A biologically stable DNAzyme that efficiently silences gene expression in cells

Abstract

Efforts to use RNA-cleaving DNA enzymes (DNAzymes) as gene-silencing agents in therapeutic applications have stalled due to their low efficacy in clinical trials. Here we report a xeno-nucleic-acid-modified version of the classic DNAzyme 10–23 that achieves multiple-turnover activity under cellular conditions and resists nuclease digestion. The new reagent, X10–23, overcomes the problem of product inhibition, which limited previous 10–23 designs, using molecular chemotypes with DNA, 2′-fluoroarabino nucleic acid and α-l-threofuranosyl nucleic acid backbone architectures that balance the effects of enhanced biological stability with RNA hybridization and divalent metal ion coordination. In cultured mammalian cells, X10–23 facilitates persistent gene silencing by efficiently degrading exogenous and endogenous messenger RNA transcripts. Together, these results demonstrate that new molecular chemotypes can improve the activity and stability of DNAzymes, and may provide a new route for nucleic acid enzymes to reach the clinic.

The DNA enzyme (DNAzyme) 10–23 (Fig. 1a), also known as Dz10–23 or simply 10–23, is the best characterized example of a Mg²⁺-dependent RNA-cleaving DNA enzyme created by in vitro selection¹. The enzyme was identified from a population of 10¹⁴ unique DNA molecules using a stringent selection strategy that was designed to favour the enrichment of individual molecules that promote the site-specific cleavage of RNA transcripts. The enzyme comprises a 15-nucleotide (nt) catalytic domain that is flanked on both sides by substrate-binding arms that can vary in length depending on the sequence of the RNA substrate². As with other oligonucleotide therapeutics, the RNA target is recognized by complementary Watson–Crick base pairing³. On binding, RNA cleavage ensues at a predefined purine–pyrimidine junction with the highest activity levels observed for G-U dinucleotides⁴. The cleavage mechanism involves metal-assisted deprotonation of a 2′-hydroxyl from the purine (R) nucleotide, followed by nucleophilic attack on the neighbouring phosphodiester bond to yield an upstream cleavage product with a 2′,3′-cyclic phosphate and a downstream cleavage product with a 5′-hydroxyl group.

Fig. 1 |. Kinetic analysis of F10–23.

a, Left: 10–23 (black) in complex with an RNA substrate (red). Right: chemical structure of DNA. b, Left: F10–23 enzyme (orange) in complex with an RNA substrate (red). Right chemical structure of FANA. c,d, Pre-steady-state kinetic analysis of RNA cleavage by 10–23 and F10–23. The reactions were performed in buffer that contained either 10 mM MgCl₂ (c) or 1 mM MgCl₂ (d) and 150 mM NaCl at 24 °C (pH 7.5) with 0.5 μM substrate and 2.5 μM enzyme. Data points indicate the percentage substrate cleavage. Error bars denote ±s.d. of the mean for n = 3 independent replicates.

Over the years, 10–23 has been chemically modified in various ways to achieve an improved efficacy in vivo and in cells. Chemical modifications used for this purpose include phosphorothioate linkages^5,6, 2′-O-methylribonucleotides^7,8, inverted 3′-3′ thymidine nucleotides^7,8, phosphoramidite linkages⁹ and locked nucleic acids (LNAs)¹⁰. The effect of these modifications on the catalytic activity of the enzyme ranges from deleterious to beneficial depending on the residue location and type of chemical modification¹¹. Most chemical modifications are directed at the substrate binding arms with the goal to increase the affinity of the reagent for the RNA target^5–10. However, this strategy poses a barrier to improving the catalytic activity of DNAzymes, as the modifications chosen for the enhanced RNA binding often lead to product inhibition with the enzyme–product complex failing to dissociate from the postcatalytic state. Thus, new molecular designs are needed for DNAzymes to efficiently cleave mRNA transcripts in cellular systems.

We hypothesized that xeno-nucleic acids (XNAs) offered a new molecular chemotype with physicochemical properties that could achieve an enhanced biological stability without sacrificing the catalytic activity under multiple-turnover conditions that typify intracellular conditions. Guided by nucleic acid chemistry, we searched for XNA residues that would provide a balanced solution to the problem of how to enhance the substrate binding kinetics but avoid the harmful effects of product inhibition. We viewed biological stability and catalytic turnover as the main obstacles that separated DNAzymes from protein-catalysed gene-silencing reagents, such as antisense or short interfering RNA reagents³. Indeed, previous clinical trials for basal cell carcinoma and asthma have already shown that 10–23 functions with a high safety profile and tolerability in human patients using topical drug administration methods that involve either direct injection or inhalation of a nebulized solution^12–14. In addition, other preclinical investigations have shown that 10–23 is non-toxic to the cell and does not elicit an innate immune response due to the presence of CpG motifs in the catalytic domain^15,16. These findings provide confidence that 10–23 variants with an improved catalytic turnover could lead to renewed interest in DNAzymes for clinical applications.

Here we report a re-engineered version of the classic 10–23 DNAzyme that mediates persistent gene-silencing activity in cultured mammalian cells and simultaneously resists nuclease digestion. The new reagent, termed X10–23, was discovered using a medicinal chemistry approach that probed each position in the DNA backbone for structural mutations that promote an enhanced catalytic activity under simulated physiological conditions. Our results demonstrate that new molecular chemotypes can greatly improve the catalytic activity of a highly evolved DNAzyme, which suggests that molecular design is a powerful approach to optimize nucleic acid enzymes with potential value as future therapeutic agents.

Results

Optimizing the substrate recognition domain.

