|Home | About | Journals | Submit | Contact Us | Français|
Signal amplification is a key component of molecular detection. Enzyme-free signal amplification is especially appealing for the development of low-cost, point-of-care diagnostics. It has been previously shown that enzyme-free DNA circuits with signal-amplification capacity can be designed using a mechanism called ‘catalyzed hairpin assembly’. However, it is unclear whether the efficiency and modularity of such circuits is suitable for multiple analytical applications. We have therefore designed and characterized a simplified DNA circuit based on catalyzed hairpin assembly, and applied it to multiple different analytical formats, including fluorescent, colorimetric, and electrochemical and signaling. By optimizing the design of previous hairpin-based catalytic assemblies we found that our circuit has almost zero background and a high catalytic efficiency, with a kcat value above 1min−1. The inherent modularity of the circuit allowed us to readily adapt our circuit to detect both RNA and small molecule analytes. Overall, these data demonstrate that catalyzed hairpin assembly is suitable for analyte detection and signal amplification in a ‘plug-and-play’ fashion.
Nucleic acids are frequently used to amplify signals during the detection of both nucleic acids and non-nucleic acid analytes including proteins and small molecules. In applications such as immuno-PCR (1,2), target-dependent rolling circle amplification (3,4), and the proximity ligation assay (5,6), nucleic acids frequently act as captured or created templates for amplification. It is also possible to rely on ribozymes, rather than protein enzymes, for amplification. For example, RNA-cleaving deoxyribozymes (7,8) and RNA-ligating ribozymes (9,10) have been used to construct autocatalytic or cross-catalytic circuits for exponential signal amplification in the detection of nucleic acid and/or small molecule analytes. While nucleic acid-alone amplification schemes have the advantage of operating under conditions that might otherwise inhibit proteins, they currently have the disadvantage of requiring a great deal of engineering for adaptation to any given analyte. For example, directed evolution of the Bartel Class I ligase was necessary in order to adapt it to detecting a particular mRNA sequence (11).
Recent advances in the field of molecular programming (12–15) have yielded DNA circuits in which the simple rules governing nucleic acid hybridization can be adapted to signal-amplification. The hybridization chain reaction (HCR) (16,17), entropy-driven catalysis (18), and catalyzed hairpin assembly (19) rely only on hybridization and strand-exchange reactions in order to achieve amplification.
Some hybridization-based, nucleic-acid-alone circuits have been adapted to analytical applications. For example, the Pierce group has successfully used the hybridization chain reaction for multiplexed imaging of endogenous mRNA in fixed zebrafish embryos (17). The application of HCR and entropy-driven catalysis in assisting in vitro nucleic acid detection has also been reported (20–22). Building on these results, we believe that there now exists an even greater potential to adapt catalyzed hairpin assembly (19) to support even more robust analytical applications.
A key feature of nucleic acid circuits for molecular amplification is that they do not require perishable protein enzymes. Although some enzymes can be stored for a relatively long period of time in a dry or semi-dry state, all oligonucleotides are amenable to such storage. Moreover, nucleic acid circuits are inherently modular and scalable, requiring only the design of base-pairing between strands. However, within this broad context there are several practical analytical features that should be achieved. First, the uncatalyzed reaction must be sufficiently slow to ensure low background. Second, the turnover of substrates must be fast enough so that a high level of signal amplification can be achieved in a relatively short period of time. Third, it should be possible to adapt the catalyst to detect various analytes. And finally, the product of the catalysis should be easily detectable by common detection modalities, such as fluorescence, colorimetric, or electrochemical detection.
In this work, we designed a simplified DNA circuit based on catalyzed hairpin assembly. Kinetic characterization of this circuit showed that the uncatalyzed background assembly was undetectable (<0.5M−1s−1). This result is particularly noteworthy since uncatalyzed background reactions have been a major challenge in the design of DNA circuits for analytical applications. In particular, the original hairpin assembly circuit (19) had an uncatalyzed rate constant as high as ~100M−1s−1. In addition, our circuit was very efficient and steady-state analysis showed that the turnover rate of the catalyst was >1min−1, yielding 50- to 100-fold signal amplification within a few hours. Finally, the modularity of the DNA circuit allowed its adaptation to detect various analytes. For example, we engineered a molecular beacon (23,24) to act as a signal transducer for virtually any nucleic acid sequence, and showed that similar aptamer beacon (25–27) transducers allowed the detection of non-nucleic acid analytes. The circuit’s modularity also allowed the output to be readily switched between fluorescence, electrochemical (28,29), or colorimetric readouts (30–33). Overall, these data suggest a paradigm for circuit-based molecular detection in which target recognition, signal processing, amplification, and transduction can be integrated via modular components.
