Structure reveals mechanisms of viral suppressors that intercept a CRISPR RNA-guided surveillance complex

Structural overview of the Csy complex

The Csy complex is a crRNA-guided surveillance complex composed of nine Cas proteins (one Cas6f, six Cas7fs, one Cas5f, and one Cas8f protein) and a single 60-nt crRNA (Figure 1) (van Duijn et al., 2012; Wiedenheft et al., 2011). Overall, the morphology of the complex is consistent with the seahorse-shape that has been described for other class 1 CRISPR-Cas complexes, with subunits referred to as the head (Cas6), backbone (Cas7), and tail (Cas5 and Cas8) (Figures 1 and 2) (Jackson and Wiedenheft, 2015; Zhang and Sontheimer, 2014). The crRNA is an integral component of the complex, making direct contact with all nine protein subunits (Figure 1). The crRNA performs an important structural role in complex assembly by tethering the protein subunits of the complex together and defining the number of Cas7 subunits incorporated into the backbone (Kuznedelov et al., 2016; Luo et al., 2016; Wiedenheft et al., 2011). The complex is assembled such that a portion of the crRNA (sometimes called the spacer) guides the complex to complementary nucleic acids found in phage and plasmid genomes, and target binding recruits the Cas2/3 nuclease-helicase protein for degradation of the target. Phage encoded anti-CRISPR proteins AcrF1 and AcrF2 (Figures 1B–D, red and green, respectively) block crRNA-guided target recognition by binding to either the Cas7f backbone (AcrF1), which blocks hybridization to a complementary target, or by inserting between the Cas8f and Cas7f subunits in the tail (AcrF2), which prevents interactions with the target DNA duplex (Figure 1) (Bondy-Denomy et al., 2015).

The CRISPR loci in P. aeruginosa are composed of 28-nucleotide repeat sequences separated by 32-nucleotide phage or plasmid-derived spacer sequences. The repeat sequences are partially palindromic, giving rise to a series of stable stem-loop structures within the long precursor CRISPR transcripts. These stem-loop structures are recognized by a CRISPR-specific endoribonuclease, Cas6f (formerly Csy4), which cleaves the CRISPR RNA at the 3′-end of each stem-loop to produce a library of 60-nt crRNAs required for assembly of the Csy surveillance complex (Wiedenheft et al., 2011). Cas6f and the 3′-stem-loop of the crRNA are not well resolved in cryo-EM density, suggesting that the linkage tethering Cas6f to the rest of the complex is flexible (Figure S2C). However, a high-resolution crystal structure of the P. aeruginosa Cas6 protein bound to the 3′ stem-loop of the crRNA was previously determined (Haurwitz et al., 2010) and the EM density in this region was sufficiently resolved to enable unambiguous rigid body fitting of the atomic model into position at the head of the complex (Figure 1). In addition, two copies of an NMR structure for AcrF1 were used to facilitate model building (Maxwell et al., 2016). Models for AcrF2 and all other Cas proteins were built de novo using the EM density. The tail of the complex, consisting primarily of Cas5f (formerly Csy2) and Cas8f (formerly Csy1) were not as well-resolved as the Cas7f backbone, so a focused classification and refinement strategy using signal subtraction was employed to improve the density in this region for de novo atomic model building (see Methods) (Figures S4 and S5).

