CRISPR-Cas12a exploits R-loop asymmetry to form double-strand breaks

Type V CRISPR-Cas interference proteins use a single RuvC active site to make RNA-guided breaks in double-stranded DNA substrates, an activity essential for both bacterial immunity and genome editing. The best-studied of these enzymes, Cas12a, initiates DNA cutting by forming a 20-nucleotide R-loop in which the guide RNA displaces one strand of a double-helical DNA substrate, positioning the DNase active site for first-strand cleavage. However, crystal structures and biochemical data have not explained how the second strand is cut to complete the double-strand break. Here, we detect intrinsic instability in DNA flanking the RNA-3′ side of R-loops, which Cas12a can exploit to expose second-strand DNA for cutting. Interestingly, DNA flanking the RNA-5′ side of R-loops is not intrinsically unstable. This asymmetry in R-loop structure may explain the uniformity of guide RNA architecture and the single-active-site cleavage mechanism that are fundamental features of all type V CRISPR-Cas systems.


INTRODUCTION 1
CRISPR-Cas systems (clustered regularly interspaced short palindromic repeats, 2 CRISPR-associated proteins) provide antiviral immunity to prokaryotes through the 3 RNA-guided nuclease activity of enzymes including Cas9 and Cas12a (Barrangou et al.  Additionally, the position of the 20-nucleotide spacer sequence in Cas12a crRNAs is 21 opposite to that in Cas9 crRNAs (3′ end versus 5′ end, respectively) (Zetsche et al. We suspected that dsDNA substrates of Cas12a would need to access a bent 1 conformation to undergo target-strand cleavage. To chemically probe the structure of a 2 Cas12a substrate in solution, we performed DNA permanganate footprinting on 3 interference complexes containing a RuvC-inactivated mutant of a Cas12a ortholog 4 from Acidaminococcus species (AsCas12a), hereafter called dCas12a. In this assay, 5 permanganate selectively oxidizes thymines in non-B-form (e.g. locally melted or 6 otherwise distorted) DNA structures, and oxidized positions are subsequently identified 7 through piperidine-catalyzed strand cleavage (which occurs specifically at thymidine 8 glycols) and denaturing polyacrylamide gel electrophoresis (PAGE) ( Fig. 2A) (Bui et al. 9 2003). To enable sensitive detection of DNA fragments, we radiolabeled either the 5′ or 10 3′ end of each DNA strand (3′-end radiolabeling of DNA, which is not a common 11 procedure, was achieved using a protocol developed for the present work, Supp. Fig.  12 2A). Consistent with previous applications of the permanganate assay to CRISPR-Cas-13 generated R-loops (Xiao et al. 2017), thymines within the portion of the non-target 14 strand displaced by the crRNA were heavily oxidized, reflecting the single-strandedness 15 of this DNA tract ( Fig. 2A). 16 Interestingly, we also observed significant oxidation at a thymine near the target-17 strand cleavage site ( Fig. 2A). To probe the region around the target-strand cleavage 18 site more thoroughly, we adjusted the sequence of the DNA substrate to contain a 19 stretch of A/T base pairs in the tract immediately adjacent to the R-loop, which we 20 denote the R-loop flank (Fig. 2B, Supp. Fig. 2B). We assessed the permanganate 21 reactivity of the R-loop flank in three states of the Cas12a cleavage pathway: prior to 22 Cas12a binding (apo DNA), after R-loop formation, and after the first set of cleavage 23 events, which yield a 5-nt gap in the non-target strand (see Appendix B for details of 1 the NTS gap). At each step, we observed an increase in permanganate reactivity on 2 both strands of the DNA that persisted seven base pairs past the end of the crRNA, 3 suggesting that Cas12a binding promotes distortion of DNA in the PAM-distal flank of 4 the R-loop (Fig. 2B, Supp. Fig. 2C, Supp. Note 1). 