All chemicals were of analytical grade and were purchased from Sigma-Aldrich (MO, USA) unless otherwise indicated. The methylene-blue labeled oligonucleotide (mH1H2-MB) was ordered from Biosearch Technologies (Novato, CA, USA). All other nucleotides were ordered from Integrated DNA Technology (IDT, Coralville, IA, USA). Oligonucleotide sequences are summarized in Supplementary Table S1. H1-E was prepared by ligating 10nmol of H1.a and 10nmol of mH1H2-MB with 16 U of T4 DNA ligase (Invitrogen, Carlsbad, CA, USA) in a 160-µl reaction in 1× T4 ligase buffer (Invitrogen) at room temperature overnight. H1, H1-E, H2, hpC1, siEGFPAS, AptC1, and AptInh were purified via denaturing PAGE (7M Urea, 1× TBE). C1, C1mut, Dz, DzInh and mH1H2 were simply ethanol-precipitated to remove salts that might interfere with 260nm absorption readings. All DNA strands were stored in 1× TE (pH 7.5) at 1–100µM concentrations. Diluted C1, mutC1, and siEGFPAS stocks (1 pM to 100nM) were stored in 1× TE (pH 7.5) supplemented with 1µM (dT)21 to prevent loss due to adsorption to plastic. Nucleosides were purchased from Alfa Aesar (Ward Hill, MA, USA).
In Figure 3c of (19), the trace ‘A+B’ showed that when 20nM hairpin A and 20nM hairpin B were mixed in the absence of catalyst, 3.5% of A (0.7nM) was hybridized with B after 5h of reaction. Assuming linearity of the reaction, this corresponds to a rate of (0.7/5=) 0.14nM h−1. The second order rate constant can be estimated to be (0.14/20/20=) 3.5 × 10−4nM−1h−1, or 97M−1s−1. Although the accuracy of this estimation might be affected by a mixed population of both fast-reacting and slow-reacting hairpins (see Discussion section), we believe it is roughly accurate.
An amount of 10µM stock of RepF:RepQ complex was prepared by annealing 10µM RepF and 20µM RepQ in 1× TNaK buffer (20mM Tris, pH 7.5; 140mM NaCl; 5mM KCl). An excess of RepQ ensures efficient quenching of RepF but does not interfere with the readout of H1:H2. All kinetic measurements were carried out at 37°C. Immediately before experiments, H1 and H2 were separately refolded in 1× TNaK buffer. This and other refolding reactions involved heating to 90°C for 1min followed by slowly decreasing the temperature to 23°C at a rate of 0.1°C s−1. All reagents were prepared in 1× TNaK buffer and were pre-warmed to 37° C for 15–30min before mixing. The reactions were started by the addition of H1. Reaction mixtures (18µl aliquots) were added to different wells of a 384-well plate, which was immediately transferred to a TECAN Safire plate reader for fluorescence measurements. With our setting, roughly 67 Raw Fluorescence Units correspond to 1nM unquenched RepF, which in turn corresponds to 1nM H1:H2.
For the detection of siEGFPAS, different concentrations of siEGFPAS were annealed with a constant concentration of hpC1 (final concentration 5nM) before the addition of other components of the circuit. For the detection of adenosine, AptC1 and AptInh were annealed at 8× final concentration in 1× TNaK, and then mixed with equal volumes of adenosine (or other nucleosides) at 4× final concentration in 1× TNaK for 2h at 37°C. Some 5µl of this mixture was mixed with H1, H2 and RepF:RepQ in a total volume of 20µl (each component at 1× final concentration).
The Au electrode (1.2mm in diameter) was polished with 1.0, 0.3 and 0.05μm α-Al2O3; washed ultrasonically with pure water three times; and then electrochemically cleaned in 0.1M H2SO4 by potential scanning between 0 and 1.6V until a reproducible cyclic voltammogram was obtained. The electrode was then sonicated and rinsed with copious amount of pure water, and blown dry with nitrogen.
