Oligonucleotides were prepared by automated DNA synthesis using cyanoethylphosphoramidite chemistry (Keck Biotechnology Resource Laboratory, Yale University; Mobix Central Facility, McMaster University). DNA and RNA oligonucleotides were purified by 10% preparative denaturing (8 M urea) PAGE and their concentrations were determined by spectroscopic methods.

Nucleoside 5′-triphosphates and [γ-³²P]ATP were purchased from Amersham Pharmacia. T4 DNA ligase, T4 polynucleotide kinase (PNK), calf intestine alkaline phosphatase and T7 RNA polymerase were purchased from MBI Fermentas. All chemical reagents were purchased from Sigma. The 50 nt RNA substrate (S) was produced by in vitro transcription using T7 RNA polymerase and an appropriate double-stranded DNA template generated by PCR as described previously (33), and standard in vitro transcription protocols were used with pairs of complementary synthetic DNA oligos to produce the 26 nt SS substrate, and 18 nt ST substrate. Nucleotide sequences of RNA substrates are as follows: 50 nt S = 5′-ggagagagaugggugcguuacguaaacuuacaucuacgaaucagguucga; 26 nt SS = 5′-ggagagagaugggugcguuacguaaa; 18 nt ST substrate = 5′-gggagagaugggugcguu.

The 50 nt or 26 nt RNA substrates were first ligated to 14 nt A1 (5′-ggaaactagacaga) in a separate large-scale ligation reaction. Typically, 2500 pmol of RNA substrate was combined with 2750 pmol ligation template (5′-tctctctcctctgtctag) and 3000 pmol of A1, heated for 30 s at 90°C and allowed to cool at room temperature for ∼10 min. After cooling, 10 µl of 10× ligase buffer (supplied by Fermentas) and 10 µl of T4 DNA ligase (5 Weiss U/µl) were added to initiate the ligation reaction. The reaction mixture (100 µl, total volume) was incubated at room temperature for 4 h or overnight. The ligated A1-RNA substrate was recovered by standard ethanol precipitation and purified by 10% denaturing PAGE. The A1-RNA substrate construct was 5′-³²P-labelled with PNK, extracted twice with phenol/chloroform/isoamyl alcohol (25:24:1), and ethanol precipitated before being used in separate small-scale ligation reactions with specific deoxyribozymes. Thirty pmol of A1-substrate was combined with 90 pmol of 5′-phosphorylated deoxyribozyme, 36 pmol of ligation template (5′-gcgtacgtgtcgaacctgattcg), and 45 pmol of substrate cleavage blocker (5′-tttacgtaacgcacccatctctctcctctgtctag), heated at 90°C for 30 s and allowed to cool at room temperature for ∼5–10 min. The substrate cleavage blocker is synthetic DNA complementary to a section of the RNA substrate surrounding the cleavage site, and was used to prevent deoxyribozyme-mediated cleavage during ligation. After cooling, 1.25 µl of 10× ligase buffer (supplied by Fermentas), 4 µl of T4 DNA ligase (5 Weiss U/µl) and DEPC treated H₂O were added to a final volume of 25 µl. The reaction was incubated at 37°C for 1 h, then ethanol precipitated and purified by 10% denaturing PAGE.

A typical intramolecular self-cleavage assay consisted of the following steps: (i) heat denaturation of the deoxyribozyme-substrate construct in water for 30 s at 90°C, followed by slow cooling at room temperature for ∼10 min, (ii) the addition of selection buffer (final concentration, 400 mM NaCl, 100 mM KCl, 7.5 mM MgCl₂, 7.5 mM MnCl₂, 50 mM HEPES, pH 7.0 at 23°C) and incubation at room temperature for a designated period of time, (iii) addition of EDTA to 30 mM to stop the reaction, (iv) separation of cleavage products by denaturing 10% PAGE and (v) quantitation using a PhosphorImager and ImageQuant software. Time courses for each deoxyribozyme were conducted at least twice using 13–16 different time points for each. The experimental data was fit to either a single, Y = Y_max(1 − e^(−k_obst)), or double, Y = Y_max1(1 − e^(−k_obs1t)) + Y_max2(1 − e^(−k_obs2t)), exponential equation using non-linear regression analysis in GraphPad Prism 4, from which the observed rate constant (k_obs) and maximum cleavage yield (Y_max) were determined.