Recognizing that XNAs harbour physicochemical properties that are distinct from those found in natural DNA and RNA¹⁷, we wondered whether XNA residues could be used to enhance the RNA-cleavage activity of 10–23 under physiological conditions. We began by replacing all of the DNA residues in the binding arms of the substrate recognition domain with 2′-fluoroarabino nucleic acids (FANA, Fig. 1b). FANA is a close structural analogue of DNA that contains a fluorine atom at the 2′ position of a 2′-deoxyarabinose sugar¹⁸. FANA was a reasonable choice for increasing the association kinetics of the enzyme as previous hybridization studies showed that FANA–RNA duplexes are typically ~1.2 °C more stable per base pair than the equivalent DNA–RNA duplexes¹⁹. Apart from thymidine, which was substituted for uridine, each DNA nucleotide was replaced with the corresponding FANA nucleotide. LNA was also considered as a possible XNA modification due to its high thermal stability with RNA (~2–3 °C per base pair), but concerns over its cellular toxicity led us to focus our efforts on FANA²⁰. This version of 10–23, termed F10–23, functioned with an observed pseudo first-order rate constant (k_obs) of 0.57 min⁻¹ under single-turnover conditions in a buffer that contained 10 mM MgCl₂ and 150 mM NaCl (pH 7.5, 24 °C), which is nearly twofold faster than that of the unmodified parent enzyme (Fig. 1c and Supplementary Fig. 1a). The enhanced activity of F10–23 over 10–23 was maintained under physiological conditions in which the Mg²⁺ concentration was reduced to 1 mM (k_obs of 0.015 min⁻¹ versus 0.009 min⁻¹, respectively) (Fig. 1d and Supplementary Fig. 1b)^21,22.

Cleavage of long RNA substrates.

To provide further evidence of RNA cleavage activity, F10–23 was challenged to cut a much longer RNA substrate that was more reminiscent of a biological RNA molecule found in nature. For this example, we chose a 103 nt segment of the ribosomal modification protein rimK, generated as an unlabelled RNA transcript by in vitro transcription with T7 RNA polymerase. The catalytic activity of F10–23 was compared those of standard 10–23 under single-turnover, steady-state and multiple-turnover conditions in a physiological buffer that contained 1 mM MgCl₂ and 150 mM NaCl (pH 7.5, 24 °C). Here, steady-state kinetic measurements are viewed as a more rigorous test of catalytic activity than the more common single-turnover reaction. Analysis of the reaction products by denaturing polyacrylamide gel electrophoresis (PAGE) (Supplementary Fig. 2) with ethidium bromide staining revealed the appearance of upstream (37 nt) and downstream (66 nt) cleavage fragments produced by sequence-specific cleavage of a central G-U dinucleotide junction. Consistent with our previous observations on the shorter RNA substrate, F10–23 was faster than the parent 10–23 enzyme in all the cases tested, which suggests that F10–23 has the potential to invade folded RNA structures found in cellular systems.

Optimizing the catalytic core.

We next asked whether the activity of 10–23 could be further enhanced by introducing chemical modifications into the catalytic domain. Although this region of the sequence represents an evolutionary optimum in which almost any nucleotide change leads to a mutant enzyme with reduced catalytic activity¹, substantially less is known about the tolerance of the catalytic core towards chemical modifications that alter the sugar moiety. We systematically replaced each DNA residue in the catalytic core with the corresponding FANA nucleotide (for example, dA was replaced with fanaA). The complete set of 15 single-point mutant enzymes was assayed for RNA cleavage activity under single-turnover conditions in a physiological buffer that contained 1 mM MgCl₂ and 150 mM NaCl (pH 7.5, 24 °C). The catalytic profile (Fig. 2a and Supplementary Fig. 3a) indicates that residues G2 and T8 are highly tolerant to substitution with the FANA residues, as they maintain ~80% activity relative to the parent enzyme. This result is consistent with a previous report that showed G2 and T8 accept conformational distortions caused by 2′-5′ phosphodiester linkages²³. Of the remaining positions, substitutions made to the T4 and A9 positions showed moderate activity (50–60%), whereas the C9, G14 and A15 positions had low activity (20–30%). The remaining eight positions were each inactive when the wild-type DNA residue was replaced with FANA. Surprisingly, a designed version of 10–23 that carried both the G2 and U8 FANA mutations was nearly 50% more active than the parental enzyme (Supplementary Fig. 3b), which suggests that these two substitutions function with a synergistic activity.

Fig. 2 |. Engineering of the X10–23 nucleic acid enzyme.

a, Structure–activity mapping of the catalytic core of 10–23 by FANA substitutions. Data points indicate the relative substrate cleavage normalized to the wild-type 10–23 DNAzyme. Error bars denote ±s.d. of the mean for n = 3 independent replicates. b, The X10–23 enzyme in complex with an RNA substrate. RNA, red; DNA, black; FANA, orange; TNA, blue. c,d, Pre-steady-state kinetic analysis of RNA cleavage by F10–23 and X10–23. Reactions were performed in a buffer that contained either 10 mM MgCl₂ (c) or 1 mM MgCl₂ (d) and 150 mM NaCl at 24 °C (pH 7.5) with 0.5 μM substrate and 2.5 μM enzyme. Data points indicate percent substrate cleavage. Error bars denote ±s.d. of the mean for n = 3 independent replicates. e,f, Representative PAGE gels showing RNA cleavage by the engineered 10–23 variants in a buffer that contained either 10 mM MgCl₂ (e) or 1 mM MgCl₂ (f). S, full-length substrate; P, 5′ cleavage product. Molecular weight markers are indicated to the right of the gel.

Designing the X10–23 enzyme.

