Nanopore microscope identifies RNA isoforms with structural colours

Identifying RNA transcript isoforms requires intricate protocols that suffer from various enzymatic biases. Here we design three-dimensional molecular constructs that enable identification of transcript isoforms at the single-molecule level using solid-state nanopore microscopy. We refold target RNA into RNA identifiers with designed sets of complementary DNA strands. Each reshaped molecule carries a unique sequence of structural (pseudo)colours. Structural colours consist of DNA structures, protein labels, native RNA structures or a combination of all three. The sequence of structural colours of RNA identifiers enables simultaneous identification and relative quantification of multiple RNA targets without prior amplification. Our Amplification-free RNA TargEt Multiplex Isoform Sensing (ARTEMIS) method reveals structural arrangements in native transcripts in agreement with published variants. ARTEMIS discriminates circular and linear transcript isoforms in a one-step, enzyme-free reaction in a complex human transcriptome using single-molecule read-out. A method has been developed to identify RNA transcript isoforms at the single-molecule level using solid-state nanopore microscopy. In this method, target RNA is refolded into RNA identifiers with designed sets of complementary DNA strands. Each reshaped molecule carries a unique sequence of structural (pseudo)colours that enables identification and quantification using solid-state nanopore microscopy.

S ingle-molecule identification of multiple transcript isoforms in parallel without preamplification is critical for understanding transcriptome diversity and gene expression networks 1 . Identification and quantification of structural arrangements in native transcripts are both challenging, and current methods do not necessarily yield results reflecting innate transcriptome diversity 2,3 . Although identification of long RNA molecules is possible with existing nucleic acid detection methods [4][5][6] , these methods lack specificity and simplicity. In addition, common approaches mainly rely on enzymatic reactions and require preamplification. These lead to inevitable biases and loss of information [7][8][9] . RNA sequencing approaches require extensive and intricate adaptations to achieve the sequencing of transcript variants and to test their circularity 2,10,11 . These widely used techniques face amplification and reverse transcription biases, and detection of transcript variants is affected by short reads in RNA sequencing 10,11 . Recently, nanopore sequencing introduced direct RNA read-out 10 ; however, access to single-molecule information of gene expression level in combination with low-quality reads and uncertainty about the 5′ end of the transcript remain major challenges 12 . As in previously established RNA sequencing methods, nanopore RNA-seq also suffers from enzymatic biases 13,14 . Additionally, secondary structures in both RNA and complementary DNA (cDNA) contribute to biases by obstructing the binding of primers and sequencing adaptors 15 . In a vast majority of cases for identifying disease-specific transcripts, looking for a transcriptome subset is sufficient rather than sequencing the whole transcriptome 16 . Accurate annotation of transcripts is still challenging, and numerous transcripts remain uncatalogued 17 . We are in need of an enzyme-free method for targeted identification of native transcript isoforms, that avoids notoriously laborious protocols and that reveals structural arrangements on the single-molecule level.
Here, we introduce our ARTEMIS method. ARTEMIS relies on the molecular design of identifiers (IDs) for native RNA targets that reshape the RNA 'scaffolds' into unique structures inspired by the DNA origami technique 18 . IDs in ARTEMIS consist of an RNA-DNA nanostructure. Each ID is composed of sequence-specific structural (pseudo)colours that identify the target RNA and its isoforms.
With ten colours we can potentially create ten billion unique identifiers that are adaptable to both nanopore microscope and super-resolution microscopies that can be based on DNA-based point accumulation for imaging in nanoscale topography (DNA-PAINT). Nanopore microscopy works via a voltage-driven translocation of negatively charged RNA IDs through a small orifice towards a positively charged electrode in an electrolyte solution. It translates the designed RNA ID into a current signal. Colours placed on each RNA ID produce downward current spikes that correspond to the size of a colour. Larger colours induce a larger blockage of current. Notably, we identified RNA targets of interest in a complex human transcriptome enabling targeted enzyme-free transcriptomics. In addition, ARTEMIS identifies physical rearrangements of transcript isoforms including the order of their structural elements and their orientation, length and circularity by using short DNA oligonucleotides (also called oligos) as inexpensive scaffolding material. In addition to DNA, we used targeted RNA origami self-assembly, that is, uncomplemented RNA as additional structural colours that can be located along the molecule or at its ends. ARTEMIS solves long-read sequencing problems of strongly structured regions or too long transcripts since our designed RNA IDs allow for efficient identification of full-length RNA. We identified isoforms of a messenger RNA and a long non-coding RNA and show that we can perform relative isoform quantification. ARTEMIS will enable studies of native RNA diversity, gene expression analysis, RNA motif nanofolding and RNA interactions with proteins and small molecules.

