Structural insights into CasA function

To further understand the critical role that CasA plays in the targeting step of CRISPR immunity, we determined the structure of CasA from T. thermophilus at 2.3 Å resolution using X-ray crystallography (Fig. 1d, Table 1). The structure reveals two separate domains: an N-terminal domain (NTD, residues 1-364) that adopts a fold not found in the Protein Data Bank when searched using the Dali server (Holm and Rosenstrom, 2010); and a C-terminal four-helix bundle (CTD, residues 365-502). Two helices of the CTD bundle pack against the NTD, maintaining the relative orientation of the domains in the two molecules present in the asymmetric unit. Each domain contains a loop (L1, residues 130-143 and L2, residues 401-410) that was disordered in the crystal, indicating flexibility of these regions.

Table 1

Data collection and refinement statistics

	Native	SeMet
Data collection
Space group	P21	P21
Cell dimension
a, b, c (Å)	94.36, 48.11, 129.78	94.54, 47.82, 129.26
α, β, γ (°)	90.0, 97.0, 90.0	90.0, 97.3, 90.0
		Peak	Remote
Wavelength (Å)	1.11587	0.97968	0.95704
Resolution (Å)*	46.83–2.30 (2.36–2.30)	46.89–2.90 (2.98–2.90)	46.89–2.90 (2.98–2.90)
Rsym (%)*	5.6 (45.6)	11.5 (52.7)	12.0 (55.1)
I/δI*	18.9 (2.9)	10.3 (2.7)	10.0 (2.5)
Completeness (%)*	99.6 (96.8)	99.8 (100.0)	99.8 (99.9)
Redundancy*	3.9 (3.6)	3.8 (3.8)	3.8 (3.8)
Refinement
Resolution (Å)	46.8–2.3
No. reflections	51896
Rwork/Rfree	0.196/0.247
No. atoms
Protein	7323
Water/ligands	360
B-factors
Protein	44.4
Water/ligands	53.4
R.m.s deviations
Bond lengths (Å)	0.009
Bond angles (°)	1.1

Open in a separate window

^*Values in parentheses are for highest-resolution shell.

Outside the context of Cascade, purified CasA is incapable of binding to DNA (Fig. 1b) (Jore et al., 2011b); consistent with this observation, no prominent positively charged regions are present in the electrostatic surface map for CasA (Fig. S1c). To better understand the function of CasA in DNA binding by Cascade, we docked the high-resolution structure of CasA into cryo-EM reconstructions of Cascade from E. coli (Fig. 1e–g) (Wiedenheft et al., 2011a). The T. thermophilus and E. coli CasA sequences are ~50% similar, and both are predicted to have very similar secondary structures based on PSIPRED prediction (Jones, 1999). The crystal structure fits well into the EM density for CasA, indicating that the tertiary architecture of CasA is unaltered in Cascade (Fig. 1e). In the structure of Cascade with no target bound, CasA L1 appears to contact the 5′ end of the crRNA (Fig. 1f). The EM density in the region of L1 is contiguous with the density for the crRNA, suggesting that L1 becomes ordered upon interaction with the RNA.

Upon target binding, Cascade undergoes a concerted conformational change in which the CasA, CasB1-2 and CasE subunits shift and rotate along the CasC—crRNA backbone (Wiedenheft et al., 2011a). The shifted position of CasA in the target-bound structure of Cascade places L1 just below the 5′ end of the crRNA-target duplex (Fig. 1g). In contrast to the non-target bound EM density, no density is present for L1 in this reconstruction, suggesting that the loop no longer interacts with the crRNA and becomes flexible. We note that no PAM sequence is present in the target-bound Cascade reconstruction, but the trajectory of the crRNA—target duplex suggests that a longer target would come into close proximity of L1. Based on these observations, we hypothesized that L1 may contact the PAM sequence when a non-self target is bound.