Encouraged by the structural mutagenesis study, we synthesized a new version of the enzyme in which the catalytic core of F10–23 was modified to contain both G2 and U8 FANA substitutions and non-complementary α-l-threofuranosyl thymidine (tT) residues were added to the 5′ and 3′ terminal positions to protect the oligonucleotide against nuclease digestion (Fig. 2b). α-L-threofuranosyl nucleic acid (TNA) is an artificial genetic polymer in which the natural five-carbon ribose sugar found in RNA has been replaced with an unnatural four-carbon threose sugar²⁴. TNA is an ideal choice to stabilize the backbone structure against nuclease digestion as TNA is completely recalcitrant to DNA and RNA degrading enzymes²⁵. This new version of the enzyme, termed X10–23, contains three different classes of nucleic acids (DNA, FANA and TNA) and functions with pseudo first-order rate constants of 0.68 min⁻¹ and 0.018 min⁻¹ in a reaction buffer (150 mM NaCl (pH 7.5, 24 °C)) that contains 10 and 1 mM MgCl₂, respectively (Fig. 2c–f). These rates are at least twofold faster than those of the parent enzyme.

We next asked whether the enhanced chemical diversity of X10–23 enabled a higher multiple-turnover activity in vitro. We began by evaluating 10–23, F10–23 and X10–23 under steady-state conditions with equimolar concentrations of substrate and enzyme. Kinetic measurements revealed that F10–23 and X10–23 (Fig. 3a) are about threefold faster than 10–23, which is consistent with their activity under pre-steady-state conditions. However, striking differences were observed under multiple-turnover conditions when the RNA substrate was present in tenfold molar excess over the enzyme. Under these conditions, F10–23 and X10–23 are ~50-fold more active than the parent enzyme, which suggests that important structural differences exist between the pre- and post-catalytic state of X10–23 versus the parent 10–23 DNAzyme. No differences in catalytic activity were observed when the binding arms of the parent enzyme were extended by 1 nt (7 + 7 rather than 6 + 6) (Fig. 3a). Furthermore, product inhibition was not observed when the analogous 7 + 7 construct was tested for F10–23 (Supplementary Fig. 4), which indicates that FANA provides a balanced solution to the problem of how to enhance substrate binding kinetics and yet avoid the harmful effects of product inhibition.

Fig. 3 |. Functional activity and biostability of X10–23.

a, Representative PAGE gels showing RNA cleavage activity under steady-state and multiple-turnover conditions. RNA cleavage reactions were performed in a buffer that contained 50 mM Tris-HCl (pH 7.5), 1 mM MgCl₂ and 150 mM NaCl at 24 °C with 0.5 μM substrate and either 0.5 μM (steady-state) (left) or 50 nM (multiple-turnover) (right) enzyme. b, Time-dependent biostability assay evaluated by denaturing PAGE. RNA-cleaving enzymes were evaluated in DMEM that contained 1 μM enzyme in the presence of 2 mg ml–1 of human liver microsome (left), 50% human serum (v/v) (middle) or 10 mU ml of SVPDE (right) at 37 °C. Molecular weight markers are indicated to the right of the gel.

Evaluating the biostability of X10–23.

Along with efficient catalytic activity, biostability is a critical parameter to achieve an improved efficacy in cellular systems that contain strong DNA and RNA degrading enzymes. For this assay, we analysed the stability of the 10–23, F10–23 and X10–23 scaffolds in concentrated human liver microsomes and 50% human serum in DMEM. Both assays provide a rigorous test of oligonucleotide stability due to the abundance and diversity of nucleases present in the media²⁶. In addition, we also evaluated each scaffold against snake venom phosphodiesterase (SVPDE), an aggressive enzyme with strong 3′-exonuclease activity commonly employed to evaluate the stability of oligonucleotide therapeutics²⁷. The results (Fig. 3b and Supplementary Fig. 5) clearly show that X10–23 displays a markedly enhanced biostability under all the conditions tested, with almost no degradation observed after either 21 hours of incubation in human liver microsomes or human serum, or after 90 minutes of incubation in SVPDE. By comparison, 10–23 and F10–23 showed substantial degradation, with F10–23 being slightly less stable than 10–23. This result validates the utility of TNA as a capping agent to protect the 5′ and 3′ termini against nuclease digestion.

Comparing X10–23 with 2′-O-methyl RNA and LNA versions of 10–23.

We next asked how X10–23 compared with other chemically enhanced versions of 10–23 that have been previously evaluated as gene-silencing reagents. Among the various combinations, 2′-O-methyl ribonucleotides and LNA have received notable attention as chemical modifications that function with an enhanced activity and biostability^7,8,10, which is consistent with their broad deployment in other classes of therapeutic oligonucleotides²⁸. For this study, we synthesized OME10–23 and LNA10–23, two 10–23 analogues with substrate binding arms that are complementary to the RNA substrate (Fig. 4a,b)^7,10. OME10–23 is an analogue of 10–23 in which 16 DNA residues are replaced with 2′-O-methyl ribonucleotides (OME) (ten in the substrate binding arms and six in the catalytic core), whereas LNA10–23 is a 10–23 analogue in which three terminal DNA residues in each binding arm are replaced with LNA.