For single turnover experiments, a 100-fold excess of deoxyribozyme was mixed with 5′-³²P-labelled substrate. Substrate and deoxyribozyme were heated together at 90°C for 30 s, and allowed to cool at room temperature for ∼10 min. A 2×selection buffer was added to initiate the reaction giving a final concentration of 1.66 µM deoxyribozyme to 0.0166 µM substrate. The reaction was terminated after a designated period of time by the addition of EDTA to 30 mM. The rate constant was determined from at least two independent experiments that differed by <10%. Each time course consisted of 16 time points. A graph of fraction cleaved versus time was plotted, and the experimental data fit to a single or double exponential equation using non-linear regression analysis in GraphPad Prism 4.

We previously conducted an in vitro selection experiment to isolate RNA-cleaving deoxyribozymes (33), using the library design illustrated in Figure 1A. In addition to 80 random deoxyribonucleotides, this library featured an RNA substrate composed of 50 fixed-sequence ribonucleotides. After performing sixteen rounds of selection, ∼50 clones were sequenced from each of generations 7, 8, 10, 13 and 15, which were subjected to increasing selection pressure in the form of decreasing reaction times of 5 h, 30 min, 5 min, 30 s and 5 s, respectively. From a total sample set of 245 sequenced clones, 110 unique sequence classes were identified. On average, different classes shared only ∼30% nucleotide identity, but clones within the same class boasted >90% identity. Further analysis revealed the presence of the 8–17 motif in ∼94% of all classes. For simplicity, we use the term 8–17 to describe both the canonical motif (32) and all related sequence variants.

The discovery of a motif as simple and well defined as 8–17, in so many different sequences, provided a unique opportunity to gain further insight into the role of excess sequence elements during in vitro selection. Toward this end, 11 sequence classes were characterized in greater detail. All eight classes identified in the terminal G15 population were investigated, including one that does not contain 8–17 or any other previously characterized motif. These deoxyribozymes presumably represent the best catalysts, because they survived under the most stringent selection time. For comparison, we also selected three classes that were observed only transiently in the preceding generations 8, 10 or 13. Figure 1B shows the sequence of a specific clone from each of the ten 8–17 containing classes that were examined, as well as the frequency and generation from which they were isolated (the non-8–17 class is addressed separately in a subsequent section). The 8–17 motif has been studied extensively (27,30–32,34,35), and is readily identifiable near the 5′ end by its characteristic sequence requirements.

The secondary structure and sequence requirements of the 8–17 motif are summarized in Figure 2A. The nucleotide numbering system used in Figure 2A is analogous to the system originally proposed for the hammerhead ribozyme (36), and subsequently adopted for the 8–17 deoxyribozyme (31,34). The canonical 8–17 motif as first described by Santoro and Joyce (32), was composed of 14–15 nt (plus 7–8 nt binding arms on either side) and characterized by a G-T wobble base pair at the cleavage site, a 3 bp stem of which at least two were G-C base pairs, an invariant A₆G₇C triloop, and a single-stranded region of sequence WC₁₃G₁₄R (W = A or T; R = A or G or AA). Subsequent studies have suggested that the sequence requirements of 8–17 are far more versatile. A study by Cruz et al. (30) reported that only four residues are strictly conserved (A₆ and G₇ in the triloop, C₁₃ and G₁₄ in the single-stranded turn region), the triloop region can also accept a tetraloop, the 3 bp stem can accommodate up to 1 or 2 mismatches and/or single-nucleotide bulges, and the single-stranded region has a more relaxed sequence requirement for positions W and R. Most recently, Peracchi et al. (34) have conducted a comprehensive mutational analysis to show the effect of different mutations on the catalytic activity of 8–17. From the results of Peracchi's study, the approximate relative activity at different nucleotide positions (in Mn²⁺ buffer) is predicted to be as follows: T_2.1>A>>G,C; C₈>T>>G,A; T₁₂>G>A>>C; A₁₅A_15.0>A>T>C>G. The magnitude and relative order of activities vary depending on the type and concentration of metal ion cofactors used. Furthermore, the composition of the 3 bp stem is also associated with varying levels of activity. The magnitudes of these effects are omitted since the assay conditions in reference (34) are not directly comparable to our own.

