Selectivity and self-assembly in the control of a bacterial toxin by an antitoxic noncoding RNA pseudoknot

Abstract

Bacterial small RNAs perform numerous regulatory roles, including acting as antitoxic components in toxin–antitoxin systems. In type III toxin–antitoxin systems, small processed RNAs directly antagonize their toxin protein partners, and in the systems characterized the toxin and antitoxin components together form a trimeric assembly. In the present study, we sought to define how the RNA antitoxin, ToxI, inhibits its potentially lethal protein partner, ToxN. We show through cross-inhibition experiments with the ToxIN systems from Pectobacterium atrosepticum (ToxIN_Pa) and Bacillus thuringiensis (ToxIN_Bt) that ToxI RNAs are highly selective enzyme inhibitors. Both systems have an “addictive” plasmid maintenance phenotype. We demonstrate that ToxI_Pa can inhibit ToxN_Pa in vitro both in its processed form and as a repetitive precursor RNA, and this inhibition is linked to the self-assembly of the trimeric complex. Inhibition and self-assembly are both mediated entirely by the ToxI_Pa RNA, with no requirement for cellular factors or exogenous energy. Finally, we explain the origins of ToxI antitoxin selectivity through our crystal structure of the ToxIN_Bt complex. Our results show how a processed RNA pseudoknot can inhibit a deleterious protein with exquisite molecular specificity and how these self-contained and addictive RNA-protein pairs can confer different adaptive benefits in their bacterial hosts.

Bacteria possess an extensive set of small, noncoding RNAs, which are used in housekeeping, regulatory, and defensive roles (1). The majority of bacterial small RNAs act at the mRNA level to modulate gene expression, whereas a more specialized subset functions by binding directly to proteins (1). These include RNAs which contribute functions to ribonucleoprotein particles, such as the antiviral clustered, regularly interspaced, short palindromic repeats (CRISPR) locus RNAs (crRNAs) (2), and RNAs that antagonize the activity of proteins by sequestering them away from their substrates, such as CsrB (3) and the 6S RNA (4).

Toxin–antitoxin (TA) systems are nearly ubiquitous in prokaryotic genomes (5, 6), and small RNAs function as the antitoxin components in two of the five TA system types: type I antitoxins are small antisense RNAs that prevent translation of the toxin transcript, and type III antitoxins are small RNAs that inhibit their cognate protein toxins by direct interaction (7). Type II, type IV, and type V TA systems all use protein antitoxins with different mechanisms of action (7–9). Toxin targets are varied, although many act to degrade cellular RNAs. A canonical TA locus consists only of the genes for antitoxin and toxin arranged in a single operon. Toxins typically are more stable than antitoxins, so continued expression of the genes is required to maintain the antitoxin at protective levels; when the balance between the components is perturbed, the toxin is released and induces bacterial cell stasis or death. TA systems have been implicated in diverse cellular processes including plasmid stabilization, persistence, and resistance to viruses (10–12).

The prototype type III TA system is the plasmid-encoded ToxIN from Pectobacterium atrosepticum (hereafter ToxIN_Pa), which originally was discovered through its ability to confer bacteriophage resistance as an abortive infection system (12, 13). ToxIN_Pa consists of a protein toxin (ToxN_Pa) and a small RNA antitoxin (ToxI_Pa), which have a kill/rescue phenotype when overexpressed in Escherichia coli. These elements occur genetically as a series of 5.5 tandem toxI_Pa repeats followed by the toxN_Pa gene. Both toxI_Pa and toxN_Pa are transcribed from the same promoter, and a transcriptional terminator between the two genes regulates the relative levels of toxin and antitoxin synthesis (Fig. 1A) (12). The purified complex of ToxN_Pa and ToxI_Pa cleaves housekeeping RNAs in vitro, suggesting that ToxN_Pa is toxic in vivo by virtue of a general ribonuclease activity which is antagonized by the RNA antitoxin ToxI_Pa (14). Bioinformatic searches subsequently identified numerous putative type III TA systems in diverse bacteria; these systems showed variation in their protein sequences and in the length, number, and sequence of their associated tandem repeats (12, 15). The toxin–antitoxin function of several of these systems was validated in E. coli, including that of plasmid-encoded ToxIN from Bacillus thuringiensis (hereafter ToxIN_Bt). The toxIN_Bt locus comprises 2.9 toxI_Bt antitoxic repeats together with the toxN_Bt gene, which encodes a protein with 30% amino acid identity to ToxN_Pa.

Fig. 1.

ToxN_Pa degrades RNAs and is inhibited by ToxI_Pa in vivo. (A) Schematic representation of ToxIN_Pa with toxIN_Pa genetic organization, processing of ToxI_Pa, and complex formation indicated. ToxN_Pa is shown in blue and ToxI_Pa in orange. (B) ToxN_Pa degrades the ompA transcript and is inhibited by ToxI_Pa in vivo. E. coli cells containing separately inducible ToxN_Pa-FLAG and ToxI_Pa plasmids were grown to log phase, and the effect of ToxN_Pa expression and subsequent coexpression of ToxI_Pa on ompA transcript levels was analyzed by Northern blot (Top). Expression of ToxI_Pa (Middle) and ToxN_Pa-FLAG (Bottom) also was assessed by Northern and Western blot, respectively. The symbols “+” and “−” represent induction and repression of ToxN_Pa or ToxI_Pa expression. Induction of a negative control is indicated by “0.” Because the RNA purification method used here excludes small RNAs, it was not possible to detect individual ToxI_Pa repeats. Note that because ToxN_Pa-FS expression did not affect growth, cells reached stationary phase and showed natural down-regulation of ompA transcription over the course of the experiment.

ToxI_Pa is a rare example of a naturally occurring small RNA which functions to counteract the activity of an enzyme. The crystal structure of ToxN_Pa bound to ToxI_Pa provided major insights into the mechanism of this antitoxic activity: three ToxI_Pa RNAs, which are themselves cleaved from their repetitive precursor by ToxN_Pa, are bound head-to-tail by three ToxN_Pa monomers to form a heterohexameric, triangular assembly in which the ToxN_Pa active site is occluded (Fig. 1A) (14). Although illuminating, the structure of the ToxIN_Pa complex naturally raised further questions about the system. First, how does ToxN_Pa, which has activity against a range of RNAs, recognize its ToxI_Pa RNA antitoxin and assemble into the triangular ToxIN_Pa complex? Second, how do the processes of ToxI_Pa cleavage and complex assembly relate to the inhibition of ToxN_Pa?