Before use, the thiol-labeled oligonucleotide S was treated with 6.7mM Tris(2-chloroethyl) phosphate (TCEP) in DP buffer (20mM Tris–HCl, pH 7.4; 100mM NaCl; 5mM MgCl2) for 2h at 28°C. The Au-S conjugation was performed by placing 20μl of the 2μM S in DP buffer on the Au electrode. The electrode was capped with a 1.5ml Eppendorf tube to protect the solution from evaporation. After 16h of conjugation at room temperature, the electrode was rinsed with pure water several times. Then the S-conjugated Au electrode was treated with 1mM 6-mercapto-1-hexanol (MCH) in DP buffer and kept at room temperature for 1h, followed by rinsing with pure water.
The catalyzed hairpin assembly reaction was allowed to proceed for 1h at 37°C. Reaction mixtures (or mH1H2-E solution; 15µl) were placed on the S/MCH-modified Au electrode, and incubated at 37°C for another 1.5h. Electrodes were then rinsed with DP buffer at 37°C and square wave voltammetry (SWV) was used to detect the MB near the surface of the electrode. SWV was performed on a DY-2000 Series Multichannel Potentiostat (Digi-Ivy, Inc. Austin, TX, USA) using a conventional three-electrode electrochemical cell with an Ag-AgCl (1M KCl) electrode (CH Instruments, Austin, TX, USA) as the reference electrode, a Pt coil as the counter electrode, and an Au disk (1.2mm in diameter) as the working electrode. All the electrochemical measurements were carried out at room temperature. SWV measurements were performed in DP buffer under an oscillatory potential between −0.5 and 0V, with a frequency of 50Hz.
An 8µM stock of Dz:DzInh complex was prepared by annealing 8µM Dz and 16µM DzInh. An excess of DzInh ensures efficient inhibition of Dz but does not interfere with the readout of H1:H2.
The catalyzed hairpin assembly reactions were set up as was described for fluorescence measurements, except RepF:RepQ was not included. In these assembly reactions, the concentrations of H1 and H2 were 600 and 900nM, respectively. The assembly reactions were incubated at 37°C for 4h. Then 35µl of the assembly reaction was mixed with 2.5µl of 8µM Dz:DzInh complex and 2.5µl of 20µM hemin (in DP buffer supplemented with 0.1% Triton-X100) and incubated for another 2h at 37°C. To develop color, 36µl of this mixture was mixed with 4µl of substrate solution (38mM ABTS and 20mM H2O2). The 20µl final mixtures were transferred to different wells of a 384-well plate. The absorption at 410nm was measured using a Synergy-HT plate reader (Bio-TEK, Winooski, VT, USA). After ~15min of data collection the plate was removed from the plate reader and a picture of the plate was also taken.
The general principle of catalyzed hairpin assembly was introduced in the seminal work by Pierce, Yin, and co-workers (19). One implication of this work is that for any unstructured ssDNA, a pair of hairpins can be designed so that the two hairpins do not initially interact with each other but can catalytically form a duplex in the presence of a ssDNA input. We designed a simple amplification circuit based on this model, shown in Figure 1. As in earlier approaches, we describe the sequence of DNA molecules in terms of numbered domains, each of which represents a short fragment (usually <12nt) of DNA sequence. Complementarity between numbered domains is denoted by an asterisk. Hairpins H1 and H2 can potentially hybridize to form a H1:H2 duplex, since H1 contains a segment that is complementary to a segment of hairpin H2 (see the H1:H2 duplex in Figure 1). However the spontaneous hybridization of the two hairpins is kinetically hindered by occluding complementary regions within intramolecular hairpin secondary structures. Domain 1* of an unstructured DNA input, C1, serves as a ‘toehold’ to initiate interactions with domain 1 of H1. Branch migration then opens hairpin H1 and forms a C1:H1 intermediate (Figure 1, reaction a). In the C1:H1 intermediate, domain 3* of H1 is no longer occluded and can bind to domain 3 of H2, again initiating a branch migration reaction to form a H1:H2:C1 complex (Figure 1, reaction b). This complex is inherently unstable, and C1 dissociates from the H1:H2 complex, completing the reaction and allowing C1 to act as a catalyst to trigger the hybridization of additional pairs of H1 and H2 hairpins. The overall reaction is at least partially driven by the enthalpy decrease resulting from the formation of more base pairs.
While the original circuit (19) was viable, it was not suitable for analytical applications, because of high background reactivity in the absence of input. In order to improve the circuit, we attempted to optimize the design of the individual components at two levels: domain organization and sequence.