Fig. 4 |. Alternative 10–23 designs.

a,b, OME10–23 (a) and LNA10–23 (b) in complex with an RNA substrate. RNA, red; DNA, black; OME, green; LNA, magenta. c, Representative PAGE gels showing RNA cleavage activity under steady-state and multiple-turnover conditions. RNA cleavage reactions were performed in a buffer that contained 50 mM Tris-HCl (pH 7.5), 1 mM MgCl₂ and 150 mM NaCl at 24 °C with 0.5 μM substrate and either 0.5 μM (steady-state) (left) or 50 nM (multiple-turnover) (right) enzyme. Molecular weight markers are indicated to the right of the gel. d, Structure–activity map of 10–23 activity for different chemical substitutions. Data points indicate the percentage substrate cleavage. Error bars denote ±s.d. of the mean for n = 3 independent replicates.

Kinetic measurements indicate that LNA10–23 is markedly faster than OME10–23 under all the conditions tested. Under single-turnover conditions, LNA10–23 functions with a rate of 0.26 min⁻¹ in the presence of 10 mM MgCl₂ and 0.03 min⁻¹ when the concentration of Mg²⁺ is reduced to 1 mM (Supplementary Fig. 6). These values compare favourably against OME10–23, which achieves rates of only 0.011 and 0.001 min⁻¹ under identical conditions of high and low Mg²⁺ ions, respectively (Supplementary Fig. 6). The superior activity of LNA10–23 over OME10–23 was maintained under steady-state conditions in which the substrate and enzyme are present in equimolar concentrations (Fig. 4c). However, the kinetic profile changed dramatically under multiple-turnover conditions where LNA10–23 shows clear signs of product inhibition (Fig. 4c). Thus, even though LNA10–23 is a faster enzyme than X10–23 under single-turnover conditions, X10–23 is a better candidate for cellular applications (Fig. 4d) due to its robust multiple-turnover activity in vitro.

Intracellular reduction of green fluorescent protein.

Next, we investigated the activity of X10–23 in cultured mammalian cells using the green fluorescent protein (GFP) as an optical reporter for gene-silencing activity. We measured GFP expression in the presence and absence of two X10–23 reagents that were designed to target G-U dinucleotides in the coding (internal) and 3′ untranslated region (3′-UTR) of the GFP mRNA transcript (Fig. 5a). Both reagents were independently validated in vitro using synthetic RNA oligonucleotides that matched the GFP segments targeted in the cellular assays (Supplementary Fig. 7). Cellular assays were performed in multiple formats using HEK293T (HEK) cells that were transfected with a GFP expression plasmid driven by a CMV (human cytomegalovirus) promoter. Fluorescent images collected after 24 hours of incubation post-transfection revealed a loss of GFP signal for cells that were co-transfected with the GFP plasmid and an X10–23 reagent targeting either the internal site or 3′-UTR site, or both sites simultaneously, as compared with cells transfected with the GFP plasmid only (Fig. 5b). Quantitative reverse transcription PCR (qRT-PCR) measurements confirm that loss of the GFP signal is due to a drop in mRNA template copy number for GFP (Fig. 5c), which demonstrates a reduction of both protein and mRNA levels in the cell.

Fig. 5 |. GFP inhibition activity of X10–23 in HEK293 cells.

a, Schematic representation of X10–23 molecules used for the intracellular inhibition of GFP. HEK293 cells were co-transfected by X10–23 molecules that targeted two different sites in the GFP mRNA transcript. Levels of gene-silencing activity were measured by GFP fluorescent cell imaging and qRT-PCR. (A)_n refers to the polyA tail on the mRNA strand. b, GFP fluorescent cell images collected prior to harvesting the cells at 24 h post-transfection. Scale bar, 100 μm. c, qRT-PCR analysis of DNA-free total RNA isolated 24 h post-transfection using GFP-specific and GAPDH (glyceraldehyde 3-phosphate dehydrogenase) loading control primers. Two biological replicates and three technical replicates were collected for each condition, with one representative biological replicate shown in b and c. Error bars are ±s.d. of the mean for n = 3 technical replicates.

Given the strength of the CMV promoter, we hypothesized that each X10–23 reagent must be engaging multiple mRNA templates in the cytoplasm to maintain strong gene-silencing activity under constitutive GFP expression conditions. Recognizing that constitutive gene expression from a CMV promoter produces larger quantities of RNA than those from endogenous gene expression, we administered a dose-dependent treatment of actinomycin D (ActD) for 4 hours after 20 hours of incubation post-transfection to inhibit the RNA transcription. ActD is a transcriptional inhibitor that prevents the continued expression of GFP in the cell^29,30, which allows X10–23 to engage only those GFP transcripts that are present when the antibiotic is administered to the cells. qRT-PCR analysis of cellular GFP transcripts shows a threefold reduction in template copy number by X10–23 when the cells are treated with 40 μM ActD, as compared with cells that are co-transfected with the GFP plasmid and X10–23 but not treated with the antibiotic (Supplementary Fig. 8). These results compare favourably against the parent 10–23 DNA enzyme synthesized with an 3′ inverted dT cap (Supplementary Fig. 5) as well as the inactive X10–23 variant and a fully complementary FANA antisense strand capped at the 5′- and 3′ ends by unmatched terminal tT residues (Supplementary Fig. 9). Of the controls, the inactive X10–23 enzyme shows no reduction in GFP protein and mRNA template copy number, whereas the parent DNAzyme and antisense strand yield slightly elevated values of GFP mRNA and protein (Supplementary Fig. 9).

Intracellular reduction of endogenous KRAS.

