Programmable design of isothermal nucleic acid diagnostic assays through abstraction-based models

The concept of the reaction graph

Our reaction graph concept entails simple representations of functional hairpin-based nucleic acid isothermal amplification mechanisms, enabling scientists to quickly establish the main reactions and the amplicons, with the ability to design strategies addressing situations that are currently challenging when using existing methods. In order to achieve this goal and demonstrate the amplification methods for developing basic DNA nodes and circuits, the process was simplified into five components (Fig. 1a): the functional hairpin motif self-folding assembly process (solid arrows); the primer extension process (dashed arrows); the formation of functional hairpin motifs generated through amplification (blue box); numeral marked primers and functional hairpin motif self-folding products (green triangle for forward strands and green square for reverse strands).

We built the basic nodes and circuits around two core isothermal amplification reaction schemes performed using a ssDNA containing a hairpin motif (green triangle), with one priming site at the loop domain, as illustrated in Fig. 1a. The amplification process is described via the state of each port in a reaction graph¹² and the secondary structure formation mechanism, simultaneously, also shown by Fig. 1a. As the primer that encodes the complementary sequence of the loop region (2a in circuit S and 1s in circuit A of Fig. 1a) hybridises to the loop domain, it initiates an extension reaction that opens the hairpin structure with a strand-displacing DNA polymerase (dashed arrows). The newly exposed domain (3s) at the stem serves as an initiation site that permits the generation of a double strand product, represented by the functional hairpin motif in the blue box.

Subsequently, an autonomous self-folding disassembly reaction of the functional hairpin motif occurs (solid arrows of Fig. 1a) to form two self-folding products (green triangle and green square). The formation of these intramolecular structures then separates the dsDNA product into two appropriate hairpin structures, exposing the priming sites. When primers bind to these exposed sites, the two complementary self-folding molecules serve as targets to start a new round of catalytic amplification and thus achieve signal amplification. The closure of the circuit between the blue box and triangle/square, Fig. 1, defines the continuation of the amplification process.

Mechanistic studies

In order to understand the mechanism and thermodynamics of the self-folding of hairpin structures and the competing folding of DNA oligonucleotides with partially complementary sequences, studies were performed using melting analysis to characterise the dependence of each phenomenon with the concentration of the constituent molecules¹³. To illustrate this behaviour in our system, we explored the thermodynamics of the amplification reactions, monitoring the melting temperature, T_m, of reactions involving oligos H1 and H2, which can both bind to each other and self-fold as hairpins individually (see Supplementary Fig. S1). Building upon our general design rules, established previously¹⁴, we are able to use the hairpins as a fuel for amplification reaction (noting improved stability when the neck is long, Supplementary Fig. S2).

The reaction graph, described in Fig. 1a, allows the generation of size-limited amplicons, which was confirmed by the smaller bands in Fig. 1d, showing the different structures predicted by the mechanisms as designed (and confirmed by sequencing in Fig. 1c). However, the longer structures are not predicted directly, as is often the case in isothermal amplifications (including LAMP, for example). The longer structures arise as a consequence of interactions of Bst polymerase, primers and amplification product, as described recently for LAMP using high resolution melting analysis¹⁵, and demonstrated here by restriction digest sequencing in Supplementary Fig. S3.

Computational design tools

The design of functionalities into an isothermal amplification system requires significant expertise¹⁶ and is often an iterative approach. Our reaction graph abstraction method provides the means to enable the design of the sequences necessary for an assay, based on relationships between different sequences of building blocks, enabling us to automate the compilation of the functional motifs for hairpin-mediated isothermal amplification. We were able to demonstrate an automated primer design system which only requires as its input specific target sequence information, but which provides as an output ranked sets of primers and an isothermal amplification process with specified dynamic behaviours, Fig. 2.

We developed such a program (described in detail in the Supplementary Text) for a user to input a target sequence (e.g. from a candidate pathogen in a diagnostic question) and obtain primer sequences to enable detection through amplification. The program is extensible, designed as a flexible tool for reaction graph primer design and can meet various design requirements in a high-throughput informatics environment. It is implemented in Golang, which can be deployed on major operating systems (Windows, Linux, Mac), to achieve high throughput analysis based on multithreading, high running efficiency, native high concurrency and powerful fault-tolerance.

To simplify the primer design discussion, we refer to different primer fragments as sequences (such as 1S/2S/US), as shown in Fig. 1. As an example of the flexibility of the approach, Supplementary Fig. S4 shows how a similar reaction graph to that used in Fig. 1 can lead to a different amplification mechanism by simply removing a priming site in between the hairpin structures. The basic model of a hairpin structure-based isothermal amplification mechanism described in Fig. 1 and Supplementary Fig. S4 encompasses methods depicted for both CPA⁴ and LAMP⁵, as illustrated in Supplementary Fig. S5.

We now demonstrate the utility of these basic models by experimentally executing two different and nucleic acid isothermal amplification strategies, each of increasing complexity, and each illustrating the different mechanisms by which the interactions within the functional motifs catalyse the amplification reactions.

