Enzyme-free synthesis of cyclic single-stranded DNA constructs containing a single triazole, amide or phosphoramidate backbone linkage and their use as templates for rolling circle amplification and nanoflower formation

Three different chemical cyclisation reactions yield biocompatible cyclic oligonucleotide templates for use in RCA and DNA nanoflower formation.


Introduction
Rolling circle amplication (RCA) is an isothermal enzymatic technique that is widely used to make very long single-stranded DNA and RNA. 1,2 The product contains a specic, tandemrepeating sequence which is encoded by the complementary cyclic template. 3,4 This technique has been harnessed as a simple and powerful method of signal amplication in the development of sensitive detection methods for a variety of nucleic acid, protein, cellular and small molecule targets for diagnostic, biosensing and genotyping purposes. [5][6][7][8][9][10][11][12][13][14] RCA has also attracted widespread interest as a tool in the synthesis of functional DNA nanomaterials 15 including origami, 16 nanosprings, 17 nanotubes, 18 templating nanoscaffolds, 19,20 hydrogels 21,22 and DNA nanoowers. 23 Proposed future applications of these materials include biosensors, targeted imaging and drug delivery agents, 24 and components of nanoscale computers and electronic circuits. An RCA reaction requires four components: a circular DNA template, a short DNA primer, a polymerase enzyme and deoxynucleoside triphosphates (dNTPs). The cyclic template is most commonly obtained via an enzyme-catalysed cyclisation of a linear DNA strand using DNA ligase enzymes such as T4 ligase or CircLigase. The efficiency of the cyclisation reaction is sensitive to the length of the linear DNA precursor, and is inefficient if the nucleobase sequence leads to secondary structures. 25,26 In addition, the scale on which the cyclisation reaction can be carried out is limited by the cost of the enzyme. In principle these limitations could be circumvented by the use of chemical ligation ( Fig. 1), which is readily scalable, and would allow the use of a wide variety of buffers, including those in which DNA structures are denatured. However, the creation of a canonical phosphodiester linkage by chemical methods in aqueous media is inherently difficult because the nucleophilicity of the attacking hydroxyl group is similar to that of the surrounding water molecules, which are present in great excess. Although chemical methods for carrying out this transformation have been reported, such methods oen rely on the use of acutely toxic reagents such as cyanogen bromide. 27 For these reasons articial DNA backbones which are less challenging to form chemically are an attractive alternative. 28 Prevalent amongst these are the triazole, 29 amide 30,31 and phosphoramidate [32][33][34] linkages ( Fig. 1(ii)). When incorporated into linear oligonucleotides these phosphodiester backbone analogues can be read through by DNA polymerases during linear copying and the polymerase chain reaction (PCR), with accurate transfer of genetic information. 28,32,35 They could potentially be incorporated into cyclic RCA templates provided that they can also be read through by the specic DNA polymerases that are compatible with efficient RCA. However, the use of chemically modied templates in RCA is particularly demanding because it requires the polymerase to accurately read through the modied linkage repeatedly in order to generate very long amplication products. In contrast, during PCR the polymerase needs only to copy the chemically modied template once during the rst cycle, aer which the reaction can proceed in an exponential manner, with the enzyme almost exclusively copying backbone-unmodied products in all subsequent cycles. Given the widespread importance of the RCA reaction, we undertook to study the cyclisation of DNA using chemical approaches and to examine the suitability of the chemically modied cyclic products as templates during RCA. We then demonstrated that RCA of the cyclic products can be used to prepare DNA nanoowers. 23,[36][37][38][39][40][41] Finally, we compared the size distribution, morphologies and extent of DNA loading of the DNA nanoowers assembled from the RCA products of the chemically modied templates.