5 In general, enhanced permanganate reactivity could reflect any of a variety of 6 departures from B-form DNA duplex geometry, ranging from slight helical distortion to 7 complete strand separation (Bui et al. 2003). As a result, the precise conformational 8 ensemble of the R-loop flank cannot be absolutely determined from permanganate 9 reactivity measurements. However, reactivity in the probed region was on the same 10 order of magnitude as that of a fully single-stranded control, suggesting that the 11 detected distortion involves substantial nucleobase unpairing and unstacking (Fig. 2B,  12 Supp. Fig. 2C, Supp. Note 1). Additionally, permanganate reactivity decreased with 13 distance from the R-loop edge (Fig. 2B), consistent with NTS:TS fraying events that 14 initiate from the R-loop edge (Supp. Note 1). The increase in bulk permanganate 15 reactivity in response to NTS cleavage may be due, at least in part, to increased binding 16 occupancy of dCas12a/crRNA on the DNA substrate, as the NTS gap creates a high-17 energy interruption in the DNA rewinding pathway that boosts the stability of the 18 ribonucleoprotein:DNA interaction (Supp. constructs reflect variations in DNA structure rather than differential Cas12a/crRNA 1 binding occupancy. 2 By implementing the permanganate assay on these complexes, we found that 3 the A/T tract was highly reactive in the full R-loop but had limited reactivity in the 4 truncated R-loop, suggesting that the distorted region had migrated with the edge of the 5 R-loop (Fig. 3A). We also observed this effect for a second set of crRNA/DNA 6 sequences with equivalent base-pairing topology, demonstrating that this result is not 7 unique to the originally tested sequence (Supp. Fig. 3B). Positions 19 and 20 of the 8 DNA substrates were G/C base pairs, so DNA conformation at these nucleotides could 9 not be assessed by permanganate reactivity. Nonetheless, these results show that 10 nucleotide unpairing near the target-strand cleavage site depends not only on Cas12a 11 binding and stable R-loop formation, but also on the extent of crRNA strand invasion 12 (i.e., the size of the R-loop). Thus, by altering R-loop size, we can manipulate which 13 nucleotides become unpaired upon Cas12a binding. 14 To test the hypothesis that distortion in the R-loop flank is linked to RuvC-15 mediated target-strand cleavage, we assembled wild type (WT) Cas12a with R-loops of 16 various sizes and determined the distribution of target-strand cut sites. For these 17 experiments, we used DNA substrates with an intact non-target strand that was 18 mismatched with respect to the target strand throughout the region of crRNA 19 complementarity (Fig. 3B)

3C). 23
As the R-loop edge was shifted toward the PAM, the target-strand cut-site 1 distribution shifted toward the new R-loop edge (Fig. 3B, Supp. Fig. 3D). When 2 compensatory mutations were made in the crRNA to restore the original R-loop size, the 3 target-strand cut sites moved back toward their original distribution (Supp. Fig. 3E). The 4 observed shifts in the cut sites were not due to general destabilization of the R-loop, as 5 a single crRNA:TS mismatch at an internal position of the target sequence (position 9) 6 slowed cleavage without affecting cut-site distribution (Supp. Fig. 3E). The cut-site 7 distribution shared by the 1-17*, 1-16*, and 1-15* substrates, along with the lack of 8 cleavage of the 1-14* substrate, may reveal a geometric limit on bent DNA 9 conformations that still permit active site association (Fig 3B). Additionally, the broader 10 target-strand cut-site distribution in the DNA target lacking NTS:TS mismatches (labeled 11 "1-20" in Fig. 3B) could reflect bending events that initiated from partially rewound R-12 loop conformations. Notably, the non-target-strand cut-site distribution did not change 13 markedly as the R-loop was truncated, suggesting that non-target-strand cleavage is 14 unrelated to nucleotide unpairing in the R-loop flank (Supp. Fig. 3F). These results 15 imply that the site of Cas12a-mediated target-strand cleavage is tied to, and perhaps 16 dictated by, the location of weakened base pairing. 17 This principle predicts that R-loop flank sequences with greater nucleobase 18 stacking energy should limit the depth of fraying and, consequently, favor target-strand 19 cleavage events that are closer to the PAM. In agreement with this prediction, of three 20 DNA targets that differed only in the sequence of their R-loop flank-native protospacer, 21 AT-rich, or GC-rich-the GC-rich substrate was cleaved most PAM-proximally (Fig. 3C). 22 Additionally, eliminating NTS:TS base pairing at positions 21-23 led to fast and PAM-23 distally shifted cleavage of the target strand in all cases (Fig. 3C, Supp. Fig. 3G). 1 Together, these results suggest that DNA distortion in the R-loop flank is an important 2 enabler of Cas12a-catalyzed target-strand cleavage. 