Here we demonstrate through cross-inhibition experiments with ToxIN_Bt that ToxI RNA antitoxins are selective inhibitors of specific toxin partners. Both ToxIN systems have an “addictive” plasmid-maintenance phenotype. Specific inhibition of ToxN_Pa ribonuclease can be mediated by both the processed and precursor forms of ToxI_Pa in vitro and is linked to the self-assembly of the heterohexameric ToxIN_Pa complex—a spontaneous process which occurs without requirement for exogenous energy or chaperones. We explain the basis for toxin–antitoxin specificity based on our crystal structure of the ToxIN_Bt complex. Finally, we define the sequence-specific ribonuclease activity responsible for toxicity and antitoxin processing of two ToxN proteins.

Results

ToxN_Pa Is a General Ribonuclease That Is Inhibited by ToxI_Pa in Vivo.

We first tested the general ribonuclease activity of ToxN_Pa. Previous results suggested a general mRNA interferase function for ToxN_Pa; however, this activity was not shown directly, and ToxN_Pa cleaved housekeeping gene transcripts in vitro even though it was present in complex with its antitoxin (14). In contrast, studies of mRNA interferases from type II TA systems showed that the addition of the antitoxin in a 1:1 stoichiometry completely inhibits the toxin’s activity in vitro (16–18). To confirm the mechanism of ToxN_Pa toxicity, Northern blots of the highly expressed housekeeping genes ompA, dksA, and lpp were performed following overexpression of ToxN_Pa and the subsequent co-overexpression of ToxI_Pa. As shown in Fig. 1B, ToxN_Pa overexpression caused a sharp reduction in the level of the ompA transcript, and subsequent overexpression of ToxI_Pa restored ompA transcript levels. The degradation was not seen when an inactive, frameshifted ToxN_Pa variant, (ToxN_Pa-FS) (12), was expressed, and ompA RNA levels were not restored in the ToxI_Pa vector-only control strain. The same pattern of ToxN_Pa-mediated RNA degradation and ToxI_Pa-mediated rescue was seen with the dksA and lpp RNAs (Fig. S1). Overexpression of ToxN_Pa also produced a broad size distribution of ToxI_Pa products, showing that ToxI_Pa is indeed processed by ToxN_Pa in vivo. These results confirm the ribonuclease activity of ToxN_Pa in vivo directed both to general cellular targets and to its own antitoxin transcript and the capacity of ToxI_Pa to suppress this activity.

ToxI Antitoxins Are Selective.

After confirming the ribonuclease activity of ToxN_Pa in vivo and the action of ToxI_Pa to neutralize this activity, we wished to explore the specificity of the ToxI RNA antitoxin. To do so, cross-inhibition experiments were performed with the toxIN_Pa components and those of the related toxIN_Bt system (Fig. 2A). Although the two ToxN proteins share 30% sequence identity, the corresponding cognate toxI RNA sequences are unrelated. In an E. coli kill/rescue assay, ToxI_Pa counteracted ToxN_Pa but not ToxN_Bt, and vice versa; each ToxI RNA antitoxin was active only against its own toxin partner (Fig. 2B). These experiments show that ToxI RNA antitoxins are highly selective inhibitors with the capacity to distinguish between closely related proteins.

Fig. 2.

ToxI antitoxins are selective for their cognate toxin. (A) Schematic of the toxIN_Pa and toxIN_Bt loci with the antitoxin and toxin components indicated. (B) Counts of viable E. coli DH5α following induction of ToxN_Bt or ToxN_Pa expression together with either ToxI_Bt or ToxI_Pa. Results shown are mean and SD for three biological replicates.

ToxIN Systems Promote Plasmid Maintenance.

Many TA systems can mediate plasmid stabilization by postsegregational killing, in which the rapid degradation of the antitoxin after plasmid loss results in the passive activation of the toxin to kill plasmid-free segregants (10). To determine whether ToxIN_Pa and ToxIN_Bt also have this activity, we performed long-term plasmid-loss experiments. ToxIN_Pa completely prevented loss of plasmid pRBJ200 in E. coli W3110 over the duration of the experiment, whereas ToxIN_Bt had no effect (Fig. 3A). However, ToxIN_Bt did promote retention of the test plasmid pHCMC05 in Bacillus subtilis YB886 (Fig. 3B). Because the Bacillus test vector is based on the low-copy number pBS72 replicon (19), this stabilization activity is likely to apply to ToxIN_Bt in its native context on B. thuringiensis plasmid pAW63 (20). This plasmid-stabilization function may represent the biological role of ToxIN_Bt, which, unlike ToxIN_Pa, did not have a detectable phage-resistance phenotype. The reason for the host dependence of this activity probably is that ToxN_Bt is not toxic enough in E. coli to mediate postsegregational killing when expressed from its native promoter on a single-copy vector; ToxN_Bt showed lower toxicity than ToxN_Pa in E. coli (Fig. S2A). These results show that ToxIN systems are addictive modules that can enhance plasmid retention.

Fig. 3.

ToxIN systems can stabilize plasmids. (A) Retention of ToxIN_Pa- and ToxIN_Bt-carrying plasmids in E. coli W3110. The percentage of cells retaining the plasmid before and 24 h after growth without selection is shown for ToxIN_Pa, ToxIN_Bt, and the vector-only control. (B) Retention of plasmids carrying ToxIN_Bt or a frameshift ToxIN_Bt-FS negative control in B. subtilis YB886. The percentage of cells retaining the plasmid is plotted as a function of the number of hours of growth without selection. Both A and B show the mean and SD for three biological replicates.

ToxN_Pa Is Inhibited by both Processed and Precursor ToxI_Pa.

