Intracellular logic computing with DNA tetrahedron processors enables precision cancer theranostics Authors: Yang Gao, Yun Wang, Yunlong Qin, Geru Zhang, Shaojingya Gao, Wen Chen, Siyi Yang, Xiao Yang, Yuanlin Tang, Yao He, Yunfeng Lin, Itamar Willner, Taoran Tian, Xiaoxiao Cai Published: 2026 ABSTRACT AbstractThe integration of molecular computation with nanomedicine holds transformative potential for precision cancer theranostics. However, achieving an intelligent, multi-input decision-making process within living cells remains a formidable challenge because of the complexity of intracellular signaling networks. Here, we present a set of DNA tetrahedron logic processors that enable smart intracellular computing through a programmable three-input architecture capable of executing seven Boolean operations—OR, AND, NOR, NAND, XOR, majority (MAJ), and OR-AND—in response to endogenous miRNA signals. Building upon the cell-internalizable tetrahedral framework nucleic acid, this processor functions as an autonomous sense-and-treat module: it decodes combinatorial biomarker states and conditionally releases therapeutic siRNA only when a predefined logical condition is met. As a proof of concept, in precision oncology, we engineered a MAJ-gated nanoplatform (siR@MAJ) that selectively silences survivin in MCF-7 breast cancer cells that exhibit a specific tri-miRNA signature while sparing other cell types. This approach has achieved potent tumor suppression both in vitro and in vivo with high specificity and minimal off-target effects. By integrating multi-target detection into a single logic circuit, this processor significantly improves the discriminative capability within complex biological environments. Our work establishes a foundational platform for intelligent intracellular diagnostics and therapy, paving the way toward adaptive, logic-driven nanomedicine.Subject terms: Drug delivery, Nanobiotechnology Introduction Living cells continuously process multivariate molecular signals to make fate decisions—a form of natural intracellular computation. The cellular microenvironment represents an extraordinarily complex and dynamic landscape, teeming with a dense network of signaling molecules that not only coexist but also engage in sophisticated, interdependent interactions. This high-dimensional molecular crosstalk acts as a biological circuit, integrating both stimulatory and inhibitory cues to regulate critical cellular functions such as proliferation, differentiation, and apoptosis. This intricate system poses significant challenges for the deployment of theranostic agents—diagnostic and therapeutic tools designed to operate with high precision within this milieu. Conventional diagnostic strategies1–3 and responsive drug delivery systems4,5 often rely on targeting one or two key molecules as indicators or effectors. While this approach can provide certain insights and benefits, it falls short in regard to tackling the complex cellular behavior in reality. The reliance on single or dual biomarkers frequently results in high rates of false positives or negatives, as it ignores the combinatorial logic that governs true pathological states. By focusing on limited targets, these methods risk oversimplification of the disease state, leading to reduced accuracy in diagnosis and suboptimal therapeutic outcomes. Hence, there is a pressing need to develop methodologies capable of assessing the logical state between multiple target molecules within cells, which would allow for more accurate diagnostics and personalized therapies. The concept of logic operations has become a cornerstone in molecular diagnosis and treatment,6,7 with DNA oligonucleotides emerging as key players because of their exceptional editability and scalability.8–10 Through precise sequence design, DNA can be programmed to perform Boolean logic, mimicking the computational functions of electronic circuits but at the nanoscale. These molecules facilitate the design of biological logic gates, including branched structures11,12 and three-dimensional framework nucleic acids.2,13 For example, DNA nanomachines capable of performing Boolean operations can achieve multidimensional identification of target cells for tumor therapy, cell marking, and tissue imaging.14,15 Most studies have focused on two-input DNA logic gates,14,16,17 which recognize the logical states of two targets via OR or AND gates. Although a few pioneering efforts have demonstrated three-input DNA logic circuits in controlled, cell-free environments18,19 or have utilized cascaded two-input gates to process multiple signals,20 these systems have not been adapted for direct operation within the complex milieu of living cells. The biological barriers, including nuclease degradation, off-target binding, and the difficulty of intracellular delivery, have severely hampered the transition of multi-input logic systems from test tubes to living tissues. The ability to perform robust, multi-input logical analysis of cytoplasmic contents is crucial for clinical disease classification and precision medicine, where accurate diagnosis and therapeutic decisions increasingly rely on the combined detection of multiple biomarkers.21,22 However, developing a stable, biocompatible, and systematically programmable three-input logic nanoplatform that functions reliably in living systems remains a significant challenge. To overcome this barrier, we developed a DNA-based logic architecture capable of executing complex multi-input operations directly within cells and in vivo. To establish a logic processor featuring good scalability and easy intracellular operation, we designed logic operation structures based on tetrahedral framework nucleic acids (tFNAs), the simplest three-dimensional DNA nanostructure that has excellent endocytosis capacity.23,24 The structure of tFNA consists of several elements, known as vertexes, edges, three-way junctions, triangles, and nicks, which are also present in more complex framework nucleic acids, such as DNA polyhedrons, 3D origami, and DNA tiles. Therefore, the key reaction domains of the tFNA-based three-input logic gate system (TIGS) developed in this study can be effectively applied to other DNA nanostructures. In particular, the use of tFNAs as functional nanocarriers has enabled the smart release of loaded nanodrugs through structural reconfiguration in living cells,25–27 indicating the excellent potential to integrate logic recognition modules and responsive actions within simple devices. Furthermore, the precisely defined geometry of tFNAs allows for the spatial orientation of logic components, minimizing steric hindrance and maximizing the efficiency of molecular recognition events. Here, we demonstrate a comprehensive set of three-input logic processors (Fig. 1), including basic ones—OR, AND, NOR, NAND, and XOR gates—and two combinational gates—Majority (MAJ) and OR-AND (OA)—which are not typical but practical. All seven gates perform the desired logic operations and display computational outputs across the eight different states of the three inputs. Moreover, all gates retain the simple tetrahedral framework structure with excellent cell entry performance to ensure the successful completion of logic computations in living cells. As a proof of concept, we demonstrate that the MAJ gate can achieve precise regulation of survivin expression in tumor cells by combining the recognition of multiple miRNAs with the delivery of a specific siRNA. By requiring a consensus of multiple disease markers before activation, the MAJ gate ensures high-fidelity targeting, effectively distinguishing cancer cells from normal cells with similar but non-identical expression profiles. This therapeutic strategy is highly effective at suppressing breast tumor cell growth both in vitro and in vivo. Thus, TIGS represents a promising strategy for constructing DNA-based recognition modules and is therefore believed to provide insights into logical theranostics in complex intracellular environments. Schematic illustration of the DNA-based Threethree-input logic gate system (TIGS). Three various ssDNAssDNAs serve as three inputs and can be recognized by logic gates in TIGS. After structural changes occur during the logic operation, the output port would displaydisplays the result of a