Ordered presentation of the crRNA-guide

Phylogenetic and structural studies suggest that multi-subunit crRNA-guided surveillance complexes found in type I and type III immune systems evolved from a common ancestor (Jackson and Wiedenheft, 2015; Makarova et al., 2015). A unifying feature of these systems, which are now collectively recognized as class 1 immune systems, is the helical assembly of a Cas7 family protein along the crRNA (Figure 2A). Consistent with previously determined Cas7 structures, the Cas7f protein from P. aeruginosa shares a “right-hand” morphology composed of fingers-, palm-, web-, and thumb-shaped domains (Figure 2B). In P. aeruginosa, crRNA binding by Cas7f is mediated by non-sequence specific contacts between the sugar-phosphate backbone and residues on the palm (R35, H275, Q277, K278, N281, R284) and web (R169, Q248). The thumb of Cas7.1f is anticipated to be part of the flexible tether that connects the Cas7 backbone to the Cas6 head, but due to the flexibility of the head, the density for the thumb is not well resolved. In contrast, thumbs of the remaining Cas7f subunits (Cas7.2f–7.6f) are well-ordered and fold over the top of the crRNA and across the palm of the adjacent subunit (Figure 2A and 2B). Similar to what has been reported for the other class 1 complexes, Cas7f oligomerization along the crRNA introduces distortions in the sugar-phosphate backbone, resulting in “kinks” at regular 6-nucleotide intervals (Taylor et al., 2015). The thumb of each Cas7f subunit folds over the top of each kink, burying one nucleobase at positions -1, 6, 12, 18, 24, and 30 of the crRNA (Figures 2A–C). The five nucleotides between each kink are ordered in a pseudo-A-form configuration and the sugar-phosphate backbone of each helical segment superimposes on the backbone of each of the other segments (e.g. nucleotides 1–5 superimpose on nucleotides 7–11, etc.) with an RMSD of less than 0.37 Å for equivalently positioned phosphates.

Presentation of the crRNA guide in helical segments of 5-nucleotides is a conserved feature of class 1 surveillance complexes. In fact, the sugar-phosphate backbones for helical segments of the crRNA from Csy (type I-F), Cascade (type I-E), and Cmr (type III-B) superimpose with an RMSD of less than 1.0 Å (Figure 2C). While the structure of each helical segment is conserved among class 1 surveillance complexes, torsion angles at each kink differ significantly between these systems. Torsion angles are determined by specific interactions between the crRNA phosphate backbone and the palm, web, and thumb domains of the subtype-specific Cas7 proteins. The resulting differences in the torsion angles at each kink correspond with large-scale structural differences in the helical pitch of the Cas7 backbone (Figure 2D). The backbone of the IF Csy complex adopts a substantially tighter pitch (80.85 Å) compared to the I-E (127.63 Å) or III-B complexes (158.94 Å). The tighter spiral of the Csy backbone results in the head and tail subunits being positioned in close proximity, creating a nearly closed ring-like structure. The functional significance of these structural differences are currently not well understood, but we do know that most anti-CRISPRs are subtype-specific (Pawluk et al., 2014; Pawluk et al., 2016), and these preferences likely reflect structural features that are unique to each of the different subtypes.

While the “right-hand” morphology used to describe Cas7 family proteins is conserved in Cas7f, the Cas7f fold is distinct (Figure S6). Previously determined Cas7 structures contain palm domains with the βαββαβ topology of an RNA recognition motif (RRM), where the two antiparallel α-helices pack against one face of a four-stranded antiparallel β-sheet that has a specific β₂β₃β₁β₄ arrangement (Topuzlu and Lawrence, 2016). In contrast, the Cas7f palm domain is an antiparallel β-sheet with a notably different β₃β₂β₄β₁ arrangement, and a structural homology search performed using the Cas7f palm domain reveals similarity to the PAD domain of eukaryotic polymerases, rather than other Cas7 family proteins (Holm and Sander, 1993; Trincao et al., 2001). Furthermore, previously determined Cas7 structures contain highly decorated RRMs with multiple insertions that form the fingers and thumb domains, while the Cas7f palm contains only one insertion, with the thumb and fingers domains formed by N- and C-terminal extensions, respectively (Figure S6). In addition to the previously described domains, Cas7f contains a 38-residue loop in the C-terminal extension, termed the “extended web”, which packs against the thumb, creating a prominent trough connecting the web and thumb domains. Structurally equivalent loops exist in other type I Cas7 proteins, although they are significantly smaller.

Organization of the tail

The tail of the Csy complex is composed of a stable Cas5f/Cas8f heterodimer, which recognizes a conserved S-shaped RNA structure, called the 5′-handle, which is formed by the final 8-nucleotides on the 5′-end of the crRNA (Figures 3A–D and S3D). The Cas5f protein adopts a “left-handed fist” morphology, consisting of a canonical RRM (i.e. β₂α₂β₃β₁α₁β₄) that forms the palm domain, flanked by a four-stranded antiparallel beta sheet (β₂β₁β₄β₃) that resembles fingers, and a loop that forms a thumb-like feature (Figure 3B). Each of these anatomical features plays a role in binding Cas8f, the 5′-handle of the crRNA, or both (Figures 3A–D).