We first simplified the hairpin designs by removing unnecessary domains. In the original circuits (19), the hairpin substrates usually contained from 10 to 13 domains. Among these domains, only four to five were necessary to dictate the progression of reaction, while the rest were ‘clamp’ domains designed to reduce uncatalyzed (background) reactions. Paradoxically, a fast, uncatalyzed reaction (~100M−1s−1) was still observed for the original circuit, which led us to question the effectiveness of these ‘clamp’ domains. Therefore, we removed these domains and only kept those domains that were necessary to dictate the progression of the reaction. Our domain organization also differed from the original circuit (19). For example, the loop of H1 (domain 4) does not contain sequences that are complementary to the overhang (domain 1), which might help prevent unwanted H1 dimerization.
The lengths of the domains were chosen based on kinetic and thermodynamic considerations. In general, toehold binding should be strong enough to efficiently initiate strand displacement (34) and weak enough to spontaneously dissociate and regenerate the catalyst. Binding should also be poised to prevent unwanted DNA hybridization (e.g. a stable interaction between domain 3* of C1 and domain 3 of H2). By estimating the stability of DNA hybridization in the chosen buffer (see ‘Materials and Methods’ and Discussion sections), we eventually chose 8nt as the length the of domains 1–3, 5 and 6. Since H2 has two mutually exclusive foldings (Figure 2a), we chose 11nt as the length of domain 4 so that the correct folding would be greatly favored.
The sequence design process is summarized in Figure 2b. The sequence of C1 was chosen within the constraints of avoiding secondary structure and maintaining a GC content of 50% for each domain. This fixed the sequences of domains 1, 2 and 3, and left domains 4, 5 and 6 to be designed. In order to avoid alternative foldings, the NUPACK package (35) was used to design the sequence of domains 4*, 5* and 6* in H1, which in turn defined the sequence of H2. The expected correct folding of H2 was also verified using NUPACK.
To monitor the assembly of H1 and H2 in real time, we designed a fluorescent reporter by hybridizing a FAM-labeled strand (RepF) and an Iowa Black FQ-labeled strand (RepQ; Figure 3a). Hybridization of these two strands did indeed result in efficient quenching. The domain 2* of H1 in the H1:H2 duplex can bind the domain 2 of RepF in the RepF:RepQ complex and initiate the branch migration reaction to displace RepQ, leading to an increase in fluorescent signal. Since the domain 2* of H1 alone is occluded, H1 alone should not interact with the RepF:RepQ complex.
Using these fluorescent reporter constructs, it can be observed that in the presence 5nM C1, 50–200nM H1, and 200nM H2, the two hairpins rapidly form a duplex (Figure 3b). The reactions approached completion within 2h. In contrast, in the absence of C1 no H1:H2 hybridization could be observed (Figure 3b, black line) even with prolonged incubation (up to 15h, data not shown). Considering the sensitivity of this assay (2nM H1:H2 should be readily detectable), the second-order rate constant of uncatalyzed reaction can be estimated to be <0.5M−1s−1, which is at least 200-fold slower than the original circuit (19).
To gain a more quantitative understanding of the efficiency of catalysis, we systematically varied the concentration of H1 and H2 and recorded the initial reaction rates v (Figure 4a). By fitting the dependence of initial rates on substrate concentrations (Figure 4b) with the equation:
kcat was found to be 1.6±0.3min−1.
Although other constants in the equation could not be accurately determined, it can be semi-quantitatively estimated that in the presence of 50–400nM H1 and 100–400nM H2, the observed turnover rate (v/[C1]) was with the range of 0.3–0.9min−1 (Figure 4b). Such a turnover rate implicates that 20–50 molecules of product (H1:H2 complex) can be produced per molecule of catalyst (C1) per hour. In other words, the signal can potentially be amplified by 20- to 50-fold per hour, suggesting that non-enzymatic DNA circuitry may be useful for sequence detection in many analytical applications.
In order to demonstrate the utility of this amplification method for analytical applications, we first attempted to adapt the system to multiple inputs. The system comprising H1, H2 and RepF:RepR can be thought as a DNA circuit with signal-amplification capacity to detect the presence of C1. Considering the extremely slow uncatalyzed reaction, the appreciable turnover rate of C1, and the ability to precisely measure the rates of catalysis enabled by real-time fluorescence readouts, we expected that very low concentrations of C1 could be detected using this circuit (Figure 4c). In fact, the presence of as low as 5 pM C1 could be unambiguously detected in our assays containing 50nM H1, 400nM H2 and 50nM RepF:RepR complex within just a few hours (data not shown).