comprehensive response as a fluorescence signal. According toIn accordance with the four types of core reaction domains we designed, a total number of seven logic gates were constructed in TIGS, including OR, AND, NOR, NAND, XOR, Majoritymajority (MAJ), and OR-AND (OA). All gates ownhave similar tetrahedral framework structures and start operationoperations from a single nanostructure Results Design strategies of the logic operation domain We utilized a toehold-mediated strand replacement reaction as the basic operation process. The length of the toehold was set as 6 nt, which was enough to complete a rapid displacement reaction.28 To transfer the response of DNA logic operation into easily detectable output signals, fluorophore groups at specific sites were set as the output port and activated by the structural changes of stepwise separation induced by strand displacement. All logic gates were unified to use fluorescence as the output signal, and the length of the input strands was programmed to be 26 nt. As shown in Fig. 2a, b, the simplest YES logic gate built by double-stranded DNA (dsDNA) can be transformed into “Y”-shaped YES logic through structural reconfiguration, which results in the formation of a three-way junction (Supplementary Fig. 1). Subtly, by extending three toeholds at each end, three-input OR gates could be successfully constructed (Fig. 2c). In the OR gate, the absence of all inputs (in000) led to a low output signal, and any 1 input led to a high signal (Supplementary Fig. 2). To construct its logic operation domain, we arranged the operation core in a three-way DNA junction whose trident framework was conducive to designing a recognition reaction activated by three ssDNAs. One pair of FAM fluorophores and a BHQ1 quencher served as the signal converter, and the fluorescence intensity of the FAM was used to analyze the output result. The strand displacement caused by any input could trigger the separation of FAM and BHQ1 and then result in an obvious increase in the fluorescence signal, which was defined as the logic 1 output (Fig. 2d, Supplementary Fig. 3). Schematic rearrangement of fluorophore/quencher-modified core logic domains subjected to diverse multi-levelmultilevel input compositions and their characteristic fluorescence outputs. a, b Rearrangement of a fluorophore/quencher-modified Y-shaped module by a single input and its fluorescence output. c The coreCore logic domain of the three-input OR gate. d Fluorescence output of the OR gate in the presence of variable combinations of three inputs, and the corresponding truth table. e The coreCore logic domain of the three-input MAJ gate. f Fluorescence output of the MAJ gate in the presence of various combinations of three inputs, and the corresponding truth table. g The coreCore logic domain of the three-input OA gate. h Fluorescence output of the OA gate in the presence of variable combinations of three inputs, and the corresponding truth table. i The coreCore logic domain of the three-input AND gate. j Fluorescence output of the AND gate in the presence of variable combinations of three inputs, and the corresponding truth table Similarly, by further introducing three toeholds and one more fluorophore, a majority (MAJ) gate could be successfully constructed (Fig. 2e). The MAJ gate was triggered to present a high output when the number of positive inputs was greater than that of the passive inputs (Supplementary Fig. 4). In other words, a three-input MAJ gate should output a positive signal when any two inputs are present. The output fluorescence intensities of the three-input MAJ gate upon interaction with the set of inputs were in good agreement with the corresponding truth table (Fig. 2f and Supplementary Fig. 5). However, it should be noted that the positive fluorescence Cy5 signals varied greatly across the different input sets, with an approximate ratio of in111:in110/in011:in101 = 4:2:1. The significant difference and peculiar ratio of output fluorescence intensities were attributed to stoichiometrically released free fluorophores in reconfigured DNA structures and the self-quenching effect29,30 of Cy5 fluorophores in close proximity in S9/S10 duplex complexes (see Supplementary Figs. 6–7 and accompanying discussions). In the OR-AND (OA) gate, input1 with any one of the other two inputs could lead to a positive output (Fig. 2g, Supplementary Fig. 8). In other words, the presence of input1 is necessary to output logic 1 in the OA gate, which is also not a basic logical rule in logic operation but is widely applicable in various scenarios.8,31 The MAJ gates and OA gates share a similar underlying logic; both require two target inputs to trigger high-output results. Therefore, we fabricated an OA logic operation domain on the basis of the MAJ gate by eliminating one Cy5 to disable the high output formerly triggered by in011 in the MAJ. Thus, the high output was restricted to the simultaneous presence of one premise input and one or more optional input(s) (Fig. 2h, Supplementary Fig. 9). In the AND gate, three positive inputs (in111) lead to a logic 1 output, and the absence of any input leads to a logic 0 output. As any point along the DNA strand is linked to up to two sequences (Supplementary Fig. 10), the translocation of any point needs up to two inputs. Therefore, to design a three-input AND logic gate based on DNA nanomaterials, a chain reaction must be adopted (Fig. 2i, Supplementary Fig. 11). The recognition and combination of input1 could further trigger the unlock of a toehold for input3. To avoid loss of operation speed and signal intensity,8 we applied another independent strand replacement for the recognition of input2. The AND gate was confirmed to be successfully synthesized and emitted obvious Cy5 fluorescence only after incubation with all the inputs, showing operational consistency in accordance with the theoretical design (Fig. 2j, Supplementary Fig. 12). Three-way DNA junctions give off a unique charm in recognition of three inputs, as three branches can respond to the respective input and interact with each other under the constraint of the middle junction. Inspired by the finding of structural changes in point-to-point separation in basic strand displacement, we converted the complicated reaction activated by three inputs into simpler one/two-input logic gates with additional internal logic relations. Thus, a total of four three-input logic operation domains, including OR, AND, MAJ, and OA, were successfully developed through a scalable strategy. tFNA-based three-input logic gate system After the key operation domain for three-input DNA logic gates was constructed, we explored the establishment of a three-input logic gate system (TIGS) on the basis of the tetrahedral framework of tFNA, which has demonstrated great cell entry capacity in related biological research.23,24 A total of seven three-input logic gates were designed, including OR, AND, NOR, NAND, XOR, MAJ, and OA gates. By extending the editable strand ends in the core operation domain and arranging specific sequences, it is feasible to construct OR, AND, MAJ, and OA gates within the framework of tFNA (Fig. 3a, d, g, j, Supplementary Fig. 13). We tracked the changes in the fluorescence of the OR gates over time under eight different input states (Fig. 3b, c, Supplementary Fig. 14). The input-dependent kinetics revealed a rapid reaction of the OR gate, and positive output could be read when the measurement just started, with the results reaching the platform period within 30 min. The minimum concentration of in1 for a detectable positive output after 30 min of operation was 20 nM, and a concentration of 40 nM led to a very clear signal distinguished from noise (Supplementary Fig. 15). Similarly, the AND gate was confirmed to be successfully synthesized to output a high signal only after it was incubated with all the inputs, showing a consistent operation in accordance with its logical design (Fig. 3d–f, Supplementary Fig. 16). As the required number of inputs increased, the sensitivity of the AND gate decreased (Supplementary Fig. 17). However, nontarget inputs did not cause false outputs even at a high concentration of 2 μM, which suggested that the AND gate is especially suitable for scenarios that require low false-positive