The Cas5f fingers “grip” a long β-hairpin on Cas8f (residues 175–191) that extends through the Cas5f subunit. We refer to this β-hairpin as the Cas8f harpoon (Figure 3B). The last two nucleotides of the crRNA 5′-handle (positions -8C and -7U) are sandwiched between the Cas8f harpoon and the first α-helix of the Cas5f RRM, while the next nucleotide at position -6A is displaced and the phosphate backbone is kinked by interactions with lysine 176 (K176) on Cas8f and asparagine 21 (N21) on Cas5f (Figures 3C–E). This network of interactions creates the first curve of the S-shaped 5′-handle, which transitions into the next curve consisting of fours bases (positions -5A, -4G, -3A, and -2A) that are arranged in a pseudo-A-form configuration (Figures 4C and 4D). The phosphate backbone these four nucleotides is stabilized by non-sequence specific interactions with Cas7.6f, while bases on either end of the helical stack are stabilized by interaction with Cas5f (i.e. R271 and N86).

The thumb of Cas5f is formed by a large insertion between β₂ and β₃ of the RRM (residues 48–109). This thumb functions similarly to the Cas7 thumbs by folding over the crRNA to mask the displaced base at position -1A (Figure 4C). In the type I-E surveillance complex, interactions between the Cas5 thumb and the fingers domain of Cas7.6 coincide with a ~180° rotation of the Cas7 fingers. This rotation creates a “lysine-rich vise” between Cas7.5 and 7.6 that is critical for binding dsDNA (Jackson et al., 2014a; van Erp et al., 2015). Notably, no comparable rotation is observed in the sixth Cas7f subunit of the Csy complex (Cas7.6f), suggesting that this complex uses an alternative mechanism for binding dsDNA.

With the exception of a recently described type I-F variant (Type I-Fv) (Gleditzsch et al., 2016), crRNA-guided surveillance complexes contain a large tail protein called Cas8, but their amino acid sequences are diverse, and Cas8f of P. areuginosa has novel fold. Unlike the Cas8 tail in Cascade, which comprises a large globular domain and a C-terminal four-helix bundle, Cas8f has an extended architecture with three discrete domains - an N-terminal “hook” shaped domain (residues 1–166), a central domain (residues 167–264), and a C-terminal helical bundle that occupies the belly of the Csy complex (residues 265–435) (Figure 3B). Although helical bundles have been observed at the C-termini of Cas8 family members, the central location of this helical bundle in P. aeruginosa is unique, contacting Cas6, Cas7.2f, and Cas7.3f at the head and upper spiral of the complex. By contrast, the corresponding helical bundle within the I-E system only contacts Cas5, the crRNA, and the tail-proximal Cse2 “belly” protein. The density for this region of the reconstruction was not as well-ordered as the rest of the structure, indicating structural flexibility.

AcrF1 proteins prevent DNA hybridization

AcrF1 binds to the Cas7f backbone and prevents target binding, but the mechanism and location of these interactions has not yet been determined (Bondy-Denomy et al., 2015). The structure presented here explains how two AcrF1 proteins (i.e. AcrF1.1 and 1.2) bind to one Csy complex through interactions with four of the six Cas7f subunits (Figures 1 and 4). AcrF1 is a small protein (78 residues) composed of a four-stranded antiparallel β-sheet, flanked on one side by two α-helices (β₁β₂β₃β₄α₁α₂) (Figure 1D and 2). The AcrF1 structure was previously determined by NMR, and mutational studies identified three residues (Y6, Y20, and E31) on one face of the β-sheet that are critical for AcrF1-mediated repression of type I-F immune response (Maxwell et al., 2016). Here we show that these three residues interact with a single, conserved lysine (K85) on the Cas7f thumb, and that a lysine to alanine mutation at this position (K85A) results in a faster dissociation rate from the complex (k_d1=3.1x10^-4/sec) as compared to the dissociation rate of AcrF1 from the wild type Csy complex (k_d1=2.1x10^-7/sec) (Figures 4D, S7, S8, and S9). In addition to interaction with the thumb, a series of acidic residues on the C-terminal α-helix (α₂) of each AcrF1 are wedged between positive charges on the extended web domains of adjacent Cas7f molecules (Figure 4C). We predict that these interactions might be important for AcrF1 binding. To test this, we mutated lysine 254 and 257 to alanines (i.e. Cas7f K254A/K257A), which resulted in a disruption of AcrF1 binding to the Csy complex (Figures 4D and S9).