The suspected secondary structure interaction between the 8–17 containing DNA domain and the RNA substrate are illustrated in Figure 2B for each class. Approximately 91% of all the observed 8–17 motifs were located at the 5′ end of the DNA domain, which allowed them to exploit the 5′ primer-binding site (P1) as one binding arm. This unintentional sequence complementarity likely favoured the selection of catalysts that cleave the adjacent G-G dinucleotide junction. Although the observed 8–17 motifs satisfy the sequence requirements established by previous studies, formation of the 8–17 motif is not predicted to be the most energetically favourable according to secondary structure prediction software, such as mfold (37). This discrepancy is maintained even when full-length molecules are truncated to remove excess sequence elements. Therefore, we synthesized 3′ truncated enzymes corresponding to the nucleotide sequences shown in Figure 2B (containing only the suspected 8–17 motif and 8–10 nt binding arms), and ligated them to a truncated RNA substrate lacking 24 consecutive nucleotides from the 3′ end (corresponding to the loop region that is not expected to interact with the deoxyribozyme domain). Figure 2C indicates that these truncation derivatives are still catalytically active, but the same constructs containing either a G₇ to T substitution, or a C₁₃ to T substitution are completely inactive. These experimental results are consistent with theoretical expectations (both G₇ and C₁₃ are known to be invariant nucleotides), and strongly support a catalytic role for the embedded 8–17 motif.

To assess the effect of excess sequence elements on the activity of the 8–17 motif, four different constructs were made for each sequence class, which are depicted in Figure 3. These constructs include: (i) the full-length enzyme, denoted by ‘L’, (ii) a 3′ truncated enzyme containing only the 8–17 motif and an 8–10 nt 3′ substrate binding arm (corresponding to the sequences shown in Figure 2B), denoted by ‘S’ (iii) the same truncated enzyme from part ii with a truncated RNA substrate lacking 24 nt from the 3′ end, denoted by ‘SS’ and (iv) a 3′ truncated enzyme containing only the 8–17 motif and a perfect 3′ binding arm containing 10 consecutive Watson–Crick base pairs, denoted by ‘PBA’. All rates were obtained using single-turnover cis constructs. However, each class is also active in a bimolecular trans format (Supplementary Figure 1). Table 1 summarizes the results of these truncation experiments in terms of their effect on the catalytic rate constant, k_obs. Since a deoxyribozyme's ability to fold into an active conformation is critical for function, we have also summarized the results of the truncation experiments in terms of their effect on the maximum cleavage yield, denoted by Y_max in Table 2.

The results shown in Table 1 indicate that the catalytic rate of each class closely parallels its abundance in the population. For instance, classes such as G15-1, G15-16 and G15-39 that boast the fastest rates also represent the largest fraction of the G15 population (see Figure 1B for class frequency). In contrast, the classes that died out in earlier generations (i.e. G13-20, G10-47 and G8-1), not surprisingly, have the lowest rates. Similarly, the most abundant classes may benefit from a slight folding advantage over others, as reflected by the higher Y_max values shown in Table 2. The relative activity of the different 8–17 motifs is also consistent with previous reports (34).

The average effect of the excess 3′ sequence in the deoxyribozyme domain is neutral, although an even distribution of both beneficial and detrimental effects is observed (Table 1, column S/L). The magnitude of these effects, however, is relatively modest with a 50% increase or decrease in activity representing the maximum change. This modest change in activity suggests that the excess 3′ sequence is acting only to mediate the stability of 8–17's active fold, likely through the formation of the 3′ substrate-binding arm. The average effect of deleting excess sequence from the RNA substrate domain is a 40% increase in activity (Table 1, column SS/S). The activity of seven classes were inhibited by the excess ribonucleotides in varying amounts, one was neutral (G8-1), and the remainder were associated with a slightly beneficial effect (G15-1, G15-39). This variation between classes is noteworthy, since the RNA sequence is constant and presumably can form the same 5′ substrate-binding arm with P1 in each class. Therefore, the magnitude of the effect is likely influenced by downstream elements including the stability of the 3′ substrate-binding arm and perhaps even the type of 8–17 motif. On average, the activity of the G15 classes appear to be less affected by the presence of excess sequence elements in both the deoxyribozyme and substrate domains, than classes that died out in earlier generations (k_{rel,S/L,G15avg} = 0.9 vs. k_{rel,S/L,<G15avg} = 1.3; k_{rel,SS/S,G15avg} = 1.3 vs. k_{rel,SS/S,<G15avg} = 1.7). This effect closely parallels the greater stability associated with the 3 substrate binding arm of G15 classes (k_{rel,PBA/S,G15avg} = 1.4 versus k_{rel,PBA/S,<G15avg} = 1.8).

