Advancing synthesis-free and enzyme-free rewritable DNA memory through frameshift encoding and nanopore duplex interruption decoding

Abstract

DNA data storage is a promising alternative to conventional storage due to high density, low energy consumption, durability, and ease of replication. While information can be encoded into DNA via synthesis, high costs and the lack of rewriting capability limit its applications beyond archival storage. Emerging “hard drive” strategies seek to encode data onto universal DNA templates without de novo synthesis, using methods such as DNA nanostructures and base modifications. However, these approaches face challenges including complexity, low data density, enzymatic constraints, and reliance on costly instrumentation. Here, we introduce a DNA memory system based on frameshift encoding, inspired by viral ribosomal frameshifting, to enable rapid, cost-effective, and parallel data writing on a universal DNA template, without synthesis, enzymatic processing, or labeling. Information is encoded as checkpoint frameshifts by annealing microstaples of varying lengths at predefined sites along a long template strand. Data are decoded using MspA nanopore duplex interruption sequencing, which leverages a novel unzipping marker we discovered and frameshift-induced current signatures to resolve individual bits while sequentially unzipping tandem template–microstaple duplexes. Importantly, the duplex structure enables efficient, bit-specific rewriting through toehold-mediated strand displacement. This approach presents a scalable and versatile framework for DNA-based hard drives, with potential applications extending into in-memory computing, encryption, and dynamic biomolecular sensing.

Significance Statement.

This proof of concept for enzyme-free, parallel, and rewritable nonsynthesis-based DNA memory was achieved through the integration of noncovalent DNA manipulation and nanopore biosensing methodologies, showcasing a versatile and innovative application of gene material as a medium for advanced information technologies. It paves a way for future research in two key directions: (i) the development of real-time, robust, and high-capacity DNA memory systems capable of precise dynamic information processing while encoding millions of bits of data and (ii) the exploration of transformative applications across interdisciplinary fields, such as in-memory computing, secure data encryption and decryption, diagnostic biomarker profiling, and synthetic biology gene recording.

Introduction

DNA, as a coded biopolymer, serves as a versatile information carrier with potential applications in computing (1–4), cryptography (5–7), multiomics detection (8–11), and gene recording (12–15). DNA data storage is an emerging technology that encodes digital information into DNA sequences, offering significant advantages over conventional storage methods, including exceptionally high data density (16–28), low energy consumption (16), long-term durability, and ease of replication. Current DNA data storage approaches write information through direct synthesis of DNA fragments (19, 21–25, 28–30) and retrieve information using sequencing tools (31, 32) such as Illumina next-generation sequencing (33–35) and Oxford nanopore sequencing (36–42). These synthesis-based methods achieve remarkable data densities (16, 18, 29), but are costly and slow, and lack the ability to recode or rewrite data, limiting their use to static, long-term archival storage.

To address these limitations, nonsynthesis-based DNA data storage has been advanced that encode information onto presynthesized universal templates through DNA manipulation without the need for de novo synthesis. One strategy utilizes DNA nanostructures to encode information. For instance, DNA origami shapes have been characterized to act as barcodes for biomolecular detection (43, 44), and distinct label patterns on DNA origami have been engineered to encodes 48-bit digital data (45) and 9-bit data for cryptographic applications (7) on single DNA molecules. These methods enable information encoding on a universal template such as M13 DNA, but face limitations in data capacity while requiring costly tools such as atomic force microscope (AFM) and superresolution imaging for information retrieval. Alternatively, M13 DNA encoded by proteins or DNA nanostructures can be decoded in a glass nanopores (26, 46), enabling rewriting but still with low data capacity (8 bits). Recently, novel methods have been developed (47–49) that use oligonucleotides as guides to introduce data-specific modifications on template DNA. For example, the “DNA punch card” method encodes information by creating single-strand nicks at designated bit sites on bacterial genome DNA for data writing (47) while erasing information by removing these nicks for rewriting (48). During the review process of our current work, the “epi-bits” approach was developed that employs oligonucleotide-guided methylation to encode information on designed long templates, with nanopore sequencing used for data retrieval (49). But DNA modifications in all these multistep encoding methods rely on enzymatic reactions, posing challenges related to special conditions (e.g. high temperature), encoding efficiency, completion, specificity, a limited number of target sites, and other enzyme-associated limitations (50). Overall, while each approach has its own advantages and limitations, as analyzed below, they collectively inspire the exploration of novel encoding and decoding methods with the potential to advance applicable dynamic DNA memory.