rates. In the MAJ gate (Fig. 3g, i, Supplementary Fig. 18), to decrease internal self-quenching, we moved the Cy5 fluorophore of S7 from the spacer by a distance of two bases (Supplementary Fig. 19) so that the measured intensity under in101 improved to close to that under in110 and in011 (Fig. 3h, Supplementary Fig. 20). We also tracked the changes in the output signal under different concentrations of in110, which revealed that MAJ operation reached a plateau after approximately 60 min of incubation, which was slightly slower than that of the OR gate, and the minimum input concentration to output a detectable signal was 40 nM (Supplementary Fig. 21). Given the same framework structure, the OA gate was successfully synthesized (Fig. 3j, l, Supplementary Figs. 22–23) and exhibited a reaction manner similar to that of the MAJ gate. The fluorescence output stabilized after 30 min of reaction (Fig. 3k), and the lowest input required for logic 1 output was 40 nM in110, approximately 1/50 the concentration compared with the detection threshold of the OA gate (Supplementary Fig. 24). Fluorophore/quencher-modified three-input logic gates in TIGS. a Schematic bridging of the core logic domain into the 3D tetrahedral nanostructure of the OR logic gate. b Input-dependent kinetics of the OR logic operation showing that high output could be rapidly stimulated by any one input. c Schematic showing the structural responses of the OR logic gate under eight different states of inputs. d Schematic bridging of the core logic domain into the 3D tetrahedral nanostructure of the AND logic gate. e Input-dependent kinetics of AND logic operation showing that high output could only be stimulated underonly in the presence of all inputs. f Schematic showing the structural responses of the AND logic gate under eight different states of inputs. g Schematic bridging of the core logic domain into the 3D tetrahedral nanostructure of the MAJ logic gate. h Input-dependent kinetics of the MAJ logic operation showing that high output could only be achieved underonly in the presence of any two inputs. i Schematic showing the structural responses of the MAJ logic gate under eight different states of inputs. j Schematic bridging of the core logic domain into the 3D tetrahedral nanostructure of the OA logic gate. k Input-dependent kinetics of the OA logic operation showing that high output could only be achieved underonly in the presence of both in1 and any one other input. l Schematic showing the structural responses of the OA logic gate under eight different states of inputs The split between a donor–acceptor pair for fluorescence resonance energy transfer (FRET) could destroy the FRET signal and turn positive signals into passive signals. The logic operations of the NOR and NAND gates are opposite to those of the OR and AND gates, respectively. Therefore, NOR and NAND gates can be easily constructed by replacing the fluorescence-quencher group of the OR and AND gates with a Cy3-FAM FRET pair (Fig. 4a, c). To analyze the measured results, we calculated the difference in the emitted fluorescence at 566 nm and 522 nm as the readout value. It represents the difference in intensity of the Cy3 fluorescence triggered by FAM and residual FAM fluorescence, an amplified value other than the direct measurement of Cy3 emission light. When the same framework was used, the NOR gates and NAND gates were present in a similar concentration-dependent manner as the OR gates and AND gates were (Fig. 4b, d, Supplementary Figs. 25–28). The NOR gate exhibited a fast response speed, with a minimum detectable concentration of 25 nM. Under low concentrations of inputs, the NAND gate required far more reaction time and still maintained resistance against false-positive reactions under high nonpurposive input conditions. In addition, another type of NAND gate can also be realized using the framework of the OR gate and FRET phenomenon (Supplementary Fig. 29). The crosswise arrangement of the two pairs of FRET fluorophores maintained positive FRET results for the NAND-2 gate until the recognition of three inputs completely separated all the fluorophores. The successful construction of the NAND-2 gate served as an example of the high scalability of the OR gate in TIGS. FRET-based three-input logic gates in TIGS. a Schematic diagram showing the structure of the NOR gate in TIGS, which was designed based on the basis of the OR gate. b Input-dependent spectrum of the NOR logic operation. The FRET phenomenon was set as the output signal and indicated by the fluorescence difference in fluorescence at 566 nm and 522 nm. c Schematic diagram showing the structure of the NAND gate in TIGS, which was designed based on the basis of the AND gate. d Input-dependent spectrum of the NAND logic operation. The FRET phenomenon was set as the output signal and indicated by the fluorescence difference in fluorescence at 566 nm and 522 nm. e Schematic diagram showing the logic operation spiredinspired by the FRET quencher group. f Schematic diagram showing the structure of the XOR gate in TIGS, which was designed based on the basis of FRET quencher groups. g Schematic showing the structural responses of the XOR logic gate under eight different states of inputs. h, i Input-dependent spectrum and Fluorescencefluorescence intensity profile of FRET output signals in the XOR logic operation. The FRET phenomenon was set as the output signal and indicated by the fluorescence difference in fluorescence at 561 nm and 547 nm. n = 5. j Truth table consistent with aan XOR TDN logic gate As shown in Fig. 4e, by combining the fluorophore-quencher pair and FRET donor-receptor pair, the split of the quencher could induce FRET, whereas further splitting of the fluorophore could extinguish FRET. The concept of reversible output in the donor-receptor-quencher group was applied to construct the XOR gate in TIGS (Fig. 4f, Supplementary Fig. 30). Three fluorophore groups, Cy3, FAM, and BHQ1, were modified as three reaction cores to execute the XOR logic operation. In the initial state, no FRET was detected because of the quenching effect of BHQ1 on all the FAM donors. As shown in Fig. 4g–i, any 1 input could lead to the destruction of a reaction core by separating its Cy3 and FAM and simultaneously trigger the FRET phenomenon of another reaction core by dislodging its BHQ1. In the same way, two inputs could destroy two reaction cores and activate the remaining reaction core. The recognition of three inputs could therefore lead to the destruction of all reaction cores and thus result in the absence of the FRET phenomenon, which presents as a fake but similar form of nonresponse such as in000. Input concentration levels of 10 nM were sufficient to initiate a detectable reaction and produce high output (Fig. 4j, Supplementary Fig. 31), indicating the relative sensitivity of the XOR gate. Overall, we successfully completed the construction of TIGS by subtly combining the design strategies of the three-input logic operation domain with different fluorophore matches. All gates start reacting from a single structure without the need for additional reagents halfway, and signal readouts are achievable in one pot at room temperature. The editability of the nick position, edge sequences, toehold length and sequence, and different types of fluorescent groups further provide a substantial possibility to create multi-input logic gates (Supplementary Figs. 32–39). Intracellular logic operation of TIGS The application potential of tFNA lies in its ability to enter various cell types through endocytosis, thereby facilitating the intracellular delivery of therapeutic agents, diagnostic tools, or genetic material.32,33 Notably, given the ability to perform complex operation tasks, all gates in TIGS maintain similar and simple tetrahedral structures, which is conducive to further scalability and helps preserve the superior cell entry property of tFNA.27,34 We next investigated the cellular internalization of TIGS. Both linear and tetrahedral three-dimensional AND logic gates were synthesized, retaining the fluorescent moiety while the quencher group was removed. This allows visualization of the intracellular distribution and localization of AND logic gates via fluorescence signaling (Fig. 5a). After incubation for 24 h, the