Choice of chemical linkages and template sequences
In the present study we investigate RCA of cyclic oligonucleotides containing a single triazole, amide or phosphoramidate analogue of the DNA phosphodiester backbone. All three modications have been previously used in PCR amplication of linear DNA. 28,32,35,[42][43][44][45] Of these, the triazole linkage ( Fig. 1(ii)A) is the most thoroughly studied articial DNA backbone. 29 It can be read through by both DNA and RNA polymerases and has been used in the construction of genes by chemical ligation. Importantly, DNA sequencing experiments did not reveal any increase in the level of mutations observed for cells transformed with the triazole-modied gene compared to those transformed with the canonical analogue. 35 There are two reports of cyclisation of linear DNA that contains 5 0 -azide and 3 0 -alkyne followed by RCA of the cyclic template. The rst one, from our laboratory, 45 shows that cyclisation and subsequent RCA of a 100-mer DNA sequence designed to be free of secondary structure is possible, despite the presence of the 1,4-triazole linkage. The second study used a triazole-containing DNA template in an RCA-based uorogenic assay for microRNA, but the authors did not compare the efficiency of the RCA reaction with that of an unmodied template. 46 In this work we also investigate amide and phosphoramidate chemical linkages ( Fig. 1(ii)B and C), both of which can be formed by di-imide coupling chemistry. The phosphoramidate linkage is of particular interest as it closely mimics the steric and electronic properties of the canonical phosphodiester backbone.
In order to study the above three non-canonical DNA linkages in RCA we designed three cyclic templates of differing base composition, all of which are predicted to incorporate secondary structures, and two of which include cancer cell surface targeting aptamer sequences (oligonucleotide sequences are given in Table 1, and the predicted structures are shown in Fig. S1, ESI †). Template 1 is based on a randomly generated 50-mer sequence with approximately 50% GC content; template 2 incorporates a dumbbell-shaped double hairpin structure and serves as a template for an aptamer sequence which targets the mucin 1 (MUC1) glycoprotein 47 -an important class of tumour surface marker that is overexpressed on a range of epithelial cancer cells; template 3 is a cytosine-rich sequence encoding a complementary trimeric G quadruplex aptamer which recognises the human epidermal growth factor receptor 2, a major therapeutic target in human cancer treatment. 48 All four versions of each template were made: an unmodied control containing a natural phosphodiester (PO 4 ) backbone and analogues containing triazole (Tz), amide (Am) and phosphoramidate (PA) linkages at the point of cyclisation. ‡ In addition we prepared the unmodied and triazole-modied cyclic templates 4_PO 4 and 4_Tz, which incorporate a palindromic sequence. The RCA product of 4_PO 4 has been reported to assemble into a three-dimensional interwoven 'nanoclew' structure that can be used to deliver anticancer drugs. 39 Following cyclisation of the linear sequences all of the cyclic templates were puried then subjected to RCA using f-29, a highly processive polymerase with strand displacement activity. 49

Cyclisation of linear oligonucleotides
Linear oligonucleotides incorporating the required 5 0 -phosphate/3 0 -hydroxyl, 5 0 -azide/3 0 -alkyne, 5 0 -amine/3 0 -phosphate and 5 0 -amine/3 0 -carboxylate modications were synthesised using automated solid phase phosphoramidite chemistry, as described in the ESI. † The linear unmodied control sequences were cyclised to give the cyclic templates 1_PO 4 , 2_PO 4 , 3_PO 4 and 4_PO 4 using T4 DNA ligase in the presence of a splint oligonucleotide. The splint oligonucleotides incorporate a region which is complementary to each end of the corresponding linear oligonucleotide and hence act as cyclisation templates by bringing the two reacting ends of the linear substrate into close proximity. Cyclisation of the chemically modied linear oligonucleotides was carried out under aqueous CuAAC or di-imide coupling conditions. In almost every case the linear precursor was successfully converted to the desired cyclic product with moderate to excellent efficiency. § The linear 5 0 -azide/3 0 -alkyne oligonucleotides were cyclised by treatment with CuSO 4 in the presence of sodium ascorbate and trishydroxypropyltriazole (THPTA) in 0.2 M aqueous NaCl or denaturing organic/aqueous solvent mixtures. In all four cases the cyclisation reaction proceeded smoothly in the absence of a templating splint oligonucleotide. The 5 0 -amine/3 0 -phosphate oligonucleotides were cyclised by treatment with 1-ethyl-3-(3dimethylaminopropyl)carbodiimide hydrochloride (EDC$HCl) and 2-(hydroxyethyl)imidazole in aqueous 0.2 M HEPES buffer (pH 7.2) using an adaptation of our previously reported phosphoramidate ligation method. 