Our long-term goal is to develop rapid and low-cost DNA memory systems that operate through biologically inspired mechanisms. Such systems can write, read, erase, and rewrite information on universal templates without requiring enzymatic reactions or labeling. With these capabilities, such DNA memory can extend beyond static archival storage to enable broader applications. A well-known biological mechanism that informs our approach is programmed ribosomal frameshifting in retroviruses, where a shift in the codon reading frame allows a single mRNA to encode distinct protein sequences (51–53). Inspired by this natural mechanism, we introduce frameshift encoding—a synthesis-free, enzyme-free, and label-free writing method. This approach uses the simple annealing of microstaples of varying lengths to a template strand to introduce checkpoint frameshifts that represent distinct bit values. As a result, it enables rapid, cost-effective, and parallel writing and rewriting of both binary and multinary information. Data encoded on the template are decoded using nanopore duplex interruption (DI) sequencing (54, 55), a platform capable of distinguishing individual frameshift bits by sequentially unzipping tandem template–microstaple duplexes, without requiring full-template sequencing. We discovered a novel nanopore unzipping marker that significantly enhance the accuracy of DI sequencing, enabling the resolution of hundreds of consecutive unzipping events along a long DNA template, even when reading repeated bits with identical or indistinguishable current signatures. Finally, this DNA memory system achieves efficient, bit-specific erasure and rewriting within minutes through toehold-mediated strand displacement (TMSD). Overall, this work significantly advances the development of dynamic DNA-based storage systems with broad potential applications.

Results

Frameshift encoding and nanopore decoding on designed templates

This DNA memory system encodes information onto universal DNA templates by annealing a set of microstaple oligos, without requiring any modifications to the template (Fig. 1a). Each microstaple represents a single bit, with bit values (0/1) determined by the binding of microstaples of different lengths at the 3′ end (Fig. 1b). Encoded information can be decoded using enhanced nanopore DI sequencing (54), which read bit values by identifying each microstaple at its designated bit site (Fig. 1b). Prior studies on MspA sequencing (37, 56), DI sequencing (54) and its applications (55), along with our single-nucleotide frameshift experiment (Text S1, Fig. S1, Table S1), suggest that the template–microstaple duplex modulates ion current through a “checkpoint” region—four unpaired nucleotides preceding the duplex and adjacent duplex bases in the constriction. Length differences between “0” and “1” microstaples form a frameshift in the checkpoint sequence, resulting in distinct current changes for differentiating them (Fig. 1c). As voltage-driven unzipping progresses, each successive duplex enters the constriction, generating a new current state. This sequential unzipping process produces a string of current states, which identify all the microstaples on the template for decoding the entire data (Fig. 1d).

Fig. 1.

DNA memory system utilizing frameshift encoding and nanopore decoding. a) Parallel encoding achieved by annealing a set of microstaples to a universal template; b) frameshift encoding and nanopore decoding: the binding of “0” and “1” microstaples, which differ by a single nucleotide in length, induces a ± 1 frameshift in checkpoint sequences. When pulled into the nanopore, these checkpoint sequences occupy the nanopore constriction, generating distinct current states that enable differentiation between “0” and “1” microstaples for data decoding; c) representative scatter plots of blocking levels (I/I₀) versus block durations (t), illustrating distinct current states for “0” (squares marked by 0) and “1” (circles marked by 1) microstaples, preceding current states (unmarked squares and circles), and unzipping markers (triangles); d) nanopore sequential bit decoding procedure: i, a microstaple is identified based on the corresponding current state, primarily determined by the checkpoint sequence in the nanopore constriction; ii, unzipping of the template–microstaple duplex produces an unzipping marker (triangles) that separates adjacent microstaple current states; iii, the next checkpoint moves into the constriction, allowing identification of the next microstaple from its current state.

One of the challenges for the nanopore decoding method described above was that identified current state patterns alone are limited in separating repeated bits, such as 000 and 111, since consecutive bits produce highly similar or even indistinguishable current signals. This issue can be addressed by incorporating unique “unzipping markers” we discovered, which potentially enhance nanopore decoding sensitivity and accuracy. The unzipping marker is characterized by a distinct reduction in blocking level (Fig. 1c). It was consistently observed at the end of an unzipping current state, indicating the completion of one duplex and the initiation of the next microstaple reading (Fig. 1d). The generation of unzipping markers may be associated with the presence of the R118 ring in the commonly used MspA M2 pore, as previous DI sequencing and its application studies (54, 55) that utilized MspA variants lacking the R118 ring did not report the appearance of such unzipping markers. This explanation aligns with our earlier findings that the MspA pores containing R118 can dock single aptamers within the pore lumen (57), enabling the detection of DNA and RNA conformational changes in response to neurotransmitters and small-molecule therapeutics (57). Together, these observations suggest that once a microstaple is unzipped from the template, it may be temporarily retained within the pore, cooccupying the lumen alongside the subsequent duplex. This interaction may explain the distinctly lower blocking level of unzipping markers compared with configurations where only the duplex is present in the pore. Regardless of the exact mechanism, the unzipping marker is essential for accurately identifying each microstaple bound to the template, particularly in distinguishing identical or highly similar current states generated by repeated checkpoint sequences, thereby enhancing decoding precision.

