Background

Important agronomic traits in crops, such as yield, quality, and stress resistance, are typically complex traits controlled by multiple genes. Genome-wide association studies (GWAS) are a primary tool for dissecting the genetic basis of these traits. However, in many crops like rice, extensive linkage disequilibrium (LD) and the polygenic nature of traits mean that a single GWAS locus can harbor dozens of genes, making the identification of the true causal gene a formidable challenge. This significantly limits the practical application of GWAS findings.

The advent of transcriptomics has opened new avenues to tackle this issue. Gene expression, as an intermediate layer between genotype and phenotype, is itself under genetic control. Methods like expression quantitative trait loci (eQTL) analysis and TWAS link genetic variants to gene expression levels, providing functional evidence to prioritize candidate genes. Recent TWAS frameworks (e.g., PrediXcan- and FUSION-based approaches) and eQTL-guided colocalization analyses provide functional evidence to link genes with complex traits and have been applied in both human and plant genetics [1–3]. These models predict gene expression from the genotype and test its association with phenotypes, offering a mechanistic layer beyond SNP-trait correlations. Colocalization-based methods further evaluate whether GWAS and eQTL signals share a common causal variant, helping prioritize functional loci. While these approaches have proven valuable, their performance in crops is often hindered by extensive LD, complex population structure, and the difficulty of distinguishing traits driven by local regulation from those influenced by distal regulatory networks.

To address this challenge, we developed the CisTrans-ECAS (cis- and trans-expression component-based association study) method. Its core concept is that the expression variation of a gene can be partitioned into two components: 1) the cis-expression component (cis-EC), in which expression variation is explained by local genetic variants (e.g., within 100 kb upstream and downstream); and 2) the trans-expression component (trans-EC), in which expression variation is driven by distal regulatory factors or other genetic influences not captured by the cis-EC.

Unlike traditional TWAS methods that rely on total gene expression or predicted expression levels, CisTrans-ECAS explicitly decomposes expression into cis- and trans-regulated components. This separation allows the method to distinguish between local regulatory effects and broader regulatory network influences, providing higher specificity for identifying causal genes in regions of extensive LD. Our rationale is that if a gene is a true functional participant in a biological process affecting a trait, its association with the trait should be evident from two independent sources. Its local regulation (cis-EC) should be linked to the trait, and its regulation within a broader network (trans-EC) should also be associated with the trait.

Therefore, a gene whose cis-EC and trans-EC are both significantly associated with a target trait is a much stronger causal candidate than a gene with only one type of association. This dual-evidence strategy effectively filters out false positives arising from LD, enabling the precise identification of causal genes, even in regions with weak GWAS signals. Furthermore, by treating the expression of other genes as molecular phenotypes (e-traits), this method can efficiently map upstream and downstream regulatory relationships, providing robust support for establishing reliable gene regulatory networks.