The detection of C1 by the circuit should be sequence-specific, since a mismatch in toehold or branch migration regions should in general impede toehold-mediated strand displacement by at least an order of magnitude (34,36,37). As expected, mutC1, which differs from C1 by just 1nt, is 10-fold less active than C1 (Figure 4d).
The modularity of the design process (Figure 2) suggested that it should be possible to design a DNA circuit to detect virtually any single-stranded DNA input. However, redesigning the entire circuit each time is unnecessarily arduous. In addition, the amplification mechanism requires that the analyte DNA (in this case C1) essentially be structure-free, which poses a challenge for the detection of natural nucleic acids. Therefore, we wished to show that we could simply reuse the working circuit to detect various inputs, including structured nucleic acids, through the use of a transducer. We designed a molecular beacon-like molecule that could bind to a new nucleic acid sequence and in so doing create a toehold in the transducer that would mimic the catalytic activity of C1 (Figure 5a). The signal transducer (named hpC1) was generated by extending the 3′-end of C1 to form a hairpin loop where domain 1* of hpC1 is occluded by domain 1. Therefore hpC1 should be catalytically inactive. The loop of hpC1 is complementary to the target nucleic acid. As in a molecular beacon (23,24), when the target nucleic acid is present, the loop should hybridize with the target, forming a rigid duplex and disrupting the stem formed by domain 1 and 1*. Even structured nucleic acid targets can be annealed to the loop of hpC1. The exposed domain 1* of hpC1 thereby serves as a toehold for initiating the amplification reaction. Similar molecular beacon-like transducers have also been used in the design of allosteric ribozymes (38,39) and deoxyribozymes (13,40).
Initially we chose the antisense strand of a siRNA directed against enhanced green fluorescent protein (named siEGFPAS) as the nucleic-acid input. This RNA molecule is predicted [by mfold (41)] to fold into a hairpin structure with stability of −2.9kcal/mol. As shown in Figure 5b, while 5nM of hpC1 alone showed extremely low catalytic activity toward 50nM H1 and 400nM H2, the presence of the appropriate target RNA molecule substantially activated hpC1 in a dose-dependent manner. In the presence of 5nM siEGFPAS, hpC1 was activated by ~400-fold. The low background activity of hpC1 and the precise measurement of rates of catalysis allowed the detection of as little as 20 pM of siEGFPAS (data not shown).
In principle, non-nucleic acid could also be employed as inputs, similar to the design of allosteric ribozymes and deoxyribozymes (26,42). In particular, Wieland et al. (43) have recently shown that through the use of structure-switching aptamers, catalyzed hairpin assembly can be controlled by small molecule inputs, although these authors did not quantify the extent of amplification. We wished to similarly show that aptamers could be used in conjugation with our circuit for the detection of non-nucleic-acid analytes. To this end, we created an aptamer beacon in which an anti-adenosine aptamer (44), AptC1, was inhibited by a complementary strand, AptInh (Figure 6a). AptC1 also contained the necessary information (C1) to initiate the amplification reaction, and AptInh was also complementary to part of C1. When AptC1 is bound by AptInh it can neither bind its ligand nor catalyze the hybridization of H1 and H2. In contrast, free AptC1 can do both. As a result, when the ligand (in this case, adenosine) binds to and stabilizes AptC1 it shifts the equilibrium such that the duplex with AptInh is destabilized and the single-stranded toehold is available to activate the amplification reaction. We first observed that the 51-nt long AptC1 is an efficient catalyst (Figure 6b, sample ‘No AptInh’), which is consistent with previous findings (45) that appending sequences to the catalyst of an enzyme-free DNA circuit does not qualitatively affect the efficiency of the catalyst, as long as the appended sequences do not form secondary structures that block the catalyst region. Next, as shown in Figure 6b, in the absence of adenosine, a combination of 13nM AptC1 and 42nM AptInh exhibited relatively low activity, whereas the presence of adenosine activated AptC1 in a dose-dependent manner. For example, in the presence of 110, 220 and 560µM adenosine the catalytic activity of AptC1 was activated by 1.2-, 1.5- and 4.8-fold, respectively, whereas 560µM guanosine, cytidine or uridine did not activate AptC1. The sensitivity and signal-to-background ratio can be systematically tuned by adjusting the length, sequence, and concentration of AptC1 (27), in accordance with the theoretical framework we developed previously (42).