AND gates in TIGS entered living cells quickly and massively. However, the linear DNA logic gates exhibited minimal uptake into MCF-7 cells, as evidenced by the absence of a Cy5 fluorescence signal within the cells (Fig. 5b). Compared with that of linear controls, the cellular uptake of AND is significantly greater, primarily because of the inherent endocytic efficiency of the tetrahedral DNA nanostructure, which enhances membrane interaction and intracellular trafficking. While both constructs exhibit good stability in culture medium (Supplementary Fig. 40), the structural advantage of tetrahedrons accounts for the observed difference in bioavailability. Logic operation of TIGS in living cells. a, b Confocal fluorescence images confirming the excellent endocytosis ability of the tetrahedral structure of the AND gate in TIGS compared with the AND gate with a linear structure. c Schematic application of the logic gate array for the output pattern identifying the three-input composition in MCF-7 cells. d, e Confocal fluorescentfluorescence images and their average cellular fluorescence intensity presented aare consistent output result as the designwith the results of the three-input OR gate. f, g Confocal fluorescentfluorescence images and their average cellular fluorescence intensity presented aare consistent output result as the designwith the results of the three-input AND gate. h, i Confocal fluorescentfluorescence images and their average cellular fluorescence intensity presented aare consistent output result aswith the design of the three-input MAJ gate. j, k Confocal fluorescentfluorescence images and their average cellular fluorescence intensity presented aare consistent output result aswith the design of the three-input OA gate. Scale bar = 40 μm. N = 5 To evaluate the structural stability of siR@MAJ under varying Mg²⁺ concentrations. siR@MAJ labeled with a Cy5–BHQ2 quencher pair was diluted in buffers containing 0.02–10 mM Mg²⁺ and 100 mM KCl (Supplementary Fig. 41). After 24 h, no significant fluorescence dequenching was observed, indicating that structural disassembly did not occur. In addition, the MAJ@siR nanostructure exhibited sufficient serum stability, with ~75% integrity retained after 24 h in 10% FBS (Supplementary Fig. 42). After 48 h at temperatures ranging from 0 to 37 °C (Supplementary Fig. 43), siR@MAJ exhibited no detectable degradation, demonstrating robust stability under storage, handling, and physiological conditions. We subsequently verified the logical operation outcomes of TIGS in living cells in a preliminary biological scenario (Fig. 5c). We transfected three inputs into MCF-7 cells to trigger the respective intracellular computations of the four gates, namely, the OR, AND, MAJ, and OA gates. As shown in Fig. 5d–k, the cellular fluorescence images and statistical charts all presented consistent outputs, as expected, which demonstrated that these four gates successfully maintained the original advantages of tetrahedral DNA nanostructures to execute intracellular logic operations efficiently. While a low level of baseline fluorescence is observed in the absence of target inputs, this signal arises primarily from technical factors, including minor stoichiometric imbalances during nanostructure assembly and the high sensitivity of detection instrumentation. Unlike catalytic DNA circuits, which suffer from autonomous amplification-driven leakage, our nonamplifying, stoichiometric design results in minimal cross-reactivity. The consistency between the observed fluorescence patterns and expected logical outcomes highlights the robustness and reliability of TIGS in executing complex computational tasks within the intricate environment of living cells. It also sets the stage for further investigations into more sophisticated intracellular logic computations. Precise logical regulation of survivin in living cells Building upon the foundational insights gained from our initial verification studies, we subsequently embarked on an exploration of the applicational potential of smart intracellular computing. As a proof of concept, we demonstrated a MAJ-responsive survivin regulation nanoplatform in combination with siRNA delivery. Many stimuli-responsive nanocarriers have been designed for realizing the smart and precise release of siRNA. However, such a unimodal switch mechanism is typically activated by a singular stimulus (such as RNA, protein molecules, pH, or enzymatic activity)35,36 but lacks the sophistication required to interpret and adapt to the multifaceted conditions in the complex intracellular microenvironment. To address these challenges, the development of multi-input, logic-gated nanocarriers that can integrate multiple stimuli to make more informed decisions is urgently needed. First, we designed a MAJ-miR gate to deploy a regulatory nanoplatform that targeted miR-155, miR-21, and miR-181 as inputs (Fig. 6a, Supplementary Fig. 44). These three miRNAs are highly expressed in MCF-7 breast cancer cells,13 among which only miR-21 is upregulated in HepG2 cancer cells.13 A critical consideration for multi-input molecular logic systems is the quantitative match between endogenous biomarker abundance and the operational sensitivity of the device. While intracellular miRNA concentrations can vary across orders of magnitude depending on cell type and pathological state, our MAJ logic gate was specifically engineered to respond to the coupregulated triad of miR-21, miR-155, and miR-181—well-established oncomiRs consistently reported as highly expressed in MCF-7 breast cancer cells (40,000–80,000 copies per cell, or approximately 50–100 nM). Experimental titration of the logic gate revealed an activation threshold of ~40–60 nM for the target miRNA, which falls within this endogenous expression range (Supplementary Fig. 45). This ensures that siR@MAJ can be efficiently triggered by the natural miRNA landscape of target tumor cells. Logically- controlled release of siRNA in living cells. a Schematic diagram showing the structure of MAJ-miR that appliedwhen three miRNAs were used as inputs. b Confocal fluorescence images confirming successful logiclogical recognition of miRNA targets in MCF-7 cells. c statisticalStatistical analysis of fluorescentthe fluorescence intensity of the images in (b). d Confocal fluorescence images confirming that MAJ-miR could not conduct logic recognition onlogically recognize the miRNA by decreasing the intracellular miRNA level. e Schematic diagram showing the structure of siR@MAJ, which was designed to logically control the release of siRNA inside the carrier. f Survivin mRNA expression level in MCF-7 cells after incubation with siR@MAJ. g Survivin protein expression levellevels in MCF-7 cells and HepG2 cells after incubation with siR@MAJ observed under a confocal microscope. h statisticalStatistical analysis of fluorescentthe fluorescence intensity of the images in (g). i Live/dead cell staining of MCF-7 cells after incubation with siR@MAJ. j, k statistical k Statistical analysis of fluorescentfluorescence intensity and dead cell counts in j To assess specificity under biologically relevant noise, the MAJ gate was challenged with a 100-fold excess of single target miRNAs, nontarget miRNAs, and total cellular miRNAs. As shown in Supplementary Fig. 46, no significant activation occurred with any single target input, which is consistent with the requirement for at least two target inputs in MAJ logic. Nontarget miRNAs and total miRNA induced only minimal background signals. Notably, the temporal alignment between molecular device activation and biomarker dynamics is critical for reliable intracellular computation. In our system, the MAJ logic gate generates a robust fluorescence output immediately upon the initiation of real-time monitoring after mixing with target miRNAs, indicating rapid conformational switching—which is consistent with the fast kinetics typically observed for toehold-mediated strand displacement under physiological conditions.37 This activation occurs on a timescale of minutes to tens of minutes, which is substantially shorter than the reported half-lives of miR-21, miR-155, and miR-181, ranging from ~3 to > 12 h in breast cancer cells.38,39 Thus, the logic gate can reliably “read” the miRNA state before significant