Access to the crRNA guide is crucial for target DNA hybridization and subsequent degradation. However, the thumb of each Cas7 protein folds over the crRNA at regular 6-nucleotide intervals, and in Cascade (type I-E) the thumb prevents base pairing at these positions (Jackson et al., 2014a; Mulepati et al., 2014). In the structure of Cascade bound to a single-stranded DNA, the target hybridizes to 5-nucleotide segments of the guide, while the 6^th-nucleobase is displaced and the phosphate backbone of the target traverses over the thumb and into the next accessible segment. AcrF1 proteins sits on top of the Cas7.6f and Cas7.4f thumbs, blocking the transition of target DNA from 1 to 2 and segment 3 to 4, respectively. Furthermore, the extended web of Cas7.6f and Cas7.4f is ~6 Å and ~4 Å (respectively) closer to the crRNA-guide than the web of the other subunits, suggesting that AcrF1 binding may induce a conformational change that restricts access to the guide RNA (Figure 4E), though a structure of Csy complex without the anti-CRISPRs will be necessary to know that the AcrF1 protein is directly responsible for this observation.

Since efficient AcrF1 binding relies on interactions with a few key residues on Cas7f (Figure 4D), we expected these residues to be under strong selective pressures. However, using a combination of evolutionary models we were unable to find evidence for positive selection at these positions (Pond et al., 2005). This suggests that K85, K254, and K257 in Cas7f are important for Csy function, which restricts mutations at these positions. Based on the location and charge of these residues, we hypothesized that they would be involved in DNA binding. To test this hypothesis, we recombinantly expressed and purified the wild type Csy complex, and complexes containing a mutation in Cas7f at position 85 (K85A) or a double mutation at positions 254 and 257 (i.e. Cas7f K254A/K257A). The mutants express and purify like wild type, suggesting that these mutations do not perturb assembly of the complex (Figure S7). Next, we used the purified complexes to perform electrophoretic mobility shift assay (EMSAs) to determine if these mutations perturb DNA binding (Figure 4F and S8). Wild type Csy complex binds to a 72-base pair double-stranded DNA target containing a protospacer and PAM with high affinity (K_D = 1.1 nM), while the mutants result in a severe binding defect (K_D > 1000 nM) indicating that the lysine residues targeted by AcrF1 are also critical for DNA binding.

AcrF2 is a DNA mimic

AcrF2 is a small acidic protein wedged between positively charged residues in the N-terminal hook of Cas8 and the thumb of Cas7.6f (Figures 5A and 5B). The AcrF2 fold is composed of a pair of antiparallel α-helices packed on either side of antiparallel β-sheet (α₁α₂β₁β₂β₃β₄α₃α₄), and acidic residues on AcrF2 (i.e., D30, E77, D76, E94, and E91) are positioned in close proximity to numerous lysines on either Cas7.6f (i.e., K79 and K77) or Cas8f (i.e., K247, K28, and K31). While the density in this region is not sufficient to confidently model the side chain position of these residues, we did noticed a pseudo-helical display of acidic residues on the surface of AcrF2 mimics the negative charge distribution on the helical backbone of a DNA duplex (Figure 5C), and that distribution of positively charged residues on the N-terminal hook of Cas8 and the thumb of Cas7f.6 form a “lysine-rich vise-like” structure that has been shown to be crucial for DNA binding by Cascade (van Erp et al., 2015). While the “lysine-rich vise” in Cascade is composed of positively charged secondary structures on finger domains of Cas7.5 and Cas7.6, our structure suggests that the Csy complex may have evolved a functionally analogous DNA vise composed of structurally distinct features. To test the functional role of the positively charged residues on the putative lysine-rich vise, we made charge-swapped mutations on Cas7f (Cas7f K77E/K79E) and Cas8f (Cas8f K247E and K28E/K31E). These mutants express and purify like wild type Csy complex, suggesting that they do not perturb assembly of the Csy complex (Figures S7, S8, and S9). However, mutations in either Cas7f or Cas8f result in severe binding defects for AcrF2 and DNA targets (Figures 5D–E, and S9). Taken together, these results suggest that AcrF2 is a double-stranded DNA mimic that blocks target recognition by competing for a critical DNA binding site (i.e. lysine-rich vise), and that the lysine-rich vise is structurally plastic and functionally conserved between the I-E and I-F crRNA-guided surveillance systems.