An amplifier circuit would be most useful if it could be shown to be compatible with multiple analytical readouts. Having already shown compatibility with fluorescence sensing, we next asked whether the circuit could be readily adapted to an electrochemical readout. Domain 6* from H1 was truncated, and the 3′-end of the truncated H1 was then labeled with a methylene blue (MB) (28,29) moiety to form H1-E (Figure 7a). At the same time, a 16-nt capture oligonucleotide (named S) that was complementary to domains 2*–5* (red in Figure 7a) of H1-E was conjugated to a gold electrode. The rationale was that H1-E can only transiently bind S (via 8bp), whereas a H1-E:H2 complex could stably bind S (via 16bp). Once H1-E:H2 was more strongly bound to S, the MB moiety should be brought closer to the electrode over a great period of time, and this change could be detected via electrochemical methods such as square wave voltammetry (SWV).
We first proofed pieces of the electrochemical readout platform. The difference between 8-bp transient binding and 16-bp stable binding was first verified using fluorescent probes (Supplementary Figure S1). It should be noted that the absolute limit-of-detection (LOD) is highly dependent on assay conditions and devices. Therefore, direct comparisons of LODs between different experimental set-ups can be very misleading, especially when the purpose is to evaluate the amplifier. To estimate the potential level of signal amplification, we used a short MB-labeled oligonucleotide (mH1H2-MB, inset of Figure 7a) to mimic the amplification product H1-E:H2.
To test the performance of the amplifier, 200nM H1-E and 300nM H2 were mixed for 1h in the presence of different concentration of C1. Then the reaction mixtures (15μl) were added to the electrode modified with strand S (see Supplementary Figure S2 for electrode characterization) and incubated for another 1.5h before SWV measurement. In a parallel experiment, 15μl aliquots of mH1H2-MB at various concentrations were added to the electrode. As shown in Figure 7b, 1nM mH1H2-MB did not elicit an observable current due to MB oxidation, whereas 1nM of C1, through the DNA circuit, elicited an obvious MB oxidation peak at ~–0.28V (versus Ag/AgCl, 1M KCl). This signal was nearly half of the signal generated by 200nM of mH1H2-MB (non-saturating, Supplementary Figure S3), which suggests a roughly 100-fold amplification by the amplifier circuit over the course of 2.5h. The peak height and area were both positively correlated with the concentration of C1, and as low as 10pM C1 could be detected using this method (Figure 7c).
As a second demonstration of the adaptability of the amplifier circuit, we attempted to facilitate the colorimetric detection of nucleic acid inputs, an urgent need for equipment-free, point-of-care diagnostics (46,47). Unfortunately, such methods usually suffer from low sensitivity, due in part to the lack of adequate signal amplification. While some signal amplification can be obtained via enzymatic turnover [such as the deoxyribozyme/ABTS reporter system (30–33)] it should be possible to further couple enzymatic turnover to the amplifier circuit, providing an additional level of signal generation. The quadruplex deoxyribozyme peroxidase that had previously been utilized was fused to domains 6-5-2. This reporter, Dz, was then inactivated by a complementary inhibitor strand, DzInh (Figure 8a). This approach is similar to that previously used for adaptation of the anti-adenosine aptamer to the input portion of the amplifier circuit, demonstrating the general ease with which these circuits can be modularly engineered. The H1:H2 complex can displace DzInh from Dz through toehold-mediated strand displacement and this will in turn allow the DNAzyme to fold into its active conformation.
To test this design, we incubated 600nM of H1 and 900nM of H2 with various concentrations of C1 for 4h followed by the addition of Dz:DzInh complex, hemin, ABTS, and H2O2. Again, we used a short oligonucleotide (mH1H2, inset of Figure 8a) to mimic the amplification product H1:H2, and different concentrations of mH1H2 were added to the Dz:DzInh complex, hemin, ABTS and H2O2. As shown in Supplementary Figure S4 and Figure 8b, when peroxidase activity was assayed on a spectrophotometer, concentrations of mH1H2 <50nM did not elicit detectable signal, whereas 1nM of C1, through the circuit, elicited discernable signal, indicating a roughly 50-fold amplification of the input. Peroxidase activity was positively correlated with the concentration of C1. Moreover, 1–5nM C1 produced sufficient peroxidase activity to cause a marked, visible color change that could be readily detected by eye (Figure 8c). In contrast, even 50nM mH1H2 did not yield such a visible color change.