turnover alters the input signature. To observe the logical recognition of miRNA targets in living cells, we incubated MAJ-miR with MCF-7 cells and HepG2 cells for 24 h. Confocal microscopy images revealed obvious fluorescent signals in the cytoplasm of the MCF-7 cells but negative signals in the cytoplasm of the HepG2 cells (Fig. 6b). This outcome reflects the expected intracellular behavior of the MAJ-miR logic gate following cellular uptake: in MCF-7 cells, coexpression of all three target miRNAs triggers structural reconfiguration of the gate, leading to fluorophore–quencher separation and fluorescence activation; in HepG2 cells, the presence of only miR-21 fails to satisfy the ≥2-input requirement, leaving the gate in its quenched state. Statistical analysis of the intracellular fluorescence intensity further confirmed the difference in the output between the two types of cells (Fig. 6c), which suggested that the MAJ-miR gate successfully achieved the goal of cell identification, as determined by logical analysis of various biomarkers. To further evaluate the logical response capacity of the MAJ-miR gate in MCF-7 cells, we attenuated the intracellular inputs by transfecting three miRNA inhibitors. As shown in Fig. 6d, along with the downregulation of miRNA inputs, the MAJ-miR gate could no longer perform intracellular logical operations and thus showed weaker fluorescence output. These results proved that the MAJ-miR gate can enter cells and further implement a logical response according to various intracellular inputs, indicating the unparalleled advantage of TIGS in smart drug delivery.38 To assess potential cytotoxicity, we evaluated the biocompatibility of TIGS and related logic gate nanostructures in both normal (MCF-10A) and cancerous (MCF-7) cells using CCK-8 assays. As shown in Supplementary Fig. 47, none of the constructs induced significant reductions in cell viability over 48 h, confirming their excellent safety profile. Therefore, we next constructed a logically controlled siRNA-loaded TIGS nanocarrier (siR@MAJ) by introducing survivin siRNA into MAJ-miR (Fig. 6e). Survivin is often overexpressed in MCF-7 cells, contributing to cancer cell resistance to chemotherapy-induced apoptosis. Survivin siRNA effectively and specifically degraded survivin mRNA to silence the expression of survivin in MCF-7 cells. The encapsulated siRNA was designed to link to the eighth nucleotide site of the MAJ-miR gate, making the cargo oriented inside the carrier framework in our previous study.39,40 Thus, the switch status of the MAJ gate determined by intracellular miRNA could further control the release of the cargo siRNA. We also used siR@tFNA as the control, which had the same structure and base sequence as siR@MAJ except for three toeholds. After the MCF-7 cells were incubated with siR@tFNA or siR@MAJ for 24 h, we examined the expression of survivin mRNA via qPCR analysis (Fig. 6f), which revealed that siR@MAJ significantly silenced survivin expression, whereas no pronounced change was detected in cells treated with siR@tFNA. In addition, we verified the expression of survivin at the protein level via immunofluorescence staining (Fig. 6g) under a confocal microscope, and the results were consistent with those of survivin mRNA expression. Moreover, siR@MAJ failed to achieve signal recognition and siRNA delivery within HepG2 cells and therefore did not affect intracellular Survivin expression in HepG2 cells (Fig. 6h, i). Owing to the significant downregulation of the survivin gene caused by siR@MAJ, the proliferation of MCF7 cells was markedly inhibited (Supplementary Fig. 48). Moreover, the proportion of dead MCF7 cells significantly increased (Fig. 6j, k). In contrast, siR@tFNA did not have a noticeable effect on MCF7 cells. These results collectively suggested that the target gene in living cells could be precisely regulated by the MAJ-based logic nanoplatform. Compared with conventional responsive siRNA nanocarriers, our system allows for the controlled release of siRNA in a manner sensitive to the logical relationship of multiple targets, which greatly enhances therapeutic specificity. The modular nature of TIGS also facilitates customization for diverse biological contexts, supporting advanced applications in personalized medicine. In vivo gene therapy effect of logical siRNA release via the MAJ gate To evaluate the antitumor efficacy of siR@MAJ via the logical delivery of siRSur, an in vivo study was conducted in MCF-7 tumor-bearing nude mice (Fig. 7a). All the nude mice were subcutaneously injected with MCF-7 cells to establish tumor xenografts. Two weeks later, when the tumors reached an average volume of approximately 120 mm³, the mice were randomly assigned to four treatment groups for intravenous administration via the tail vein. The control group received saline, while the other groups were administered free siRSur, siR@tFNA, or siR@MAJ. The siRNA dose for each group was normalized to body weight and set at 1 mg/kg. Treatments were administered every two days for a total of 10 doses, after which the mice were sacrificed, and samples were collected for analysis. In vivo gene therapy effect of logical siRNA release via the MAJ gate. a Schematic diagram illustrating the treatment procedures ofused for the in vivo experimentexperiments. b In vivo imaging showing the biodistribution of Cy5-labeled siRSur after intravenous injection in tumor-bearing mice. c Representative ex vivo images of tumors isolated from mice 6 h after i.v. administration of Cy5-labeled siRSur. d Body weight changes of the mice during the treatment period. e TumorTumoour volumes measured and calculated over time during treatment. f MCF-7 tumors were excised from nude mice after completion of treatmenttreatment was complete. g H&E staining of tumor tissues, showing evident necrosis in the siR@MAJ group. Scale bar = 100 μm. h Ki-67 immunofluorescence staining of tumor tissues reflecting changes in cell proliferation. Scale bar = 50 μm. i TUNEL staining of tumor tissues indicating levels of apoptosis. Scale bar: 200 μm (main panels), 100 μm (magnifiedmagnification view). j, k Western blot analysis of Survivin expression in tumor tissues and corresponding quantification. l qPCR analysis of Survivin mRNA expression levels in tumor tissues across groups. m Immunofluorescence staining of Survivin in tumor tissues. Scale bar: 200 μm (main panels), 100 μm (magnifiedmagnification view) In vivo fluorescence imaging revealed distinct biodistribution patterns of siRSur across the groups. As shown in Fig. 7b, compared with free siRSur, both siR@tFNA and siR@MAJ resulted in tumor accumulation, with siR@MAJ demonstrating the strongest fluorescent signal at the tumor site, indicating superior and rapid intratumoral retention of siRSur loaded with the MAJ gate. This observation was further corroborated by ex vivo imaging of excised tumors at 6 h post-injection, which revealed a progressive increase in the amount of siRSur from the free state to carrier-mediated delivery, with the highest signal intensity observed in the siR@MAJ group (Fig. 7c). The tumor accumulation of siR@MAJ is mediated by the EPR effect, leveraging the ~15 nm size of the TDN scaffold for passive targeting—a strategy validated by multiple clinical nanodrugs.41–43 Critically, therapeutic activity is further restricted by the MAJ logic gate, which ensures that siRNA is released only in tumor cells expressing the target miRNA triad. This combination of passive delivery and conditional activation provides dual-layer specificity, mitigating off-target risks inherent to EPR-based systems. To evaluate the systemic pharmacokinetic behavior and renal clearance of siR@MAJ, we monitored the time-dependent fluorescence signal in serum and urine following intravenous administration. As shown in Supplementary Fig. 49, the total Cy5-associated fluorescence in serum decreased rapidly within the first 10 min post-injection, decreasing by more than 70% from the initial value, followed by a gradual reduction at 6 h. A slight increase was observed at 24 h, likely attributable to the re-entry of fluorescent metabolites into circulation following hepatic, renal, or tumor processing of the nanostructure. In addition, fluorescence signals of siR@MAJ were rapidly detected in urine, with