3 4 Duplex instability is intrinsic to DNA in the RNA-3′ flank of R-loops 5 Next, we wondered what role the Cas12a protein plays in distortion of the R-loop flank. 6 To assess the contribution of the protein, we formed a protein-free mimic of the nucleic 7 acid structure immediately prior to target-strand cleavage. This artificial R-loop 8 contained a pre-cleaved non-target strand that was mismatched with respect to the 9 target strand in the 20-nt stretch adjacent to the PAM, and the same stretch of the target 10 strand was hybridized to a complementary 20-nt RNA oligonucleotide (Fig. 4A). When 11 we probed the permanganate reactivity of this protein-free R-loop, we found that the A/T 12 tract was slightly more reactive than in the Cas12a-generated R-loop (Fig. 4A Duplex instability is not a feature of DNA in the RNA-5′ flank of R-loops 5 While Cas9 also conducts R-loop-dependent DNA cleavage, its R-loop topology is 6 inverted with respect to that of Cas12a as a result of their opposing crRNA 7 architectures-Cas12a crRNAs occur as 5′-repeat-spacer-3′, whereas Cas9 crRNAs 8 occur as 5′-spacer-repeat-3′ (Fig. 1). Given the instability of the Cas12a R-loop flank 9 (referred to as a 3′ R-loop flank because it contains a 3′ RNA terminus), we wondered 10 whether the PAM-distal flank of the Cas9 R-loop (a 5′ R-loop flank) would also be 11

unstable. 12
To test this question, we assayed flank distortion in an R-loop created by a 13 catalytically inactive mutant of Cas9 from Streptococcus pyogenes (dCas9) and in the 14 corresponding protein-free mimic (the non-target strand was pre-cleaved analogously to 15 a Cas12a substrate). Remarkably, we found that the flank experienced nearly 16 background oxidation levels both in the protein-bound R-loop (with dCas9 at a 17 saturating concentration, Supp. Fig. 4B) and in the protein-free mimic, suggesting that 18 unlike 3′ R-loop flanks, 5′ R-loop flanks are not naturally unstable (Fig. 4A, Supp. Fig.  19 4A). The 5′ R-loop flank behaved consistently with expectations about coaxial duplex 20 stacking, as the RNA oligonucleotide protected the DNA:DNA duplex terminus as 21 compared to the bubbled control (Supp. Fig. 4A). Thus, an RNA:DNA hybrid can either 22 stabilize or destabilize a juxtaposed DNA:DNA duplex terminus, depending on whether 23 the hybrid terminus contains a 5′ RNA end or a 3′ RNA end, respectively. These results 1 suggest a fundamental energetic difference in the conformational landscapes of 3′ 2 versus 5′ R-loop flanks (Fig. 4B). 3 The conformational difference between 3′ and 5′ R-loop flanks is intrinsic to 4 strand polarity, as the trends in permanganate reactivity were robust to changes in 5 nucleic acid sequence, end chemistry, and non-target-strand cleavage state (Supp. Fig.  6 4C-F). Additionally, we detected the same polarity dependence when we measured 7 fluorescence intensity of a single 2-aminopurine nucleotide present at position 21 of the 8 original protospacer sequence, indicating that the conformational difference is not an 9 artifact of the permanganate reactivity assay or of the AT-rich sequence of the modified 10 protospacer (Supp. Fig. 4G). Therefore, while Cas12a does not seem to actively 11 destabilize the R-loop flank, the protein forms R-loops with the topology that natively 12 yields a greater degree of flexibility in the region beyond the end of the crRNA. To further explore the unequal stability of 3′ versus 5′ R-loop flanks, we hypothesized 17 that the asymmetry may emerge from energetic differences in the coaxial stacking of a 18 DNA homoduplex on either end of an RNA:DNA hybrid (Fig. 5). Notably, in the 19 structural dataset that most clearly resolves the PAM-distal (RNA-5′) boundary of a 20  Therefore, weaker interhelical stacking energy should lead to more frequent excursions 8 to unstacked or frayed states that promote target-strand cleavage (Protozanova et al. 9

2004). 10