Results and discussion
Structural colours enable RNA isoform identification with nanopore microscope. As a first example, we introduce the concept of structural colours for three RNA target isoforms that are complemented with DNA oligonucleotides as shown in Fig. 1a. In this example, the molecular design of each isoform-specific ID is represented by a molecular pattern with two sites that have structural colour '1' , '2' or '3' . We create a structural colour by interspacing an integer number of structural units. Each structural colour has a linearly increasing molecular weight with its 'colour number' (design Nanopore microscope identifies RNA isoforms with structural colours Filip Bošković and Ulrich Felix Keyser ✉ Identifying RNA transcript isoforms requires intricate protocols that suffer from various enzymatic biases. Here we design three-dimensional molecular constructs that enable identification of transcript isoforms at the single-molecule level using solid-state nanopore microscopy. We refold target RNA into RNA identifiers with designed sets of complementary DNA strands. Each reshaped molecule carries a unique sequence of structural (pseudo)colours. Structural colours consist of DNA structures, protein labels, native RNA structures or a combination of all three. The sequence of structural colours of RNA identifiers enables simultaneous identification and relative quantification of multiple RNA targets without prior amplification. Our Amplification-free RNA TargEt Multiplex Isoform Sensing (ARTEMIS) method reveals structural arrangements in native transcripts in agreement with published variants. ARTEMIS discriminates circular and linear transcript isoforms in a one-step, enzyme-free reaction in a complex human transcriptome using single-molecule read-out.
details of structural colours are illustrated in Supplementary Fig. 1). The basic design of the structural unit ( Fig. 1b and Supplementary  Fig. 1) consists of a docking strand (shown in black), complementary to the specific sequence on the target RNA, and the overhang sequence (red). The imaging strand is complementary to the docking strand's overhang and contains a structure of fixed molecular   Fig. 1 | ARtemis identifies multiple RNA targets using structural colours and nanopore microscopy. a, rNA isoform-specific ID fabrication using structural colours (red, 1; orange, 2; and green, 3). Three rNA isoforms are linearized with complementary DNA oligos and labelled with an 'exon-specific' structural colour, thus creating rNA IDs (rNA-DNA hybrid) '12', '13' or '23'. each structural colour is composed of an integer number of structural units. b, The ID with ten different structural colours is read by passing it through the nanopore microscope. A structural colour consists of an integer number of structural units (0-10) that are placed sequentially and read as one structural colour. each structural unit is composed of the part that binds to a target and has the overhang (docking strand, black) and the imaging strand (red) that is complementary to the overhang and has a terminal structure (monovalent streptavidin or DNA cuboid). c, The nanopore microscope detects up to ten structural colours within the same molecular ID. An example nanopore event in which each structural colour is identified by the nanopore microscope is shown. I, current. d, Single-molecule read-out of structural colours and their identity as assigned in example nanopore events. e, Ten-colour ID heatmap indicates error rates for assigning structural colours with >85% accuracy. Sample size (N) is 60 nanopore events.
The sequence of structural colours made from RNA-DNA hybrids (RNA ID) can be conveniently read using a solid-state nanopore microscope 20 (Fig. 1b) as a rapid, enzyme-free and affordable alternative for both short-and long-read sequencing. ARTEMIS with its direct nanopore read-out avoids the technical artefacts of RNA-seq and the imperfections of the motor proteins used in nanopore sequencing 21 . RNA ID fabrication and identification do not require reverse transcription or preamplification; hence, it offers rapid one-step direct detection of native RNA targets in the whole transcriptome.
Using the well-characterized ability of nanopore microscopes to detect molecular weight 20,22 , we show, as in fluorescent microscopy, identification of multiple colours (Fig. 1b-e). We designed and tested structural colours with up to ten levels as shown in the schematic in Fig. 1b Tables 2 and 3). Our ten-colour palette is shown in Fig. 1b-e, demonstrating the simultaneous detection of ten colours at the single-molecule level in a nanopore microscope (Fig. 1c,e). Example events and design details for the ten-colour ID are shown in Supplementary Figs. 3 and 4; oligonucleotides are listed in Supplementary Tables 2 and 4. The number of possible colour combinations depends on the length of the RNAs. We estimate that up to 10 10 colour combinations are possible combining all structural colours (Fig. 1b-e), highlighting the feasibility of ARTEMIS for transcriptome profiling. We validated the fabrication of IDs with biotinylated 'imaging strand' using polyacrylamide gel electrophoresis (PAGE) with and without the addition of neutravidin ( Supplementary Fig. 4). Additionally, we verified the correct assembly of ten colours with fluorescence quenching using fluorescein (6-FAM)-labelled structural units (Supplementary Fig. 5a). IDs were fabricated in equimolar concentration, each containing only one colour from '1' to '10' (Supplementary Figs. 5 and 6a). The fluorescence output of each separate colour indicates accurate fabrication of structural colours ( Supplementary Figs. 5b and 6). We show that assigning the accuracy of a detected colour to a designed colour is >85% accurate and that for colours '1' , '2' , '3' and '4' is 97% (Fig. 1e). Our ability to design IDs with specific colours at the predicted positions enables the low error rate (Supplementary Fig. 7).