excretion peaking at 1 h postinjection, which indicated rapid renal elimination of the siR@MAJ nanostructures or their degradation products. These results also demonstrate no significant retention of MAJ nanostructures in the urinary route, supporting their favorable biocompatibility profile. During the treatment period, no significant body weight loss was observed across any of the groups, suggesting good systemic tolerance and minimal toxicity of all the formulations (Fig. 7d). In contrast, tumor growth was most effectively suppressed in siR@MAJ-treated mice, as reflected by the smallest increase in tumor volume over time (Fig. 7e). Treatment with siR@MAJ also led to the longest median survival time among all the groups, with a significant portion of the mice surviving past 50 days (Supplementary Fig. 50). At the end of the study, the tumors were harvested, and visual inspection confirmed that the tumors from the siR@MAJ group were substantially smaller than those from the other groups were (Fig. 7f). Histopathological analysis of tumor tissues by hematoxylin and eosin (H&E) staining revealed extensive areas of necrosis in the siR@MAJ-treated group, whereas tumors from the siRSur and siR@tFNA groups exhibited predominantly viable tumor cells (Fig. 7g). Consistent with these findings, Ki-67 immunofluorescence staining revealed a dramatic reduction in proliferating cells in the siR@MAJ group, underscoring the potent antiproliferative effect of siR@MAJ-mediated Survivin silencing (Fig. 7h, Supplementary Fig. 51). Furthermore, TUNEL staining revealed the highest level of apoptotic cells in the siR@MAJ group, indicating robust induction of programmed cell death (Fig. 7i, Supplementary Fig. 52). Western blot analysis also confirmed that survivin protein expression was most effectively downregulated in the siR@MAJ group (Fig. 4j, k). Quantitative real-time PCR further revealed a corresponding decrease in Survivin mRNA levels, indicating successful gene silencing at the transcriptional level (Fig. 7l). These findings were supported by immunofluorescence staining of Survivin in tumor sections, which revealed minimal residual expression in the siR@MAJ group compared with the other groups (Fig. 7m, Supplementary Fig. 53). To assess immunotoxicity, we measured complete blood counts (Supplementary Table 7) and serum CRP levels (Supplementary Fig. 54) in nude mice 24 h postinjection. All hematological parameters remained within normal ranges, and the CRP levels did not significantly differ from those of the controls (p > 0.05), indicating the absence of a systemic inflammatory response. This aligns with the established low immunogenicity of TDNs. Collectively, these results demonstrated that siR@MAJ not only enhanced the intratumor retention of siRSur via rapid responsive release upon miRNA recognition but also achieved superior therapeutic outcomes through efficient Survivin inhibition, leading to increased tumor cell death, reduced proliferation, and significant inhibition of tumor growth. Discussion The development of intelligent systems capable of simultaneously recognizing multiple biomarkers and performing programmable responses is crucial for advancing precision medicine and early disease diagnosis. In response to the growing demand for multitarget identification, herein, we proposed TIGS, a series of three-input logic processors built upon tetrahedral framework nucleic acids. This system can logically recognize three specific molecular inputs and distinguish among all eight possible combinations of their states. Owing to the DNA tetrahedral structure, the TIGS logic gate retains the excellent cellular uptake and tissue permeability characteristics of tFNAs, enabling rapid entry into cells. This also lays the foundation for its application in precise gene regulation. The inherent editability of DNA provides considerable potentialfor the development of logically controlled delivery systems. In particular, the potential broad applicability benefits from the following characteristics. First, TIGS is currently the most comprehensive three-input logic gate system with available OR, AND, XOR, NOR, NAND, MAJ, and OA gates. Compared with the individual logic gates in previous studies,10 TIGS offers a versatile approach applicable for diverse diagnostic scenarios. Second, all gates in TIGS are built upon the framework of the simplest three-dimensional DNA nanostructure, known as tFNA. Basic elements required for fabricating core reaction domains, such as vertices, nicks, edges, and three-way junctions, are fundamental modules in more complex DNA framework nanomaterials. Notably, the structural rigidity of the tFNA scaffold ensures precise spatial orientation of the logic strands, minimizing steric hindrance and maximizing the efficiency of strand displacement reactions. TIGS and its design principles might serve as a useful and promising toolbox that can be adapted to more complicated DNA nanostructures, such as a connecting recognition system and amplification system at the front end or an additional readout system at the back end. Our results demonstrate that TIGS successfully achieves multiplexed recognition of three intracellular miRNAs—key regulators of gene expression and valuable biomarkers for various diseases, including cancer. This multiplexed sensing capability allows for highly specific and accurate detection of disease-associated miRNA signatures, reducing false positives that may arise from single-target sensing approaches and enabling more accurate molecular diagnostics. Building upon this logical recognition, we further engineered an embedded siRNA delivery platform integrated within the tFNA structure. The carrier remains structurally closed under nontargeting or partially matched conditions, minimizing off-target effects and premature release. Only upon full recognition of the intended three-target miRNA could the logic gate become activated, triggering the rapid release of encapsulated siRSur. This ensures that therapeutic intervention is executed exclusively in target cells that exhibit a tumor-specific miRNA fingerprint. This “MAJ” logic-gated release mechanism effectively transforms the carrier from a passive delivery vehicle into an active molecular classifier, significantly improving the signal-to-noise ratio in heterogeneous biological environments. While the successful in vivo implementation of TIGS represents a meaningful advance toward intelligent molecular diagnostics and therapy, several inherent challenges associated with DNA nanotechnology in physiological settings warrant careful consideration. Furthermore, an often underappreciated challenge lies in the very nature of DNA as a biomaterial: its molecular identity closely resembles that of endogenous nucleic acids. This intrinsic similarity is a double-edged sword—it confers excellent biocompatibility and programmability but simultaneously complicates precise pharmacokinetic (PK) and biodistribution analyses. Unlike small-molecule drugs or protein therapeutics, which can be directly quantified via mass spectrometry or immunoassays, DNA nanostructures are typically tracked indirectly through covalently attached fluorophores or radiolabels. Such proxies may not fully reflect the fate of the intact nanostructure, as label detachment or quenching can occur independently of scaffold degradation. Consequently, the lack of direct, structure-specific analytical methods introduces uncertainty into the optimization of dosing regimens, clearance pathways, and long-term safety profiles—key parameters required for clinical translation. Beyond the technical optimization of the hardware itself, the true power of TIGS lies in its potential to redefine clinical decision-making. By shifting from a “one-size-fits-all” therapeutic model to a “compute-then-act” paradigm, this technology allows for the stratification of patients at a molecular level that was previously inaccessible. For instance, the ability to distinguish between the subtle differences in miRNA expression profiles—differentiating between early-stage malignancy, benign inflammation, and healthy tissue—could significantly reduce the overtreatment often seen in current oncological practices. Furthermore, the modular nature of the tFNA framework suggests that TIGS could be adapted