In principle, toxin inhibition by ToxI RNA could require cleavage of the repetitive elements, for instance by linking the energy of cleavage with stable assembly. To test this possibility, stop-point RNA degradation assays were performed in vitro using purified ToxN_Pa ribonuclease with ompA RNA as a substrate, and ToxI_Pa RNA was added either as the long repetitive precursor, which was transcribed in vitro, or as precleaved, 36-nt pseudoknot repeats, which were purified from dissociated ToxIN_Pa complex. ToxN_Pa alone degraded the test substrate ompA to generate four major products (Fig. 4A, lanes 2–5), and addition of processed single repeats of ToxI_Pa to the reaction in a 1:1 molar ratio of ToxI_Pa:ToxN_Pa drastically reduced ompA degradation (Fig. 4A, lanes 6–8). Degradation of ompA RNA by ToxN_Pa also was inhibited by addition of the long ToxI_Pa precursor RNA, again in a 1:1 ratio of ToxI_Pa repeats to ToxN_Pa (each precursor RNA contains four copies of the functional ToxI_Pa repeat). The precursor ToxI_Pa was cleaved into progressively smaller units during the reaction and appeared to protect the ompA substrate from degradation completely (Fig. 4A, lanes 9–11), suggesting that the repetitive ToxI_Pa RNA is a preferred substrate of ToxN_Pa. Addition of the ToxI_Bt precursor did not prevent ToxN_Pa cleavage of ompA (Fig. 4A, lanes 12–15), although the ToxI_Bt RNA was cleaved by the toxin, further highlighting the selectivity of ToxI antitoxins observed in vivo. Within the resolution of this experiment, the repetitive ToxI_Pa precursor appeared to be a more effective ToxN_Pa inhibitor than its processed counterpart. These results indicate that the inhibitory action of the ToxI_Pa RNA is entirely self-contained and occurs without cofactors or exogenous energy and that the energy of RNA cleavage is not necessary to form the inhibitory structure.

Fig. 4.

ToxI_Pa inhibits ToxN_Pa and self-assembles the ToxIN_Pa complex in vitro. (A) In vitro degradation of ompA RNA by ToxN_Pa with different forms of added ToxI. Reactions (2 pmol ompA + 6 pmol ToxN_Pa) were incubated at 25 °C, and samples were taken at the times indicated. ToxN_Pa protein and single ToxI_Pa monomeric repeats were purified by FPLC. The full-length repetitive ToxI_Pa and ToxI_Bt precursor RNAs were transcribed in vitro. (B) Size-exclusion chromatography of ToxIN_Pa assembled in vitro. ToxN_Pa was incubated with ToxI_Pa single repeats or full transcript for 1 h at 37 °C, and the reactions were analyzed by size-exclusion chromatography on an S200 13/30 column. Scaled absorbance traces are shown for ToxN_Pa + single ToxI_Pa, ToxN_Pa + full ToxI_Pa, ToxIN_Pa complex, and each of the individual reaction components. The elution volume and calculated molecular weight of each peak are given in Table S1.

ToxI_Pa inhibition of ToxN_Pa appears to work in two ways: First, the ToxI_Pa precursor is a preferred substrate of ToxN_Pa which diverts the enzyme away from cellular RNAs when present (as also observed in vivo; Fig. 1B Middle), and second, the processed 36-nt ToxI_Pa unit is active as an inhibitor of ToxN_Pa, independently of its own cleavage. Free ToxN_Pa also appeared to have higher activity than the purified ToxIN_Pa complex in vitro (Fig. S2B). That ToxI_Pa could inhibit ToxN_Pa in vitro suggested that the heterohexameric ToxIN_Pa complex may have assembled in these reactions; this possibility was examined next.

ToxIN_Pa Complex Can Self-Assemble in Vitro.

ToxN_Pa was incubated with either processed ToxI_Pa single repeats or with the in vitro transcribed ToxI_Pa precursor, and the reactions were analyzed by size-exclusion chromatography (Fig. 4B and Table S1). Buffer conditions were kept the same as in the in vitro inhibition experiment above. The ToxN_Pa plus precursor ToxI_Pa reaction eluted as three peaks, one at the same size as the ToxIN_Pa complex, one at the elution volume of both ToxN_Pa and ToxI_Pa, and an additional peak eluting at ∼15 mL which contained the promoter and terminator sequences cleaved from either side of the ToxI_Pa repeats in the long precursor RNA. The combination of ToxN_Pa plus an overabundance of ToxI_Pa-processed single repeats gave a small peak of ToxIN_Pa complex and a large peak of ToxI_Pa monomer. All the detectable ToxN_Pa in this reaction was contained in the complex peak. These results confirm that the ToxIN_Pa complex can self-assemble in vitro from active ToxN_Pa combined with either processed ToxI_Pa or its long precursor RNA; the requirements for ToxI_Pa cleavage, folding, and assembly into a stable complex are all intrinsic to the system. The inhibition of ToxN_Pa by ToxI_Pa observed in vitro therefore is linked to the spontaneous formation of the ToxIN_Pa complex.

Structure of ToxIN_Bt Reveals the Basis for Antitoxin Specificity.

To understand better the selective inhibition displayed by ToxI antitoxins, we solved the structure of the ToxIN_Bt complex to 2.2 Å by X-ray crystallography (see crystallographic statistics, Table S2). A modified ToxN_Pa structure was used as the search model to obtain phase information by molecular replacement, and ToxI_Bt RNA then was built into the omit map (Fig. S3A). Because the antitoxic repeats of toxIN_Bt are not identical, the structure was solved and refined using the consensus ToxI_Bt RNA repeat sequence, which is offset −8 nt relative to the genetic toxI_Bt repeat (Fig. S3B). ToxI_Bt nucleotides are numbered 1→34 based on the RNA observed in the structure.

ToxIN_Bt is a heterohexameric, triangular assembly of three ToxN protomers and three ToxI RNAs (Fig. 5A), which is generated by a threefold rotational symmetry operation on the crystal asymmetric unit of 1ToxN_Bt:1ToxI_Bt. The bound ToxI_Bt units are cleavage products of ToxN_Bt, as shown by their pseudocontinuous arrangement and by the presence of a 2′-3′-cyclic phosphate at the 3′ end of each ToxI_Bt, which was observed in the omit map with density at a contour level >3 σ. The processing of ToxI by ToxN and the core architecture of the resulting complex are shared between ToxIN_Bt and ToxIN_Pa.

Fig. 5.