To investigate whether differences in interhelical stacking energy could explain 11 the difference in flank stability of the two R-loop topologies, we designed dumbbell 12 substrates that reduced each type of R-loop boundary to a single chimeric 13 oligonucleotide that contains both an RNA:DNA hybrid and a DNA:DNA homoduplex 14 ( Fig. 5A, Supp. Fig. 5A). A stronger interhelical stack in these dumbbells should 15 increase the thermal stability of the folded state (Erie et al. 1987). Through temperature-16 dependent hyperchromicity measurements, we determined that the RNA-5′ dumbbell 17 (resembling the PAM-distal R-loop edge of Cas9) had a melting temperature 9°C higher 18 than that of the RNA-3′ dumbbell (resembling the PAM-distal R-loop edge of Cas12a) 19 ( Fig. 5A, Supp. Fig. 5A). The observed difference in melting temperature supports the 20 idea that the resistance of 5′ R-loop flanks to permanganate oxidation may emerge from 21 a more stable interhelical stack. 22 To probe the structural and energetic features of the interhelical stacks in atomic 1 detail, we built models of coaxially stacked interhelical junctures of the two types: one 2 containing an RNA-3′ end and one containing an RNA-5′ end. We performed a total of 3 500 nanoseconds of molecular dynamics simulation on each model, and we performed 4 a second set of simulations on models of a different nucleotide sequence. Strikingly, the 5 3′ R-loop junctures frequently unstacked over the course of these short simulations, 6 while the 5′ R-loop junctures remained relatively stable (Fig. 5B, Supp. Fig. 5B).  Fig. 5C). Conversely, in the Cas9-like R-loop, both 1 spacer mimic variants increased the permanganate reactivity level toward that of the 2 bubbled DNA control (Supp. Fig. 5C). This result suggests that the identity of the sugar 3 can affect the interaction of the two juxtaposed helices in ways besides direct steric 4 interference of the 2′-OH (which is pointed away from the juncture in the 5′ R-loop 5 flank). 6 Additionally, the experiments with fully-DNA spacers (Supp. Fig. 5C) revealed 7 that flapped structures, in general, allow juncture-adjacent DNA to explore distorted 8 conformations, consistent with prior nuclear magnetic resonance experiments that 9 detected enhanced local flexibility at DNA nicks (Roll et al. 1998). Given this result, 5′ R-10 loop flanks, whose permanganate reactivity levels approach those of a B-form duplex, 11 stand out as exceptionally rigid structures (Supp. Fig. 4A,C,E-F). We speculate that 12 higher-order features of the two duplexes juxtaposed in 5′ R-loop flanks, such as their 13 helical geometry, encourage unusually stable interhelical stacking, which leads to 14 greater base-pairing stability in the DNA homoduplex terminus. to cleave outside the R-loop (Fig. 1). In type I CRISPR interference complexes, which 23 have the same R-loop topology as Cas12a, the single-strand-specific DNase Cas3 is 1 used to nick the displaced portion of the non-target strand after R-loop formation 2 (Westra et al. 2012), similar to the initial non-target-strand nicking event in Cas12a. 3 Cas3 eventually gains access to the PAM-proximal R-loop flank for processive DNA 4 degradation, but, importantly, it uses an ATP-driven helicase to do so (Mulepati and 5 Bailey 2013). Thus, DNA cleavage in the 3′ R-loop flank of the Cas12a interference 6 complex seems to be a solution that maximizes the utility of its minimal enzymatic While these evolutionary speculations cannot be experimentally verified at 6 present, our findings also provide valuable mechanistic information that will support the 7 development of Cas12-based genome manipulation technologies. For instance, it has 8 been proposed that Cas12-family proteins could be used as RNA-guided DNA 9 "nickases" through selective ablation of target-strand cleavage activity, but those reactivity results revealed that target-strand distortion in Cas12a-generated R-loops can 13 be largely explained by conformational dynamics intrinsic to the nucleic acids (Fig. 4A), 14 arguing against a major role for the protein in substrate contortion. Thus, engineering a 15 type V CRISPR nickase that performs fast non-target-strand cleavage but undetectable 16 target-strand cleavage will likely require modifications that shield or distance the RuvC completely single-stranded, or will it also cut small distortions in dsDNA, as has been 12 observed for S1 nuclease, an unrelated DNase with reported specificity for single-13 stranded substrates (Wiegand et al. 1975)? To compare the substrate range of Cas12a 14 RuvC to that of S1 nuclease, we tested the susceptibility of various radiolabeled DNA 15 structures (including a single strand, a duplex, a nicked duplex, and duplexes with gaps, 16 bubbles, and bulges) to cleavage by the two enzymes, used at concentrations with 17 comparable specific activity (Supp. Fig. 7A). In contrast to S1 nuclease, which 18 exhibited minimal discrimination against even the fully duplex substrate, Cas12a 19 discriminated strongly against substrates with up to 8-nt tracts of non-duplex DNA 20 Fig. 7A). This stringent substrate preference suggests that non-ssDNA 21 structures are either sterically excluded from the RuvC active site or unable to assume 22 catalytic geometry once bound. However, strand discontinuities as small as a nick were 23 sufficient to permit low levels of internal strand cleavage by Cas12a (Supp. Fig. 7A). 