Soybean DICER-LIKE2 regulates seed coat color via production of primary 22-nucleotide small interfering RNAs from long inverted repeats

In plants, 22-nucleotide small RNAs trigger the production of secondary small interfering RNAs (siRNAs) and enhance silencing. DICER-LIKE2 (DCL2)-dependent 22-nucleotide siRNAs are rare in Arabidopsis (Arabidopsis thaliana) and are thought to function mainly during viral infection; by contrast, these siRNAs are abundant in many crops such as soybean (Glycine max) and maize (Zea mays). Here, we studied soybean 22-nucleotide siRNAs by applying CRISPR-Cas9 to simultaneously knock out the two copies of soybean DCL2, GmDCL2a and GmDCL2b, in the Tianlong1 cultivar. Small RNA sequencing revealed that most 22-nucleotide siRNAs are derived from long inverted repeats (LIRs) and disappeared in the Gmdcl2a/2b double mutant. De novo assembly of a Tianlong1 reference genome and transcriptome profiling identified an intronic LIR formed by the chalcone synthase (CHS) genes CHS1 and CHS3. This LIR is the source of primary 22-nucleotide siRNAs that target other CHS genes and trigger the production of secondary 21-nucleotide siRNAs. Disruption of this process in Gmdcl2a/2b mutants substantially increased CHS mRNA levels in the seed coat, thus changing the coat color from yellow to brown. Our results demonstrated that endogenous LIR-derived transcripts in soybean are predominantly processed by GmDCL2 into 22-nucleotide siRNAs and uncovered a role for DCL2 in regulating natural traits.