Structure of the ToxIN_Bt complex. (A) The trimeric ToxIN_Bt complex. ToxN_Bt is shown in cartoon representation in teal, and the ToxI_Bt RNA backbone is orange with the bases colored in a gradient from orange to blue. The surface of two ToxN_Bt monomers is shown also, with blue for positively charged and red for negatively charged regions. In each ToxI_Bt protomer the nucleotide A34 and its 2′,3′-cyclic phosphate are shown as white sticks. (B) Schematic of internal and external bonding of a single ToxI_Bt monomer. Canonical Watson–Crick A:T or G:C base pairs are represented by a single, black horizontal bar. Noncanonical base pairs of ToxI_Bt are shown in Leontis–Westhof symbols (49) with the interacting edges indicated as follows: Watson–Crick edge, ●; sugar edge, ►; Hoogsteen edge, ▪. Filled and open symbols indicate the cis or trans orientation of the glycosidic bonds, respectively. The vertically aligned letters indicate stacked bases. Black dashed lines indicate single hydrogen-bond interactions. Bonds numbered 1–4 involve backbone atoms as follows: 1, G9 (N3) to G23 (O2); 2, G9 (N7) to A11 (O2); 3, G9 (O2) to A11 (PO2); 4, G9 (N2) to U24 (PO2). The ToxN_Bt monomers at key interaction sites are indicated, and interactions between ToxI_Bt nucleotides and ToxN_Bt are indicated with a black outline to show base–ToxN_Bt interactions or a gray highlight to indicate backbone–ToxN_Bt interactions. See Fig. S4 for comparison with the ToxI_Pa structure. (C) A single ToxI_Bt pseudoknot, bound by two monomers of ToxN_Bt (labeled ‘”A” and “B”). ToxI_Bt is shown as a cartoon with nucleotides colored according to their location as in panel B. ToxN_Bt monomers are shown as a cartoon beneath a surface representation with positively and negatively charged regions colored in blue and red, respectively.

The processed ToxI_Bt RNA folds into a compact pseudoknot core, containing three internal base triplexes, with two single-stranded tails. This architecture creates two large surfaces for interaction with ToxN_Bt, each formed from one face of the pseudoknot core and one of the single-stranded tails (Fig. 5 B and C). Although toxI_Bt did not have any detectable sequence similarity to toxI_Pa a priori, the key structural features of the processed inhibitory ToxI_Bt RNA, and the nucleotides involved, are conserved with ToxI_Pa (Fig. S4). The two base-paired stems of the pseudoknot are separated by a single-tiered G:U:U base triplex (triplex 3, G23:U16:U24; see Fig. 7A), and these three elements stack in a quasi-continuous helical core, which is stabilized further by interdigitation of the G23 purine ring between U8 and U9 at the base of stem I. The precise tertiary structure of ToxI_Bt is maintained by an extensive internal hydrogen-bonding network involving all but two nucleotides of the central pseudoknot. This network includes several noncanonical base pairs and backbone–base interactions in addition to the Watson–Crick pairs of the pseudoknot stems (Fig. 5B). The absolute conservation of both triplex 3 and the guanine base interdigitation between ToxI_Bt and ToxI_Pa suggests that these interactions are required for the formation or maintenance of the ToxI fold. Importantly, conservation of these core structural elements does not preclude changes to ToxN-interactive regions of ToxI, which could impart specificity to the ToxI–ToxN interaction.

Fig. 7.

ToxI_Bt triplex structures and ToxIN_Bt mutagenesis study. (A) Details of the three base triplexes of ToxI_Bt, shown as pale pink sticks with black dashed lines to indicate hydrogen bonds. Lys148 of ToxN_Bt is shown in blue. The geometric classification of each triplex is indicated (22). (B) Analysis of effects of ToxI_Bt mutations in vivo. Growth of E. coli DH5α following co-overexpression of ToxN_Bt with mutants of ToxI_Bt is shown and is representative of three biological replicates. (C) Effect of ToxN_Bt mutations in vivo. Viable counts of E. coli cells overexpressing ToxN_Bt mutant constructs are shown as mean ± SD.

The toxin protein ToxN_Bt comprises a highly twisted, antiparallel β-sheet core flanked by several helices, including the long, kinked helix H3 (Fig. 6A), the same core fold as in ToxN_Pa. Both ToxN proteins are structural homologs of the MazF/Kid family of type II TA system toxins, with additional features to facilitate binding to ToxI (21). However, an overlay of the two protein structures does show five key variable regions (V1–V5) (Fig. 6A): the ToxN_Bt active site loop S1→S2 (V1; residues F29 R37); loop S3→S4 (V2; R58–Q66), which is α-helical in ToxN_Pa but a short 3₁₀ helix in ToxN_Bt; loop S4→S5 (V3; D73–K77); and helices H2 and H4 (V4 and V5; residues I98–Q109 and T140–V155, respectively), which are insertions in ToxN_Pa S7→H3 and H3→H5, respectively. Strikingly, these five variable regions are the main sites for interaction of the protein with ToxI. A protein sequence alignment of ToxN_Bt and ToxN_Pa with six additional ToxN-family proteins also shows a largely conserved secondary structure, with the variability between homologs clustered in these five loop regions (Fig. S5). It appears that the same core fold is shared across the entire ToxN family, but each protein has a unique set of variable loops reflecting the different antitoxin specificity of each member of the ToxN family.

Fig. 6.

ToxIN_Bt and ToxIN_Pa have different protein–RNA interfaces. (A) Least-squares superimposition of the B. thuringiensis ToxN_Bt and P. atrosepticum ToxN_Pa (PDB ID code 2XD0) protein structures. The superimposed protein monomers are each shown with one associated ToxI RNA pseudoknot (corresponding to interface 2). ToxN_Bt is shown in teal, ToxI_Bt is shown in orange and blue, and both ToxN_Pa and ToxI_Pa are shown in silver. Variable loop regions are labeled V1–V5 and are numbered according to the corresponding residues of ToxN_Bt. (B–E) Comparison of ToxI–ToxN interfaces in the ToxIN_Bt and ToxIN_Pa complex structures. ToxI RNAs are shown as cartoons with key nucleotides represented as sticks in pale pink (ToxI_Bt) or pale yellow (ToxI_Pa). ToxN proteins are shown as cartoons in teal (ToxN_Bt) or magenta (ToxN_Pa) with key residues represented as sticks. Structural regions of ToxN are indicated. Black dashed lines indicate hydrogen bonds of 2.6–3.4 Å.