for use in non-oncological fields, such as neurodegenerative diseases or autoimmune disorders, where complex biomarker panels are increasingly recognized as critical for accurate diagnosis. In terms of scalability and versatility, the TIGS strategy demonstrates remarkable flexibility. The modular design principle allows for the rapid construction of logic gates with higher input complexity. For example, we have successfully leveraged the TIGS framework to design and implement a four-input logic gate in our ongoing research, highlighting the system’s capacity to handle more sophisticated diagnostic scenarios. Additionally, the bio-orthogonality of the DNA framework ensures minimal interference with endogenous cellular machinery. Unlike protein-based sensors or viral vectors, which often trigger immune responses or integrate unpredictably into the genome, our DNA-based logic processor operates as a transient, non-genotoxic agent. This bio-orthogonality not only enhances safety but also ensures that the computational output is solely dependent on the designed logic, rather than confounding biological variables, paving the way for reliable clinical translation. Future efforts will need to address these translational barriers through engineering refinements, the development of more faithful tracking strategies, and comprehensive preclinical validation. Concurrently, expanding the logic framework to incorporate non-nucleic acid inputs (e.g., proteins or small molecules) could further broaden its diagnostic scope. Moreover, the constructed TDN arrays exhibit limited scalability and sensitivity toward target miRNA biomarkers, primarily because of the stoichiometric nature of DNA strand displacement-based signal transduction. To improve the scalability and sensitivity of the TDN logic computing platform, an engineered TDN logic frameworks that interface with cellular machinery, such as transcription/translation machinery44,45 or endogenous enzymes,46 could be envisaged. Alternatively, the incorporation of exogenous signal amplification strategies, such as CRISPR‒Cas machinery47 or DNA hybridization chain reaction (HCR) modules,48 represents a promising route to substantially increase signal output and detection performance. Additionally, the integration of machine learning algorithms to analyze the combinatorial output of these logic gates could provide a more nuanced interpretation of complex disease states, moving beyond binary decisions to probabilistic risk assessments. By progressively overcoming these challenges, TIGS and similar platforms hold promise as foundational components of next-generation precision medicine systems capable of autonomous, context-aware decision-making at the molecular level. Materials and methods Design and synthesis of logic gates in TIGS TIGS included seven three-input logic gates designed on the basis of the structures of a single tFNA with edges of 21 base pairs. Each gate’s important design principles and ideas are explained in the Results section. Therefore, in this section, we introduce mainly the structural parameters of logic gates. In accordance with the various logic operation requirements, the types and positions of fluorophores in all gates differed. The sequences of the DNA oligonucleotides used in this study are shown in Supplementary Fig. 1 and Supplementary Table 1. All DNA oligonucleotides were purchased from Sangon Biotech (Shanghai) Co., Ltd., were purified by high-performance liquid chromatography (HPLC) and were delivered as powders. Before synthesis, the DNA powder was dissolved in DNase-free water, and the concentrations were determined using a Nanodrop (Thermo Fisher Scientific). 1×TM buffer (50 mM MgCl2 and 10 mM Tris base, pH 8.0) was used as the synthesis buffer. All tFNA logic gates (including MAJ-miR and siR@MAJ) were synthesized by one-pot self-assembly via a thermal cycler (95 °C for 10 min, then 4 °C for 20 min after fast annealing). Notably, the stoichiometric ratios of all fluorophore/quencher labels to the associated DNA strands are 1:1 in all the systems. In this study, the synthesis concentration of the logic gates was 1000 nM.49 Dynamic analysis of logic gate reactions To monitor the reaction kinetics of the four gates with a fluorophore-quencher group as the output indicator (OR, AND, MAJ, and OA), we used a qPCR instrument (QuantStudio 6; Thermo Fisher Scientific, Inc., USA) to continuously detect the fluorescence signal intensity under different input concentrations and different inputs. The concentrations of the input strands in the different experiments are shown in Supplementary Table 2. A 25 μL reaction system included 200 nM logic gates, input strands of relevant concentrations, 5 mM Mg2+, and DNase-free water. Measurements were taken at 25 °C every minute for the first 30 min, every 2 min between 32 and 90 min, and every 5 min after 95 min. Threshold Definition We used the background fluorescence measured under nontarget input conditions as the baseline to eliminate the effect of potential signal leakage. Furthermore, the direction of the threshold depends on the output logic type: For signal-on gates (OR, AND, MAJ, and OA), where target recognition leads to an increase in fluorescence, the threshold for “output = 1” was defined as follows: Thresholdon=μbg+3σbg. For signal-off gates (XOR, NOR, and NAND), where target binding induces fluorescence quenching, the threshold for “output = 0” was defined as follows: Thresholdoff=μbg+3σbg. Here, μbg and σbg denote the mean and standard deviation, respectively, of the background fluorescence from nontarget input controls (n ≥ 5). Measurement of Fluorescence Resonance Energy Transfer (FRET) The FRET phenomenon served as the output signal during the structural transformation of the NOR, NAND, and XOR gates. Therefore, these three gates cannot be accurately detected by a qPCR instrument; thus, we conducted direct fluorescence spectrum measurements to observe the FRET level. A 100 μL reaction system included 200 nM logic gates, input strands of relevant concentrations, 5 mM Mg2+, and DNase-free water. The reaction mixture was incubated in a black 96-well plate at room temperature (RT) on a shaker at 100 rpm. After incubation, measurements were conducted by a fluorescence microplate reader (VarioskaLUX multimode microplate reader, Thermo Fisher, Massachusetts, USA). The incubation duration for the logic operation was 1 h. The incubation duration for concentration-dependent detection was also 1 h, except for the NAND gate, which required a more prolonged incubation of 6 h to accommodate low concentrations of inputs. For Cy3-FAM FRET, λex: 492 nm, λem: 508–620 nm; for Cy5-Cy3 FRET, λex: 550 nm, λem: 565–700 nm. The measurement duration was 100 ms. The FAM fluorophore showed an emission peak at 521 nm after an excitation wave at 492 nm, which was downregulated by FAM-Cy3 FRET, followed by a peak at 566 nm. Nucleic acid electrophoresis 8% Native polyacrylamide gel electrophoresis (PAGE) was used to prove the synthesis and reaction of logic gates in TIGS. One 6 mL PAGE gel included 1.2 mL of 40% acrylamide/bis-acrylamide, 60 μL of 10% ammonium persulfate, 0.6 mL of 10×TAE (1×TAE: 40 mM Tris-acetate, 1 mM EDTA, pH 8.0), 0.6 mL of 50 mM MgCl2, 4.2 ml of pure water and 6 μL of TEMED. The gel was run at 80 V for 70 min at RT using 1×TAE buffer with an additional 5 mM MgCl2 in running buffer. The fluorescence images were captured immediately after running the ChemiDoc MP Imaging System (Bio-Rad, USA). Afterward, the gels were stained with 1× gel under slight shaking for 20 min before imaging. Images of different channels of one gel were colored and merged for further analysis. For the electrophoresis of the logic reaction results, 5 μL of 1 μM logic gates was mixed with 1 μL of 30 μM input strands and incubated in a thermal cycler at 25 °C for 60 min before running. Cellular uptake of logic gates MCF-7 cells were used to analyze the cellular uptake of the AND gate. High-strength DMEM supplemented with 10% fetal bovine serum (FBS) and 1% penicillin‒streptomycin was used for cellular culture and adherence. After cell adhesion occurred, the original culture medium was replaced with 1 mL of 250 nM AND gates with tetrahedral structures and linear structures in FBS-free high-grade DMEM. After incubation