We first designed a 3-bit DNA memory to demonstrate the encoding and decoding processes. Each bit site accommodated “0” or “1” microstaple (Fig. 2a). The “1” staple is 4-nt shorter than the “0” staple at the 3′ end. Upon binding to the template, this length difference creates a 4-nt frameshift between their checkpoint sequences (TTTT for “0” and CCCC for “1”). We encoded all eight possible 3-bit numbers (000 to 111) by annealing the corresponding microstaples at designated bit sites and then read these numbers by differentiating “0” and “1” microstaples using the nanopore.

Fig. 2.

Writing and reading on a short-templated DNA memory. a) Sequences showing the binding of three “0” or “1” microstaples to a universal template, representing encoding 0 or 1 at each bit. Binding “0” or “1” microstaples generates checkpoint sequences TTTT and CCCC respectively through frameshift. b) Storing eight 3-bit numbers in the DNA memory by hybridization of corresponding “0” or “1” microstaples to the universal template. c) Nanopore signatures for decoding eight 3-bit numbers. The unzipping markers (indicated by triangles) separated each signature into three tandem states, and the nanopore blocking level of each state generated by the checkpoint sequence read the identity of each microstaple, decoding the value of each bit (0 or 1). d) Nanopore decoding of an 8-bit number (10110010). Unzipping markers can separate consecutive microstaples that generate the same checkpoint sequences, enabling decoding of bit repeats, such as 00, 11, 000, and 111. Nanopore currents were recorded at 100 mV in 1 M KCl and 10 mM Tris (pH 7.4). The sequences and nanopore current state blocking levels are provided in Table S2.

The resulting nanopore signatures for all eight 3-bit numbers clearly exhibited two distinct current states: level 0 (I/I₀ = 0.184 ± 0.006 to 0.194 ± 0.002) for the “0” microstaple and level 1 (I/I₀ = 0.225 ± 0.008 to 0.242 ± 0.006) for the “1” microstaple (Fig. 2b, Table S2), demonstrating accurate differentiation of the microstaples encoding 0 and 1 at each bit site. Further, through combined analysis of current state patterns and unzipping markers, we found that all nanopore signatures consistently exhibited three consecutive current states, separated by two unzipping markers (Fig. 2b). Each state corresponded to either level 0 or level 1, indicating the binding of a “0” or “1” microstaple to the template, enabling precise reading of all eight 3-bit binary numbers. Moreover, this analytical pipeline ensured accurate decoding in an 8-bit DNA memory system, even when repeated bit sequences were present (Fig. 2c, Table S2). In conclusion, these findings validated the efficiency and accuracy of both frameshift encoding and nanopore decoding processes in this DNA memory.

Encoding and decoding different datasets on M13 genome DNA template

To explore the feasibility of encoding and decoding bit-level data across a long, native DNA scaffold, we employed linearized viral M13mp18 genomic DNA (7,219 nt) as the template (Text S2, Figs. S2 and S3). Based on the M13 sequence, we designed two sets of 284 microstaple oligonucleotides (15–25 nt each), Set1 and Set2, to simulate two distinct datasets (Text S2, Fig. S4, Table S3). These microstaples were optimized for melting temperature uniformity, balanced GC content, and minimized off-target hybridization. The key difference between Set1 and Set2 lies in a subset of 29 “frameshift microstaples,” each differing by a single nucleotide at the 3′ end. When annealed to the M13 DNA scaffold, these frameshift microstaples introduce either a + 1 or −1 nucleotide shift in the checkpoint sequence. These staples are evenly distributed along the M13 backbone, spaced approximately every 9–11 microstaples from the 5′ to 3′ end. Successful differentiation of these frameshift microstaples using nanopore DI sequencing would demonstrate the system's ability to write and read discrete, user-defined data values at specific positions along a long DNA template. The remaining microstaples are identical in both sets. Although these staples do not encode information in this experiment, their detection via nanopore DI sequencing supports accurate sequence localization, alignment, and assessment of overall writing efficiency and full data recovery.