Two extended interfaces sustain the trimeric ToxIN_Bt assembly, each formed from a section of the kinked helix H3 and two variable surface loops of ToxN_Bt interacting with one binding groove of ToxI_Bt. Interface 1, where ToxI_Bt groove 1 interacts with the N-terminal portion of H3 (Fig. 6B), is maintained by U10 binding in a hydrophobic pocket formed by the ToxN_Bt loop V4 and helix H3 and the binding of the single-stranded ToxI_Bt tail across the surface groove of ToxN_Bt. This interface, which buries ∼1,000 Ų, involves far fewer interactions than interface 1 of ToxIN_Pa (Fig. 6C). At ToxIN_Bt interface 2, the RNA chain of ToxI_Bt groove 2 wraps around the surface of ToxN_Bt at loop V2 and across helix H3 (Fig. 6D). The RNA is positioned by base-specific hydrogen bonding and hydrophobic interactions between sidechains of loops V2 and V5 and bases C19 and U5 and by an extended series of hydrogen bonds between the positively charged side chains along helix H3 and the phosphate backbone of ToxI_Bt G20–A17 (Fig. 6D and Fig. S4C). This interface also differs notably from its equivalent in ToxIN_Pa (Fig. 6E), because a shortening of the ToxI_Bt groove 2 loop together with a nine-residue insertion in ToxN_Bt loop V5 allows the two components to pack more closely. The buried surface area of interface 2 is ∼1,180 Ų. An effect of the changes to the interface interactions in ToxIN_Bt is that each ToxI_Bt unit leaves its bound ToxN_Bt monomer at a different angle than in ToxI_Pa, so the central axis of each ToxI_Bt pseudoknot deviates further from the triangular frame of the three ToxN_Bt monomers, although the overall triangular complex architecture is retained (Figs. 5A and 6A).

In summary, both ToxI_Bt and ToxN_Bt have structures similar to their P. atrosepticum equivalents, but subtle, complementary changes to both toxin and antitoxin result in substantial differences in the protein–RNA interfaces that maintain the inactive complex.

ToxI_Bt Contains a Rare C:G:G cis Watson–Watson/trans Sugar–Hoogsteen Triplex.

The three base triplexes of ToxI_Bt are shown in Fig. 7A along with their geometric classification (22). Triplex 1 is a notable point of difference between ToxI_Pa and ToxI_Bt: In ToxI_Pa triplex 1 is a type II A-minor motif of the stem I base pair G2:C15 with A19 of the following loop. In ToxI_Bt triplex 1 the loop following the stem I base pair G6:C19 is absent. Instead, the adjacent nucleotide G20 interacts with the minor-groove edge of the base pair via its Hoogsteen face to form a rare C:G:G cWW/tSH cis Watson–Watson/trans Sugar–Hoogsteen (cWW/tSH) triplex, which has been observed only once before, in the E. coli 16S rRNA (22, 23). Triplexes 2 and 3 are conserved between ToxI_Bt and ToxI_Pa, and both are common RNA tertiary motifs. Triplex 2 is a type I A-minor motif (24) formed from the stem 1 base pair G7:C18 with A22, whereas Triplex 3 comprises the U24:U16 pair, with G23 interacting with the major groove edge of this pair (Fig. 7A).

ToxI_Bt Triplex 1 is stabilized by hydrogen bonds from Lys148 of ToxN_Bt to C19 and G20. Because Lys148 is not conserved in other ToxN homologs (Fig. S5), this interaction may be unique to ToxIN_Bt. The importance of interaction with Lys148 for antitoxicity could not be assessed, because mutation of Lys148 to alanine abolished toxicity (Fig. 7C). Mutation of G20 had only a minimal effect on antitoxicity, suggesting other nucleotides can form a functional platform in this position (Fig. 7B). Specific mutations to other ToxI_Bt nucleotides confirmed the importance of the intercalated base G23, its stacking partners U8 and G9, and the ToxN_Bt-binding U10, with these mutants showing reduced activity in a kill/rescue assay with ToxN_Bt (Fig. 7B).

ToxN_Pa and ToxN_Bt Are Sequence-Specific Endoribonucleases with Different Substrate Preferences.

ToxN_Pa and ToxN_Bt are both bacteriostatic ribonucleases that cleave their own antitoxins. ToxN_Bt has a ribonuclease active site similar to that of ToxN_Pa (Fig. 8 A and B). The 2′-3′-cyclic phosphate product is held by an extensive hydrogen-bond network formed from the basic residues Lys31, Arg37, and Arg58 and the conserved Ser57 and Thr56 close to the 2′O of the cyclic phosphate. The purine ring of A34 is coordinated in the anti- conformation by Tyr110 and Gln117, as in ToxIN_Pa. Several of the active-site residues were shown by mutagenesis to be required for toxicity (Fig. 7C). ToxN enzymes are proposed to cleave their substrates through a metal-independent RNase T1-like mechanism, based on their active-site architectures, the presence of the 2′-3′-cyclic phosphate product in the active site, and their structural homology to Kid, whose reaction mechanism has been studied in detail (21, 25). The RNA fragment patterns produced by ToxN_Pa degradation in vitro (Fig. 8C and ref. 14), together with the precise ToxI–ToxN interactions observed in both complex structures, suggested sequence-specific ribonuclease activity. Both ToxIN_Pa and ToxIN_Bt cleaved several highly expressed E. coli test substrates—rpoD, ompF, ompA and dksA—into defined patterns of RNA fragments in vitro (Fig. 8C). Sites of ToxN-mediated cleavage then were identified by performing 5′ RACE on the cleaved RNA substrates, and sequence-specificity profiles were generated from a total of 14 unique 5′ fragment ends for ToxIN_Pa and 12 unique ends for ToxIN_Bt. ToxN_Pa cleaves RNAs at AA↓AU sequences, and in two instances, at AA↓AG (Fig. 8D). ToxN_Bt recognizes the sequence A↓AAAA with some tolerance for different nucleotides at positions +2 and +4 (Fig. 8D). The in vitro sequence specificity matches the active-site interactions seen in both ToxIN complexes; in ToxIN_Bt A34 is specifically coordinated; the purine rings of A1→A3 bases form a hydrophobic stack supported by Phe29, which is required for toxicity; and A1 and A3 also make base-specific hydrogen bonds to ToxN_Bt (Fig. 8A). In ToxN_Pa all three adenines of the AA↓AU sequence are held by specific hydrogen bonds to the protein, whereas the U pyrimidine ring stacks between A3 and Phe88 (Fig. 8B). Therefore both ToxN ribonucleases have sequence-specific activity consistent with a general mRNA interferase mechanism of toxicity. Unlike the ribonuclease toxins of type II TA systems, this activity also is required to generate the mature antitoxin.

Fig. 8.