CasA L1 is required for viral resistance

Our structural data indicate that CasA L1 could interact with both the crRNA and the PAM, suggesting that this sequence may be very important for Cascade function and viral resistance. We wished to probe the importance of the L1 sequence in a lambda phage (λ) challenge experiment using a previously described over-expression system that relies on arabinose for the tunable expression of cas genes under control of a T7 promoter. Previously, this system has been shown to confer >10⁷-fold resistance against λ infection, even when the spacers targeted λ sequences with incorrect PAMs (Brouns et al., 2008). We hypothesized that overexpression of Cascade could allow the system to overcome the stringent PAM requirements, as high concentrations of Cascade can bind targets with incorrect PAM sequences (Brouns et al., 2008; Semenova et al., 2011). To avoid masking defects in viral resistance upon mutation of L1, we attenuated expression of the immune system by decreasing the amount of induction reagent by 10-fold (0.01 mM IPTG and 0.02% L-arabinose). Reduction of the amount of inducer weakened the immune system relative to previous reports (Brouns et al., 2008; Semenova et al., 2011) but permitted distinction between the WT immune system, mutation of CasA or the role of the PAM in CRISPR-mediated defense (Fig 2a). Although these results suggest that overexpression of the CRISPR system can overcome the requirement for a correct PAM sequence, the CRISPR-bearing plasmid seems to escape destruction under these conditions based on continued antibiotic resistance and immunity against phage. While we are unsure how the plasmid is preserved, it is possible that complete complementarity between the crRNA and target may inhibit Cascade binding or subsequent cleavage by Cas3.

Open in a separate window

Figure 2

CasA L1 is important for Cascade assembly and target binding. (a) Phage sensitivity of E. coli strains expressing a CRISPR/Cas system with either WT CasA or the indicated CasA mutant. Cas genes were co-expressed with a CRISPR containing spacers that are not complementary to lambda phage, a CRISPR containing spacers complementary to the coding strand of lambda phage genes J,O, R, and E but no PAM; or a CRISPR containing spacers complementary to the coding strand of lambda phage genes J,O, R, and E, and including a PAM. The presence (+) or absence (−) of a PAM and/or a targeting CRISPR (TC) is indicated. See also Table S1. The number of plaques were counted and values from three replicates were averaged. Error bars represent one standard deviation. (b) Incorporation defects for CasA L1 mutants. The amount of CasA relative to CasC was determined from denaturing protein gel band intensities and normalized against relative amount of CasA in low salt Cascade immunoprecipitation. Aggregate data from three replicates are shown, with error bars representing one standard deviation. For WT, F129A, V130A and N131A CasA, reconstituted CasA + CasBCDE-crRNA (CasB-E) complex was formed and washed in the presence of low (100 mM) or high (500 mM) concentrations of NaCl. (c–e) Target binding defects for CasA L1 and PAM mutants (PAM mutations labeled in red). (c–d) Bar graph plotting K_d values for (c) dsDNA or (c–d) ssDNA targets. Average values from three replicates are shown, with error bars representing one standard deviation. Targets according to nomenclature in Figure S1a are as follows: Bars 1–4: O1—O2 duplex; Bar 5: O3—O4 duplex; Bar 6: O3—O2 duplex; Bar 7: O1—O4 duplex; Bars 8–11: O1; Bar 12: O3. (e) Gel shift assay for non-specific O5—O6 dsDNA (Fig. S1a) target binding by WT and mutant CasA + CasBCDE-crRNA (CasB-E).

Using this CRISPR/Cas expression system, we probed the importance of the L1 sequence by mutating several similar amino acids present in both the T. thermophilus and E. coli L1 CasA sequences and tested the effect of these mutations on resistance to λ infection. For cells expressing a fully functional CRISPR immune system, we observed a ~95% reduction in the sensitivity to phage infection compared to a strain containing a non-targeting CRISPR (Fig. 2a, S2a–c and Table S1). In contrast, strains containing CasA mutations in a three amino acid sequence of L1 (Phe129-Asn131) were far more sensitive to phage. In particular, cells expressing CasA with mutations at either F129A or N131A showed a ~80% and ~45% increases in phage sensitivity relative to wild-type (WT), respectively. In comparison, a strain expressing crRNAs that target λ phage at sites containing incorrect PAM sequences showed similar levels of phage sensitivity as observed with CasA F129A and N131A mutations (~60% more sensitive than WT), indicating that these residues are as important for viral resistance as the PAM itself. Because our structural data suggested that CasA L1 might interact with both the target and the crRNA (Fig. 1f–g), we hypothesized that the observed loss of viral resistance may be caused by defects in Cascade assembly, target binding, or both.