The nanopore signature revealed a series of consecutive current states. Most of them terminated with an unzipping marker (Video S1, Fig. 3b), consistent with observations from 3- and 8-bit DNA memory (Fig. 2). By analyzing current state patterns, such as amplitude and noise level, and unzipping markers, we determined each unzipping-induced current state. A state-to-state comparison between Set1 and Set2 allowed us to assign the observed current states to their corresponding microstaples. Frameshift microstaples generated distinct current states, while nonframeshift microstaples exhibited identical current state patterns between Set1 and Set2. For example, microstaples 264–284 were identified from a truncated nanopore signature (Fig. 3b). Distinct blocking levels of two current states in Set1 (top trace) and Set2 (middle trace) were identified, and they should be assigned to frameshift microstaples 271 and 283, which have +1 or −1 frameshift in their checkpoint sequences. In contrast, no significant blocking level difference was found for other observed current states, enabling the assignment of these states to the remaining microstaples 264–270, 272–282, and 284, which form identical checkpoint sequences in both datasets. Additionally, comparing the nanopore signatures of Set1 and Set1_Sub, in which the last 12 staples were removed from Set1 (Fig. 3a, Table S3), showed that the nanopore signature of Set1_Sub properly terminated at staple 272 instead of 284, while maintaining the same microstaple assignment as Set1 for all preceding microstaples (Fig. 3b, bottom trace). The difference identified for the last 12 microstaples between Set1 and Set1_Sub confirms the nanopore’s capability to decode the entire dataset on the M13 DNA.

Fig. 3.

Nanopore identification of frameshift staples for decoding different datasets encoded on the M13 DNA template. a) Out of 284 staples that bind to the M13 DNA, 29 were selected to construct frameshift staples, which form single-nucleotide frameshifted checkpoint sequences codes at each address in Set1 and Set2, representing different values (e.g. 0 and 1) written to 29 bits in two datasets. Set1_sub is a subset of Set1 without staples 273–284, used to confirm staple reading accuracy. The staple properties, including the sequences, checkpoint sequences, and melting temperature are provided in Table S3, and the sequences showing the binding of Set1 and Set2 to the M13 DNA is illustrated in Fig. S4. b) Truncated nanopore signatures for reading staples 264–284 in Set1 (top trace), 264–284 in Set2 (middle trace), and 264–272 in Set1_sub (bottom trace). Identified current states and corresponding staple IDs are labeled. The current states for frameshift staples 271 and 283 are marked in Set1 and in Set2. Frameshifted checkpoint sequences generated by the two staples are also illustrated. Data were recorded at 100 mV in 1 M KCl and 10 mM Tris (pH 7.4).

By aligning the complete signatures of Set1 (n = 24) and Set2 (n = 23), we identified 246 current states associated with staple binding (Fig. 4a and b, Table S4). This corresponds to 87% of the total staples that are designed to bind the M13 DNA. Analysis of the I/I₀ for these current states (Fig. 4a) allowed us to calculate the blocking level difference (ΔI/I₀) for each current state between Set1 and Set2 and identify frameshift staples feathering significant ΔI/I₀ values (P ≤ 0.05, Fig. 4b, Table S4). Through this method, we successfully identified 21 unique current states generated by the frameshifting staples, which exhibited significantly different blocking levels between Set1 and Set2, as identified from their nanopore signatures (Fig. 4c, Table S4). Notably, complete data reading necessitates the analysis of multiple signatures (n > 10) since not all staple current states were present in every nanopore signature (P_Hybrid < 100%, Fig. 4d).

Fig. 4.

a) Profiles of blocking levels (I/I₀) of identified current states for staples in Set1 and Set2. b) Blocking level difference (ΔI/I₀) between Set1 and Set2, calculated from the profile in a). Frameshift staples were identified based on significance in difference (P-value). c) Representative single pore current traces showing the identification of frameshift staples in Set1 (left) and Set2 (right), indicated by lines. Upward and downward arrows above the current states for frameshift staples in Set2 indicate an increase or a decrease in blocking levels compared with those in Set1. All current recordings shown in the figure were extracted from full signatures for unzipping of all staples from the M13 DNA template. Single pore currents were recorded at 100 mV in 1 M KCl, 10 mM Tris (pH 7.4). d) M13 DNA/staple hybridization probability (P_Hybrid) for identified staples. P_Hybrid is defined as the fraction of signatures that contain the current state for a specific staple. e) Calculated single-strand probability (P_SS) of staple-binding domains in the M13 DNA. f) A trend of positive correlation between P_Hybrid and P_SS.