1 Notably, the nicked structure resembles the juxtaposed duplexes of an R-loop 2 boundary, and the increased sensitivity of this substrate to trans cleavage may emerge 3 from the same phenomenon that enables cis cleavage of the target strand. Elucidating the mechanism of target-strand cleavage requires an understanding of its 8 interplay with non-target-strand cleavage. To precisely determine the location of 9 Cas12a-catalyzed NTS cleavage, we monitored the formation of DNA cleavage 10 products over time by denaturing PAGE, and we distinguished between different DNA 11 fragments by placing radiolabels on the 5′ or 3′ end of each strand (Fig. 7, Supp. Fig.  12 7B, Supp. Fig. 2A). These experiments were conducted on a timescale and at an 13 enzyme concentration at which cis cleavage events (i.e., events in which a Cas12a 14 molecule cuts the DNA molecule to whose PAM it is bound) are the primary contributor 15 to the observed DNA cutting signal (Supp. Fig. 7C,D). Additionally, while trans 16 cleavage events (i.e., events in which a Cas12a molecule cuts free DNA or DNA bound 17 to another Cas12a molecule) may minorly contribute, the concentration-dependence of 18 the trans cleavage mode allows it to be distinguished from cis cleavage processes 19 Fig. 7E,F). 20 According to these mapping experiments, the non-target strand has two major 21 cleavage sites, at dinucleotides 13/14 and 18/19 (numbers denote distance from the 22 PAM), suggesting the formation of a 5-nt gap within the tract of DNA displaced by the 23 crRNA (Fig. 7B). The evolution of the cleavage pattern over time indicates that, in at 1 least some fraction of the molecules assayed, the NTS is first cut between the two 2 major sites and achieves its final state through two or more "trimming" events. A 3 Cas12a ortholog from Francisella novicida also produced a gap in the NTS, implying 4 that this phenomenon may be conserved across type V-A enzymes (Supp. Fig. 7G). 5 Because our ensemble biochemical assay is blind to the occurrence of additional 6 cuts that occur farther from the radiolabel than the first cut, we cannot unambiguously 7 assign cleavage states to individual interference complex molecules. Nevertheless, by 8 comparing the 5′-and 3′-mapped cut-site distributions at a given timepoint, we can 9 roughly assess the predominant cleavage state of individual molecules. For example, a 10 population of DNA molecules cut exactly once would yield completely overlapping 5′-/3′-11 mapped cut-site distributions, while a population with a gap would yield non-overlapping 12 peaks in the two distributions. 13 Therefore, in the target-strand mapping experiments, overlap of the 5′-and 3′-14 mapped distributions is consistent with (although not uniquely explainable by) a 15 population of interference complexes that have cleaved the TS exactly once (Fig. 7A). 16 However, separation of the two distribution peaks (at dinucleotides 22/23 and 24/25) 17 indicates that most individual complexes perform at least one additional cut in the TS, 18 yielding a small TS gap prior to dissociation of the PAM-distal cleavage product (Fig.  19 7A) (Singh et al. 2018). 20 These results show that while Cas12a activity does generate staggered cuts in 21 dsDNA targets as previously reported (Zetsche et al. 2015), its trimming activity, most 22 notably on the NTS, destroys the capacity of released DNA products to be trivially re-23 ligated in a restriction-enzyme-like procedure. In this particular DNA target, the major 1 Cas12a cleavage products contain a 9-nt overhang on the PAM-proximal fragment and 2 a 6-nt overhang on the PAM-distal fragment, with 4 nt of complementarity between 3 these overhangs (Fig. 7). These findings may aid in the study of eukaryotic DNA repair 4 pathways elicited by Cas12a cleavage events and in biotechnological applications of 5 Cas12a that exploit its staggered cuts.  Fig. 8C). 