As expected, optimized 3′ binding arms resulted in an average increase in catalytic activity (∼50%), although three classes (G15-1, G15-39 and G15-35) actually decreased in activity (∼10–20%), contrary to expectations (Table 1, column PBAA/S). This decrease can probably be attributed in part to experimental error, but also to the peculiarities associated with G15-1 and G15-35 (and related classes G15-48, G13-20 and G10-47). These classes have only a single adenosine residue in the terminal A₁₅ position of the single-stranded bulge region, which may base pair with the uridine residue immediately 5′ to the GG cleavage site. The extra stability conferred by the additional downstream base pairs in the perfect binding arm (PBA construct), may promote the A-U base pair to form and thereby constrain the 8–17 motif into a sub-optimal fold. Consistent with this theory, the original weaker binding arm (S construct) is associated with higher activity in these classes. Moreover, the insertion of another A residue adjacent to the existing one in the perfect binding arm (PBAA column in Tables 1 and 2) nearly restores activity to wild-type level.

Interestingly, the excess sequence elements in both the DNA and RNA domains do not seem to have a significant impact on the maximum cleavage yield. The general effect appears to be neutral, and the minor changes that are observed are likely due to inherent experimental variation. However, one class (G15-49) experiences a 70% decrease in Y_max after deletion of its excess 3′ DNA sequence, the magnitude of which cannot be explained by simple experimental variation. Instead, we suspect the disruptive effect of the two T bulges in the 5′ and 3′ binding arms is attenuated by the presence of the downstream DNA sequence, and intensified in its absence. The downstream sequence elements may form additional stabilizing contacts with A1. Consistent with this theory, Y_max is ∼3.7-fold larger when both T bulges are removed in G15-49 PBA.

We wondered if the results of our 10 deoxyribozyme sample set were representative of the global 8–17 population from our selection experiment. On average, a neutral or modest inhibitory effect was observed with excess sequence elements in the deoxyribozyme and substrate domains, respectively. However, we might expect that the G15 classes would have been selected, in part, for their ability to minimize any negative effect associated with excess sequence elements. To address this issue, we looked for the six different 8–17 variants from G15 (G15-35 and G15-48 have the same 8–17 core motif) in the other ∼100 sequence classes that were eliminated during earlier generations under more permissive selection conditions. Most of these fast 8–17 variants were unique, and not observed in other generations. However, the 8–17 catalytic core from G15-1 was also present in G13-E78, and G15-35 shares the same core motif with G15-48, G13-20, G10-47 and G7-E33. We determined that the full-length G13-E78 deoxyribozyme has a k_obs of 1.09 min⁻¹, and a Y_max of 66%. Therefore, G13-E78 was not eliminated because of an uncharacteristically large inhibitory effect from excess sequence elements. We suspect G13-E78's absence from the generation 15 population may be due to sampling error, PCR bias, sub-sampling effects or some combination thereof.

Classes G15-35 and G15-48 are 2–3 times faster than G13-20 and G10-47 even though they contain the same 8–17 catalytic core. The disparity between their rates, however, can be explained by the difference in their 3′ substrate binding arms. Both G13-20 and G10-47 have relatively poor 3′ substrate binding arms as reflected by the ∼2-fold higher activity observed upon optimization (Table 1, PBAA/S column). This contrasts with G15-35 and G15-48, which experience an approximately neutral effect upon optimization of their 3′ substrate binding arms. It should be noted that G15-48 and G13-20 appear to follow biphasic kinetics, which are characterized by a fast initial phase followed by a slower second phase. Though not particularly common, biphasic kinetics have been observed previously in other deoxyribozymes (38), and ribozymes like the hairpin (39,40) in which it was attributed to the formation of alternative conformations. Interestingly, the 3′-truncated versions of each deoxyribozyme exhibit monophasic kinetics, which suggests that the extra nucleotides may facilitate the formation of alternative structures and the observed biphasic kinetics (Figure 4).