To explain the remaining 13% of undetected staples whose current states could not be identified in any signature, we first ruled out the possibility of off-target cleavage during linearization by comparing the signatures for Set1 with Set1_Sub (Fig. 3b and d), as Set1_Sub properly terminated at staple 272 instead of 284, the last staple in Set1. Instead, we hypothesize that these undetected staples likely failed to bind to the template due to competitive self-hybridizations in the template (Text S2, Fig. S5). To test this hypothesis, we calculated the probabilities for staple-binding domains to be single-stranded (P_SS), which indicates their availability for staple binding (Fig. 4e, Text S2, Table S3). Domains without self-hybridizations are more accessible for staple binding. We observed a positive trend between P_Hybrid and P_SS (Fig. 4f), and the undetected staples exhibited notably low P_SS values, indicating a low probability of binding for these staples. These findings highlight that self-hybridizations throughout the long template could impede the binding of microstaples by competing for hybridization.

In summary, we have showcased the effectiveness of frameshift encoding and nanopore decoding on long templates. Specifically, (i) through combined analysis of current states and unzipping markers, the nanopore can efficiently unzip sequential template–microstaple duplexes across the entire template chain, achieving complete retrieval of the encoded data; (ii) nanopore can sensitively distinguish between microstaples generating single-nucleotide frameshifts at a specific site, enabling accurate reading of different values stored in a bit; and (iii) the future study can improve microstaple binding through careful design of long template sequences or selection of native genomic sequences with minimal self-hybridization.

Multinary DNA memory through single-nucleotide incremental frameshift encoding

Building on the above study of binary DNA memory, we now explore the use of multiple microstaples of varying lengths to introduce multiple frameshifts at the same site. If these frameshifts generate distinct nanopore current states, it would allow differentiation of their corresponding microstaples, enabling the decoding of multiple values from a single digit (Fig. 5a). This multinary DNA memory system (58) holds potential for increasing data storage density and applications in DNA computing, including the implementation of fuzzy logic circuits (59, 60).

Fig. 5.

Multinary data reading using nanopore recognition of multiple single-nucleotide frameshift staples at each address. a) Sequence of a four-address template and six sets of 4 staples. The binding of each set of staples to their addresses represents the encoding of four data bits. Vertically, six staples binding at each address form six codons with frameshift at 1-nucleotide increment (quadramers), representing writing six different values to each bit. b) Nanopore signatures showing sequential recognition of the four staples via unzip-sequencing for the six sets of staples. The tandem current states for the four sequential staples are marked by different level lines. Current amplitude histograms are shown for extracting the average blocking levels for the four checkpoint sequences. c) Blocking levels (I/I₀) for the six codons at each address formed by six single-based frameshifting staples. The nanopore discrimination capability of the codon pairs among the six codes was analyzed by Tukey’s multiple comparison test and ranked as highly discriminable (**P < 0.001), discriminable (*P < 0.05), and indiscriminate (NS, not significant) (Table S4). Codon pair comparisons not shown in the plot are highly discriminable.

We utilized a four-address DNA memory to study multinary encoding/decoding (Fig. 5a, Table S5a). At each address (i = 1 to 4), six staples (j = 1 to 6) were designed. These staples were successively elongated or shortened by one nucleotide from the 3′ end. Upon binding to the address, they generated six distinct checkpoints that were frameshifted in increments of a single nucleotide (Fig. 5a). For instance, when the six staples (staples 11–16) were bound to address 1, they formed six single-nucleotide frameshifted checkpoints: GTTT, TTTT, TTTA, TTAC, TACA, and ACAA (Fig. 5a), representing the encoding of each address as six different values. A total of 6 × 4 staples were employed to encode six 4-digit datasets, namely Set1–Set6 (Fig. 5a, Table S5). Each dataset was encoded onto the template by binding one of the six staples to each address. For example, Set1 was encoded using staples 11, 21, 31 and 41, while Set6 was encoded using staple 61 through staple 64.