DNA circuits provide interesting opportunities for diagnostic applications. However, in order to function in assays the circuits must be engineered to take into account a number of practical considerations. While HCR and entropy-driven catalysis have begun to be characterized (45) and used in molecular detection applications (20,21), little was previously known about the efficiency, robustness, and application potential of catalyzed hairpin assembly.
Fortunately, since catalyzed hairpin assembly is based on DNA hybridization, secondary structure formation, and branch migration, a basic understanding of these processes also allows us to rationally design effective circuits. For example, since we envisioned executing the circuit as part of a point-of-care detection system where nuclease contamination is likely to be present, we decided not to include Mg2+ in the reaction buffer, as most nucleases required magnesium or another divalent ion for function. In consequence, DNA hybridization was weaker in our buffer than in buffers typically used in previous DNA circuits [(15,18,34,45,48); reactions at ~10mM Mg2+]. For this reason, we chose 8nt as the length of the toehold, longer than toeholds previously used (typically 6–7nt). The absence of Mg2+ should also disfavor non-specific reactions, which might have contributed to the more than 200-fold lower background reactivity of our circuit.
The elimination of ‘clamp’ domains greatly simplified the design process (Figure 2), since the overall length of designed sequences was reduced from 75–100nt to 40–60nt (even as the toehold length was increased to 8nt). Moreover, oligonucleotides with such lengths can be faithfully synthesized as one piece, eliminating the necessity of preparing hairpins by enzymatic ligation of two shorter oligonucleotides (19). Simple denaturing PAGE purification of directly synthesized hairpins proved sufficient to achieve near-zero background with our designs. This modest methodological improvement will greatly reduce the turnaround time for designing and testing DNA circuits, and thus should begin to make these technologies much more widely accessible.
In our circuit, the kcat of C1 is >1min−1, making it one of the fastest known DNA catalysts. The kcat values of the widely used trans-acting DNAzymes 10–23, 8–17 and E6 are 3min−1 (49), 0.6min−1 (50) and 0.04min−1 (51), respectively. Although this comparison is purely numerical rather than mechanistic, the rate of reaction is the primary consideration when a catalyst is to be used as a signal-amplifier (52). Thus, catalyzed hairpin assembly is in fact an excellent contender when a catalytic, DNA-based amplification is being considered.
The relatively fast reaction yielded ~100-fold signal amplification within just a few hours, a value that moves enzyme-free, colorimetric detection methods much closer to point-of-care applications. As an example, without using spectroscopic equipments or enzyme-based signal amplification (31–33), the well-developed peroxidase DNAzyme/ABTS reporter usually requires sub-micromolar concentrations of analyte to achieve visible signals (53). The signal amplification provided by the DNA circuit allowed us to readily detect target nucleic acid in low nanomolar range (Figure 8c), marking a two orders of magnitude improvement in sensitivity. Further kinetic characterizations, especially using pre-steady state methods, may reveal the rate-limiting step and thereby direct the development of even faster DNA circuits.
The programmability of DNA allows the product of one amplification reaction to act as the template or activator of another amplification reaction in order to achieve higher-level signal amplification. This principle has been extensively exploited in enzyme-dependent DNA amplification schemes such as PCR, loop-mediated isothermal amplification (LAMP) (54), nucleic acid sequence based amplification (NASBA) (55) and invader squared assays (56) to achieve quadratic or exponential amplification of initial signal. Similarly, catalyzed hairpin assembly reactions can also be cascaded (19) (see Supplementary Figure S5 for an example) to achieve higher levels of amplification.
However, a great challenge in cascading amplifiers is the amplification of background (also known as circuit leakage): the downstream amplifier can pick up the signal generated by the upstream amplifier in the absence of the input. Therefore, it is worthwhile to investigate the origin of circuit leakage and consider whether it can be eliminated. In toehold-mediated enzyme-free DNA circuits (such as the one described in this work) there are two major sources of circuit leakage: impurities and toehold-independent reactions. Using our circuit as an example, some sub-population of H1 may be mis-synthesized and/or mis-folded and can therefore react with H2 in the absence of C1. On the other hand, even if all molecules are perfectly synthesized and folded, it is still possible that the stem of hairpin H1 can ‘breathe’ (even at very low frequency and with very short duration), leaving part of domain 3* transiently unpaired and able to interact with domain 3 of H2, leading to consequent strand displacement reactions and ultimately to the formation of a fully extended H1:H2 duplex. When the kinetics of a single amplifier are observed, leakage induced by impurities usually results in a fast initial increase in product concentration, which stops increasing as impure molecules are used up. In contrast, leakage induced by toehold-independent reactions can lead to a slow but steady increase of product concentration until all the reactants are used up.
