Rice is an important cereal crop and provides over 21% of caloric intake for more than half of the global population (Zhang et al., 2014; Vishnukiran et al., 2020; Zhao et al., 2020). Rapid population growth remains a major threat to food security worldwide, thus there is a need to augment efforts to increase the yield potential of rice. Spikelet fertility, panicle number per plant, number of grains per panicle, and thousand-grain weight are key indicators of grain yield and continue to be major targets in breeding programs (Sakamoto and Matsuoka, 2008; Huang et al., 2013; Li et al., 2019). Grain yield is quantitatively inherited and controlled by multiple loci (genes) (Xing and Zhang, 2010; Bai et al., 2012); therefore, it is imperative to ascertain the genetic bases of grain yield to boost rice productivity. With the advancement of QTL mapping, breeders have been able to identify genomic regions influencing complex agronomic traits (Daware et al., 2017) and have expedited the process of improving quantitative traits through marker-assisted selection (Zhang et al., 2010). However, the application of QTLs by breeders in a broader spectrum has major setbacks due to the genetic background of the populations used in QTL mapping and phenotype environment (Arcade et al., 2004; Daware et al., 2017). Therefore, a comprehensive analysis of identified QTLs is required to make these QTLs applicable in rice breeding and to decipher the underlying genetic factors conferring grain yield. Essentially, QTL meta-analysis has been shown to be a reliable and effective instrument (Goffinet and Gerber, 2000).

An integrated meta-QTL analysis uses QTLs from independent trials regardless of their genetic backgrounds, cross types/parents, size of populations, years, and locations to identify consensus QTLs (Goffinet and Gerber, 2000; Arcade et al., 2004; Sosnowski et al., 2012). This approach has been evidenced to reliably refine QTLs location and narrow confidence intervals (CI) which enhance the precision of marker-assisted selection (MAS) and discovery of candidate genes (Martinez et al., 2016; Zhang et al., 2017). Furthermore, a meta-analysis of QTLs attempts to clarify if QTLs encompassing different loci from different studies are the same loci or if they represent the same position on a linkage map of species under study. To date, numerous MQTLs for important agronomic traits have been identified in different crops such as wheat (Acuña-Galindo et al., 2015; Darzi-Ramandi et al., 2017; Soriano and Alvaro, 2019; Liu et al., 2020; Saini et al., 2021; Amo and Soriano, 2022; Miao et al., 2022; Saini et al., 2022; Yang et al., 2021), soybean (Zhao-Ming et al., 2011; Hwang et al., 2016; Qin et al., 2018), maize (Semagn et al., 2013; Wang et al., 2013; Martinez et al., 2016; Chen et al., 2017; Zhao et al., 2018; Kaur et al., 2021), barley (Zhang et al., 2017; Khahani et al., 2019), peanut (Lu et al., 2018), and sorghum (Aquib and Nafis, 2021). However, reports on Meta-QTLs for grain yield in rice under field conditions are limited. More so, the understanding of the molecular basis of QTLs/genes controlling grain yield is insufficient. Sandhu et al. (2021) detected 76 MQTLs governing grain number per panicle, spikelet fertility, and panicle number per plant on 29 QTL studies published from 2000 to 2022. Also, Wu et al. (2016) reported ten MQTLs associated with spikelet fertility on 82 QTL studies from 1996 to 2013. In another study, Khahani et al. (2021) integrated 101 studies in rice and confined these QTLs to meta-QTLs with reduced confidence intervals for use in MAS and discovered candidate genes for future functional validation.

This study performed a comprehensive meta-QTL analysis examining QTLs for spikelet fertility (SF), panicle number per plant (PN), grain number per panicle (GNP), and 1000-grain weight retrieved from 47 QTL studies published between 2002 and 2022 in 101 biparental populations evaluated across diverse environments and under field condition. The main objective of this study was to conduct meta-QTL analysis and identify stable and robust MQTLs and candidate genes (CGs) to aid rice breeding through MAS and enhance molecular dissection of grain yield targeted at improving the yield potential of rice.