The analysis of nanopore signatures for six datasets revealed four consecutive current states, identified by examining both the unzipping markers and blocking levels of nanopore current states (Fig. 5b, Table S5b). To investigate the nanopore’s ability to recognize the six single-nucleotide length-varying staples at each address, a Tukey’s multiple comparison test was conducted, aiming to determine the nanopore’s capability to distinguish the blocking levels of current states for each pair of staples. The results were ranked as highly distinguishable (P < 0.001), distinguishable (P < 0.05), or indistinguishable (not separable) (Fig. 5c, Table S5c). Specifically, for address 1, all staple pairs, except the staples 12/16 and staples 16/14 pairs, were found to be highly distinguishable or distinguishable. Therefore, except for staple 16, the other five staples demonstrated distinguishable nanopore signatures. Regarding address 2, all staple pairs, except for the staples 23/24/25 combinations, were highly distinguishable or distinguishable. This suggests that staples 21, 22, 26, and either staples 23, 24, or 25 can be distinguished from each other. For address 3, only the staple 33/36 pair was found to be indistinguishable, indicating that the five staples at this address, including staples 31, 32, 34, 35, and either staple 33 or 36, are distinguishable by the nanopore. Lastly, for address 4, the indistinguishable pairs between staples 43, 44, and 45 suggest that a set of four staples, including staples 41, 42, 46, and either staples 43, 44, or 45, can be distinguished by the nanopore.

This result indicates that the nanopore can differentiate most of the six single-nucleotide frameshift microstaples at each address. This demonstrates the feasibility of using these microstaples to encode and decode multiple values per digit, highlighting the potential for multinary data encoding and decoding. At a minimum, it is possible to implement a quaternary DNA memory by utilizing four distinguishable staples at each address to represent four values for encoding each digit (Table S5d).

“Formatting” DNA memory for rapid, parallel, bit-specific rewriting

Frameshift encoding offers the advantage of writing diverse data on a universal DNA; however, it confronts challenges when applied to DNA memory. Firstly, frameshift encoding lacks the ability to rewrite data. Secondly, the annealing-based writing procedure involves a time-consuming temperature ramp spanning several hours or longer, rendering it unsuitable for rapid, dynamic data processing essential in memory applications. Addressing these limitations, TMSD has emerged as an enzyme-free approach enabling fast, high-yield strand exchange at room temperature in nucleic acid hybridization reactions (61–63). This technique finds extensive utility ranging from DNA computing (3, 4, 64, 65), to gene expression manipulation in synthetic biology (66–68) and clinical biosensing (69–72) (including solid state nanopore (72)). In this context, we have combined frameshift encoding with TMSD to achieve transient, parallel, and specific bit-level data writing, erasing, and rewriting (Fig. 6, Table S6), opening up avenues for diverse profound applications.

Fig. 6.

DNA memory with frameshift encoding and TMSD for quick, bit-specific, and all-in-one rewriting. a) A 4-bit template with three data bits and a terminal reference bit. Frameshifted codons CAGC and CCAG are formed by “0” and “1” staples at each address, generating different current state blocking levels for recognizing “0” and “1” staples. “1” staple with a toehold can transiently replace the “0” staple through TMSD to erase “0” and rewrite with “1.” Nanopore signatures for a blank (or “all-zero”) DNA memory constructed by annealing three “0” staples and one reference staple at their addresses (b), writing “1” by erasing “0” with the “1” staple at addresses 1 (c), 2 (d), and 3 (e), respectively, and at all three addresses (f). “0” and “1” staples can be discriminated by their current state blocking levels at level 0 and level 1. Data were recorded at 100 mV in 1 M KCl and 10 mM Tris (pH 7.2). g) Blocking levels of level 0 and level 1 at addresses 1, 2, and 3. h) 0→1 Rewriting efficiency (E_RW) at each address.

The model system consists of three data bits and one reference bit (Fig. 6a). We initially fabricated a “formatted” or “blank” DNA memory by hybridizing the template with the “0” staples, resulting in all three data bits being set to “0.” The “all-zero” DNA memory exhibits four tandem nanopore current states separated by the unzipping markers (Fig. 6b). The first three “0” staples formed a common code CACG, generating three level 0 states (I/I₀ = 0.190–0.197, Fig. 6g, Table S6), while the reference staple produced a terminal state with a step-up level (I/I₀ = 0.281 ± 0.002, Fig. 6g). Therefore, the blank DNA memory reads out as 000.

