Transcriptome-wide association analyses reveal the impact of regulatory variants on rice panicle architecture and causal gene regulatory networks

Rice, *Oryza sativa*, is a crucial food crop. Panicle architecture, a key factor in determining grain yield, involves complex genetic and regulatory mechanisms. Previous research, often based on mutant analyses or bi-parental populations, has identified several genes involved in panicle development (e.g., *OSHI*, *LAXI*, *Gnla*, *OsSPL14*, *DEPI*, *FZP*). However, a comprehensive understanding of their regulatory networks and the favorable alleles for enhanced yield remains elusive. Genome-wide association studies (GWAS) have explored natural variations in panicle traits, but the large linkage disequilibrium (LD) decay distance in the rice genome and the complexity of the genetic basis have limited the identification of genes under quantitative trait loci (QTLs). Gene expression variation, regulated by *cis*-variants or *trans*-factors, significantly contributes to phenotypic diversity. In rice, several genes (*Gnla*, *IPA1*, *FZP*) exhibit *cis*-variants regulating gene transcription and influencing panicle architecture. Inspired by the human Genotype-Tissue Expression (GTEx) project, the authors aimed to develop an analysis strategy suitable for crops to understand gene regulation related to rice panicle traits using a population transcriptome approach.

Existing research has identified several key genes impacting rice panicle architecture through various approaches, including mutant analysis and bi-parental population studies. Genes like *OSHI*, *LAXI*, *Gnla*, *OsSPL14*, *DEPI*, and *FZP* have been implicated in this process. However, a complete understanding of their regulatory networks and the identification of favorable alleles remain incomplete. GWAS studies have been conducted to uncover natural variation influencing these traits; however, the complexity of the genetic architecture and the extensive LD decay distance in the rice genome pose significant challenges in pinpointing causative genes within QTL regions. The significance of gene expression variation regulated by both *cis*- and *trans*-acting factors is increasingly recognized. Several rice genes, including *Gnla*, *IPA1*, and *FZP*, have been shown to harbor *cis*-variants that directly influence their expression levels, thus affecting panicle development. The study draws inspiration from the successful human GTEx project which highlights the power of population transcriptome data in uncovering genetic variations underlying gene expression differences.

This study employed a multi-faceted approach integrating field experiments, transcriptome sequencing, and advanced statistical analyses. Field experiments were conducted across multiple locations and years using a population of 529 representative rice varieties. Panicle traits, including spikelets per panicle (SPP), number of primary branches (NPB), and panicle length (PL), were meticulously measured. Best linear unbiased prediction (BLUP) models were used to integrate phenotypic data across experiments. Genome-wide association studies (GWAS) were performed using linear mixed models to identify QTLs associated with panicle traits. Transcriptome data were generated for 275 representative rice varieties using 1–2 mm young panicle samples. RNA sequencing was performed using the Illumina HiSeqX platform. Transcriptome-wide association studies (TWAS) were conducted by associating gene expression levels with panicle traits. A novel method combining information on cis- and trans-expression components of genes was developed to identify causal genes even at small-effect loci. This involved estimating cis-expression components (cis-EC) using GCTA and calculating trans-expression components (trans-EC) as residuals after removing cis-EC. Cis- and trans-ECAS (expression component-based association studies) were performed. The derived allele frequencies (DAF) at *cis*-eQTLs were analyzed to assess the impact of selection on gene expression and panicle traits. Functional validation of key genes, including *OsMADS17*, was performed using CRISPR/Cas9 gene editing and transient transcriptional activity assays.

GWAS revealed a complex genetic architecture for panicle traits, with SPP and NPB primarily influenced by numerous small-effect loci. TWAS identified thousands of genes significantly associated with panicle traits. Among the top SPP-associated genes were several transcription factors, including MADS-box and YABBY genes. Known panicle development genes, such as *OsSPL14*, *OSH1*, *OsPID*, and *PLAI*, were also identified. Positive selection on derived alleles at *cis*-eQTLs of SPP-associated genes was observed, favoring alleles increasing SPP. The newly developed method, integrating cis- and trans-expression components, pinpointed 36 putative causal genes for SPP. *SDT* (*MIR156j*) and *OsMADS17* were identified as causal genes; *OsMADS17* was experimentally validated to regulate *SDT* expression. Analysis of *cis*-eQTL DAF revealed that high-frequency derived alleles tend to upregulate PAGs or downregulate NAGs, demonstrating positive selection for alleles improving SPP. Functional validation of *OsMADS17* using CRISPR-Cas9 showed a significant increase in SPP and NSB in knockout lines. Analysis of natural variation in *OsMADS17* identified a superior allele associated with increased SPP, which was successfully introgressed into a breeding line, further enhancing NSB and SPP.

The findings address the research question by revealing the complex interplay of numerous small-effect loci and regulatory variants in shaping rice panicle architecture. The identification of key genes, such as *SDT* and *OsMADS17*, and the elucidation of their regulatory interactions offer crucial insights into the underlying mechanisms. The significance of the results lies in the potential for targeted molecular breeding strategies. The integration of cis- and trans-eQTL data significantly improved the power of detecting causal genes, even those with weak GWAS signals, demonstrating the utility of the developed method. The observed positive selection on derived alleles affecting SPP highlights the role of regulatory variations during rice domestication and improvement. The study provides a framework for identifying and validating causal genes in other crops.

This study provides a comprehensive understanding of the genetic architecture and regulatory networks governing rice panicle architecture. The identification of key genes like *SDT* and *OsMADS17*, along with the development of a novel method to pinpoint causal genes at small-effect loci, offers valuable tools for future molecular breeding efforts. Future research could focus on further elucidating the entire regulatory network, exploring the functional roles of other identified genes, and developing marker-assisted selection strategies based on identified superior alleles.

The study's limitations include the potential for confounding factors in field experiments despite the use of BLUP models and the focus on a specific developmental stage (1–2 mm young panicles). While the developed method enhances causal gene identification, it relies on the availability of high-quality transcriptome data and robust phenotype data. The generalizability of findings to other rice varieties and diverse environments warrants further investigation.