To enhance understanding, a schematic flow chart has been drawn and presented in Figure 1 , which illustrates the approach adopted in executing this study. Further description of these steps is as ensued:

Increasing the productivity of rice is one of the foremost constraints rice breeders face to ensure adequate food supply to a rapidly growing global population. Most important agronomic traits such as spikelet fertility, number of grains per panicle, panicle number per plant, and thousand-grain weight are complex and modulated by numerous polygenic loci. Over the years, the success in the introgression of QTLs in breeding programs by marker-assisted selection has been slow due to inconsistency in QTL genetic backgrounds and phenotype environment. Integration of QTLs information from diverse independent studies and exploiting meta-analysis to identify MQTL regions has proven to be an ideal approach to accelerating QTLs fine mapping and gene cloning (Wang et al., 2013; Wang et al., 2015; Martinez et al., 2016; Wu et al., 2016; Chen et al., 2017).

To decipher the molecular mechanism underlying grain yield, MQTL analysis was conducted based on the QTLs conferring grain yield collected from previous independent studies. The initial step in the meta-analysis of QTL is the QTL projection onto the reference map. A characteristics of the reference map utilized in the current study showed that the highest marker saturation was reported on chromosome 1 followed by chromosomes 3 and 2 with the lowest marker density found on chromosome 10. Individual chromosomes had uneven marker distribution, and the density of markers at either end of the chromosome differed. This is primarily due to the use of independent genetic maps with different marker types and number of markers. Interestingly, similar cases have been reported in earlier studies on meta-analysis of QTLs (Venske et al., 2019; Liu et al., 2020; Saini et al., 2021; Yang et al., 2021; Saini et al., 2022). More so, the QTLs investigated were not evenly distributed across the entire rice chromosomes. Chromosomes 1, 2, 3, and 5 reported the highest number of QTLs which is consistent with the previous studies (Xing and Zhang, 2010; Swamy and Sarla, 2011; Swamy et al., 2011; Wu et al., 2016; Khahani et al., 2020). However, chromosomes 8, 9, and 10 reported the lowest number of QTLs. The possible reason for this phenomenon could be the low level of polymorphism associated with these chromosomes. Out of the total QTLs projected, only 32.47% constitute PVE >10%, indicating that the individual QTLs contributed very little to phenotypic variability. This shows that grain yield is primarily determined by minor effects loci, representing a complex genetic architecture.

