CRISPR–Cas immune systems must discriminate between self and non-self to avoid an autoimmune response⁹. In type I and II systems, foreign DNA targets which contain adjacent PAM sequences are targeted for degradation, whereas potential targets in CRISPR loci of the host do not contain PAMs and are avoided by RNA-guided interference complexes^3,5,6,10. Single-molecule and bulk biochemical experiments showed that PAMs act both to recruit Cas9–guide RNA complexes (Cas9–gRNA) to potential target sites and to trigger nuclease domain activation⁷. Cas9 from Streptococcus pyogenes recognizes a 5’-NGG-3’ PAM on the non-target (displaced) DNA strand^4,6, suggesting that PAM recognition may stimulate catalysis through allosteric regulation. Moreover, the HNH nuclease domain of Cas9, which mediates target strand cleavage^4,5, is homologous to other HNH domains that cleave RNA substrates^11,12. Based on the observation that single-stranded DNA (ssDNA) targets can be activated for cleavage by a separate PAMmer oligonucleotide⁷, and that similar HNH domains can cleave RNA, we wondered whether a similar strategy would enable Cas9 to cleave ssRNA targets in a programmable fashion (Fig. 1a).

Using S. pyogenes Cas9 and dual-guide RNAs (Methods), we performed in vitro cleavage experiments using a panel of RNA and DNA targets (Fig. 1b and Extended Table 1). Deoxyribonucleotide-comprised PAMmers specifically activated Cas9 to cleave ssRNA (Fig. 1c), an effect that required a 5’-NGG-3’ or 5’-GG-3’ PAM. RNA cleavage was not observed using ribonucleotide-based PAMmers, suggesting that Cas9 may recognize the local helical geometry and/or deoxyribose moieties within the PAM. Consistent with this idea, dsRNA targets were not cleavable, and RNA–DNA heteroduplexes could only be cleaved when the non-target strand was composed of deoxyribonucleotides. Interestingly, we found that Cas9 cleaved the ssRNA target strand between positions 4 and 5 of the base-paired guide RNA-target RNA hybrid (Fig. 1d), in contrast to the cleavage between positions 3 and 4 observed for dsDNA^3-5 likely due to subtle differences in substrate positioning. However, we did observe a significant reduction in the pseudo-first order cleavage rate constant of PAMmeractivated ssRNA as compared to ssDNA⁷ (Extended Data Fig. 1).

We hypothesized that PAMmer nuclease activation would depend on the stability of the hybridized PAMmer–ssRNA duplex and tested this by varying PAMmer length. As expected, ssRNA cleavage was lost when the predicted melting temperature for the duplex decreased below the temperature used in our experiments (Fig. 1e). In addition, large molar excesses of di- or tri-deoxyribonucleotides in solution were poor activators of Cas9 cleavage (Extended Data Fig. 2). Collectively, these data demonstrate that hybrid substrate structures composed of ssRNA and deoxyribonucleotide-based PAMmers that anneal upstream of the RNA target sequence can be cleaved efficiently by RNA-guided Cas9.

We investigated the binding affinity of catalytically inactive (dCas9; D10A/H840A) dCas9–gRNA for ssRNA targets with and without PAMmers using a gel mobility shift assay. Intriguingly, while our previous results showed that ssDNA and PAMmer-activated ssDNA targets are bound with indistinguishable affinity⁷, PAMmer-activated ssRNA targets were bound >500-fold tighter than ssRNA alone (Fig. 2a,b). A recent crystal structure of Cas9 bound to a ssDNA target revealed deoxyribose-specific van der Waals interactions between the protein and the DNA backbone¹³, suggesting that energetic penalties associated with ssRNA binding must be attenuated by favorable compensatory binding interactions with the provided PAM. The equilibrium dissociation constant measured for a PAMmer–ssRNA substrate was within 5-fold of that for dsDNA (Fig. 2b), and this high-affinity interaction again required a cognate deoxyribonucleotide-comprised 5’-GG-3’ PAM (Fig. 2a). Tight binding also scaled with PAMmer length (Fig. 2c), consistent with the cleavage data presented above.

It was known that Cas9 possesses an intrinsic affinity for RNA, but sequence specificity of the interaction was not explored⁵. Thus, to verify the programmable nature of PAMme-rmediated ssRNA cleavage by Cas9–gRNA, we prepared three distinct guide RNAs (λ2, λ3, and λ4) and showed that their corresponding ssRNA targets could be efficiently cleaved using complementary PAMmers without any detectable cross-reactivity (Fig. 3a). This result indicates that complementary RNA–RNA base-pairing is critical in these reactions. Surprisingly, though, dCas9 programmed with the λ2 guide RNA bound all three PAMmer–ssRNA substrates with similar affinity (Fig. 3b). This observation suggests that high-affinity binding in this case may not require correct base-pairing between the guide RNA and the ssRNA target, particularly given the compensatory role of the PAMmer.