CasA L1 is important for incorporation into Cascade

To determine whether the three amino acid motif in L1 is important for CasA incorporation into Cascade, we incubated WT or mutant CasA with the CasBCDE-crRNA subcomplex and performed pull-down assays using an affinity tag on the CasB subunits (Fig. 2b and S2d). In comparison to co-expressed Cascade, all four CasA constructs were incorporated at the same levels when the complex was formed at low salt concentrations (100 mM NaCl). By contrast, the amount of incorporated mutant CasA decreased by 40–80% in the presence of 500 mM NaCl, while incorporation of WT CasA was unaffected. These data indicate that residues in L1 (Phe129-Asn131) are important, but not essential, for CasA binding to the CasBCDE-crRNA subcomplex.

Importantly, the CasA mutants remain stably associated with the complex at low salt concentrations, and the stoichiometry and size exclusion chromatography elution profiles are indistinguishable from purifications of the complex that include the WT CasA protein (Fig. S2e–f). We predict that mutations in L1 may disrupt interactions with the crRNA, but that contacts with other protein subunits of Cascade are sufficient for incorporation of CasA into the complex. At cellular protein concentrations, it is possible that the instability of the interaction between L1 mutant CasA and the CasBCDE-crRNA subcomplex could be at least partially responsible for the observed loss of viral resistance.

CasA L1 is important for dsDNA binding

We next asked whether CasA L1 plays a role in target binding by measuring the affinities of Cascade containing WT or mutant CasA for the 71-bp dsDNA with correct PAM and proto-spacer sequences (O1—O2 duplex, Fig. S1a). DNA binding experiments were performed using excess CasA (2 μM) in a low salt buffer to ensure that the WT and mutant CasA subunits were all stably associated with the CasBCDE-crRNA complex. While the V130A mutation has little effect on DNA binding, mutation of Phe129 or Asn131 to alanine causes substantial defects in dsDNA binding (Fig. 2c and Table S2), with ~500- and ~60-fold increases in K_d for F129A and N131A, respectively. In contrast, we observe only a 2–4-fold increase in K_d for ssDNA binding (O1, Fig. S1a) for all three mutants (Fig. 2c,d and Table S2). These data indicate that the L1 sequence is far more important for binding dsDNA than ssDNA. This difference suggests that CasA L1 may destabilize dsDNA targets, allowing the crRNA to access the seed and nucleate crRNA—target duplex formation.

An additional factor that could affect the affinity of Cascade for dsDNA is the ability of the complex to localize on DNA by non-specific binding prior to crRNA base pairing to a target sequence. Indeed, both F129A and N131A CasA mutations substantially reduce the amount of binding observed for a 71-bp dsDNA target containing a non-complementary crRNA binding sequence (O5—O6 duplex, Fig. S1a), indicating that L1 is the source of the previously observed CasA-dependent non-specific binding activity of Cascade (Fig. 2e) (Jore et al., 2011b).

Notably, similar to the CasA L1 sequence, we observe a difference in the importance of the PAM sequence for double-stranded versus single-stranded target binding (Fig. 2c–d and Table S2). Mutations in the PAM cause only a ~2-fold defect in binding to ssDNA targets (O3, Fig. S1a), while binding to dsDNA with an incorrect PAM (O3—O4 duplex, Fig. S1a) is significantly impaired. The observed binding of this target likely occurs mainly through non-specific interactions, as binding is reduced substantially in the presence of a non-specific competitor DNA (Fig. S2h). Importantly, mutation of the PAM sequence on only the target strand (O3—O2 duplex, Fig. S1a) has a similar effect as mutation of both strands, while mutation of the PAM sequence on the non-target strand (O1—O4 duplex, Fig. S1a) causes only a ~15-fold decrease in binding affinity (Fig. 2c and Table S2). These data suggest that only the target strand of the PAM is actively recognized by Cascade, while the non-target strand may indirectly promote PAM recognition by positioning the target strand through base pairing interactions.