32 Similarly the 5 0 -amine/3 0carboxylate were cyclised using EDC$HCl in combination with N-hydroxysuccinimide in aqueous NaCl or HEPES buffer, also using an adaptation of a previously reported method. 28 The phosphoramidate-and amide-modied cyclic constructs 1_PA and 1_Am formed readily in the absence of a templating splint oligonucleotide; for the amide and phosphoramidate sequences 2_PA, 2_Am, 3_PA and 3_Am the cyclisation reactions were found to be less favourable under non-templating conditions and a splint oligonucleotide was therefore added in order to achieve more efficient cyclisation. Crude cyclisation reactions were analysed by polyacrylamide gel-electrophoresis (PAGE). Representative gel images for the template sequences 1_PO 4 , 1_PA, 1_Tz and 1_Am are shown in Fig. 2 (see Fig. S2-S9, ESI † for PAGE analysis of all other cyclisation reactions).
During the EDC-mediated cyclisation reactions to prepare the phosphoramidate-modied templates we observed the formation of adducts of the cyclic products with molecular weights which are 155 mass units higher than those for the expected products by UPLC-mass spectrometry. This is consistent with the chemical addition of EDC to the oligonucleotides. Similar observations have been documented previously: the adducts are proposed to arise from a chemical reaction between the di-imide coupling reagent and the imide group of guanine and thymine DNA bases. [50][51][52][53][54] In support of this we did not observe formation of these adducts upon exposure of polyadenine and poly-cytosine DNA sequences to 0.1 M EDC in 0.2 M HEPES buffer (pH 7.2) at room temperature, while in contrast a signicant degree of EDC adduct formation was observed when the same experiment was performed with sequences containing guanine and thymine bases. Pleasingly it was possible to reverse this unwanted side-reaction by gently heating the crude product mixtures in 0.1 M NaOH aer performing the phosphoramidate cyclisation reactions. Interestingly, we did not observe any evidence of the formation of the same EDC adducts during the reactions to prepare the amide-modied cyclic constructs, which were carried out under slightly different conditions and required a lower concentration of the di-imide coupling agent.
All of the modied and unmodied cyclic products were readily puried by PAGE. For the splint-mediated cyclisation reactions it was necessary to use denaturing PAGE conditions during purication, while for the non-templated reactions either denaturing or non-denaturing conditions were selected, depending on which gave better separation between the linear oligonucleotide precursor and cyclic oligonucleotide product. Full details of the cyclisation and purication protocols along with analytical data for all of the cyclic constructs are provided as ESI (Fig. S10-S23, ESI †).
While the efficiency of each cyclisation method described above was observed to vary depending on the individual linear oligonucleotide sequence, it is noteworthy that the chemical ligation methods can offer signicant improvements in cyclisation efficiency in cases where the more commonly used enzymatic approach proves problematic. For example, while the T4-DNA ligase-mediated cyclisation of the 'nanoclew' template 4_PO 4 was low-yielding, producing a number of higher molecular weight oligomeric side-products, the analogous CuAACmediated cyclisation reaction-when performed in the presence of denaturing organic solvents to disrupt secondary structure-proceeded more cleanly, generating fewer byproducts (Fig. 3). The efficiency of this chemical cyclisation reaction appears comparable to that of the reported CircLigase II-mediated cyclisation of the same template; 39 however, the chemical cyclisation approach is not limited in scale by the high cost associated with the CircLigase enzyme. The chemical ligation methods discussed here therefore provide a greatly expanded toolkit for the cyclisation of linear oligonucleotides, with the potential to carry out the reactions under varied conditions, in the presence of denaturing agents, at varying salt concentration and temperature, and with signicantly reduced cost. In addition, chemical ligation reactions are more robust and more scalable than enzymatic reactions, with attendant enhancements in the ease of purication, isolation and characterisation of the cyclic products.