When encoding data, the user can simply erase specific “0” staples and rewrite them with a collection of “1” staples via a transient, all-in-one TMSD reaction. Each “1” staple first binds to the single-stranded section of the template via a 5-nt toehold and then immediately displaces the corresponding “0” staple. The binding of each “1” staple results in a −1 frameshifted code CCAC (Fig. 6a), leading to an elevated current state at level 1 (Fig. 6c–f). To test fast, address-specific rewriting, we incubated each “1” staple, 1t, 2t, or 3t, with the “formatted” DNA memory template at room temperature for 5 min. The nanopore signatures immediately reveal an increase in conductance to level 1 for the first, second, or third current state (I/I₀ = 0.222–0.240, Fig. 6c–e and g), while other current states, including the terminal reference state, remain unchanged. This result verifies the 0→1 staple displacement at each address (or bit), allowing for the quick writing of 100, 010, and 001 datasets in the DNA memory within minutes. To test the capability of multibit rewriting, we followed the same incubation procedure as above to anneal three “1” staples together with the blank DNA memory. This time, all three of the first current states were elevated to level 1 (Fig. 6f), while the reference state remained unchanged. Thus, the 0→1 displacement occurred in all three data bits, verifying the fast writing of 111 in the DNA memory.

Finally, we assessed the strand displacement efficiency Eff_SD from “0” to “1.” Eff_SD was measured as the populational fraction of the nanopore signatures with the 0→1 displacement among all identified signatures with the four addresses bound by the staples. E_RW was found to be 0.92 ± 0.05, 0.91 ± 0.07, and 0.85 ± 0.12 for all three data bits, respectively, at a template/staple concentration ratio of 1:2 (Fig. 6h, Table S6). These results indicate that averagely 90% of the microstaples can be displaced in just a few minutes, verifying TMSD as an enzyme-free, label-free, fast-kinetics, and high-yield approach for efficient, room-temperature writing and rewriting at specific addresses. In our TMSD rewriting model (Fig. 6), the interstaple strand is designed to enable a single rewriting event at a given bit by binding to a toehold extension on the microstaple, typically around six bases in length. To support multiple rounds of rewriting at the same position, the intermicrostaple region should be sufficiently long to accommodate multiple toeholds, one for each rewriting cycle. However, this design requirement implies a trade-off: increased rewriting capability comes at the expense of reduced data storage density.

Discussion

Recently, novel synthesis-free DNA data storage methods have been developed (47, 49). These approaches use oligonucleotides to guide enzymes in introducing data-specific modifications on template DNA. For instance, the “DNA punch card” method encodes information by creating nicks at designated bit sites on the template (47), while the “epi-bits” approach utilizes oligo-guided base methylation to encode bit values (49). In contrast, frameshift encoding employs a parallel mix-then-write process, eliminating challenges associated with DNA modifications, such as variable modification yields that can affect writing completion. Therefore, this method enhances accuracy and reliability while remaining simple and rapid. In addition, the use of unmodified microstaple oligos in frameshift encoding significantly improves writing cost-efficiency by at least several times compared with approaches such as epi-bits storage ($0.00325/bit–0.00916/bit) (49), which requires a large number of methylated oligonucleotides, and currently is higher than synthesis-based DNA data storage methods ($0.0007/bit) (28). Importantly, after encoding, the resulting structure of template/microstaples hybrids supports single-step strand displacement via toehold design. This enables efficient (within minutes), high-yield (∼90%), room-temperature, and bit-specific information rewriting. These findings support further research on rewriting capability and applications on long DNA templates.

Our frameshift encoding achieves a data density of ∼25 bases per bit, depending on microstaple length and intermicrostaple distance. This density is comparable to that of published oligonucleotide-mediated DNA data storage methods (47, 49) around tens of bases per bit, but significantly higher than that of earlier M13 DNA nanostructure-based encoding methods, which store 8 bits (7, 73) or 64 bits (45). Currently, synthesis-based data storage methods remain achieving the highest density of 0.5 bases per bit (21, 28). We successfully identified 246 out of 284 microstaples designed to hybridize with the M13 DNA scaffold. Of these, 29 were frameshift microstaples intended to encode individual data bits, with 21 out of 29 bits accurately encoded and decoded. While the data density of our frameshift encoding approach is comparable to that of recently developed DNA punch card (47) and Epi-bit (49) systems, those platforms have achieved significantly greater storage capacities (10⁴–10⁵ bits) by using native long genomic DNA or custom-designed barcoded templates. These comparisons highlight the need to develop custom long DNA templates for frameshift encoding, enabling its scalability toward high-capacity DNA-based memory systems.

Information encoded on the template can be decoded through nanopore-based sequential unzipping of template/microstaple duplexes. Numerous studies have investigated the unzipping mechanisms of double-stranded nucleic acids in nanopores (37–39, 42, 56, 74–84). The multiple DNA duplex unzipping mechanism in the MspA pore serves as the foundation of nanopore DI sequencing (54) and its applications such as barcoding for biomolecular detection (55). However, these studies were limited to distinguishing only a few current states in short DNA fragments and faced challenges in resolving consecutive, indistinguishable states in longer DNAs. In this study, the identified “unzipping marker,” combined with current pattern attributes such as blocking levels, provides an effective strategy to identify large-scale microstaple bits sequentially distributed along long templates, including those with indistinguishable current states (Text S2).