The 8–17 motifs of all sequence classes are listed in Supplementary Figure 2, and are grouped into four primary phenotypic classes (based on the relative activity scheme described in Figure 2A). Although we have listed 22 other 8–17 variants in the same general phenotypic class as G15-1, G15-16 and G15-39, it should be noted that these classes are still likely to be diverse in activity, and further subdivision has not been attempted due to insufficient mutational data. However, we do not expect that any of these classes died out due to an uncharacteristically large inhibitory effect from excess sequence elements. We identified two classes (G7-E129 and G10-E36.2) that closely resemble the fastest characterized 8–17 variant (i.e. identical to G15-16 but lacking one of the terminal adenosine residues and containing a different 3′ substrate-binding arm), yet died out in generation 7 and 10, respectively. The full-length version of G7-E129 was assayed for self-cleavage yielding a k_obs of 0.58 min⁻¹ and Y_max of 61%, which provides further evidence that the results from our sample set are likely representative of the general population. At this time, we can only speculate as to why G7-E129 was eliminated so early in the selection, despite having catalytic properties comparable to G15-49 (k_obs = 0.56 min⁻¹, Y_max = 65%). Other factors, such as PCR amplification bias, random genetic drift, and sampling error may have contributed to the loss of this sequence class. We have previously initiated an investigation into the population dynamics of in vitro selection, and expect future reports to yield even greater insight into the significance of these factors on the outcome of selection (41).

From a total of 110 sequence classes isolated in our previous selection experiment, only seven did not appear to contain the 8–17 motif or any other previously reported RNA-cleaving deoxyribozyme motif. Five of these classes were eliminated by generation 8, and another class was observed as a single clone in G13. Only one class (G15-30) survived the stringent 5 s selection pressure applied in G15, and represented ∼10% of the population (from a sample of 49 clones). G15-30 was therefore selected for further examination. It should be noted that this class does possess some of the characteristic sequence requirements of 8–17 near the 5′ end, but lacks the strictly conserved G₁₄ residue from the single-stranded turn region described in Figure 2A. Despite this discrepancy we originally suggested that catalysis in G15-30 [referred to previously as DZ4 in reference (33)], might still be supported by an unusual 8–17 variant. However, subsequent evidence from a more rigorous investigation supports the contrary. Multiple truncation tests were carried out to determine the minimal sequence requirements and the effect of excess sequence elements on G15-30. Figure 5A shows the nucleotide sequence of G15-30 and the catalytic rate constant associated with different truncated versions. The full-length molecule possesses a k_obs of 0.48 min⁻¹, but deleting 60 nt from the 3′ end leads to a k_obs of 1.4 min⁻¹. Further deletion causes a reduction in activity, with no activity being observed after an additional 10 nt are removed. The relative activity of the various deletion constructs are illustrated in Figure 5B. Point mutations were also introduced into the 59 nt version at positions expected to be crucial for activity, if 8–17 was indeed responsible for catalysis. However, a 5′ G22T mutant exhibited a k_obs of 1.1 min⁻¹, and a 5′ C29T mutant exhibited a k_obs of 0.28 min⁻¹, indicating that G15-30 does not utilize an 8–17 motif. Remarkably, the 59 nt version of G15-30 in conjunction with a truncated RNA substrate (59SS) is about 6-fold more active than the original 119 nt version. We used mfold to facilitate the design of a trans-acting version, called G15-30T, that engages a shortened RNA substrate through the formation of two duplex stems flanking the cleavage site (Figure 5C). The P1 primer-binding site is predicted to act as part of the 5′ substrate-binding arm of G15-30T, thereby defining the boundaries of the minimal catalytic core. Nevertheless, additional confirmatory deletion tests were conducted on the 5′ end of this trans-acting deoxyribozyme, and the results were consistent with the proposed structure model (data not shown). Under single-turnover conditions, G15-30T exhibits biphasic kinetics with a k_obs of 2.48 min⁻¹ (Y_max = 52%) in the first phase, and 0.11 min⁻¹ (Y_max = 24%) in the second phase (Figure 5D). G15-30T is also capable of multiple substrate turnovers. Over a 60 min incubation period, in the presence of 10 nM deoxyribozyme and 1 µM substrate, ∼41% of substrate was cleaved corresponding to 41 turnovers (Figure 5E). Further characterization and optimization of G15-30T represents a future objective.