ToxN_Bt and ToxN_Pa are endoribonucleases with selective substrate preferences. (A) View of ToxIN_Bt active site showing key interactions for product recognition and substrate cleavage. ToxN_Bt is shown in teal and ToxI_Bt RNA in white. Hydrogen bonds are indicated by black dashed lines. Residues that were mutated resulting in a loss of toxicity are underlined (see also Fig. 7C). Phosphate groups of the ToxI_Bt backbone are omitted for clarity. (B) Active site of ToxIN_Pa (PDB ID 2XDB). ToxN_Pa is shown in light magenta and ToxI_Pa in white. (C) In vitro degradation products of ToxIN_Pa and ToxIN_Bt. Substrate RNAs were incubated with purified ToxIN complex at a 1:4 molar ratio, and the products were examined by agarose gel electrophoresis. (D) Sequence-preference profiles of ToxN_Bt and ToxN_Pa, generated by performing 5′ RACE on the ToxIN_Bt- and ToxIN_Pa-cleavage products of E. coli rpoD, ompF, ompA, and dksA RNAs. The sequence-preference profiles shown are from a total of 14 (ToxIN_Pa) and 12 (ToxIN_Bt) unique 5′ ends.

Discussion

The potential for small RNAs to act as protein agonists has long been appreciated, as demonstrated by the directed evolution of RNA aptamers that inhibit HIV reverse transcriptase (26) and influenza virus B hemagglutinin (27), among others (28). Despite the successful experimental generation of these artificial species, there are very few examples of naturally occurring RNAs that act directly to impede protein activity in vivo, and ToxI antitoxins are, to our knowledge, the only RNAs whose function is to inhibit their own parent-processing enzyme. The questions naturally arise as to how they work and what is the origin of their molecular specificity. We show that ToxI antitoxins are selective enzyme inhibitors and that the capacities to inhibit the protein and to assemble the ToxIN_Pa complex are determined entirely by the sequence of the ToxI_Pa RNA and its interactions with its cognate toxin. We also show that ToxIN systems, in addition to their role in phage resistance (12, 13), have an addictive plasmid-maintenance activity, which likely contributes to their evolutionary success. The structural basis for ToxI antitoxin specificity is revealed through the crystal structure of a second ToxIN system. Finally, we show that ToxN proteins have a sequence-specific ribonuclease activity responsible both for their toxicity and for processing of their own RNA inhibitors.

ToxI_Pa inhibited ToxN_Pa in vitro, in both its processed and precursor forms (Fig. 4A), and the ToxI_Pa precursor also appeared to act as a preferred ToxN_Pa substrate. Therefore the requirements for toxin inhibition are contained entirely in the sequence of the repetitive ToxI_Pa transcript and are retained after cleavage into single pseudoknot units. Whether folding of the ToxI_Pa pseudoknot occurs before or concomitantly with binding and cleavage by ToxN_Pa remains to be addressed experimentally. The ToxIN_Pa complex assembled spontaneously when its components were combined in vitro, suggesting that toxin inhibition and formation of the trimeric protein–RNA complex are inextricably linked. Our results in vitro indicate that ToxI_Pa RNA is likely to be an extremely efficient and potent inhibitor of ToxN_Pa in vivo: Formation of the inhibited ToxIN_Pa complex is robust, and the ToxI_Pa precursor diverts free ToxN_Pa away from other substrates. Furthermore, these cleavage and inhibition events in vivo would occur in the context of a constant oversupply of ToxI_Pa precursor (12). The use of an RNA (rather than a protein) inhibitor in type III TA systems may offer benefits to the cell, such as a reduced metabolic cost and increased sensitivity to global changes in rates of protein and RNA synthesis. Our results suggest that these benefits do not come at a cost of reduced protection by the antitoxin.

The ToxI_Bt and ToxI_Pa RNAs are both pseudoknots (Fig. 5 B and C and Fig. S4), and this fold is likely to be conserved across the ToxI family despite considerable variation in primary sequence (15). Pseudoknot folds are a dominant form for structured RNAs, because this architecture is generally compact but can create accommodating surfaces for interaction with other molecules (29, 30). As a result, pseudoknot folds have been identified in functionally diverse RNAs including ribozymes (31), riboswitches (32, 33), and several RNA aptamers (26, 34). ToxI antitoxins are natural examples of RNA pseudoknots that inhibit enzymes, mirroring a function originally observed in artificially generated aptamers.

ToxI-mediated inhibition is selective (Fig. 2B), and the crystal structure of a second ToxIN system showed how this specificity is achieved. The ToxI_Bt and ToxN_Bt components are broadly similar to their Pectobacterium orthologs. However, minor changes in these components lead to substantial changes in the two extended protein–RNA interfaces of the ToxIN complex. Within the framework of a 3ToxI:3ToxN triangular complex, antitoxin binding appears to be specific and mediated primarily by a few variable regions of a common ToxN scaffold (Fig. 6 A–E). In addition, the mature form of each ToxI is generated through the sequence-specific RNase activity of its cognate toxin. Note that although ToxN_Pa cleaved the ToxI_Bt precursor in vitro (Fig. 4A), cleavage at each AA↓AU sequence would not generate the same 34-nt inhibitory units observed in the ToxIN_Bt complex (see Fig. S3B). In summary, ToxN_Bt and ToxN_Pa differ in both structure and endoribonuclease specificity, and these differences are matched by variations in their cognate antitoxins. These differences reflect the coevolution of each toxin with its inhibitory RNA partner and the selective pressure to maintain processing of the antitoxic RNA and formation of a stable complex.

The biological function of ToxI antitoxins to inhibit their own parent-processing enzymes is, to our knowledge, unique. However, parallels can be drawn between ToxN proteins and type I Cas6 ribonucleases of the anti-viral CRISPR/Cas systems, which also remain bound to their own catalytic product following cleavage of their CRISPR transcript substrates into crRNAs (35, 36). However, the functional consequences of these events are different. ToxI RNAs inhibit their processing enzymes to protect the cell from a harmful general RNase, whereas binding of crRNAs to their Cas6-processing enzymes (which do not cleave other RNAs) leads to formation of the Cascade ribonucleoprotein complex and its highly specific recognition of invading nucleic acids.

ToxN_Bt and ToxN_Pa displayed sequence-specific ribonuclease activity (Fig. 8), which also has been observed for numerous type II toxins including MazF and Kid (16, 17). Although there are examples of TA system toxins that target specific RNAs in vivo, including several VapC and MazF homologs (37–39), our Northern blot results show that ToxN_Pa is unlikely to do so, because this enzyme degraded all three transcripts tested (Fig. 1B and Fig. S1). This general RNase mechanism of toxicity is assumed to be the same for ToxN_Bt, which has a longer recognition sequence but greater tolerance for variations.