During dsDNA targeting by Cas9–gRNA, duplex melting proceeds directionally from the PAM and strictly requires formation of complementary RNA–DNA base-pairs to offset the energetic costs associated with dsDNA unwinding⁷. We therefore wondered whether binding specificity for ssRNA substrates would be recovered using PAMmers containing 5’-extensions that create a partially double-stranded target region requiring unwinding (Fig. 3c). Indeed, we found that use of a 5’-extended PAMmer enabled dCas9 bearing the λ2 guide sequence to bind sequence-selectively to the λ2 PAMmer–ssRNA target. The λ3 and λ4 PAMmer–ssRNA targets were not recognized (Fig. 3d and Extended Data Fig. 3), although we did observe a 10-fold reduction in overall ssRNA substrate binding affinity. By systematically varying the length of the 5’ extension, we found that PAMmers containing 2–8 additional nucleotides upstream of the 5’-NGG-3’ offer an optimal compromise between gains in binding specificity and concomitant losses in binding affinity and cleavage efficiency (Extended Data Fig. 4).

Next we investigated whether nuclease activation by PAMmers requires base-pairing between the 5’-NGG-3’ and corresponding nucleotides on the ssRNA. Prior studies showed that DNA substrates containing a cognate PAM that is mismatched with the corresponding nucleotides on the target strand are cleaved as efficiently as a fully base-paired PAM⁴. Importantly, this could enable targeting of RNA while precluding binding or cleavage of corresponding genomic DNA sites lacking PAMs (Fig. 4a). To test this possibility, we first demonstrated that Cas9–gRNA cleaves PAMmer–ssRNA substrates regardless of whether or not the PAM is base-paired (Fig. 4b, c). When Cas9–RNA was incubated with both a PAMmer–ssRNA substrate and the corresponding dsDNA template containing a cognate PAM, both targets were cleaved. In contrast, when a dsDNA target lacking a PAM was incubated together with a PAMmer-ssRNA substrate bearing a mismatched 5’-NGG-3’ PAM, Cas9–gRNA selectively targeted the ssRNA for cleavage (Fig. 4c). The same result was obtained using a mismatched PAMmer with a 5’ extension (Fig. 4c), demonstrating that this general strategy enables the specific targeting of RNA transcripts while effectively eliminating any targeting of their corresponding dsDNA template loci.

We next explored whether Cas9-mediated RNA targeting could be applied for tagless transcript isolation from HeLa cells (Fig. 4d). The immobilization of Cas9 on a solid-phase resin is described in Methods (see also Extended Data Fig. 5). As a proof of concept, we first isolated GAPDH mRNA from HeLa total RNA using biotinylated dCas9, gRNAs and PAMmers (Extended Table 2) that target four non-PAM-adjacent sequences within exons 5–7 (Fig. 4e). We observed a substantial enrichment of GAPDH mRNA relative to a control β-actin mRNA by Northern blot analysis, but saw no enrichment using a non-targeting gRNA or dCas9 alone (Fig. 4f).

We then used this approach to isolate endogenous GAPDH transcripts from HeLa cell lysate under physiological conditions. In initial experiments, we found that Cas9–gRNA captured two GAPDH-specific RNA fragments rather than the full-length mRNA (Fig. 4g). Based on the sizes of these bands, we hypothesized that RNA:DNA heteroduplexes formed between the mRNA and PAMmer were cleaved by cellular RNase H. Previous studies have shown that modified DNA oligonucleotides can abrogate RNase H activity¹⁴, and therefore we investigated whether Cas9 would tolerate chemical modifications to the PAMmer. We found that a wide range of modifications (locked nucleic acids, 2’-OMe and 2’-F ribose moieties) still enabled PAMmer-mediated nuclease activation (Extended Data Fig. 6). Importantly, by varying the pattern of 2’-OMe modifications in the PAMmer, we could completely eliminate RNase H-mediated cleavage during the pull-down and successfully isolate intact GAPDH mRNA (Fig. 4g,h). Interestingly, we consistently observed specific isolation of GAPDH mRNA in the absence of any PAMmer, albeit with lower efficiency, suggesting that Cas9–gRNA can bind to GAPDH mRNA through direct RNA:RNA hybridization (Fig. 4f,g and Extended Data Fig. 7). These experiments demonstrate that RNA-guided Cas9 can be used to purify endogenous untagged RNA transcripts. In contrast to current oligonucleotide-mediated RNA-capture methods, this approach works well under physiological salt conditions and does not require crosslinking or large sets of biotinylated probes^15-17.

Here we have demonstrated the ability to re-direct the dsDNA targeting capability of CRISPR/Cas9 for RNA-guided ssRNA binding and/or cleavage (RCas9). Programmable RNA recognition and cleavage has the potential to transform the study of RNA function much as site-specific DNA targeting is changing the landscape of genetic and genomic research⁸ (Extended Data Fig. 8). Although certain engineered proteins such as PPR proteins and Pumilio/FBF (PUF) repeats show promise as platforms for sequence-specific RNA targeting^{18- 22}, these strategies require re-designing the protein for every new RNA sequence of interest. While RNA interference has proven useful for manipulating gene regulation in certain organisms²³, there has been a strong motivation to develop orthogonal nucleic acid-based RNA recognition systems, such as the CRISPR/Cas Type III-B Cmr complex^24-28 and the atypical Cas9 from Francisella novicida^29,30. In contrast to these systems, the molecular basis for RNA recognition by RCas9 is now clear and requires only the design and synthesis of a matching gRNA and complementary PAMmer. The ability to recognize endogenous RNAs within complex mixtures with high affinity and in a programmable manner paves the way for direct transcript detection, analysis and manipulation without the need for genetically encoded affinity tags.