Having demonstrated that X10–23 is capable of knocking down the expression of transiently transfected genes in cultured cells, we next asked whether similar effects could be achieved for endogenous mRNA transcripts. For this study, KRAS was chosen as a cellular target due to its implication in lung, pancreatic and colorectal adenocarcinomas^31–33. KRAS has been the focus of many drug-targeting campaigns and is often viewed as an ‘undruggable’ target due to the inherent difficulty of altering its cellular-expression profile³⁴. We designed, synthesized and tested two X10–23 reagents that targeted the first exon and 3′-UTR (Fig. 6a) of endogenous KRAS in cervical cancer (HeLa) and breast cancer (MDA-MB-231) cell lines. HeLa and MDA-MB-231 cells were either transfected with or without 4 μg of X10–23, and the KRAS mRNA levels were quantified by qRT-PCR after 48 hours of incubation post-transfection (Fig. 6b). Relative to the transfection control, both cell lines showed a >65% reduction of mRNA copy number for the X10–23 reagent that targeted the first exon (Fig. 6c,d). The X10–23 reagent that targeted 3′-UTR was slightly less effective, yielding a ~35–45% reduction in KRAS mRNA copy number (Fig. 6c,d). The differences in the RNA cleavage activity between the two X10–23 reagents are congruent with their in vitro activity observed with synthetic oligonucleotides (Supplementary Fig. 10), and probably reflect sequence-specific differences in the binding energetics of the two reagents. Overall, these data clearly demonstrate that X10–23 can be used to knockdown the expression of disease-causing proteins in human cells.

Fig. 6 |. Targeting endogenous oncogene KRAS by X10–23 in cancer cells.

a, X10–23 target sites located at the first exon and 3′-UTR regions of KRAS. The cleavage GU junctions are coloured red. b, Schematic representation of X10–23 molecule used for the inhibition of endogenous KRAS expressed by cancer cells. Cervical cancer cells (HeLa) and breast cancer cells (MDA-MB-231) were either transfected with individual X10–23 or with transfection carrier but no X10–23, and the KRAS mRNA copy number was quantified by qRT-PCR. c,d, qRT-PCR analysis of DNA-free total RNA extracted from HeLa cells (c) and from MDA-MB-231 cells (d) 48 h post-transfection using KRAS-specific and GAPDH loading control primers. Two biological replicates and three technical replicates were collected for each cell line, with one representative biological replicate shown. Error bars are ±s.d. of the mean for n = 3 technical replicates.

Mechanistic insights into cellular cleavage.

Previous studies have implicated RNase H as a contributor to RNA-based degradation by 10–23 variants due to the presence of complementary substrate binding arms that mimic antisense oligonucleotides³⁵. Recognizing the substantial difference in multiple-turnover activity between 10–23 and X10–23, we hypothesized that the newly engineered X10–23 reagent was sufficiently fast that it would be less susceptible to the effects of an RNase H-induced cleavage pathway. To investigate this possibility, we evaluated the catalytic activity of X10–23 under simulated physiological conditions in buffered solutions that either contained or lacked RNase H. X10–23 variants were designed to cleave segments of the GFP and KRAS transcripts that were prepared as synthetic oligonucleotides. Inactive versions of X10–23 and active versions that were non-complementary to the mRNA targets were used as negative controls. The X10–23 reagents showed a strong site-specific RNA cleavage activity in the presence and absence of RNase H (Extended Data Fig. 1a,d), which implicates Mg²⁺-dependent XNAzyme-catalysed RNA cleavage as the predominant mechanism of RNA degradation. Among the negative control sequences, the inactive X10–23 variant that targeted GFP yielded a banding pattern consistent with limited RNase H activity, whereas the equivalent KRAS targeting reagent showed no activity in the presence or absence of RNase H (Extended Data Fig. 1b,e). This observation may be attributed to differences in the hybridization efficiency of the two X10–23 reagents or, possibly, some unknown sequence-specificity preference of RNase H. As expected, the second X10–23 control with non-complementary binding arms but an active catalytic core failed to cut the RNA GFP and KRAS RNA substrates (Extended Data Fig. 1c,f). Similar results were obtained for in vivo GFP and KRAS gene-silencing experiments performed in mammalian cells that compared the active X10–23 reagent to the inactive or unpaired control reagents (Supplementary Figs. 9 and 11).

Discussion

DNAzymes are often thought of as ideal reagents for performing sequence-specific RNA cleavage reactions, as they are readily available by solid-phase chemical synthesis, amenable to directed evolution, programmable to any RNA target of interest and capable of achieving a high catalytic activity independent of accessory proteins. However, the transition from laboratory tools to clinical reagents has stalled due to their reduced activity in clinical trials. In an effort to overcome this problem, numerous chemical modifications have been evaluated with the goal to stabilize the backbone structure against nuclease digestion and increase the catalytic activity under reduced concentrations of divalent metal ions³⁶. In addition to modifications made to the sugar–phosphate backbone, researchers also pursued base-modified analogues with side chains that mimic RNase A using a Mg²⁺-independent RNA cleavage mechanism^37,38. However, despite substantial effort, these analogues have yet to effectively compete with conventional gene-silencing agents, which include antisense, short interfering RNA and CRISPR technologies³⁹.