16 Using a series of NTS variants that were chemically locked in various states of 17 cleavage (i.e., intact, a single nick, and gaps of varying sizes) (Supp. Fig. 8D), we 18 measured the extent of TS cleavage after one hour in calcium-containing buffer. TS 19 cleavage was almost undetectable in the presence of a single nick, and its extent of 20 cleavage only reached that observed with a fully phosphodiester-linked NTS when the 21 gap was widened to 5 nt (Fig. 8A, Supp. Fig. 8E). We observed a similar trend when 22 the experiment was conducted in the presence of magnesium (Supp. Fig. 8F). 23 Together, these results indicate that formation of a gap in the NTS accelerates TS 1 cleavage. Although NTS gap formation is not strictly required for TS cleavage to occur, 2 our bulk cleavage analysis suggests that the NTS gap does in fact form before TS 3 cleavage in the native Cas12a cleavage pathway (Fig. 7). Thus, for most experiments in 4 this work that probed the mechanism of TS cleavage (Figs. 2-5), we used substrates 5 that recapitulated the 5-nt NTS gap (referred to as a "pre-cleaved" or "pre-gapped" 6 NTS). 7 The dependence of TS cleavage on NTS gap formation suggests that the NTS 8 occludes the RuvC active site immediately following R-loop formation. After a NTS 9 nicking event, RuvC releases and rebinds the NTS in different registers to cleave it in 10 multiple locations, forming a gap that clears the active site for entry of the TS. 11 Consistent with a substrate-occlusion model, we determined that trans ssDNA cleavage 12 is enhanced by gap formation in the NTS and only achieves its maximal rate when the 13 TS has also been cut (Supp. Fig. 8G). These observations hint at a possible 14 evolutionary origin for non-specific trans activity-an enzyme that must loosely shuttle 15 multiple cis substrates into and out of a shared catalytic center would benefit from a 16 promiscuous and "open" active site. Therefore, the target-activated non-specific 17 ssDNase activity of type V Cas enzymes may be a mechanistic artifact of single-DNase 18 cis cleavage rather than a direct immunological necessity. 19 To further understand the interplay between non-target-strand and target-strand 20 cleavage, we investigated strand cleavage kinetics in two type V Cas enzymes that 21 have been reported to act as NTS "nickases" (i.e., cleave the NTS but not the TS): the 22  (Fig. 8B, Supp. Fig. 8H). Importantly, slow TS cleavage activity 5 was coupled to slow NTS cleavage activity for both AsCas12a R1226A and Cas12i1 6 ( Fig. 8B, Supp. Fig. 8H), suggesting that slow TS cleavage emerges mostly from low 7 overall catalytic efficiency. While this low efficiency could be explained by weak target 8 association for Cas12i1, which exhibited no detectable DNA binding activity in our filter-9 binding assay, the affinity of AsCas12a R1226A for DNA was unimpaired as compared 10 to WT AsCas12a (Supp. Fig. 8I). Still, both enzymes exhibited kTS:kNTS ratios lower 11 than that of WT AsCas12a, suggesting that there may also be more fundamental 12 differences in their DNA cleavage pathways (Fig. 8B). 13 Specifically, we wondered whether these enzymes were able to form non-target-14 strand gaps. To test this question, we began by performing cleavage site mapping on 15 the NTS of AsCas12a R1226A. At the 1-hour timepoint, the 5′-and 3′-mapped cut-site 16 distributions contained significant overlap, suggestive of a population of DNA strands 17 containing either a single nick or a small gap (Supp. Fig. 8J). In contrast, at the 5-18 second timepoint of the NTS mapping experiments for WT AsCas12a, the two 19 distributions were almost completely non-overlapping, having already developed peaks 20 at the major cut sites (Fig. 7B, Supp. Fig. 8J). Because the 1-hour timepoint for 21 AsCas12a R1226A and the 5-second timepoint for WT AsCas12a have similar values 22 (~70%) of total NTS cleavage (an unambiguous measure of the fraction of molecules 23 that have experienced at least one cut), the difference in distributions implies a 1 fundamental difference across the two enzymes in terms of their relative rates of NTS 2 nicking and trimming (i.e., R1226A has a lower ratio of ktrim:knick than WT) (Fig. 7B,  3 Supp. Fig. 8J). 