Discussion

Antagonistic interactions between predators (e.g. viruses) and their prey (e.g. bacteria) create dynamic selective pressures that drive diversification and genetic innovation (Van Valen, 1973). The remarkable diversity of CRISPR-Cas systems (two classes, six types, and nineteen subtypes) is consistent with their role in defense against rapidly evolving predators, and the discovery of diverse anti-CRISPRs with distributions that mirror these immune systems is beginning to explain the molecular basis of genetic conflict at the CRISPR-anti-CRISPR interface (Pawluk et al., 2016).

To determine the mechanistic differences that distinguish type I-F systems from other CRISPR-mediated immune systems, and to understand the molecular mechanisms of AcrF1- and AcrF2-mediated suppression, we determined the cryo-EM structure of the Csy surveillance complex bound to both AcrF1 and AcrF2. The seahorse-shaped morphology of the complex is similar to what has been observed in recent structures of other class 1 surveillance complexes (i.e. type I and III), however significant structural differences in individual subunits manifest in higher-order morphological distinctions (Figure 2). Like other class 1 surveillance complexes, the backbone of the Csy complex is composed of an assembly of Cas7-family proteins interwoven around the crRNA. While the theme of Cas7 oligomerization along the crRNA is conserved, the Cas7f fold is distinct and Cas7f-induced kinks in the crRNA result in a tight helical pitch that creates a nearly closed ring-like structure. The appearance of a closed ring is accentuated by an elongated Cas8-family tail protein that extends toward the Cas6f head, and the C-terminal helical domain of Cas8f is wedged in the middle of the Cas7f ring (Figure 1).

While the fold of Cas7f is distinct from previously determined Cas7-family proteins, the familiar right-hand morphology is largely maintained, with one prominent distinction. The web domain in Cas7f extends along the thumb, creating a conspicuous trough between the web and the thumb domains (Figure 4A, 4C, and 4E). We anticipate that this positively charged trough functions as an electrostatic cradle for the target strand as it traverses over each thumb, from one complementary segment of the crRNA-guide to the next. The structure explains how two molecules of AcrF1 interact with conserved lysine residues that line this trough and block target binding by obstructing contiguous access of the target to adjacent segments of the crRNA-guide sequences (Figure 4). In addition, the web-domains of Cas7.4f and Cas7.6f are closer to the crRNA, suggesting that AcrF1 binding may induce local conformational changes that further restrict target access. However, detection of lysines K85, K254, and K257 by AcrF1 suggests that natural variation at these positions would prevent AcrF1 binding and that these mutants would escape AcrF1-mediated suppression of the immune system. To test this hypothesis we mutated these lysines and show that they inhibit AcrF1 binding, but these mutations also result in severe DNA-binding defects. Together, these results suggest that AcrF1 has evolved to target specific features of the Csy complex that are essential for DNA binding.

In addition to binding the crRNA, structures of Cascade (type I-E) have shown that a subset of the Cas7e proteins also play a critical role in binding DNA (Hayes et al., 2016; Hochstrasser et al., 2014; Jackson et al., 2014a; van Erp et al., 2015; Wiedenheft et al., 2012). The fingers domain of Cas7.6e rotates 180 degrees creating a lysine-rich vise that makes a non-sequence specific interaction with double-stranded DNA, and mutations in Cas7e proteins that eliminate the positive charge no longer bind DNA (van Erp et al., 2015). We anticipated that this would be a conserved feature of the Cas7 backbone in type-I surveillance systems, however no such rotation is observed in the Cas7.6f subunit of the Csy complex. Instead, the structure reveals an analogous lysine-rich vise-like feature created by lysines displayed on the thumb of Cas7.6f and the N-terminal hook-domain on Cas8f. The importance of these lysines may have escaped our attention if it weren’t for the negative charge and positioning of AcrF2 directly in the vise. Interestingly, acidic residues on the surface of AcrF2 mimic the negative charge distribution on the helical backbone of a DNA duplex (Figure 5C), suggesting that AcrF2 is a double-stranded DNA mimic that competes with DNA for binding to the lysine-rich vise. Indeed, mutations in the lysine-rich vise that inhibit AcrF2 binding also prevent DNA binding by the Csy complex, which may explain why type-I systems rely on distinct structural innovations to achieve similar functions.

