[66][67][68] . Tillering is a complex process and highly affected by environmental conditions 66 . We detected only two MQTLs on chromosomes 5 and 6 which were associated with TN (Table 3). In MQTL-TN2, we identified OsAID1 as a gene associated to TN regulation in rice 69 .
MQTLs and CGs for drought tolerance. Plant water content is highly affected by water deficit conditions and in turn can contribute to drought tolerance. Plant water content can be measured by different criteria  MQTLs and CGs for root architecture. Root architecture develops through dynamic processes that effectively contribute to water deficit adaptation allowing water and nutrient uptake from deep soil 74,75 . We studied five major traits related to root architecture including RDW, RL, RN, RT and RDR under water deficit conditions and identified 23 MQTLs including seven MQTLs for RDW, four MQTLs for RL, three MQTLs for RN, six MQTLs for RT and three MQTLs for RDR (Table 3). MQTL-RL2 had the highest number of initial QTLs (six QTLs from three independent studies) and it was considered as the most stable QTL for root architecture (Table 3). Interestingly, MQTL-YLD4 for YLD under water deficit conditions on chromosome 4 overlapped with MQTL-RDR2 and RL1 (Fig. 2).
Overlapping MQTLs for different root architecture traits included MQTL-RDW1 and MQTL-RT2 on chromosome 1, MQTL-RDW2 and MQTL-RN2 on chromosome 5 and MQTL-RN1 and MQTL-RT1 on chromosome 1, suggesting the possible existence of genes with pleiotropic effects on these traits. For example, the genomic region spanning MQTL-RN2 and MQTL-RDW2 harbors OsAIR1, a gene affecting root architecture and contributing to drought tolerance 76 . In the overlap region between MQTL-RT2 and MQTL-RDW1, noteworthy is OsMGT1 which was shown to affect root architecture during salinity stress 77 . MQTL-RT1 contains OsPIN3t 78 and OsFBK1 79 genes that are reported to control root architecture under water deficit conditions. The SMOS1 gene within MQTL-RT4 determines root meristem size 80 , and the cZOGT3 81 gene within MQTL-RDR2 regulates root architecture. Additionally, AMTR1 (MQTL-RN2) affects root architecture under drought stress 82 .
The same source of favorable allele from 'IRAT109' parent derived from two independent populations was identified in QTLs located at MQTL-RL3 (Additional file 4). The FC1 gene 83 on MQTL-RL1 controls root growth and might contribute to drought tolerance under water deficit conditions. Within MQTL-RL4, we detected a cluster of small auxin-up RNA (SAUR ) genes. Over-expression of OsSAUR45 regulates root length and other related root traits 84 .
For RDW, we detected two co-located MQTLs including MQTL-RDW1 and RDW2 co-locating with OsMGT1 and OsAIR1genes, respectively. Additionally, WEG1 at MQTL-RDW5 is a novel gene that regulates root related traits 85 and may keep the same role under water deficit conditions. Ortho-MQTL mining. To investigate ortho-MQTLs for yield and yield-related traits under water deficit conditions between rice and maize as the two most important cereals with generally high water demand, the syntenic regions of all detected rice MQTLs in this study were compared with published maize MQTLs 17,86,87 . Comparative genomic analyses provide a valuable approach to transfer information across species and identify conserved genes 19 . Through synteny analysis between rice and maize, we uncovered 5 ortho-MQTLs including 4 ortho-MQTLs for YLD on chromosomes 2, 3, 4 and 8 and 1 ortho-MQTL for PH on chromosome 4 (Table 4; Fig. 3). The genes located at these syntenic regions were further investigated (Additional file 5; Fig. 4).
The orthologous genes located at these ortho-MQTLs in both rice and maize are shown in the Additional file 5 and Fig. 4. The rice genomic region subtending MQTL-PH7 harbors the NAL1 gene as a regulator of PH 55 : we identified the ortholog of this gene (Zm00001d026296) in the maize ortho-MQTL. In the syntenic region of rice MQTL-YLD2 on chromosome 5 of maize, there were two MQTLs (mQTL_GY_5 86 , MQTL5.7 17 ) containing the orthologs of OsALMT7 and SID1 genes (Zm00001d017571 and Zm00001d017560), known to affect YLD in rice 62,88 . The orthologous gene of TAC3 in maize (Zm00001d033857) in the syntenic region of MQTL-YLD3 in maize (mQTL_GY_1b) regulates tiller angle that might affect YLD under water deficit conditions 63 . In the syntenic region of rice MQTL-YLD6, there was a MQTL (MQTL6.1 17 ) on chromosome 6 of maize. This rice MQTL contains Ehd3 gene regulating flowering and consequently YLD 38 and its orthologous (Zm00001d035008) was detected in its ortho-MQTL in maize, likely to have similar functions. This approach provided better understanding of genes controlling investigated traits under water deficit conditions with similar evolutionary history and Table 4. Ortho-MQTLs in rice and maize based on the syntenic analyses.  www.nature.com/scientificreports/ conserved function between these cereals. These results can benefits breeders by tracing CGs and using markerassisted selection in breeding programs of cereals under water deficit conditions.

