Discovery of the biogenesis pathway of tiny RNAs (LINK)
MicroRNAs (miRNAs) are 20~25 nucleotide(nt)-length noncoding RNAs that are loaded into Argonaute proteins (AGOs) to assemble the RNA-induced silencing complexes (RISCs) (Our Reviews in 2022, and 2016). Although tiny RNAs (tyRNAs) shorter than 19 nt have been found to bind to plant and vertebrate AGOs, their biogenesis remains a long-standing question (Our Review in 2021 ). Our in vivo and in vitro studies show several 3′→5′ exonucleases, such as interferon-stimulated gene 20 kDa (ISG20), three prime repair exonuclease 1 (TREX1), and ERI1 (enhanced RNAi, also known as 3′hExo), capable of trimming AGO-associated full-length miRNAs to 14 nt or shorter tyRNAs. Their guide trimming occurs in a manganese-dependent manner but independently of the guide sequence and the loaded four human AGO paralogs. We also show that ISG20-mediated guide trimming makes Argonaute3 (AGO3) a slicer. Given the high Mn2+ concentrations in stressed cells, virus-infected cells, and neurodegeneration, our study sheds light on the roles of the Mn2+-dependent exonucleases in remodeling gene silencing.

Human AGO3 becomes a slicer with cityRNAs (cleavage-inducing tiny guide RNAs)
Among four human AGOs, only AGO2 has been shown to have slicer activity when target RNAs are fully complementary to the guide. We recently discovered that specific miRNAs such as miR-20a, but neither let-7, miR-16, nor miR-19b, catalytically activates AGO3, albeit the activity was much lower than that of AGO2 (LINK). Our following study revealed that AGO3 becomes a competitive slicer of AGO2 when loaded with 14-nt tiny RNAs (tyRNAs) of miR-20a whose 3’ 8~9 nt are deleted (LINK). Surprisingly, even a 14-nt tyRNA of let-7a converted AGO3 to a slicer. In contrast, AGO2 drastically decreased the slicing activity when loaded with those tyRNAs. Our findings demonstrate that AGO2 and AGO3 have different optimum lengths of guide RNA for slicing activity. We defined tyRNAs such as the 14-nt miR-20a and let-7a as cleavage-inducing tyRNAs (cityRNAs). To our best knowledge, this is the first example that miRNAs change their roles depending on the length (see also Our Review).

RNA-induced silencing complex (LINK1, and LINK2)
It has been thought that miRNAs are randomly loaded into four human Argonaute proteins (AGOs). However, deep sequencing analyses of endogenous miRNAs associated with human AGOs show partially different preferences of miRNAs, though the sorting mechanism remains unclear. Another remaining question is their target specificity. There is a
consensus that miRNAs recruit the AGOs to specific sites on the target mRNA(s) solely depending on their sequence complementarity. However, recent studies have reported the specialized function of four human AGOs, which cannot be explained simply by base pairing between the guide and target RNAs. These indicate that more than the base complementarity needs to be taken into account to understand the mechanisms of the different target specificities among the four AGOs.
We determined the first structures of human Argonaute3 and Argonaute4 in complex with guide RNA. Their structures revealed specific local structures that make their nucleic acid-binding channel different shapes.

Argonaute proteins are excellent samples to study how ribonucleoproteins are assembled (LINK)
Hydrophobic effect is the primary driving force of protein folding by excluding water molecules from the inside of protein. Our 1.9 Å crystal structure of human Argonaute4 showed 17 water molecules trapped inside, which seemed to be violating the rule of protein folding. Our combinatorial approach of biochemistry and molecular dynamics suggests that the trapped water molecules are exchangeable with solvent and that each of the water molecules moves from a specific position to the other specific ones. The observation suggests that the trapped water molecules work as a glue to tie together domains to retain the RISC structure. This water gluing may also be important to dissociate the domains when the RISC is disassembled to exchange guide RNAs.

Application to a programmable RNA-restriction enzyme (LINK)
Nature has designed many DNA restriction enzymes. Most of them take a simple strategy by which they form a homodimer to recognize a 6~8-nt palindromic sequence. DNA restriction enzymes could take the same strategy because DNAs usually remain the double-stranded helical structure, regardless of sequence. This is not the case of RNAs because RNAs can be folded into different structures depending on their nucleotide sequence. That’s why nature seems to have given up all hope of designing RNA restriction enzymes systematically. But our group made it.
To make a cost-effective system, we programmed yeast Argonaute from Kluyveromyces polysporus with 23-nt single-stranded DNAs, although this Argonaute naturally uses single-stranded RNAs as guides to cleave target RNAs. We name our deoxyribonucleoprotein complex ‘DNA-induced slicing complex (DISC).’ Our study showed that DISC cleaved structured RNAs such as the 5’ untranslated region of HIV-1, indicating that DISC unwinds even structured RNAs and cleaves the target site that is complementary to the loaded guide DNA. We continue to develop the DISC system.

Argonaute proteins do more than providing the binding site for guide and target strands (Link)
Argonaute proteins have a bilobal structure. The intervening channel between the N-and C-terminal lobes serves as the nucleic acid-binding channel to accommodate the guide and target RNAs. It is known that after loaded into Argonaute, the guides do not follow the rules of binding and dissociation that nucleic acids possess, but rather behave as if they are part of an RNA-binding protein. The idea that the binding of a guide to targets depends solely on their sequence complementarity was challenged. We succeeded in designing a stable C-terminal lobe that recognizes guides as does its full-length Argonaute, even in the absence of the N-terminal lobe. This miniature Argonaute retains a competitive guide-dependent RNA cleavage. Our studies revealed that the N-terminal lobe is essential, not to cleavage RNAs, but to modulate the slicing activity of the catalytic C-terminal lobe by recognizing mismatches between the guide and target strands. We also discussed that Argonaute and CAS9 take a similar strategy in that both ribonucleoproteins recognize only a few nucleotides within their target sequence, before forming an extensive base pairing.