Analysis of nanopore sequential unzipping allows for the direct identification of 87% of microstaple bits from M13 DNA unzipping signatures. The previously reported Epi-bit approach utilized conventional nanopore sequencing to determine 90% of methylation-encoded bits directly from raw nanopore recordings and 97% after algorithmic enhancements (49). Our findings (Fig. 4d) further support that the “missed” microstaples not identified by the nanopore could result from the formation of complex self-hybridized structures in M13 DNA, which prevent microstaples from binding to the template. An effective solution, as validated by the Epi-bit method (49), is the use of designed templates whose sequences can be optimized with minimized self-hybridization to enhance microstaple-binding probability.

Based on duplex unzipping times using synthetic templates (Figs. 2, 5, and 6), we estimate the readout rate of nanopore DI sequencing to be ∼10 bits per second, depending on the duplex length and sequence composition. This speed is comparable to that of the Epi-bits data storage method (49), which employs Oxford Nanopore sequencing (∼400 bases per second per pore (85)) to read methylation states at each bit. With an average spacing of 25 bases between bits, Epi-bits could achieve a decoding rate on the order of tens of bits per second. In contrast, unzipping all template/microstaple duplexes along the M13 DNA (∼80 s) is frequently interrupted by long, clogging-like current states of unknown identity (see Video S1). We speculate that these undesired events may be mitigated through the use of optimized, custom-designed long templates, which could minimize such interruptions and significantly accelerate the bit readout process.

At its core, this parallel, rewritable DNA memory functions as a sensor—capable of dynamically detecting numerous target oligonucleotides that are collected and exchanged on a single template molecule, much like a molecular sponge. This unique capability may open avenues for translational applications beyond data storage. For instance, in a decision-making system, the DNA memory could simultaneously quantify large populations of microstaples produced as outputs from DNA computing processes (3, 4, 65, 86), while also enabling direct in-memory computing—all integrated within a single DNA template. The DNA memory could find robust applications in encryption, where it can generate key sizes of hundreds of bits on a single template, meeting the advanced encryption standard 256-bit protocol. Additionally, the platform's versatility enables parallel detection of biomolecules, such as microRNAs (11, 80), genetic and epigenetic sequencing, and nucleic acid secondary structures (87–89).

Methods

The Materials and methods section for this work is provided in S2 Supplementary Methods, including: (i) DNA hybridization for short template experiments; (ii) M13 DNA linearization, purification, and sequencing; (iii) preparation of the MspA protein, which includes gene sequence design, plasmid synthesis, protein expression, and purification for MspA; (iv) methods for optimized design of microstaples for binding to the M13 DNA; (v) DNA memory chain assembling; (vi) nanopore single-channel recording, which covers lipid preparation, device setup, lipid bilayer formation, protein pore formation, electric recording, and bandwidth setting; (7) data analysis, which involves methods for analyzing current state properties (blocking level, duration, and noise level), alignment, unzipping marker, and statistical significance; and (8) methods for computing the single-strand probability for microstaple-binding domains.

Supplementary Material

Supplementary material is available at PNAS Nexus online.

Funding

This research was partially supported by the National Science Foundation Convergence Accelerator Phase I #0072474 and Phase II #2344877 (L.-Q.G.), the Missouri Spinal Core Injury/Disease Research Program (L.-Q.G.), the National Institute of Health R35 GM134919 (S.-J.C.) and U54 AI176060 (S.-J.C.), National Science Foundation CHE-2154924 (S.-J.C.), National Institute of Health R01 HL173025 (S.-Y.C.), R01 HL117247 (S.-Y.C.), and R01 NS069726 (S.D.). M.S. was a Research Experience for Undergraduates (REU) student supported by the National Science Foundation #1757936.

Author Contributions

L.-Q.G., K.T., and S.-J.C. conceived the research. K.T., S.Z., L.-Q.G., S.-Y.C., B.L., and S.D. designed the research. K.T. and S.Z. performed the research. K.T., S.Z., S.C., S.-J.C., L.-Q.G., R.G.C., C.H., E.M., J.R., and M.S. analyzed the data. L.-Q.G., K.T., S.-J.C., and S.Z. wrote the paper.

Data Availability

The authors confirm that the data supporting the findings of this study are available within the article and its supplementary materials.