Rolling circle amplication of the cyclic oligonucleotides RCA using the f-29 polymerase. For all fourteen cyclic templates, RCA was carried out using f-29 polymerase, a 16 or 18 nucleobase primer, 2 mM dNTPs and 20 mM Mg 2+ at 30 C.{ Analysis of the product mixtures using agarose gelelectrophoresis indicated that in all cases the RCA reactions were successful ( Fig. 4 and S24-S32, ESI †). However, there were signicant differences in the quantities and size distributions of the products, which depended on the nature of the modied linkage in the cyclic template. Each reaction produced a detectable quantity of very slow-migrating products which were retarded in the wells of the gel and, in some cases, a second discrete band with higher gel mobility was also observed to form a substantial proportion of the product distribution (Fig. 4). By analogy with a recent literature report, 55 this faster-running band was assigned to double-stranded products, while the material which is retarded in the wells is assigned to the expected very long single-stranded products. A possible mechanism for the formation of double-stranded sideproducts during RCA is illustrated in Fig. 5. It is noteworthy that the faster-running band assigned to double-stranded DNA side-products was not observed aer RCA of the cyclic templates 3_PO 4 , 3_Am, 3_PA and 3_Tz, which encode a complementary trimeric G quadruplex aptamer sequence ( Fig. S30-S32, ESI †). One possible explanation is that the mechanism of doublestranded DNA formation is suppressed by the tendency for the emerging RCA product to fold up into a G quadruplex secondary structure. The phosphoramidate modication gave very efficient RCA, consistently yielding product distributions which were indistinguishable from those of the unmodied templates (Fig. 4, S24, S27 and S30, ESI †). The amide-modied cyclic templates also appeared to be well-tolerated by the f-29 polymerase, producing similar product distributions to the unmodied control sequences in all cases (Fig. 4, S25, S28 and S31, ESI †). In contrast, the triazole-containing templates gave Fig. 3 Comparison of the enzymatic and CuAAC-mediated cyclisations used to prepare the cyclic constructs 4_PO 4 and 4_Tz respectively, whose sequences incorporate complex secondary structure. The crude reaction products were analysed using 8% denaturing polyacrylamide gels, which were visualised by post-staining with SYBR Gold    5 Proposed mechanism for the formation of double-stranded products during RCA of the triazole-and amide-modified cyclic templates. (i) In the normal RCA mechanism, the amplified singlestranded product is continuously displaced from the cyclic template, resulting in the formation of very long single-stranded products. (ii) An alternative mechanism allows the enzyme to dissociate from the cyclic template oligonucleotide and copy the emerging single-stranded product, resulting in the formation of double-stranded side-products. This process sequesters the cyclic template and inhibits RCA. This mechanism may be favoured in the presence of the unnatural triazoleand amide-backbone linkages, which the enzyme has greater difficulty in reading through.