4 To more directly probe the kinetic contribution of NTS trimming activity, we 5 measured the rate of AsCas12a R1226A TS cleavage in an interference complex with a 6 pre-gapped NTS. The observed TS cleavage rate in this complex was 40-fold higher 7 than in the one with an intact NTS (Fig. 8B), suggesting that the physical basis for the 8 disproportionately slow TS cleavage kinetics of AsCas12a R1226A actually lies in 9 disproportionate slowing of a step prior to TS cleavage (i.e., NTS gap formation). Gap-10 widening cleavage events may be slower than the initial nicking event in this mutant due 11 to the high entropic cost of associating a severed strand with the active site, as nicking 12 of the NTS is expected to boost its conformational freedom (Xiao et al. 2018). 13 Similarly to Cas12a, Cas12i1-mediated TS cleavage depends upon NTS 14 cleavage (Supp. Fig. 8A). However, cleavage site mapping for Cas12i1 revealed that a 15 NTS gap had already formed at the earliest timepoints for which cleavage was 16 detectable (Supp. Fig. 8K). Additionally, Cas12i1-mediated TS cleavage could be only 17 slightly accelerated by pre-gapping the NTS, indicating that NTS trimming is not rate-18 limiting for TS cleavage in Cas12i1 (Fig. 8B, Supp. Fig. 8H). These results suggest that 19 across diverse families of type V interference complexes, the microscopic steps of 20 double-strand break formation can vary in absolute rate and relative kinetic breakdown. 21 Still, while both AsCas12a R1226A and Cas12i1 exhibit kNTS:kTS ratios favorable 22 for "nickase" applications, their use may be limited by their low NTS cleavage rate, 23 which is 10 2 -10 3 times slower than that of WT AsCas12a under the tested conditions. FnCas12a and SpCas9 were expressed with the same protocol as for AsCas12a. 5 Purification of FnCas12a only differed from that of AsCas12a in the lack of a TEV 6 protease cleavage step, as the FnCas12a expression construct lacked a cleavable tag. 7 Purification of SpCas9 only differed from that of AsCas12a in the ion exchange buffers 8 and gel filtration buffer, which contained 10% glycerol instead of 5% glycerol. 9 10

In vitro transcription of RNA 11
Guide RNAs and some short RNAs used in R-loop mimics were produced by in vitro 12 transcription (see Supp.

Nucleic acid and interference complex preparation 10
All DNA oligonucleotides and some RNA oligonucleotides (as indicated in Supp. Table  11 2) were ordered from Integrated DNA Technologies. DNA oligonucleotides used in 12 biochemical experiments were PAGE-purified in house and resuspended in water. A260 13 was measured on a NanoDrop (Thermo Scientific), and concentration was estimated  For 3′ radiolabeling (see Supp. Fig. 2A), which was based on the mechanistic work of 20 to this annealed structure. Ligation was allowed to proceed overnight at 37°C. The 10 phosphate shuttle and splint RNA oligonucleotides were degraded by adding 150 mM 11 NaOH and incubating at 95°C for 10 minutes. The degradation reaction was stopped by 12 adding a stoichiometric amount of HCl and placing on ice. The 3′-radiolabeled DNA 13 oligonucleotide was then buffer-exchanged into 20 mM Tris-Cl (pH 7.9 at 25°C) using a 14 Microspin G-25 spin column. This protocol has ~75% yield in terms of transfer of 15 radioactivity from the phosphate shuttle RNA to the DNA oligonucleotide 3′ end. The hot 16 hydroxide treatment causes slight accumulation of depurination products, but such 17 products comprise a trivial fraction of the total population of radiolabeled DNA and do 18 not interfere with downstream analysis. See Supp. Table 3 and Supp. Table 2 for the 19 identities and sequences (respectively) of oligonucleotide reagents used in 3′-20 radiolabeling procedures. 21

Denaturing polyacrylamide gel electrophoresis and phosphorimaging 23
Radiolabeled DNA oligonucleotides were denatured (95°C in 50% formamide for 3 1 minutes) and resolved on a denaturing polyacrylamide gel (15% acrylamide:bis-2 acrylamide 29:1, 7 M urea, 0.5X TBE). Gels were dried (4 hours, 80°C) on a gel dryer 3 (Bio-Rad) and exposed to a phosphor screen. Phosphor screens were imaged on an 4 Amersham Typhoon phosphorimager (GE Healthcare). Phosphorimages were 5 quantified using ImageQuant software (GE Healthcare). 