In general, leakage induced by impurity may be eliminated by better synthesis and folding protocols, or by purification methods where substrates (in this case, H1 and H2) are first incubated together in the absence of catalyst (C1), and then the unreacted substrates are separated from each other and the product [see (15) and (18) for examples]. In contrast, toehold-independent leakage can only be mitigated by better designs. In the circuit described here, although we observe a very low level of leakage induced by impurity (compare the black and red traces in Figure 3b) we did not observe toehold-independent leakage. This suggests that, with improved purification methods, our design strategy is capable of producing amplifiers with extremely low leakage, and that these amplifiers can be cascaded to achieve better signal amplification.
An additional source of false-positive signal is mis-triggering of the catalytic circuit by irrelevant DNA or RNA present in the sample, especially when clinical or environmental samples are used. Earlier studies on entropy-driven catalytic DNA circuits showed that the addition of total cellular RNA or random DNA did not result in observable mis-triggering (18,45). We have observed similar results with catalyzed hairpin assembly (data not shown). This robustness is likely due to the remarkably high sequence specificity of toehold-mediated strand displacement. In addition, unnatural nucleobases such as isoC and isoG (57) can be incorporated into these types of enzyme-free DNA circuits to further decreases the likelihood of spurious interaction between the circuits and irrelevant nucleic acids.
The real-time kinetics of DNA circuits has been routinely monitored by fluorescent reporters similar to that shown in Figure 1a, in which the exposure of a toehold leads to toehold-mediated strand displacement and separation of the fluorophore-bearing strand and the quencher-bearing strand. A subtlety in the design of such reporters that is not commonly appreciated is where to place the fluorophore and quencher. In most DNA circuit-related works, the fluorophore is on the strand that is ultimately displaced, whereas the quencher is on the strand that accepts the invading strand (‘design I’ in Supplementary Figure S6a). The position of fluorophore and quencher can also obviously be swapped (‘design II’ in Supplementary Figure S6a).
In previous studies (58), we designed the reporter based on ‘design I’, and added the quencher-bearing strand in 1.2-fold excess to ensure quenching. This allowed us to qualitatively detect the product of the circuit, but decreased our ability to quantitate the product, since the excess quencher strand can bind both the substrate and the product of the circuit and thereby confound the interpretation of the kinetic data.
In contrast, in ‘design II’ the reporter strand, whether fluor or deoxyribozyme, was double-stranded at the conclusion of the reaction (Supplementary Figure S6). The free quencher-bearing strand could only interact with its complement to form the substrate of the strand-exchange reaction (Supplementary Figure S6a), and thus could be added in even greater (2-fold) excess, ensuring efficient quenching. The DNAzyme reporter was largely double-stranded, preventing misfolding (Supplementary Figure S6b). Avoiding mispairing and misfolding in the reporter strands should allows greater overall flexibility in sequence design.
Although one could argue that the designs I and II should have identical performance if the reporter duplex is pre-annealed and purified on a native polyacrylamide gel (15), design II reduces the purification steps required to create a functional circuit. As larger and larger circuits are designed (59) with a view towards diagnostic and other applications, fault-tolerant design principles that allow less rigorous purification while still maintaining optimum performance will be essential.
One focus in the field of biosensor development has been the engineering of ‘smart molecules’, such as allosteric ribozymes/DNAzymes, aptamer beacons, and FRET-based signaling proteins (60), that not only recognize target molecules but also transduce the molecular recognition to observable signals. However, it is usually difficult to integrate many functions into one molecule and also to fine-tune each function. Molecular circuitry offers an opportunity to distribute tasks to many molecules that work collaboratively in a circuit to achieve higher-level performance (Supplementary Figure S7). In addition to amplifiers, other DNA-based signal processors such as logic gates (13,15) and thresholding devices (59) can also be used in DNA circuits to integrate information and parallelize signal transduction. Because of the inherent modularity of signal transduction and integration in DNA circuits, general design principles can be applied separately to each of these components.
Supplementary Data are available at NAR Online.
Funding for open access charge: National Institutes of General Medical Sciences (R01GM077040) and National Institutes of Allergy and Infectious Diseases (R01AI092839).
Conflict of interest statement. None declared.
The authors thank the Electrochemical Center at UT-Austin and Ms. Xiaole Chen for electrochemical equipments, and Steven Chirieleison for help on processing of electrochemical data.