The current study aimed to narrow the gap between DNAzymes and protein-based gene-silencing tools by expanding the chemical space of nucleic acid analogues used to construct nucleic acid enzymes. We focused our efforts on XNAs, which are artificial genetic polymers with novel sugar–phosphate backbones that harbour unique physiochemical properties relative to natural DNA and RNA^40,41. We hypothesized that appropriate positioning of XNA residues in the nucleic acid backbone structure of a highly evolved DNAzyme would lead to enhanced RNA cleavage activity under physiological conditions, and simultaneously protect the molecule from the harmful effects of nuclease digestion. This hypothesis is supported by limited structural data that show subtle, yet important, differences in the helical geometry of XNA duplexes⁴². Critical to our design was the need to identify XNA residues that would (1) increase the substrate-binding kinetics without sacrificing multiple-turnover activity, (2) improve the cofactor binding and (3) minimize the exolytic activity of biological enzymes. Using a medicinal chemistry approach that systematically probed the substrate binding arms and catalytic domain of the classic 10–23, we discovered a highly efficient and biologically stable variant that comprises three different classes of nucleic acid molecules (DNA, FANA and TNA). Relative to the parent enzyme, X10–23 achieves a ~50-fold increase in multiple-turnover activity under simulated physiological conditions and enhances the biological stability of the backbone structure >100-fold under stringent nuclease conditions. In cultured mammalian cells, X10–23 imbues a >60% reduction in mRNA and protein abundance under conditions of constitutive expression, which is further enhanced on treatment with a transcriptional inhibitor. Similar activity profiles were observed for X10–23 reagents that targeted endogenous KRAS expression in human cancer cell lines, which implies that X10–23 has the potential to alter the expression profiles of proteins that are thought to be ‘undruggable’. Finally, compelling evidence shows that X10–23 does not rely on RNase H as a mechanism for RNA degradation.

Given the enhanced activity of X10–23 relative to the parent enzyme, it will be interesting to see how X10–23 compares with conventional gene-silencing technologies in practical applications that relate to human health and disease. One area in which X10–23 could excel as a gene-silencing agent involves the validation and treatment of genetic diseases that are caused by allelic mutations in protein-coding genes⁴³. Protein-based gene-silencing technologies are constrained by the thermodynamics of oligonucleotide hybridization, which makes it difficult to distinguish mRNA targets that differ in sequence by a single nucleotide^44,45. However, DNAzymes, like 10–23, can unequivocally distinguish wild-type and mutant alleles in mRNA transcripts using catalytic domains that are highly tuned for specific dinucleotide pairs. In this way, X10–23 has the potential to advance the field of precision medicine by combining its acute substrate specificity with a level of cellular activity that appears to be unmatched by any DNAzyme developed to date.

In summary, our work establishes X10–23 as a new tool in the ever expanding toolbox of gene-silencing reagents. The ability for X10–23 to function with a high activity and biological stability in vitro and in cultured mammalian cells suggests that even highly evolved nucleic acid enzymes can be optimized for an improved activity. Based on these findings, we suggest that the exploration of new molecular chemotypes provides a powerful approach to create highly active nucleic acid enzymes with potential value as future therapeutic agents.

Online content

Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41557-021-00645-x.

Methods

Materials.

2′-FANA nucleoside triphosphates (faATP, faCTP, faGTP, faUTP) were obtained from Metkinen Chemistry. DNA, FANA and 5′-hexynyl phosphoramidites, as well as Universal Support II CPG columns, were purchased from Glen Research. TNA phosphoramidites were synthesized in-house following procedures reported previously⁴⁶. Oligonucleotides containing FANA and TNA were synthesized on an ABI3400 DNA synthesizer using chemical synthesis reagents purchased from Glen Research. DNA and RNA oligonucleotides were purchased from Integrated DNA Technologies). All the oligonucleotides were purified by denaturing PAGE and quantified by ultraviolet absorbance. YM-3 microcentrifugal concentrators were purchased from EMD Millipore. DMEM was purchased from ThermoFisher Scientific. Human serum and SVPDE were purchased from Sigma Aldrich. Human liver microsome was purchased from Sekisui XenoTech, LLC.

In vitro kinetic and biostability studies.

XNA-containing oligonucleotide synthesis and preparation.

Standard β-cyanoethyl phosphoramidite chemistry and an Applied Biosystems 3400 DNA Synthesizer were used to synthesize the TNA oligonucleotides on Universal Support II CPG columns in a 1 μmol scale. Standard DNA-coupling procedures were modified for FANA- and/or TNA-containing oligonucleotides such that the coupling time for FANA and TNA amidites was increased to 360 and 1,200 s, respectively. Detritylation was performed in two cycles for TNA amidites, 60 s each. The oligonucleotides used for biostability studies were coupled with a 5′-hexynyl phosphoramidite for later IR-680 fluorophore tagging via click chemistry. Cleavage from the solid support and final deprotection of synthesized oligonucleotides were achieved simultaneously in NH₄OH (33%) for 18 h at 55 °C. Oligonucleotides were purified on denaturing (8 M urea) PAGE, recovered by electro-elution and subsequently desalted by buffer exchange using microcentrifugal concentrators, and quantified by NanoDrop. All the oligonucleotides synthesized in-house were subjected to quadrupole time-of-flight mass spectrometry for identity confirmation.

Generation of RIMKLA mRNA transcript for in vitro cleavage assay by 10–23 and F10–23.

Homo sapiens ribosomal modification protein rimK-like family member A (RIMKLA) was reversed transcribed from HeLa total RNA using SuperScript RT III (Invitrogen-Life Technologies) according to the manual instruction. RIMKLA complementary (cDNA) was subjected to a two-round nested PCR using KOD polymerase (Fisher Scientific, catalogue no. 710863) to introduce a T7 promoter upstream of the coding sequence of RIMKLA for subsequent in vitro transcription. mRNA was then transcribed in a 1× RNAPol reaction buffer, supplemented with 0.5 mM each ATP, UTP, GTP and CTP, 5 mM dithiothreitol, 1 U μl⁻¹ of RNase inhibitor using 4 μg μl⁻¹ of the purified DNA amplicon and 25 U μl⁻¹ of T7 RNA polymerase at 37 °C for 16 h. Transcription was terminated by the addition of 10 U ml⁻¹ of RNase-free DNase I and incubation at 37 °C for 15 min. The transcription reaction was resolved by 10% denaturing purification PAGE (8 M urea), and the gel was visualized by ultraviolet shadowing. The RNA transcript was excised, electroeluted, exchanged into H₂O using an EMD Millipore YM-3 microcentrifugal device and ultraviolet quantified by NanoDrop before in vitro kinetic cleavage assays by 10–23 and F10–23.