Mapping CasA L1 position on dsDNA target

The relative importance of both CasA L1 and the PAM sequence for double-stranded target binding supports our structure-based hypothesis that L1 is involved in PAM recognition. Alternatively, mutations in L1 may affect the ability of Cascade to properly undergo conformational changes upon target binding (Wiedenheft et al., 2011a), resulting in the observed dsDNA binding defects. To test these two possibilities, we performed hydroxyl radical cleavage on the 71-bp dsDNA target (O1—O2 duplex, Fig. S1a) bound by Cascade containing WT, F129A or N131A CasA (Fig. 3). Overall, the cleavage patterns for all three Cascade variants are very similar, indicating that mutations in L1 do not significantly alter the overall architecture of Cascade. Together with the observation that L1 mutation does not significantly affect ssDNA target binding, it is likely that the concerted conformational change observed upon target binding still occurs when L1 is mutated.

Open in a separate window

Figure 3

CasA L1 interacts directly with the PAM. (a,b) Representative sequencing gels for hydroxyl radical cleavage of O1—O2 dsDNA target in the absence (−) or presence of Cascade with either WT CasA, F129A or N131A. (a) Target (O1) strand or (b) non-target (O2) strand was 5′ end labeled with ³²P and annealed to cold complement (Fig. S1a). (c) Percent cleaved at each position of the PAM and proto-spacer on the target strand for WT CasA + CasBCDE-crRNA bound to dsDNA. For each band, the peak intensity from three replicates were averaged and plotted versus sequence position, with error bars representing standard deviations. (d) Difference plot of percent of the target strand cleaved for mutant versus WT CasA + CasBCDE-crRNA bound to dsDNA. Positive values indicate regions of decreased protection upon mutation of L1. Aggregate data from three replicates are plotted, with error bars representing propagated standard deviations for average mutant and WT peak intensity. (e) Percent cleaved at each position of the PAM and proto-spacer on the non-target strand for WT CasA + CasBCDE-crRNA bound to dsDNA, plotted as in (c). (f) Difference plot of percent of the non-target strand cleaved for mutant versus WT CasA + CasBCDE-crRNA bound to dsDNA, plotted as in (d).

On the target strand (Fig. 3a,c, O1 Fig. S1a), a periodic 6-nt oscillation between protection and cleavage marks the gaps between the CasC2-C6 subunits, while CasC1 and C2 appear to fully protect the target from cleavage (Fig. S3). Strikingly, the only major difference in protection between WT and mutant Cascade occurs in the PAM region, where the −2 and −1 positions of the target are protected by Cascade containing WT CasA but not by Cascade containing F129A or N131A CasA (Fig. 3a,d). The loss of protection at such a localized region of the target strongly suggest that these differences are due to a loss of interaction between L1 and the PAM sequence, and that the position of CasA is likely not significantly altered outside of the L1 region.

Footprinting of the non-target strand (O2, Fig. S1a) reveals two main regions of protection, positions 5–8 and 13–22 (Fig. 3e), which may correspond to interactions with the CasA CTD and the two CasB subunits. The CasB subunits have been previously proposed to stabilize the displaced strand in the R-loop, consistent with this observed protection pattern (Wiedenheft et al., 2011a). Similar to the target strand, we observe reduced protection in the PAM and seed region of the non-target strand when F129A or N131A CasA is bound, although the effect is not as pronounced as on the target strand (Fig. 3f). In addition, position 11 is less protected in the presence of mutant CasA. These data suggest that the loss of interaction between L1 and the target strand PAM may result in a rearrangement of interactions between CasA and the displaced strand.