Strategy 1. Generic tail strategy

Our first strategy involves the design of a functional hairpin motif through sense and anti-sense primers by encoding a synthetic generic tail (termed “us”, where “u” is for “universal”) with the same sequence at their 5′ end, but with their 3′ ends complementary to the specific target sequence, Fig. 3a. The amplification pathways specified in the reaction graph of Fig. 3a was translated into the secondary structure-based molecular implementation, Fig. 3b. These primers coexist in the absence of the target. In the reaction graph, this property was programmed by the absence of a starting point (i.e. the blue box was removed from the graph). The introduction of the target led to the generation of the amplicon containing the functional hairpin motifs (us+1s+2s+ua), followed by serial self-folding and primer extension by DNA polymerase. Although not mechanistically required, we also added two outer primers (F3, B3). In a similar strategy as for their use in conventional LAMP systems, they increase the kinetics of the reactions by providing an additional route to obtain the functional amplification products, which then enable the exponential amplification phase.

Gel electrophoresis, Fig. 3d, confirmed that no amplicon was generated in the absence of the target, whilst, on addition of the target, the amplification was initiated. DNA sequencing of the amplicon revealed the expected functional motifs, Fig. 3c. This system can be used to amplify sequences from pathogens (as templates) with short, highly conserved sequences between 40 and 60 bp.

Developing assays for shorter nucleic acid sequences (down to ca. 20 nt)¹⁷ has been challenging using existing assay mechanisms¹⁰, often requiring additional steps to extend the target sequence, as is the case in Exponential Isothermal Amplification (EXPAR¹⁸), ligation-LAMP¹¹ or RCA¹⁹. To this end, we demonstrate that our generic tail strategy (Fig. 3) can be directly applied to the detection of such small sequences and illustrate this capability with the detection of miR21 microRNA in a concentration dependant manner, down to fM concentrations (Fig. 3e), confirmed with a gel electrophoresis analysis (Fig. 3f). The limit of detection was estimated between 2 and 1 fM with 8 replicates, all detected at 2 fM (54.6 ± 1.2 min), whilst only half (4/8) were detected at 1 fM (56.2 ± 0.6 min) (Supplementary Table S2).

Part of the central challenge in this is that microRNAs are generally contained within families of closely related sequences with different biological functions²⁰. Our mechanism was able to differentiate the miR21 from miR21-A, which differs by only 2 bases (Supplementary Fig. S6), a feat which is most usually challenging using standard isothermal assays such as LAMP of RCA²¹. We also used the wider family of miR-7 and demonstrated that the assay was able to detect miR-7-a from other members of the same family at low concentrations, often required for detection in clinical samples (Fig. 3g).

Expanding further on the capabilities of our assay design to detect small miRNA sequences, we also demonstrate that this can be carried out in a multiplexed fashion for different (but closely related) targets (Supplementary Fig. S7). The universal tail strategy can be extended with different recognition sequences, coding for different target molecules (Supplementary Fig. S7b). Existing isothermal systems have been modified to tackle this complex challenge (such as RPA²² or EXPAR²³), however here we show that our strategy, involving using a universal tail, can be used to enable this design quickly and efficiently. We also demonstrate that detection can be incorporated into lateral flow strips for easy visual output (as could be required for example in resource-limited settings—Supplementary Fig. S7d).

In order to further illustrate the effectiveness of multiplexing strategies, we also explored the design and implementation of multiplexed reaction graphs to detect different sequences in a single, long, nucleic acid sequence²⁴ (Fig. 4), for example different regions of the same genome, in this case HIV. The design of such assays, particularly for the detection of such RNA viruses, requires to overcome their high frequency of mutations, which often results in limited conserved regions²⁵. We provide a strategy whereby two sequences initiate the same analytical output, so improving the efficacy of the assay, with the amplification and diagnostic output proceeding when either one or both sequences are introduced in the assay (as might be needed for a multiplex assay when two or more variants are present).

Figure 4a shows such a design of cross primers, which contain three different constructs: (i) the 3′-end (1s in the example of P1) is complementary to the conserved region of the target (1); (ii) a sequence downstream, towards 5′ (2a of P1), complementary to a priming site on the opposite strand of the same target (2s, the priming site for P2); and (iii) an incorporated sequence (3s in P1), which is the same as the priming site of another cross primer (3s of P3) that can amplify another conserved region (2) from the target. During the amplification cycling steps, the products from the cross primers formed the functional motifs. The functional motifs from Region 1 exposed the priming sites common with Region 2, which enabled the primers for Region 2 to bind and be extended by DNA polymerase, and vice versa.

The reaction graph also enabled us to predict the sequences of the amplification product and functional motifs. In the example of HIV, we used conserved regions in the gag and pol gene. Figure 4 demonstrates the ability of the reaction to generate amplification when one, or the other, or both targets are present. The predicted sequences of the amplicons were confirmed with DNA sequencing (Supplementary Fig. S8), although similar high molecular weight structures as described in Fig. 1c are also observed.