ToxIN_Pa is a powerful antiviral abortive infection system which also can stabilize plasmids (Fig. 2A and refs. 12 and 13). ToxIN_Bt can stabilize plasmids in a Bacillus host (Fig. 2B), but, unexpectedly, did not affect the replication of any of >100 environmental Bacillus phages. Therefore ToxIN_Bt likely functions in vivo to promote the maintenance of plasmid pAW63, although additional functions are possible. Type III TA systems are found in the chromosomes and plasmids of multiple prokaryotic phyla and appear to be horizontally transferred throughout these genomes (15). The broad-ranging ribonuclease specificity of the two ToxN homologs, and the demonstration of toxin inhibition and complex assembly in vitro, suggest that toxicity and antitoxicity have very little dependence on the host and so could be maintained even when these loci are transferred to distantly related bacteria. Indeed, the structurally similar ToxIN_Pa and ToxIN_Bt are both functional and able to confer adaptive benefits in their very distantly related Enterobacterium and Firmicute hosts. Our research presents a scenario in which ToxIN is an entirely self-contained, addictive, and potentially lethal molecular machine upon which evolution can act to drive distinct adaptive advantages within different populations of bacterial hosts.

Materials and Methods

Northern Blot.

E. coli DH5α strains carrying the plasmid pairs pTRB1/pTA76, pTA50/pTA76, or pTRB1/pTA100 (Table S3) were grown in 25 mL LB plus 0.2% (wt/vol) glucose as shaking cultures at 37 °C to OD₆₀₀ 0.6–0.8. Cells were spun down, resuspended in 25 mL LB plus 0.1% (wt/vol) L-arabinose to induce expression of ToxN_Pa, and grown at 37 °C for a further 2 h. ToxI_Pa expression then was induced by addition of 1 mM isopropylthio-β-galactoside (IPTG), and cells were grown for a further 4 h. RNA was extracted from cell samples before and after ToxN_Pa expression and after coexpression of ToxI_Pa, using an RNeasy kit (Qiagen). Antisense 3'-digoxigenin (DIG)–labeled RNA probes were transcribed in vitro from PCR products [αOmpA: pFLS50, primers FS61/TRB230; αToxI–pTA110: primers M13-20/TRB57 (Tables S3 and S4)] and used for Northern blotting of 3.5 μg total RNA according to the DIG user manual (Roche). Cell samples taken before and after ToxN_Pa expression and after coexpression of ToxI_Pa also were analyzed by Western blot for the FLAG-tagged ToxN_Pa protein, as reported previously (12).

ToxIN_Bt Mutagenesis and Toxicity/Antitoxicity Assays.

ToxN_Bt mutants were constructed by overlap-extension PCR and were cloned into pBAD30 (40). Single-repeat ToxI_Bt sequences were cloned into pTA100 as described (12). Single strains of E. coli DH5α were cotransformed with one ToxN_Bt and one ToxI_Bt plasmid (Table S3) and were used for overexpression-based toxicity and antitoxicity assays as described (12).ToxIN_Bt–ToxIN_Pa cross-talk experiments were performed in the same way, using pTA100-derived ToxI constructs containing the full series of repeats and pBAD30-derived ToxN constructs (Table S3).

Plasmid-Loss Assays.

Plasmid-loss experiments were performed in B. subtilis YB886 with plasmids pFLS79 and pFLS80 and in E. coli W3110 with plasmids pFLS118 and pFLS121 and the pRBJ200 vector control (Table S3). B. subtilis YB886 cells carrying test plasmids were grown overnight in LB supplemented with 10 μg⋅mL⁻¹ chloramphenicol. Fresh LB medium without antibiotics was inoculated with overnight culture at a calculated OD₆₀₀ of 5 × 10⁻⁷, and the culture was grown at 28 °C. Cultures were reinoculated into fresh LB at a starting OD₆₀₀ of 5 × 10⁻⁷ every 24 h to maintain exponential growth. The proportion of plasmid-containing cells at each time point was determined by serially diluting the cultures and plating on LB agar, then patching colonies onto LB plates supplemented with 10 μg⋅mL⁻¹ chloramphenicol. Plasmid-loss experiments using E. coli W3110 were performed as described previously (41) with nonselective exponential growth maintained for 24 h.

Purification of ToxN_Pa, ToxI_Pa and ToxIN_Pa.

ToxIN_Pa was expressed and purified from E. coli ER2566 pTRB14/pTRB18 as reported previously (14). A small proportion of the ToxIN_Pa complex dissociated between the affinity and anion exchange purification steps, so separate fractions of ToxN_Pa and ToxI_Pa were isolated by anion exchange chromatography in the same purification run as the trimeric ToxIN_Pa complex. Samples were concentrated and exchanged into gel filtration buffer (150 mM NaCl, 50 mM Tris⋅HCl, 1 mM DTT, pH 7.5) before use in RNA-degradation or complex-reassembly experiments.

In Vitro RNA Degradation Assays.

The full ToxI_Pa and ToxI_Bt precursor RNAs were transcribed in vitro from PCR products generated using the template/primer combinations pTA111:M13-20/MJ12 and pFLS66:M13-20/PF196 (Tables S3 and S4). ToxN_Pa protein and ToxI_Pa single repeats were purified by FPLC. RNA degradation reactions of 15-μL volume were set up in assay buffer [150 mM NaCl, 50 mM Tris⋅HCl (pH 7.5)] with E. coli ompA RNA (700 ng or 2 pmol) and ToxN_Pa in a 1:3 molar ratio, and ToxI RNA added at a 1:1 molar ratio of ToxI repeats to ToxN_Pa monomers. ToxN_Pa and ToxI components were allowed to react at 25 °C for 2 min before reactions were started by the addition of the ompA substrate. Time-point samples of 5 μL each were quenched in 5 μL 2× RNA loading buffer (Fermentas) and flash-frozen in liquid nitrogen. Reaction products were visualized by electrophoresis on a 6% (wt/vol) acrylamide TBE-Urea gel (UreaGel, National Diagnostics) followed by staining with SYBR Gold (Invitrogen).

ToxIN_Pa Complex Reassembly Reactions and Gel Filtration.