Crystallization and structure determination of CasA

Native CasA crystals were grown in 2.1 M sodium acetate and 0.1 M MES pH 6.5, and SeMet CasA crystals were grown at 2.25M sodium acetate and 0.1 M Tris, pH 8.0 at 18°C using hanging drop vapor diffusion by mixing 15 mg/mL CasA and reservoir solutions at 2:1 ratio. Native crystals diffracting to 2.3 Å grew after ~1 month, while selenomethionine crystals diffracting to 2.9 Å grew after ~3 months. For data collection, crystals were flash frozen in liquid nitrogen using a cryoprotectant containing 2.8 M sodium acetate and 30% glycerol.

Diffraction data were collected at 100 K on beamline 8.3.1 of the Advanced Light Source (Lawrence Berkeley National Laboratory). Two-wavelength multiple anomalous dispersion (MAD) data were collected using SeMet crystals at wavelengths of 0.97938 Å and 0.95704 Å. All datasets were processed using XDS(Kabsch, 2010). Eight selenium sites (four from each of two molecules in the asymmetric unit) were located using the Hybrid Substructure Search (HySS) module of the Phenix package(Adams et al., 2010). Substructure refinement, phasing and density modification were performed using autoSHARP (Vonrhein et al., 2007). The resulting electron density map displayed clear secondary structural elements, allowing for manual model building of ~60% of the protein. This model was completed by iterative rounds of refinement against the native dataset in Phenix.refine, followed by manual building in COOT (Emsley and Cowtan, 2004). The final model has a R_free of 24.7% and R_work of 19.6%, with 97% of torsion angles falling in the Ramachandran favored region. Each molecule in the asymmetric has one Ramachandran outlier (Arg184), but the electron density for this amino acid and the surrounding region is well defined, indicating that the geometry in this region is correct. The rmsd between the two molecules in the asymmetric unit is 0.28 Å. Rigid-body docking of crystal structures into segmented cryo-EM densities was performed using the “Fit in Map” module in UCSF CHIMERA (Goddard et al., 2007). Cross-correlation values of 0.53 and 0.55 were calculated between the atomic model of CasA and the apo and target-bound versions of Cascade, respectively. All structure figures were also generated using UCSF CHIMERA.

Viral resistance

Host sensitivity to Lambda (λ_vir) phage was performed using E. coli BL21-AI (Invitrogen) cells, essentially as described previously (Brouns et al., 2008), with the exceptions noted below. All strains (Table S1) were grown in 2YT-Lambda (2YTL) media (16 g/L tryptone, 10 g/L yeast extract, 5 g/L NaCl, 10 mM MgSO₄, 0.2 % maltose) until the optical density (OD600nm) reached 0.3. Cas protein and pre-crRNA production was induced for 45 min by adding L-Arabinose (Sigma-Aldrich) and IPTG to a final concentration of 0.02% and 0.01 mM, respectively. Following induction, the cells were pelleted by centrifugation and resuspended in 10 mM MgSO4. Plaque assays were performed in triplicate according to standard procedures (Sambrook et al., 1989), using a synthetic CRISPR locus containing four spacers that do not match the Lambda phage genome (pWUR477) (Brouns et al., 2008), that target proto-spacers located within essential genes (genes J, O, R and E) but that are not adjacent to a PAM sequence (pWUR478) (Brouns et al., 2008), or that target proto-spacers located within genes J, O, R and E that are adjacent to PAM sequence (Fig. S2a–c, Table S1). The new CRISPR construct was cloned into pACYCDuet-1 vector (Novagen) between NcoI and XhoI restriction sites. Error-bars represent one standard deviation.

Isothermal titration calorimetry

ITC experiments were measured on a nanoITC (TA Instruments) containing a 190 μL cell. All samples were purified by size exclusion chromatography using identical buffer (20 mM Tris, pH 7.5, 100 mM KCl, 1 mM TCEP, 5% glycerol), then concentrated to the indicated concentration. 10 μM CasA was titrated into the cell containing 80 μM CasBCDE-crRNA complex with 25–30 1.5 μL injections at 22°C. Concentrations of CasA and CasBCDE-crRNA complex were determined by UV absorbance (CasA ε₂₈₀: 59,525 M⁻¹cm⁻¹; CasBCDE-crRNA complex ε₂₈₀: 654,620 M⁻¹cm⁻¹). Data were baseline corrected, integrated and fit using NanoAnalyze software (TA Instruments). Titrations were performed in triplicate. Reported values are average of three replicates with errors bars reflecting one standard deviation.