Strategy 2. Progressive model

Our second strategy sought to overcome the increment in complexity in a reaction pathway, which arises when linking an additional reaction pathway onto a primary pathway, for example where one product of an amplification pathway serves as an input into a second reaction (e.g. resulting in two amplification reaction systems, as in e.g. EXPAR, the first one being linear and the second one exponential²⁶). Figure 5 shows the implementation of such a step change in the complexity of design by using the intermediates as the links between different reaction pathways. In these cases, the primer extension generated many products containing functional motifs which were used as targets in the further amplification process. All pathways described here start with the addition of genomic DNA.

By programming this progressive model into the reaction graph abstraction of Fig. 5a, we obtained an isothermal amplification system with a sense primer (3a+1s) with the 3′ end complementary to the target sequence. The 5′ end was complementary to a sequence that lies downstream on the same strand of the target and anti-sense primers (2a, 3a, 4a and 5a) complementary to the individual sequences of the target, Fig. 5a. In the corresponding molecular implementation, all the primers coexisted in the absence of target molecules. The target catalyses the generation of three different functional hairpin motifs (3a+1s+2s+3s, 3a+1s+2s+3s+4s and 3a+1s+2s+3s+4s+5s) which all contained the same functional hairpin motif (3a+1s+2s+3s).

Such functional hairpin motifs tend to form intramolecular hairpin structures after self-folding and are stabilised by the paired, helical structures that form between 3a and 3s (at the 3′ end). The antisense strand product can therefore be extended after self-folding from the 3a site by adding a 4a or 4a+5a sequence at the loop region. During the primer extension, primers 2a/3a+1s bound to the loop region of sense and anti-sense strand, respectively, formed functional motifs, and led to the opening of the hairpin structure after extension, exposing the priming sites for other primers.

Critically, the formation of intermediate structures (as 3a+1s+2s+3s+4s and a hairpin motif 3a+1s+2s+3s+4s+5s) drives an exponential amplification process, powered by this progressive model. The hairpin motif (3a+1s+2s+3s+4s+5s) has two functions, namely: (i) with primers 2a/4a and 2a/3a, it catalyses the generation of 3a+1s+2s+3s+4s and 3a+1s+2s+3s in circuit S1 (and similarly in A1 with the respective primers); and (ii) with primers 2a/5a, it produces a new copy in both circuits S1 and A1.

Similarly, the motif 3a+1s+2s+3s+4s leads to the production of 3a+1s+2s+3s with the primers 2a/3a in circuit S2 (or in circuit A2 with primer 3a+1s), and a new copy of itself with the primers 2a/4a and 3a+1s/4s in circuit S2 and A2, respectively. Hence, 3a+1s+2s+3s+4s and 3a+1s+2s+3s+4s+5s form a progressive process capable of inputting another copy of the target motif to parallel amplification processes thus increasing the amplification efficiency, Fig. 5c. The amplification pathways specified in the reaction graph of Fig. 5a are thus translated into the secondary structure-based molecular implementation of Supplementary Fig. S9.

The sensitivity of this progressive model was evaluated in the design of an in vitro diagnostic assay using tenfold serially diluted target of the genomic DNA from M. tuberculosis cells (400 to 4 cells per reaction). Evagreen fluorescent dye was used to monitor the formation of double-stranded structures and thus characterise the kinetics of the reactions, Fig. 5b, showing faster reactions as the target concentration was increased.

The implementation of the different pathways simultaneously was gated by the presence of different anti-sense primers. Using the same real-time approach as in Fig. 5b, we defined a threshold time (Tt) and compared different reaction velocities by using a measure of the time when the fluorescence signal reaches 20% of its maximum (by analogy to the cycle threshold of real-time PCR). Figure 5c confirms that as more anti-sense primer sequences were introduced, different amplification pathways were enabled, leading to exponential reaction kinetics. Systems with 5a+4a and 4a were 13% ± 6 and 10% ± 3 respectively faster than when using 3a only, as their presence led to more opportunities to form more functional hairpin motifs. We confirmed the formation of specific functional hairpin motifs (3a+1s+2s+3s) and (3a+1s+2s+3s+4s) in the reaction graph (Fig. 5a) by agarose gel electrophoresis (Fig. 5d) and sequencing (Supplementary Fig. S9). Subsequent analysis of the amplification products by sequencing show that these large amplification products are composed of tandem repeats of the expected functional domains (the full mechanism is presented in Supplementary Fig. S9). This system can be used to amplify the DNA of pathogens with long, highly conserved sequences.

We further explored the applicability of our two de novo designs (Figs. 3a, 5a) by investigating their translation into medical diagnostics, processing 54 clinical samples from Hepatitis B virus patients, with the results shown in Table 1 (individual values are presented in Supplementary Table S3). Our methods showed the results had a high coincidence (96.3%), when compared against the real-time PCR gold standard. Only two samples were missed in these systems with copy numbers at 6.35 and 14 respectively, below the PCR clinical sensitivity threshold. As with many isothermal amplification methods²⁷, our mechanisms have shown kinetics that are concentration-dependant in well-controlled systems (e.g. artificial sequences). However, clinical samples contain large amounts of inhibitors and potentially more complex sequences which limit further the applicability of the techniques as quantitative methods, for example for viral load measurements in HBC patients. In these applications qPCR should be used as the gold standard method.