Reassembly reaction 1 contained 10 μg ToxN_Pa and 17 μg ToxI_Pa transcript (total volume 140 μL), and reaction 2 contained 20 μg ToxN_Pa and 150 μg ToxI_Pa single repeats (total volume 120 μL). All components were in gel filtration buffer. Reactions were incubated at 37 °C for 1 h, diluted to a 200-μL volume in gel filtration buffer, and then were used for size-exclusion chromatography on a Superdex S200 13/30 column (GE Healthcare) at a flow rate 0.6 mL⋅min⁻¹ and a load of 100 μL. Samples analyzed in this way were the two reassembly reactions, ToxN_Pa, ToxIN_Pa, ToxI_Pa monomer and full ToxI_Pa transcript controls, and five size-calibration standards (Table S1).

ToxIN_Bt Purification, Crystallization, and Structure Determination.

ToxI_Bt and ToxN_Bt were coexpressed in an E. coli ER2566 strain (New England BioLabs) carrying the plasmids pFLS67 and pFLS44 (Table S3). Cells were grown at 37 °C to OD₆₀₀ 0.8, and ToxIN_Bt complex was expressed overnight following a temperature shift to 18 °C and induction with 1 mM IPTG. Cells were resuspended in 5 mL lysis buffer [500 mM NaCl, 50 mM Tris⋅HCl, 10 mM imidazole, 10% (vol/vol) glycerol, pH 8.0] per gram of cell mass and were lysed by sonication, and the ToxIN_Bt complex was purified using Ni-NTA (Qiagen) Ni²⁺-affinity chromatography followed by HiTrap-Q (GE Healthcare) anion-exchange chromatography with a gradient of 50 mM to 1 mM NaCl. Purified ToxIN_Bt complex was concentrated to 6.8 mg⋅mL⁻¹ and was used directly in crystallization trials in its anion-exchange elution buffer (280–300 mM NaCl, 50 mM Tris⋅HCl, 1 mM DTT, pH 7.5). Initial crystallization screens were performed using vapor diffusion at 18 °C in 96-well sitting-drop plates using JCSG+, Procomplex, Nucleix, and PACT screens (Qiagen), with drop sizes of 100 nL protein plus 100 nL precipitant against a 200-μL reservoir of the same precipitant. Crystallization optimization was performed by conducting manual screens around several of the conditions that produced hits, and crystals diffracting to 2.2 Å were obtained by streak seeding into a 1.2 + 1.2 μL hanging-drop setup with a reservoir of 0.2 M ammonium phosphate, 0.1 M Tris⋅HCl (pH 8.5), and 50% (vol/vol) 2-methyl-2,4-pentanediol (MPD) (JCSG+ screen condition 11).

X-ray diffraction data from a single crystal in spacegroup P6 were collected at wavelength 0.9795 Å at station I02 of the Diamond Light Source, Oxford, UK. Data processing and reduction were performed with iMOSFLM (42), SCALA (43), and TRUNCATE (44) in the CCP4 suite of programs (45). Initial molecular replacement attempts in PHASER (46) with the ToxIN_Pa structure [Protein Data Bank (PDB) ID code 2XDB] did not yield a solution, so a ToxN_Bt homology model was generated using MODELER. The toxN_Pa and toxN_Bt sequences were aligned using FUGUE to guide the construction of a modified homology model in COOT (47) based on the ToxN_Pa chain in PDB 2XDB, with variable loop regions deleted and nonconserved residues changed to alanine. This search model, which did not contain either heteroatoms or ToxI, was used for molecular replacement (rotation and translation) in PHASER. The initial unbiased Fo-Fc map at 2.5σ (Fig. S3A) was calculated from this solution. This map gave positive peaks with a maximum value of 0.7989 electron⋅ų at an rmsd of 6.53 σ, corresponding to the PO₄⁻ groups of the bound ToxI_Bt RNA. The continuous Fo-Fc map allowed portions of the bases and backbone of a consensus ToxI_Bt RNA to be traced unambiguously. Following building of the ToxI_Bt, the ToxI_Bt:ToxN_Bt model was used as the search model for an additional iteration of molecular replacement in PHASER (Table S2). The structure refinement was performed using restrained maximum-likelihood target function, bulk solvent, and anisotropic scaling, and group_TLS (Translation/Libration/Screw) in PHENIX (48). Simulated annealing was used in the early stages to reduce model bias. Water molecules were assigned to positive peaks with a maximum peak height of 0.2107 electron⋅ų, with an rmsd in the range of 2.5–4.17 σ in the Fo-Fc map at full occupancy. No positive peaks at rmsd greater than 4.5 σ were observed in the Fo-Fc map, suggesting there were no metal ions present in the complex at significant occupancy. The stereochemistry of the final structure was checked using SFCHECK and PROCHECK from the CCP4 program suite. The final 2.2-Å structure of the ToxN_Bt–ToxI_Bt complex contains ToxN_Bt residues 5–172, 34 nt ToxI_Bt, and one MPD molecule from the crystallization precipitant (Table S2). Structural superimpositions were done using SUPERPOSE, LSQ, and COOT, and accessible surface areas were calculated using PISA with a probe solvent molecule radius of 1.4 Å. Figures were prepared in PyMOL. Leontis–Westhof symbolism for base pairing (49) was used in the ToxI_Bt schematic figure (Fig. 5B).

Identification of RNA Cleavage Sites by 5′ RACE.

E. coli K-12 rpoD, ompF, ompA, and dksA RNAs were transcribed from PCR product templates using T7 RNA polymerase (Fermentas). RNA was added to purified ToxIN_Pa or ToxIN_Bt in a 1:4 molar ratio and incubated in 50 mM NaCl, 50 mM Tris⋅HCl, pH 7.5, at 37 °C for 2 h. The cleaved RNA was purified using a NucleoSpin II RNA cleanup kit (Macherey-Nagel). Then 200 ng of treated RNA was reverse transcribed in a SuperScriptII RT (Invitrogen) reaction using up to three gene-specific primers (from FS60, FS62, FS64, FS66, FS68, and FS79–FS96; Table S4). Products then were poly(G)-tailed with TdT terminal transferase (Roche), amplified by PCR using a polyCD forward primer and a gene-specific reverse primer (Table S4), and sequenced. Where sequencing of a PCR product gave an ambiguous result, the product was cloned into pBS KSII+ and sequenced from the recombinant plasmid. Sequence specificity profiles were generated using WebLogo (50).