Multiplexed RNA identification in one pot reaction. We next used ARTEMIS to identify various transcripts in a one pot reaction as  Supplementary Tables 5 and 6, respectively) and an external RNA ID control from MS2 bacteriophage with a known concentration (oligonucleotides are listed in Supplementary Tables 7 and 8 with the design details in Supplementary Fig. 9a). We successfully identified 18S and 28S rRNA in human total universal RNA (composition is listed in Supplementary Table 9) and human total cervical adenocarcinoma RNA. Each RNA ID was identified with the nanopore microscope; events for 18S rRNA ID with four sites ('1111'), 28S rRNA ID with five sites ('11111') and an external RNA ID control with three sites ('111') are depicted in Fig. 2b-d, respectively (additional events Supplementary Fig. 10). Expected velocity fluctuations during translocation 23 play a minor role in measuring distances between sites as we achieve the correct read-out and position sequencing of sites along the target RNA ( Supplementary Fig. 9c).
Two main obstacles for general RNA analysis are (1) degradation by nucleases assisted by magnesium ions and (2) structured regions that terminate the amplification or block hybridization. The former issue we addressed by replacing divalent ions with various monovalent ions. The removal of magnesium provides the added benefit of reducing the RNA structure stabilization and fragmentation for RNA ID fabrication ( Supplementary Fig. 11). We also determined the optimum salt concentration for RNA ID fabrication in our experimental conditions ( Supplementary Fig. 12). The stability and purity of the fabricated RNA IDs over time were assessed using nanopores and agarose gel electrophoresis ( Supplementary Figs. 13  and 14). We find that RNA IDs show no to minimal degradation with standard storage conditions, in agreement with the previous observations for RNA-DNA hybrids 24 . ARTEMIS discriminates RNA isoforms by their order, directionality, length and circularity. Based on the multiplexed detection, ARTEMIS discriminated transcript variants, which are a result of alternative transcript processing and structural arrangements in a premature transcript (pre-messenger RNA (pre-mRNA); Fig. 3). As a proof-of-principle, we designed exons and their respective isoforms (Fig. 3). We designed a sequence of three structural colours per exon (Fig. 3a), thus creating asymmetric and isoform-specific RNA IDs (designs with example events are presented in Supplementary  Fig. 15, and oligonucleotides used are listed in Supplementary  Tables 10 and 11). ARTEMIS identified order, length and circular isoforms (Fig. 3b-d, respectively). The combination of exons results in multiple transcript isoforms with the same length but different sequences, that is, IDs (Supplementary Fig. 16). In Fig. 3b, Fig. 3 | ARtemis discriminates engineered, alternative splicing isoforms resulting from any physical transcript arrangement. a, Isoform-specific labelling is achieved by labelling each synthetic exon (I, II, or III) with an asymmetric sequence of structural colours that results in unique IDs. b, example events of the ordered rNA isoforms that differ in the order and combination of structural elements (i.e. synthetic exons) as illustrated in a. c, rNA isoforms can differ in length, and so successful discrimination of the two length isoforms is shown. d, Detection of circular isoforms is shown by creating IDs on DNA scaffolds. The nanopore microscope can distinguish between circular and linear configurations. e, The nanopore microscope discriminates the linear and circular populations based on the translocation time (Δt), which is about two times shorter for the circular isoform, and the event current blockage (ΔI), which is about two times larger for the circular than for the linear ID. Sample size is 168 nanopore events.
three correctly identified isoforms with the same length but a different order of exons that contain either exons I and II ('211312'), exons I and III ('123112') or exons II and III ('312123') as shown in Fig. 3b. Even more, we can clearly determine the directionality of matching exons in isoforms by our asymmetric design of the three structural colours. Besides this, we demonstrate the identification of length isoforms for exon I (Fig. 3c and Supplementary Fig. 17). Another critical feature that is hardly achievable with RNA-seq is the discrimination of circular and linear isoforms (Fig. 3d,e). As a proof-of-concept, we used M13 phage DNA to create circular and linear IDs with the sequence of colours '111' using the same oligonucleotide mixture (Fig. 3d; event examples shown in Supplementary  Fig. 18, while oligonucleotides are in Supplementary Table 12). The scatter plot in Fig. 3e shows minimal overlap of two populations indicating the circular and linear IDs '111' . We fixed the position of colours in the circular ID conformation by using interlock oligonucleotides ( Supplementary Fig. 18c,d). Circularity discrimination is also assessed by in vitro RNA circularization ( Supplementary  Fig. 19) of linear MS2 RNA ID '111' using T4 RNA ligase I (ref. 25 ).