weaker and more variable results. While the triazole-modied cyclic template 1_Tz performed moderately in comparison to the unmodied and amide-and phosphoramidate-modied analogues, the cyclic templates 2_Tz, 3_Tz and 4_Tz consistently produced much lower yields of very long, slow-migrating RCA products ( Fig. 4; S26, S29 and S32, ESI †). The overall performance of the triazole-containing templates therefore appears to be compromised in comparison to those of the unmodied and phosphoramidate-containing analogues. This is a salient observation given that a recent publication uses a triazole-containing cyclic template with f-29 in an RCA-based uorescent switch-based diagnostic assay for microRNA detection. 46 It is also interesting that during RCA of the amide-and triazole-modied templates 1_Am, 1_Tz and 2_Am the proportion of double-stranded side-products appears to increase with extended amplication time. In contrast the relative proportions of the bands assigned to single-and double-stranded products respectively appear to stay approximately constant throughout the course of the RCA reactions for all of the unmodied and phosphoramidate-modied templates (Fig. S24-S28 and S30, ESI †).k Probing the single-stranded RCA products with complementary uorescent oligonucleotides. To obtain further information about the nature of the products from the RCA reactions on the modied phosphoramidate-, amide-and triazolecontaining cyclic templates we developed a uorescent probe assay to allow differentiation between single-and doublestranded RCA products by gel-electrophoresis. A singlestranded Cy3-labelled uorescent hybridisation probe with a sequence which is complementary to those of the single stranded products obtained from RCA of the cyclic templates 1_PO 4 , 1_Tz, 1_PA and 1_Am was synthesised. In order to validate the assay, a scrambled version of the probe was also prepared, and we performed a preliminary control experiment using the complementary and scrambled versions of the hybridisation probe in order to conrm that the complementary probe binds specically with the RCA product of the cyclic template 1_PO 4 through base paring (Fig. S33, ESI †). RCA of the four cyclic templates 1_PO 4 , 1_Tz, 1_PA and 1_Am was then performed using the f-29-polymerase enzyme and individual RCA reactions were stopped by heat inactivation of the polymerase at two hour intervals over a twenty hour time period. The product mixtures were then incubated with the uorescent hybridisation probe for two hours at room temperature before analysis using agarose gel electrophoresis. The gels were imaged with excitation at 520 nm under the Cy3 uorescence channel of a Syngene G:BOX imager (Fig. 6). Finally, the same gels were post-stained with a solution of SYBR Gold, which can sensitively detect both double-and single-stranded DNA, and re-imaged with excitation at 302 nm (Fig. S34, ESI †). The results show that a strong uorescence signal from the bound Cy3 labelled probe is detected aer amplication of the unmodied, amide-modied and phosphoramidate-modied templates 1_PO 4 , 1_PA and 1_Am. This demonstrates that the uorescent probe has successfully hybridised to the RCA products, con-rming their single-stranded nature. Conversely, aer amplication of the triazole-modied template, no appreciable signal from the Cy3 uorophore is detected, although some slowmigrating RCA products are weakly detected in the wells of the gel aer post-staining with SYBR Gold. This suggests that the products of the RCA reaction using the triazole-modied template contain a signicantly lower quantity of both singleand double-stranded DNA.
Probing the nature and quantity of the RCA products via a uorometric plate-reading assay. In order to corroborate the results from the gel-imaging we used a uorometric platereading assay to allow a semi-quantitative comparison of the rates and total DNA yields of RCA reactions performed using f-29 polymerase and the cyclic templates 1_PO 4 , 1_Tz, 1_PA and 1_Am over a 10 hour time course. Aer the desired time period the reactions were stopped by heat inactivation of the polymerase and then analysed, using three different DNA-binding dyes to provide a uorescence signal: SYBR Gold, SYBR Green I and SYBR Green II. The intensities of the measured uorescence signals are approximately proportional to the total concentrations of DNA in each reaction mixture, with some expected deviations as a result of the differing proportions of single-and double-stranded DNA amplication products. In agreement with the gel-imaging results, the reaction rates and product yields for the RCA reactions with the cyclic phosphoramidate-containing template 1_PA are essentially the same as those for the unmodied cyclic template 1_PO 4 , and the cyclic amide template 1_Am performs almost as well. The cyclic triazole template 1_Tz is shown to be inferior: at each time-point the measured uorescence intensities are weaker for the RCA reaction involving the triazole-containing template Fig. 6 Probing the nature of the amplified products from f-29mediated RCA of the cyclic templates 1_PO 4 (A), 1_PA (B), 1_Am (C) and 1_Tz (D) using a fluorescent probe hybridisation assay. Aliquots of the RCA reactions were stopped by heat inactivation at regular two hour intervals for a total period of 20 hours. A Cy3-labelled fluorescent probe with a sequence which is complementary to the RCA products was added to each aliquot and the mixtures were analysed by agarose gel electrophoresis (0.8% agarose). The gels were imaged under the Cy3 fluorescence channel. The gels were subsequently stained with SYBR Gold (Fig. S34, ESI †).