Conclusions
Through MQTL analysis this study provides an overview of genomic regions controlling YLD, yield-related traits, root architecture and plant water content including GW, HD, PH, TN, RDW, RL, RT, RN, RDR and DT under water deficit conditions in rice. This approach is useful in overcoming some limitations of single QTL mapping studies on different genetic backgrounds and environments and greatly facilitates the identification of CGs and robust flanking markers for MAS in breeding programs. The results offer a framework for future genetic studies of yield under drought conditions, e.g. through fine mapping, positional cloning, producing chromosome substitution lines, as well as validation of CGs by genome editing using Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR) and similar approaches. This study also demonstrates the value of ortho-MQTL mining among evolutionarily close crop species for identification of genomic regions and CGs controlling complex quantitative traits.  89 . In order to incorporate SNP markers of those initial QTLs with SNP markers (Table 1) into the reference map, we applied our previous approach 19 in which the genomic position of SNP markers on the rice genome were determined and in consequence the closest markers based on the physical position were used to project them on the reference map. QTL position, CI, proportion of phenotypic variance (R 2 ), log of odds ratio (LOD score), additive effects and favorable alleles were compiled for each QTL in the 67 populations (Additional file 4). In order to calculate 95% CI for QTLs, we used the following formulas: CI = 530/(N × R 2 ) for BC and F 2 lines, CI = 287/(N × R 2 ) for DH lines and CI = 163/(N × R 2 ) for RLLs lines 90,91 , where N is the population size and R 2 is the proportion of phenotypic variance of the QTL. MQTL analysis was carried out using BioMercator V4.2 11,92 . Meta-QTL analysis and ortho-MQTL mining. The MQTL analysis was conducted on integrated and re-positioned QTLs on the reference map using BioMercator V4.2 11,12,92 . The best model of MQTLs was chosen according to the prevalent value among Akaike Information Criterion (AIC), corrected Akaike Information Criterion (AICc and AIC3), Bayesian Information Criterion (BIC) and Average Weight of Evidence (AWE) criteria. Therefore, the consensus QTL from the best model was reported as a "real" QTL/MQTL 12,92 . Considering the known correlations among RWC, CT, LR, LD and DRI 3,32,33,70 , the respective QTLs were analyzed together as one trait named as DT in BioMercator V4.2. 12,92 . Mapchart V.2.32 software 93 was used to show the MQTLs and related QTLs on the reference map.

Materials and methods
The distribution of MQTLs on the rice genome (IRGSP-1.0) compared to the position of centromeric and telomeric regions and the gene density along each chromosome were surveyed and shown by heatmap using pheatmap and R 94,95 . Centromere position, gene density, SNP and structural variations (SV) and recombination rate density of each chromosome, as well as rice genome duplications were retrieved from EnsemblPlants (https:// plants. ensem bl. org/ index. html) database. Additionally, the position of identified MQTLs were compared with selective sweep regions and functional variants in coding regions with strong alteration in allele frequency between cultivated and wild rice reported by Huang et al. 25 . The distribution of aforementioned factors, number of MQTL under water deficit conditions and number of MQTLs under well-water conditions 19 over the rice genome were shown by using Circos 96 .
To detect ortho-MQTLs between rice and maize, syntenic regions between the two species were identified by using EnsemblPlants database 97 . MQTLs identified for yield and yield-related traits under drought conditions in maize 17,86,87 were compared with MQTLs detected for similar traits in our study. www.nature.com/scientificreports/ positions, the closest markers were applied for detecting the genomic coordinates of MQTL. Gene annotations within MQTL genomic regions were carefully explored by EnsemblPlants (https:// plants. ensem bl. org/ index. html) and FunRiceGenes (https:// funri cegen es. github. io/) 98 databases.

Data availability
The relevant data and additional information are available in the supplementary files and also from the corresponding author on reasonable request.