Programmable self-assembled RNA origami. Up to this point in our study, ARTEMIS has depended on the reshaping of RNA to a linear sequence of structural colours. Nevertheless, some transcripts contain strong RNA secondary structures that are challenging to remove with short oligo hybridization, or long transcripts greater than 10 kb (refs. 15,26 ). Especially the latter transcripts would require oligos that complement the whole RNA, which may be cost prohibitive. Instead, we decided to use these parts of the RNA as alternative structural colours present in the native target. We extended the ARTEMIS ID functionality by using short RNA motifs (Fig. 4a), long regions (Fig. 4b) or a combination of DNA and RNA structural colours (Fig. 4c). Terminal RNA origami ID 'AΩ'

Fig. 4 | structural colours by programmable RNA origami self-assembly. a,
The rNA origami ID is designed to have three internal rNA origami structures at the designed locations 'I', 'U' and 'Y' that represent a specific structure with a unique current downward signal. b, The ends of rNA do not need to be complemented with oligos and that can serve as terminal structural signatures. c, Only a part of the rNA is needed to assemble IDs. Here, we have self-assembled rNA terminal structures ('A' in green and 'Ω' in turquoise), rNA-DNA hybrid origami (grey-black) and self-assembled DNA double-hairpins (red). d, Predicted secondary (2D) and tertiary (3D) structures of designed rNA origamis to match letters 'I', 'U' and 'Y'. e, Heatmap indicates correct identification of 'I', 'U' and 'Y' with 99.4%, 99.1% and 99.2% accuracy, respectively. Sample size is 385 nanopore events. f, The 'A' and 'Ω' colours are identified with ~100% accuracy. Sample size is 3,099 nanopore events.
More specifically, we assembled RNA origami IDs by employing secondary structure formation in predesigned locations ( Fig. 4a; oligos are listed in Supplementary Table 13). Three structural colours were assembled by the nanoscale folding of 114-nucleotide (nt)-long, 190-nt-long and 342-nt-long single-stranded RNA to form structural colours 'I' , 'U' and 'Y' , respectively (predicted two-dimensional and three-dimensional structures 27 are shown in Fig. 4d; more details in Supplementary Fig. 20). As above, each self-assembled RNA 'origami' has a specific current signature that can be identified from nanopore events, as shown in Fig. 4a (additional events are presented in Supplementary Fig. 21). The accuracy of each structural colour identification is over 99%, as displayed in the summary in Fig. 4e.
Interestingly, RNA IDs can even be realized when only the middle part of a long RNA is linearized, as shown in Fig. 4b (oligonucleotides are listed in Supplementary Table 14). The two terminal RNA structures are 1,230 nt and 401 nt in length (' A' and 'Ω' , respectively). The asymmetry in the RNA ID is directly obvious during RNA ID translocation through a nanopore. The two terminal downward signals directly correspond to the terminal RNA colours 'Α' and 'Ω' (additional events are presented in Supplementary Fig. 22) with an accuracy in all the unfolded identification of ~100% (Fig. 4f). Finally, we designed a combination of ID '111' with a terminal RNA origami on both ends, as shown in Fig. 4c. The RNA ID traces in Fig. 4c show that DNA and RNA structural colours can be combined for RNA IDs. Hence, ARTEMIS can deal with secondary structures, and RNA length and accurate read-out is possible for any combination of structural colours. It is important to note that the addition of colours expands the number of unique RNA IDs.

RNA isoform relative quantification in a complex transcriptome.