Kinetic cleavage reactions for 10–23, F10–23, X10–23, OME10–23 and LNA10–23.

Single-turnover kinetic cleavage reactions were conducted in 50 mM Tris buffer (pH 7.5) that contained 150 mM NaCl, 1 mM MgCl₂, 0.5 μM substrate and 2.5 μM enzyme at 24 °C. Purified enzymes and substrates were annealed in 50 mM Tris buffer (pH 7.5) by heating for 5 min at 90 °C and cooling for 5 min on ice. Reactions were initiated by the addition of NaCl and MgCl₂ to the reaction. For the determination of the pseudo first-order rate constant, multiple time points were collected by quenching 1.5 μl of the reaction using 15 μl (10 equiv., v/v) of formamide stop buffer (99% deionized formamide, 25 mM EDTA) and cooling on ice. Samples were denatured for 15 min at 95 °C and analysed by 15% denaturing PAGE. Gels were visualized and quantified using a LI-COR Odyssey CLx. Values of $k_{\mathrm{o}\mathrm{b}\mathrm{s}}$ were calculated by fitting the percentage of substrate cleaved and reaction time (min) to the first-order decay, equation (1), using Prism 6 (GraphPad):

$$P_t=P_{\mathrm{\infty }}\left(1-{\mathrm{e}}^{-k_{\mathrm{o}\mathrm{b}\mathrm{s}}t}\right)$$ (1)

where $P_t$ is the percentage of cleaved substrate at time $t$ and $P_{\mathrm{\infty }}$ is the apparent reaction plateau.

For kinetic cleavage reactions under stoichiometric and multiple-turnover conditions, substrate concentrations were poised at 0.5 μM, and enzyme concentrations were adjusted to 0.5 μM and 50 nM, respectively.

Biostability measurement.

All biostability assays were performed in DMEM that contained 1 μM of tested construct with the presence of 2 mg ml⁻¹ human liver microsome, 50% human serum (v/v) or 10 mU ml⁻¹ SVPDE at 37 °C. Multiple time points were collected for each condition by quenching 1.5 μl of the reaction using 15 μl (10 equiv., v/v) of formamide that contained 25 mM EDTA. Samples were denatured for 15 min at 95 °C and analysed by 15% denaturing PAGE. Gels were visualized using a LI-COR Odyssey CLx.

In vitro RNase H activity assays with X10–23, unmatched X10–23 and inactive X10–23.

All RNase H activity assays were performed under simulated physiological buffer conditions in 50 mM Tris-HCl (pH 7.5) that contained 0.5 mM MgCl₂, 150 mM NaCl and 0.1 U μl⁻¹ of RNase H at 37 °C. 1 μM RNA substrate was mixed with 1 μM tested constructs of X10–23, unmatched X10–23 and inactive X10–23 in Tris-HCl (pH 7.5) buffer to anneal by heating for 5 min at 90 °C and cooling for 5 min on ice. Reactions were initiated by the addition of MgCl₂, NaCl and RNase H to the final concentration. Reactions were sampled by quenching 1.5 μl of the reactions using 15 μl (10 equiv., v/v) of formamide that contained 25 mM EDTA at time points of 0, 1, 5 and 20 h. Samples were denatured for 15 min at 95 °C and analysed by 15% denaturing PAGE. Gels were visualized using a LI-COR Odyssey CLx.

Intracellular GFP and KRAS reduction test.

Cell lines and mammalian cell culture conditions.

HEK and HeLa cells were cultured in DMEM (Corning, catalogue no. 10–017-CM) supplemented with 10% FBS, 1% (1 mg ml⁻¹) penicillin and streptomycin and grown at 37 °C, 5% CO₂. MDA-MB-231 cells were cultured in the same medium as the HEK and HeLa cells but supplemented with the additional component of 1 mM sodium pyruvate.

Transfection.

For titrating the amount of single or multivalent X10–23 experiments.

After 48 h of seeding of 2.5 × 10⁵ cells per well, HEK in 6-well plates were transfected with 1 μg of pCDNA3.3-EGFP only (negative control) or with 1 μg of pCDNA3.3-EGFP and 4, 6 or 8 μg of either internal or 3′-UTR GFP X10–23 in single experiments or 2/2 μg, 3/3 μg or 4/4 μg of both internal/3′-UTR GFP X10–23 in multivalent (dual X10–23) experiments by using JetPrime Transfection reagent (Polyplus Transfection) according to the manufacturer’s instructions except that we used 5× the recommended volume of JetPrime Reagent. For negative controls, the volume of JetPrime Reagent used for each well was the same as those for X10–23 to ensure the same transfection conditions in the control and experimental samples.

For comparison of active, inactive core and active unpaired X10–23.

After 48 h of seeding at 2.5 × 10⁵ cells per well (six-well plate), HeLa cells were transfected with 5.9 μg of X10–23 variants that targeted the 3′-UTR region of the KRAS transcript (active versus inactive core) or GFP transcript (active core but unpaired binding arms) using the JetPrime Transfection reagent. At 96 h post-transfection, cells were harvested and subjected to total RNA extraction and subsequently underwent DNAse treatment as described in the RNA isolation section. DNA-free RNA was subjected to qRT-PCR as described in the reverse transcription and SYBR Green qPCR analysis section.

For multivalent benchmark experiments.