The MQTL analysis identifies stable and robust QTLs irrespective of the genetic background of mapping population, phenotype environment, and marker density which are the main constraints of QTL mapping (Arcade et al., 2004; Zhang et al., 2010; Daware et al., 2017; Bilgrami et al., 2018; Khahani et al., 2019). Further, association mapping approach has also been shown as an alternative instrument with better accuracy compared to QTL mapping of complex agronomic traits (Daware et al., 2017). However, this method faces remarkable false positive outcomes because of the population structure employed in the analysis (Daware et al., 2017). Owing to this, analysis of MQTL is regarded as the most reliable approach for detection of stable loci modulating quantitative traits. In this study, a total of 313 QTLs were confined into 62 MQTLs across the 12 chromosomes for the evaluated traits. Chromosome 3 reported the highest number of MQTLs (9 MQTLs) and the least MQTL were located on chromosome 8 (2 MQTLs). Intriguingly, our findings corroborate with a previous study on MQTLs in rice (Khahani et al., 2021). More than 80% (50/62) of MQTLs harbored at least two of the initial QTLs. Furthermore, some of the MQTLs contained as high as 10 initial QTLs exhibiting MQTLs robustness. The reference map developed and utilized in this study is much more informative compared to those employed in the previous studies (Wu et al., 2016; Khahani et al., 2020; Khahani et al., 2021). We also found that six out of the total MQTLs identified had common genomic positions with previously reported MQTLs (Wu et al., 2016; Sandhu et al., 2021). This includes regions on chromosomes 2 (MQTL2.3), 4 (MQTL4.1, MQTL4.2), 7 (MQTL7.2), 9 (MQTL9.1), and 12 (MQTL12.1). It is worth noting that, the average confidence interval of MQTLs remarkably reduced up to 3.4 folds times compared to the confidence interval of original QTLs. Our study clearly shows that the meta-analysis is effective in locating consensus and exact QTLs. Previous studies on meta-analysis in wheat, rice, maize, soybean, and cotton, reported 10% to 21% reduction in the total QTLs with two to four times reduction in the average confidence interval of original QTLs (Guo et al., 2006; Rong et al., 2007; Ballini et al., 2008; Courtois et al., 2009; Lanaud et al., 2009; Löffler, 2009; Khahani et al., 2020). MQTLs identified were compared with GWAS studies via the Rice SNP-seek database which led to the identification of common significant regions. Consequently, 16 (25.81%) important GWAS peaks for spikelet fertility (SF) and 1000-grain weight (TGW) were co-located with our MQTLs. Our results demonstrate the coherence of the two approaches in pinpointing common loci associated with the studied traits. However, there are also reports which contradict the result in the present study with 63%, 61.3% and 38.66% MQTLs validated with GWAS (Yang et al., 2021; Amo and Soriano, 2022; Saini et al., 2022). These phenomena may be ascribed to the material utilized for initial QTL mapping (finally used for MQTL analysis) and GWAS.

Identification of underlying genes in genomic regions of major QTLs is of great interest in breeding programs. More so, MQTLs with narrow and precise physical regions are beneficial in candidate gene mining. According to Kaur et al. (2021), MQTLs with 5-8 initial QTL clusters with moderately high R² values could be considered important for mining candidate genes. The genes present in the 8 MQTLs regions were analyzed using rice databases (http://www.ricedata.cn, http://shigen.nig.ac/jp/rice/oryzabase/, and http://rice.uga.edu.cgi-bin/gbrowse/rice). We identified 2,030 genes and as many as 327 genes were found to encode proteins (sugar transporters, PPR repeat-containing protein-like genes, cytochrome P450, zinc fingers, F-box genes, PP2C family, MADs-Box domain-containing protein, AP2 domain-containing protein, and cytokinase oxidase genes) associated with yield (Wu et al., 2016; Chen et al., 2020; Jiang et al., 2022).

Several genes of unknown functions were also found in stable MQTLs intervals; further study is necessary to explore the functional role of those genes in regulating grain yield. A bioinformatics pipeline implementing the genomic sequences of stable MQTLs was used to identify potential candidate genes. These pipelines involved three steps, thus retrieving candidate genes from rice genome project annotation database, visualizing the molecular function of candidate genes by GO enrichment analysis, and implying candidate genes in rice yield-related tissues based on their expression pattern. Consequently, 52 genes were considered potential candidate genes ( Table 3 ). Most of the identified candidate genes are associated with the terms of binding, catalytic activity, cell, cell part, cellular, and metabolic process. These terms have functions related to yield (Mao et al., 2010; Li and Li, 2014; Elattar et al., 2021). In the stable MQTL regions, most of the candidate genes encode zinc finger family protein, ubiquitin ligase domain-containing protein, auxin-responsive gene family, AP2 domain-containing protein, cytochrome P450, receptor-like kinase protein, and MADs-box family protein. Furthermore, zinc fingers have been reported to play a vital role in plant growth, development, and response to abiotic stresses (Li et al., 2013). For instance, a zinc finger protein DST regulates the expression of OsCKX2 and increases cytokinin level in reproductive shoot apical meristem, which in turn elevates meristem activity, panicle branching, and increased grain number (Li et al., 2013). Similarly, DHHC-zinc finger protein gene has been reported to regulate plant architecture and grain yield (Zhou et al., 2017). AP2 domain-containing proteins have been reported to play essential roles in regulating plant growth and development. Overexpression of OsAP2-39 decreased biomass and grain yield in transgenic lines of rice (Yaish et al., 2010). Overexpression of OsIAA10 (auxin-responsive gene in MQTL2.6) was reported to increase rice susceptibility to rice dwarf virus (Jin et al., 2016; Qin et al., 2020). One reported gene, OsBIHD1 (encodes homeodomain protein, putative, expressed) was found in MQTL3.5. Overexpression of OsBIHD1 revealed an increased salt sensitivity and oxidative stress in rice (Luo et al., 2005). In addition, a clone gene Oshox22 (encodes homeobox associated leucine zipper, putative, expressed) was identified in MQTL4.2. Overexpression of Oshox22 increased ABA sensitivity and reduced salt drought and stress tolerance (Zhang et al., 2012).