Cascade pull-down assays

500 nM CasA and 250 nM CasBCDE-crRNA complex were incubated in 20 mM Tris, pH 8.0 and either 100 mM or 500 mM NaCl at 37°C for 10 minutes, then incubated with 15 μL Strep-Tactin Superflow Plus resin (Qiagen) for 15 minutes at 4°C. The resin was washed two times with either low or high salt buffer, then 40 μL of SDS loading dye was added to the resin. Samples were boiled and resolved by SDS-PAGE. Gels were stained with Coomassie Brilliant Blue dye and scanned using a Typhoon scanner, and densitometry was performed using ImageQuant software (GE Life Sciences). Pull-downs were performed in triplicate, and the average relative amount of CasA versus CasC is reported with error bars representing one standard deviation.

DNA binding assays

DNA oligomers were ordered from Integrated DNA Technologies (Fig. S1a). Duplexes were formed by mixing 40 μM of each strand in 40 mM Tris pH 8.0, 38 mM MgCl₂, and 1 mM Spermidine, heating at 95°C for 2 minutes, and cooling at RT for 10 minutes. All DNAs were purified on native gels containing 6% 29:1 polyacrylamide and 1X TBE. DNA bands were visualized by UV light, excised, and eluted by soaking gel pieces in deionized H₂O. Gel pieces were removed, and DNA was ethanol precipitated and resuspended in deionized H₂O. DNA samples were 5′ labeled with γ-[³²P]-ATP using polynucleotide kinase (PNK, New England Biolabs) for 30 minutes at 37°C. PNK was heat denatured at 65°C for 20 minutes, and excess ATP was removed using a G-25 spin column (GE Healthcare).

All binding assays were performed in buffer containing 25 mM Tris, pH 8.0, 130 mM KCl, 1 mM TCEP and 5% glycerol. CasA and CasBCDE-crRNA complex were pre-incubated for 10 minutes at 37°C to ensure complex formation prior to addition of DNA. CasA or CasBCDE-crRNA complex were titrated with constant final concentrations of 1 μM CasBCDE-crRNA complex (Fig. 1b) or 2 μM CasA (Fig. 2c,d, S1b and S2h), respectively. DNA was added to a final concentration of 5–20 pM, and samples were incubated at 37°C for 30 minutes, then resolved at 4°C on a 6% 29:1 polyacrylamide gel containing 1X TBE. Gels were dried and DNA was visualized by phosphorimaging and quantified using ImageQuant software (GE Life Sciences). The fraction of DNA bound (amount of bound DNA divided by sum of free and bound DNA) was plotted versus concentration of titrated species (CasA or CasBCDE-crRNA complex) and fit to a binding isotherm using Kaleidagraph software. Reported K_d values are the average of three replicates, and errors are the standard deviations between the three values.

We thank Kaihong Zhou for purification of Cascade and CasBCDE-crRNA complexes; Stan J. J. Brouns and John van der Oost for providing Cascade expression and viral resistance plasmids; Syed Habib Tahir Bukhari for help designing CRISPRs; members of the Doudna lab for critical reading of the manuscript; J. Holton and G. Meigs (Beamline 8.3.1, Advanced Light Source, Lawrence Berkeley National Laboratory) for assistance with X-ray data collection; and C. Quinn, M. Golinar and J. Berger for use of the NanoITC. This project was funded by a National Science Foundation grant to J.A.D.. D.G.S. is supported by the Damon Runyon Cancer Research Foundation. B.W. is a Howard Hughes Medical Institute Fellow of the Life Sciences Research Foundation. J.A.D. is a Howard Hughes Medical Institute investigator.