After 48 h of seeding at 2.5 × 10⁵ cells per well, HEK cells in six-well plates were transfected with 1 μg of pCDNA3.3-EGFP only (negative control) or with 1 μg of pCDNA3.3-EGFP and 4/4 μg of both internal/3′-UTR GFP X10–23 (dual X10–23), DNA10–23 (dual DNA10–23), antisense (dual antisense oligos), inactive X10–23 (dual inactive X10–23) or inactive DNA10–23 (dual inactive DNA10–23). The parameters used in the subsequent imaging and RNA isolation at 48 h post-transfection were the same as those described in the RNA isolation section.

For single or multivalent KRAS X10–23 experiments in HeLa or MDA-MB-231 cells.

After 48 h or 76 h of seeding of 2.5 × 10⁵ cells per well of HeLa or MDA-MB-231 cells in six-well plates, respectively, the cells were transfected with transfection carrier only (negative control) or with 4/4 μg of both internal/3′-UTR KRAS X10–23 in multivalent (dual X10–23) or 8 μg of either internal or 3′-UTR KRAS X10–23 experiments using the JetPrime Transfection reagent (Polyplus Transfection). The parameters used in RNA isolation at 48 h post-transfection were the same as those described in the RNA isolation section.

Cell imaging.

At 24 h or 48 h post-transfection, cells were subjected to live imaging using a 200M Axiovert Zeiss fluorescent microscope with a ×10 objective and GFP filter. After imaging, the cells were subjected to RNA extraction.

Reverse transcription.

DNA-free RNA (2 μg) was subjected to cDNA synthesis using the SuperScript III First-Strand Synthesis System (Invitrogen–Life Technologies) with random hexamer primers in a 20 μl reaction according to the manufacturer’s instructions. cDNA was subsequently purified using DNA Clean & Concentration columns from Zymo Research (catalogue no. D4003) according to the manufacturer instructions and eluted twice with 100 μl of water each time.

SYBR Green semiquantitative PCR analysis.

To quantify the copy number of the GFP transcript in the presence or absence of GFP-X10–23, cDNA was subjected to qPCR analysis using iQ(tm) SYBR(R) Green Supermix (BioRad, catalogue no. 1708880) on a BioRad CFX real-time PCR system. In each qPCR run, a known concentration of DNA standards at five serial dilutions was used to establish the standard curve and calculation of the starting quantity (SQ) of the target transcripts. Specific primers to interrogate EGFP, KRAS and GAPDH (loading control) transcripts as well as for the qPCR standards are listed in Supplementary Table 2. For each experimental sample, three replicates were performed, and each serial diluted standard was assayed in duplicate. The relative mRNA copy number of the target transcripts was calculated by multiplying each individual SQ to a corresponding scaling factor derived from the loading control GAPDH SQ. By dividing the median of the GAPDH SQ in a qPCR run to an individual GAPDH SQ, a scaling factor for that particular sample was generated. The fold reduction was calculated using 1/2^ΔΔCt.

RNA polymerase inhibitor (ActD) treatment.

In the titration of ActD concentration experiment, cultures of HEK cells in six-well plates transfected with either 1 μg of pCDNA3.3-EGFP only (negative control) or with 1 μg of pCDNA3.3-EGFP and 4 μg of internal GFP X10–23 were treated with ActD at final concentration of 0, 10, 20 and 40 μM at 20 h post-transfection and harvested 4 h later (at 24 h post-transfection). Treated cells were harvested and subjected to total RNA isolation and subsequent analyses. In the benchmark experiment, 40 μM ActD was added to HEK cells transfected with dual 4 μg internal and 4 μg 3′-UTR of X10–23, 10–23, inactive X10–23 or antisense at 44 h post-transfection and incubated at 37 °C for an additional 4 h prior to harvesting time at 48 h post-transfection. The treated cells were imaged and subjected to total RNA isolation and subsequent analyses.

RNA isolation.

To each well of the six-well plate of cells (HEK, HeLa or MDA-MB-231), the total RNA isolation was performed using 1 ml per well of Trizol Reagent (Invitrogen) according to the manufacturer’s instructions. Total RNA was treated with Turbo DNAse (20 U per reaction) at 37 °C for 30 min on a shaker, and followed by purification using an equal volume of Phenol-Chloroform, pH 4.5 (Thermo Fisher, Ambion catalogue no. AM9720). The aqueous layer was transferred to a new tube and precipitated with a one-tenth volume of 5 M NaCl and one volume of isopropanol at −20 °C overnight. The precipitated RNA was pelleted at 4 °C and 15,000 r.p.m. using bench-top centrifuge, followed by two washes with cold (−20 °C) 70% ethanol.

Extended Data

Extended Data Fig. 1 |. Mechanistic analysis of X10–23.

Representative gels showing RNA cleavage activity in the presence and absence of RNase H for an internal segment of GFP (a-c) and the first exon segment of KRAS RNA (d-f). Color code: RNA (red), DNA (black), FANA (orange), and TNA (blue). (a,d) X10–23 with an active catalytic core. (b,e) X10–23 with an inactive catalytic core. (c,f) X10–23 with an active catalytic core that does not hybridize to the RNA target. All assays were performed in buffer containing 0.5 mM MgCl₂ and 150 mM NaCl at 37 °C (pH 7.5) with 1 μM substrate and 1 μM enzyme. Nuclease reactions included 0.1 unit/μL of RNase H. S: full-length substrate, P: 5’ cleavage product. Molecular weight markers indicated to the right of the gel.

Supplementary Material

Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41557-021-00645-x.

Data availability

The authors declare that the data supporting the findings of this study are available within the article and its Supplementary Information files. Source data are provided with this paper.