Phylogenetic and structural studies have shown that Cas8 proteins are extremely diverse, which in light of the results presented here, may be in part due to selective pressures from anti-CRISPR proteins. However, all Cas8 proteins appear to be involved in tail assembly through interactions with both Cas5 and the 5′-end of the crRNA (Hayes et al., 2016; Hochstrasser et al., 2016; Jackson et al., 2014a; Jackson and Wiedenheft, 2015; Mulepati et al., 2014; Osawa et al., 2015; Taylor et al., 2015; Zhang and Sontheimer, 2014; Zhao et al., 2014). In E. coli, the thumb on Cas5e creates a cylindrical pore that serves as a docking module for a short helix on Cas8e (also called Cse1). This helix is within a loop, called loop 1 (L1) that inserts into the Cas5e helix-binding pore and makes base-specific interactions with nucleotides in the 5′-handle (Hayes et al., 2016; Jackson et al., 2014a; Mulepati et al., 2014; Zhao et al., 2014). While the structural details are different, similar themes are observed in the Csy complex, wherein the thumb and finger domains of Cas5f protein grip a β-hairpin (residues 175–191, called the harpoon) of Cas8f, which makes both specific (e.g. Q177 interacts with adenosine at position -5) and non-sequence specific interaction with nucleotides in the 5′-handle. In addition to the role of Cas8 in tail assembly, a recent structure of Cascade (type I-E) bound to a partial duplexed DNA target explains how the N-terminal globular domain of Cas8e recognizes the PAM sequence and how PAM recognition is coupled to strand separation via insertion of a β-hairpin called the glutamine wedge (Hayes et al., 2016). Initial duplex destabilization by the wedge may be coupled to the C-terminal domain of Cas8e, which contains a four-helix bundle that appears to function like a molecular pry bar to maintain separation of the two strands. The displaced strand in this structure traverses over the four-helix bundle, whereupon it is presumably presented to the trans-acting Cas3 nuclease-helicase, though the remainder of the displaced strand was not included in the DNA target used in this structural study. Although, Cas8f does not share significant sequence similarity with Cas8e, Cas8f also contains a C-terminal four-helix bundle and the central position and orientation of this structure suggests that it may also be involved in strand separation during double-stranded DNA binding.

Another conserved feature of all type-I systems, is that after DNA binding by the crRNA-guided surveillance complex, they all recruit Cas3, a trans-acting nuclease-helicase (Jackson et al., 2014b). In type I-F systems, the Cas3 nuclease-helicase is fused to an N-terminal Cas2 protein that is involved in new sequence integration, and an anti-CRISPR protein (i.e. AcrF3) that binds to Cas2/3 prevents recruitment to the Csy complex (Bondy-Denomy et al., 2015; Wang et al., 2016a; Wang et al., 2016b). Using Dali (Holm and Sander, 1993), a structure homology search program, we identified structural similarity between AcrF3 and the C-terminal four-helix bundle of Cas8f (Figure S10). Cas8 proteins in other systems have been shown to be involved in Cas3 recruitment (Hochstrasser et al., 2014; Huo et al., 2014; Xue et al., 2016), and we speculate that AcrF3 may be a molecular mimic of the domain on Cas8f that is involved in Cas2/3 recruitment.

Presumably, CRISPRs evolved in response to antagonistic interactions with molecular parasites such as phages, and phages evolved anti-CRISPRs to subvert these systems. Here we provide structural and biochemical evidence showing that anti-CRISPR proteins AcrF1 and AcrF2 evolved to target specific structural features that are essential for immune system function, which suggests these antagonists play a central role in driving the diversification of CRISPR-Cas immune systems.

Supplementary Material