In MQTL5.2 lies two reported genes, RSR1 and OsAPIP6. The overexpression of RSR1 resulted in the change of amylopectin and starch gelatinization (Fu and Xue, 2010). OsAPIP6 interacted with AvrPiz-t in vitro and reduced flg22-induced ROS generation, suppressed defense gene expression, and increased rice plant susceptibility to M. Oryza (Park et al., 2012). We also found two reported genes, OsPUB44 (encodes U-box domain-containing protein, expressed) and OsMADS4 (encodes MADS-box family gene with MIKCc type-box) in MQTL5.5. OsPUB44 positively control immune responses of rice plant to bacterial blight (Ishikawa et al., 2014). In another study, U-box domain family protein has been reported to play a vital role in rice pollen development (Chen et al., 2019). A mutant of OsPUB73 (U-box domain protein) showed low pollen fertility suggesting that OsPUB73 was responsible for pollen exine or tapetal development and led to pollen partial sterility (Chen et al., 2019). The genes, OsMADS4 interacted with rice AP3 orthologs SPW1 and facilitate stamen and lodicule development (Yao et al., 2008). In previous studies, MADs box genes have been implicated to modulate stamen development in rice (Yamaguchi et al., 2006; Hu et al., 2011; Yan et al., 2015; Wei et al., 2019), arabidopsis (Ito et al., 2004; Ito et al., 2007), and maize (Schreiber et al., 2004). Two genes, CYP94C2b (cytochrome P450, putative, expressed) and OsARF24 (auxin response factor, putative, expressed) underlie MQTL 12.2 have been reported to enhance salt tolerance in rice (Li et al., 2014) and control cell growth (Kurotani et al., 2015).

In the present study, we integrated QTLs conferring spikelet fertility, panicle number per plant, thousand-grain weight, and number of grains per panicle which led to the identification of 62 MQTLs. 16 of these MQTLs were successfully verified with GWAS studies on the studied traits. 10 MQTLs that harbored at least six of the initial QTLs were selected as the most stable and robust chromosome regions to mine CGs and positional cloning that could be useful for breeding programs. Based on the bioinformatics pipelines, 52 potential CGs were identified to be associated with grain yield. The high expression of these genes was evidenced in yield-related tissues and maybe grain yield regulatory genes. This study provides a useful framework for genetic dissection of grain yield under unstressed conditions via positional cloning, fine mapping, and CGs validation by gene editing technique. The remaining novel CGs detected in this study could be validated or functionally characterized via gene editing, overexpression, knockout techniques, and candidate gene-based association mapping to ascertain their roles in regulating grain yield in future studies.

The original contributions presented in the study are included in the article/ Supplementary Material . Further inquiries can be directed to the corresponding author.

XT, NEO, and KDA conceptualized and design the study. XT supervised the study. KDA, AA, and NEO curated data and perform the analysis. KDA wrote the manuscript. SFB, SA, and AA revised and edited the manuscript. All authors have read and approved the final version of the manuscript.

This study was supported by The Key R&D Project in Hubei Province, China (2020BBB060).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.