6 7 Electrophoretic mobility shift assay and filter-binding assay 8 In both kinds of binding assays, complexes were formed in 1X binding buffer (20 mM 9 Tris-Cl, pH 7.9 at 25°C, 150 mM KCl, 5 mM MgCl2, 1 mM TCEP, 50 µg/mL heparin, 50 10 µg/mL bovine serum albumin, 5% glycerol). Cas protein was first diluted in series in 11 binding buffer, added to a fixed concentration of guide RNA, and incubated at 37°C for 5 12 minutes, then 25°C for 25 minutes. This complex was then added to the radiolabeled 13 DNA probe and incubated at 37°C for 5 minutes, then 25°C for 1 hour. When the titrant 14 was crRNA instead of Cas protein, the Cas12a:crRNA complex was incubated at 25°C 15 for 30 minutes, added to DNA probe, and incubated at 25°C for an additional 1 hour. 16 For the EMSA, samples were then resolved on a native PAGE gel (8% acrylamide:bis-17 acrylamide 29:1, 0.5X TBE, 5 mM MgCl2), which was dried and phosphorimaged. For were air-dried and phosphorimaged. For assays testing complex assembly in calcium-1 containing buffer, 5 mM CaCl2 was substituted for MgCl2 in the binding buffer. 2 Oligonucleotide identities and sequences are shown in Supp. Table 4 and Supp. Table  3 2, respectively. Cas12i1, the 2X quench buffer also included 400 µg/mL heparin and 0.2% sodium 17 dodecyl sulfate to prevent aggregation in gel wells. For "t=0" timepoints, surveillance 18 complex was first added to quench buffer and mixed, followed by addition of DNA 19 substrate. Products were then resolved by denaturing PAGE and phosphorimaging. 20 Oligonucleotide identities and sequences are shown in Supp. Table 4 and Supp. Table  21 2, respectively.  All oligonucleotides used in these experiments were first ethanol-precipitated to remove 20 impurities from commercial synthesis that interfered with the optical spectra of interest. 21 Oligonucleotides were resuspended to an estimated 2.25 µM (the extinction coefficient 22 of a highly stacked nucleic acid structure is difficult to estimate, but the unimolecular 23 physical processes being probed are concentration-independent, in theory) in 1X 1 nucleic acid spectroscopy buffer (10 mM K2HPO4/KH2PO4, pH 6.7, 150 mM KCl, 0.1 2 mM EDTA). Samples were placed in a 1 cm CD-grade quartz cuvette (Starna Cells) 3 with a stir bar and cap, which was placed in the sample cell of a temperature-controlled 4 spectrophotometer (Cary UV-Vis 100). An equivalent cuvette containing only nucleic 5 acid spectroscopy buffer was placed in the reference cell. The stir apparatus was turned 6 on, the block was heated to 95°C, and the samples were allowed to equilibrate for 3 7 minutes. The system was cooled to 2°C at 1°C/minute, collecting an A260 measurement 8 every 0.5°C (averaging time = 2 s, slit bandwidth = 1 nm). Refolding of the Cas12a-like 9 dumbbell at a slower temperature ramp rate (0.3°C/minute) yielded results similar to 10 those pictured, indicating that the faster ramp rate (1°C/minute) was still slow enough 11 that the absorbance measurements approached their equilibrium values. 12 Oligonucleotide identities and sequences are shown in Supp. Table 4 and Supp. Table  13 2, respectively. 14 15

Molecular dynamics simulations 16
The starting conformation of each juncture was based on a relaxed structure of a 5′ juncture). These systems were solvated and minimized as before. The systems were 15 then equilibrated for 1 ns with the nucleic acid atoms held fixed. This system served as 16 the starting state for 10 separate production trajectories that were each run for 50 ns 17 with all atoms free. All equilibration and production runs were carried out in the NPT 18 ensemble at a temperature of 300 K and pressure of 1 atm. 19 The simulations were performed on XSEDE computing resources (Towns et al. For each trajectory, the coordinates of the two nucleobases at the juncture on the 7 flapped strand were isolated for further analysis at a sampling rate of 1/ps. Envelope 8 surface area (ESA), defined as the solvent-exposed surface area of the two isolated 9 nucleobases, was determined in PyMOL and serves as a metric of the degree of base 10 stacking (bases that are well-stacked have a low ESA, whereas bases that are 11 unstacked have a high ESA). All figures were prepared in PyMOL.

Model fitting 5
All models were fit by the least-squares method in Prism 7 (GraphPad Software). The 6 model used for each dataset is described in the corresponding figure legend. 7 8 ACKNOWLEDGMENTS 9 We thank the lab of Andreas Martin for use of their temperature-controlled 10 spectrofluorometer. We thank Gavin Knott, Andreas Martin, David Wemmer, and Kevan 11 Shokat for helpful discussions. We thank Lucas Harrington and Janice Chen for a 12 critical reading of the manuscript. This work used the Extreme Science and Engineering