1_Tz than for those involving the other three templates, indicating that a lower quantity of DNA is produced when 1_Tz is used as a substrate. By taking into account the differing selectivities of the three DNA binding dyes for double-stranded DNA over single-stranded DNA it also possible to draw inferences about the nature of the products. While all three dyes are known to bind both double-and single-stranded DNA to some extent, SYBR Green I has the most pronounced selectivity for doublestranded DNA over single-stranded DNA. [56][57][58] The measured uorescence intensities for the samples containing the triazole-modied template relative to those containing the phosphate and phosphoramidate templates are noticeably higher in the presence of SYBR Green I than in the presence of the SYBR Gold and SYBR Green II dyes (Fig. 7). Hence there is a strong indication that a signicant amount of double-stranded DNA is produced in the RCA reaction on the triazole-containing template, which is consistent with the prior gel-imaging results. The same phenomenon is observed for the amide-modied template 1_Am, but to a much smaller extent. This is probably because each time the polymerase encounters the modied linkage in the cyclic template, it can choose between processively reading through the modied linkage or dissociating from the cyclic template and instead copying the RCA product as it emerges, thereby producing double-stranded products (Fig. 5). 55 Since the triazole linkages are seemingly more difficult/slower to read through, it is likely that the alternative mechanism is more strongly favoured in their presence, resulting in the formation of a higher proportion of doublestranded products which would be of shorter length. This process may also sequester the cyclic template, contributing to the lower overall yield of DNA product.
RCA using the Bst 2.0 DNA polymerase. In light of the disappointing performances of the triazole-containing templates in the f-29-mediated RCA reactions, we also investigated RCA using the Bst 2.0 DNA polymerase for two of the cyclic triazole templates 1_Tz and 4_Tz and their unmodied analogues 1_PO 4 and 4_PO 4 , in order to ascertain whether the triazole linkage can be read through more efficiently by this alternative polymerase. The Bst 2.0 enzyme was previously used for RCA of the cyclic template 4_PO 4 39 and is known to efficiently incorporate modied dNTPs during RCA but its ability to read through modied backbone linkages has not yet been thoroughly tested. 59,60 Agarose electrophoretic analysis of the RCA products indicated that, while the f-29 enzyme again exhibited higher RCA efficiencies with the unmodied templates 1_PO 4 and 4_PO 4 compared to the triazolecontaining analogues 1_Tz and 4_Tz, Bst 2.0 was able to process the triazole-modied templates more efficiently (Fig. 8). For example, reactions performed using the Bst 2.0 enzyme produced similar yields and product distributions for the triazole-modied and unmodied cyclic templates 1_Tz and 1_PO 4 , which are based on a randomly generated sequence (Fig. 8B). Moreover, for the nanoclew template sequences 4_Tz and 4_PO 4 the Bst 2.0 DNA enzyme appeared to generate a signicantly greater quantity of very long, slow-migrating RCA products from the triazole-containing template 4_Tz than from the unmodied control 4_PO 4 . Although unexpected, this result exemplies the utility of having an available choice of chemical cyclisation methods in cases where the cyclisation and/or Fig. 7 Fluorescence quantification of amplified DNA products from RCA reactions using the cyclic templates 1_PO 4 , 1_Am, 1_PA and 1_Tz and the f-29 DNA polymerase over a 10 hour time course. EDTA was added to break down the precipitate releasing extended DNA prior to incubation with DNA binding dyes. Measurements are recorded in the presence of SYBR Gold, SYBR Green I and SYBR Green II fluorescent DNA binding dyes. Error bars represent the standard deviation of three measurements taken from a single RCA reaction. The reactions were also repeated in duplicate and the same trends were observed in both cases. amplication of a specic canonical sequence proves to be inefficient. The RCA products from the reactions using the Bst 2.0 polymerase were also quantied via a uorometric platereading assay, which corroborated the gel-imaging results. However it is notable that in all four reactions the Bst 2.0 enzyme generated a higher proportion of double-stranded products than the f-29 enzyme (Fig. 9A and B)-reecting the Bst enzyme's inferior processivity and strand displacement activity under the conditions used in our RCA experiments.