Finally, we use ARTEMIS for the targeted identification of enolase (ENO) isoforms in commercially available human cervix adenocarcinoma total RNA (Fig. 5a,b). The ENO gene is known to have multiple transcript isoforms that differ in length or sequence as a result of alternative splicing of pre-mRNA 28,29 . We employed three colours to identify four transcript isoforms ( Fig. 5a; oligonucleotides are listed in Supplementary Table 15). RNA isoform ID designs and example events are illustrated in Fig. 5a. We determined the expression level of each ENO1 transcript isoform (Fig. 5b). An internal reference ID can further improve the transcript isoform-level quantification (more details in Supplementary  Fig. 8), and so we chose 18S rRNA as an intersample reference 30 . We confirmed that the nanopore event frequency is independent of the level of complementarity between the target RNA and oligonucleotides ( Supplementary Fig. 23). The absolute concentration may be calculated from the nanopore frequency of RNA ID events using a previously introduced model 31 .
As further demonstration, we used ARTEMIS to target X-inactive specific transcript long non-coding RNA (Xist lncRNA). Here we used terminal RNA, internal RNA motifs and DNA structural colours to identify length isoforms in the native transcriptome ( Fig. 5c-e). We targeted part of Xist RNA to fabricate ID '111111' (the design of Xist lncRNA ID is schematized in Supplementary  Fig. 24, and oligonucleotides used for its fabrication are enumerated in Supplementary Table 16). The part of the sequence that differs among long (L-isoform) and short (S-isoform) isoforms is left unpaired (Fig. 5c). The expected ID read-out should depict the six sites with a structural colour '1' , the terminal unpaired RNA coil ' A' and an internal self-assembled RNA origami colour 'L' as predicted from the sequence (Fig. 5c-e). We show typical examples of Xist lncRNA isoform IDs that match the predicted design and previously identified Xist lncRNA isoforms ( Supplementary Fig. 24) 29,32 .
In this study, we introduced ARTEMIS, an approach that reshapes an RNA target into a sequence of structural colours that we call an ID, using the subnanometre precision of DNA nanotechnology 33,34 . ARTEMIS omits amplification and enzyme-based steps and identifies multiple native RNA transcripts and alternative 50 pA 1 ms   a, Four identified ENO-like transcript isoform ID designs and example nanopore events. b, Detected events for each ENO1 transcript variant (T) for three individual nanopore measurements over 20 hours with total of 39,521 detected events; 18S rrNA '1111' was used as the internal control with 107 ± 12 events per hour. c, The Xist lncrNA L-isoform ID depicts six sites with colour '1', middle colour 'L' and terminal rNA coil 'A'. Nanopore events match their ID as depicted with colours above the downward signals. d, The S-isoform ID has the same design as the L-isoform except for the terminal rNA coil, which creates a difference in length between these two isoforms. e, both isoforms have between the third and fourth sites an internal self-assembled rNA origami indicated with colour 'L'. Predicted secondary (2D) and tertiary (3D) structures of colour 'L' are shown.
splicing variants in parallel using the nanopore microscope. As an electric measurement device, a nanopore microscope has a spatial resolution comparable to that of complex optical microscopies with higher throughput and straightforward origami assembly 20,33 .
Most diseases are classified by a change of a few transcripts. Thus, accurate identification of RNAs of interest has to bypass prior amplification and reverse transcription biases 7,11 . Our approach has the potential to identify extensive RNA diversity from a gene of interest without the need to align to reference transcriptomes. It is known that reference transcriptomes neglect unreferenced RNA diversity, and hence the extensive RNA variation is lost 35 . ARTEMIS is complementary to RNA-sequencing-based approaches as RNA IDs rely on RNA or genomic sequence information and do not provide de novo sequence information. We demonstrated multiple isoform identification by using the same oligo mix to identify order isoforms and length isoforms, as well as transcript circularity. ARTEMIS features are promising for characterizing therapeutic RNA uniformity and circularity using minimal sample amounts 2,36 . Amplification-based RNA sequencing technologies enable identification of RNA transcripts at the single-cell level 37 . Further developments are required to enable single-cell RNA characterization with ARTEMIS. Nanopore read-out of RNA at the single-cell level would require the integration of ARTEMIS and RNA handling with droplet-based techniques 37 .
We employed RNA origami self-assembly [38][39][40] as an additional way to identify transcripts of interest or to overcome some challenges of RNA analysis. Transcript IDs are assembled by using stable RNA structures as structural colours that can be either within the molecule or at its ends. We believe that ARTEMIS opens avenues for single-molecule mapping of RNA motifs. In addition, our study demonstrates that highly abundant natural RNAs may serve as scaffolds for DNA origami assembly with a wider length range and yield 38 .
Our multicolour palette paves the way towards targeted isoform profiling in the whole transcriptome that excludes enzymatic and amplification biases. ARTEMIS has the potential to create ~10 10 unique RNA IDs that are readable using nanopore microscopy or imaging methods that rely on super-resolution microscopy, including DNA-PAINT.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41557-022-01037-5.