Nanoower formation
Long RCA products are known to spontaneously self-assemble into densely-packed DNA-inorganic hybrid nanoower (DNA-NF) structures in the presence of a magnesium pyrophosphate co-precipitant, which is also generated as a by-product of the RCA reaction. The DNA-NF structures are thought to comprise an inorganic magnesium pyrophosphate core onto which the long single-stranded DNA RCA products are adsorbed. These nano-ower structures have several attractive features which are driving their investigation as potential diagnostic and therapeutic agents: they are easy to prepare; they are size-tuneable and biodegradable, yet resistant to digestion by exonuclease enzymes; they can be designed to incorporate a functional aptamer, DNAzyme, restriction enzyme, antisense or drug-loading sequence for targeted recognition and delivery; and they have been shown to be capable of cellular transfection. 23,[36][37][38][39][40][41] Scanning electron microscopy (SEM) experiments were used to map the size and morphology of the nanoowers formed at different time points during the RCA reactions performed using the unmodied and modied cyclic templates. Aer time periods of eight and twenty hours aliquots of the reactions were removed and heated to inactivate the enzyme. The precipitated nanoowers were then collected by centrifugation, thoroughly washed with deionised water and subjected to SEM imaging experiments. The extent of DNA loading was also examined by agarose gel electrophoresis ( Fig. S35-S36, ESI †). The SEM images ( Fig. 10 and 11) indicated that there was no signicant difference in the size and morphology of the DNA nanoowers generated from the four types of cyclic template, even though the triazole containing sequences are less efficient as RCA templates. In all cases DNA-NF structures with diameters of >1 mm and petal-like surface morphologies were formed in the RCA reaction mixture within eight hours, and their sizes were observed to have marginally increase aer 20 hours. Variation of the Mg 2+ concentration was found to have a signicant effect on the particle size and DNA loading. At Mg 2+ concentrations of #10 mM the particles were found to contain negligible quantities of DNA; as the Mg 2+ concentration was increased from 10-25 mM the extent of DNA loading steadily increased, with a concomitant reduction in the size of the particles from $3 mm to $0.8 mm (Fig. 11 and S37-S46, ESI †). Although the reason for the contraction of the nanoparticles with increasing Mg 2+ concentration is not immediately obvious, it is possible that this reects an increased rate of nucleation at higher metal ion concentration, which might be expected to produce a higher number of particles with a smaller average diameter. Fig. 9 Fluorescence quantification of amplified DNA products from RCA reactions using the cyclic templates ((A) 4_PO 4 , and 4_Tz; (B) 1_PO 4 , and 1_Tz) and enzyme over a 10 hour time course. The performances of the f-29 and Bst 2.0 DNA polymerases are compared. EDTA was added to break down the precipitate releasing extended DNA prior to incubation with DNA binding dyes. Measurements are recorded in the presence of SYBR Gold, SYBR Green I and SYBR Green II fluorescent DNA binding dyes. Error bars represent the standard deviation of three measurements taken from a single RCA reaction. The reactions were also repeated in duplicate and the same trends were observed in both cases.

Conclusions
We have shown that single-stranded DNA templates can be efficiently cyclised using three different chemical ligation methods. The resulting cyclic constructs contain a single unnatural triazole, amide or phosphoramidate linkage at the point of ligation. These chemical cyclisation methods offer several advantages over the more limited enzymatic approach: they are cheaper and easier to carry out, allowing for the largescale synthesis and purication of the cyclic oligonucleotide products; they can also be performed under a wider variety of conditions, including denaturing ones, and are therefore particularly useful for cyclising linear oligonucleotides which contain problematic secondary structure. The chemical cyclisation methods described here therefore provide a versatile expanded toolkit for the preparation of cyclic oligonucleotides, which have an expansive range of potential applications across molecular biology 61-65 and nanotechnology. [66][67][68][69] Here we investigated the use of the chemically modied cyclic products as templates for the production of long single-stranded DNA concatemers via the highly demanding rolling circle amplication reaction, an important technique which is currently attracting widespread interest. The cyclic templates containing a non-canonical phosphoramidate backbone linkage performed particularly well during RCA-with overall yields, product distributions and reaction rates which were indistinguishable from those of the analogous unmodied templates. In contrast, the triazole-modied templates were less well tolerated by the f-29 polymerase, showing a tendency to produce doublestranded DNA side-products and varying overall yields. The performance of the templates containing an amide modication falls in between those of the phosphoramidate and triazole linkages. Interestingly, for one of the cyclic template sequences, the Bst 2.0 polymerase enzyme was found to amplify the  triazole-modied template with greater efficiency than the unmodied analogue, suggesting that different combinations of chemical ligation method and polymerase enzyme may nd application where RCA proves challenging. Of all the cyclic templates investigated, the phosphoramidate-containing versions are the most straightforward to synthesise and could be produced in commercial DNA synthesis facilities from readily available precursors. We propose that phosphoramidate cyclisation is a practical, scalable and convenient method for the production of biocompatible cyclic DNA constructs for use in many applications, including RCA.

Conflicts of interest
There are no conicts to declare. Notes and references ‡ Where necessary slight modications were made to the ordering of the terminal nucleobases in order to make the sequences compatible with the chemistry used to introduce the required amino, azido or alkyne functionality at the 5 0 and 3 0 ends. Each phosphoramidate sequence contains 5 0 -amino dT; each amide sequence contains both 5 0 -amino dT and 3 0 -carboxy dT; and each triazole sequence contains 5 0 -azido dT and 3 0 -propargyl dC. See the ESI † for further details of the oligonucleotide synthesis chemistry. § The only exception is that the amide cyclisation reaction to give the cyclic template 3_Am was inefficient, yielding a signicant quantity of higher molecular weight side-products. Although we were able to isolate a sample of the cyclic template 3_Am, it was found to contain impurities by HPLC mass spectrometry analysis (see Fig. S20, ESI †). Nonetheless, this sample was successfully used as a template in RCA, with promising results. { A lower Mg 2+ concentration of 10 mM is recommended by the enzyme supplier.

Acknowledgements
Our experiments indicated that, although RCA proceeds efficiently over a wide Mg 2+ concentration range of 10-25 mM, the preparation on DNA-NFs is more sensitive to Mg 2+ concentration: at 10 mM Mg 2+ concentrations the extent of DNA loading is oen negligible, whereas a higher 20 mM Mg 2+ gives higher and more consistent levels of DNA loading. Taking this into account, we chose to use a 20 mM concentration of Mg 2+ throughout our RCA experiments. k We also investigated whether the formation of double-stranded amplication products could be suppressed by addition of single-stranded DNA binding protein to f-29 mediated RCA reactions of the cyclic templates 1_PO 4 and 1_PA, as previously described by Högberg and co-workers. 55 However, when using our smaller, chemically modied templates we did not observe any signicant differences in the product distributions of the RCA reactions carried out in the absence and presence of the single-stranded binding protein (see Fig. S47, ESI for further details †).