for 24 h, the logic gate solution was removed, the cell nuclei were stained with Hoechst 33342 for 10 min at RT, and the confocal dishes were covered with complete high-strength DMEM during observation (FV3000, Olympus confocal microscope). Cell culture Human breast adenocarcinoma MCF-7 cells and human hepatoblastoma HepG2 cells were obtained from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). MCF-7 cells were maintained in minimum essential medium (MEM; Gibco, Thermo Fisher Scientific) supplemented with 10% fetal bovine serum (FBS; Gibco), 1% penicillin–streptomycin (P/S; 100 U/mL penicillin and 100 μg/mL streptomycin), and 0.01 mg/mL recombinant human insulin (Sigma‒Aldrich). HepG2 cells were cultured in Eagle’s minimum essential medium (EMEM, or MEM with nonessential amino acids; NEAA; Gibco) supplemented with 10% FBS (Gibco) and 1% P/S. For both cell lines, the cultures were incubated at 37 °C in a humidified atmosphere containing 5% CO₂. Cells were passaged upon reaching 80–90% confluence. Confocal imaging of logic operations in living cells For the experiments shown in Fig. 5d, f, h, and j, MCF-7 cells were seeded onto glass-bottom confocal dishes and cultured overnight. The following day, the cells were transfected with the corresponding DNA input strands using Lipofectamine™ 3000 (Thermo Fisher Scientific) following the manufacturer’s instructions. The transfection complexes were freshly prepared immediately before use; for each input, 2 μL of a 1000 nM stock solution of the input strands was used per dish. After 24 h of incubation, the transfection medium was removed, and the cells were gently washed twice with phosphate-buffered saline (PBS) to eliminate residual nucleic acids. Fresh complete high-glucose DMEM was then added, and the cells were incubated for an additional 3 h to allow recovery. Subsequently, 1 mL of serum-free high-glucose DMEM supplemented with 200 nM of the target logic gate was added to each dish. Following another 24 h of incubation, the logic gate solution was removed. For live-cell imaging, the nuclei were stained with Hoechst 33342 (10 min at room temperature), and the dishes were replenished with 1 mL of complete high-glucose DMEM prior to confocal microscopy. Confocal imaging of MAJ-miR in living cells As shown in Fig. 6b, MCF-7 and HepG2 cells were seeded in confocal dishes and cultured overnight. The next day, the culture medium was replaced with fresh high-glucose DMEM or MEM (containing 0.01 mg/mL recombinant human insulin) supplemented with 200 nM MAJ-miR. After 24 h of incubation, the MAJ-miR-containing medium was removed. Cells were then fixed with 4% paraformaldehyde (PFA) for 15 min at room temperature, followed by sequential staining with phalloidin (to label F-actin) and DAPI (for nuclear counterstaining). Fluorescence images were acquired using a confocal laser scanning microscope. As shown in Fig. 6c, MCF-7 cells were first transfected with a cocktail of miRNA inhibitors targeting miR-21, miR-155, and miR-181. Briefly, after being seeded overnight in confocal dishes, the cells were transfected with the inhibitors using Lipofectamine™ 3000 according to the manufacturer’s protocol. After 24 h, the transfection medium was removed, and the cells were washed twice with PBS. Subsequently, 1 mL of serum-free high-glucose DMEM containing 200 nM MAJ-miR was added. Following another 24 h of incubation, the MAJ-miR solution was removed, and the cells were fixed with 4% PFA, followed by phalloidin and DAPI staining as described above. Confocal images were then captured under identical imaging conditions. qPCR After sample harvest, the total RNA of the MCF-7 cells/tumor tissue was collected via a TRIpure Total RNA Extraction Reagent Kit (Bioteke), and reverse transcription was completed via a PrimeScript RT reagent Kit (Takara Bio). Real-time analysis was performed using a SYBR Premix Ex Taq TM II kit (Takara Bio) on a QuantStudio 6 (Thermo Fisher). To determine the miRNA expression level in the MCF-7 cells, total small RNA was extracted and purified using a SanPrep Column microRNA Extraction Kit (Sangon). Reverse transcription of small RNA was completed with a poly(A) tailing reaction. A MicroRNA qPCR Kit (Sangon) was used for the final qPCR analysis. The miRNA levels were normalized to the U6 levels. All the primers used for qPCR are listed in Supplementary Table 6. Survivin protein expression analysis in MCF-7 cells Flow cytometry and immunofluorescence imaging were performed to explore the effect of siR@MAJ on survivin gene expression at the protein level. MCF-7 cells and HepG2 cells were cultured in 6-well plates and confocal dishes for 24 h. Then, all the original medium was replaced with 1.5 mL of the corresponding complex of interest diluted in serum-free basic medium with the same final concentration of 250 nm siRNA. After being incubated for 48 h, the cells on the confocal dishes were fixed with 4% paraformaldehyde and then sequentially stained with a survivin antibody (2808 T; CellSignal Technology), a fluorescent secondary antibody, phalloidin and DAPI before being captured by fluorescence imaging. Cells in 6-well plates were collected and fixed in 70% ethanol. After being stained with survivin antibody and fluorescent secondary antibody, the sample cells were thoroughly resuspended in 500 µL of PBS before being run on a flow cytometer (CytoFLEX). Assessment of cell viability CCK-8 and live/dead cell staining were performed to assess the therapeutic effect of siR@MAJ. MCF-7 cells were cultured and processed as described above. After 48 h of incubation, the cells were detected according to the instructions. For CCK-8 analysis, CCK-8 reagent working solution (C6005; NCM Biotech), which was diluted with basal medium, was added to the wells. After 1 h, the OD450 nm was measured using a fluorescence microplate reader. For live/dead cell staining, following the instructions provided with the Calcein/PI cell staining kit (C2015S; Beyotime Biotechnology), a staining working solution was prepared. The cells were incubated with this solution for 30 min. Subsequently, the wells were gently washed twice with PBS, and images were observed and captured using a fluorescence microscope. Animal study Female BALB/c nude mice (approximately 20 g) were purchased from Chengdu Dossy Experimental Animal Co., Ltd. (Chengdu, China) and housed under specific pathogen-free (SPF) conditions. All animal experiments were approved by the Ethics Committee of West China Hospital of Stomatology (WCHS), Sichuan University (WCHSIRB-AT-2025-475). Biodistribution To evaluate the in vivo biodistribution of siRNA, female nude mice (approximately 20 g) bearing MCF-7 xenograft tumors were intravenously injected with Cy5-labeled siRNA in the form of free siRSur, siR@tFNA, or siR@MAJ via the tail vein. Whole-body fluorescence images were acquired at 5 min and 6 h post-injection using an in vivo imaging system (AniView100, Bltlux) to monitor the distribution of Cy5-labeled siRNA. The tumors were subsequently excised from the mice and further analyzed for fluorescence signal intensity using an imaging system (ChemiDoc, Bio-Rad). In vivo circulation and excretion profiling Male BALB/c nude mice were intravenously injected via the tail vein with 100 µL of siR@MAJ solution at a concentration of 1 µM. At predetermined time points (0, 10 min, 30 min, 1 h, 6 h, and 24 h postinjection), urine and blood were collected after anesthesia. Serum was separated by centrifugation at 1000 × g for 30 min at 4 °C. Serum and urine from age-matched naïve mice (without injection) were used as a baseline for background subtraction and signal normalization. For fluorescence quantification, 100 µL of serum or 50 µL of urine was transferred to a black 96-well plate. The fluorescence intensity was measured using a microplate reader with an excitation wavelength of 644 nm and an emission wavelength of 664 nm. All measurements were performed in triplicate, and background fluorescence from naïve mouse serum/urine was subtracted from each reading. In vivo antitumor efficacy To establish tumor xenografts, female nude mice (approximately 20 g) were subcutaneously injected in the dorsal flank with a 200 μL mixture containing 100 μL of MCF-7 cells (1 × 10⁷ cells) and 100 μL of Matrigel. Tumor growth was monitored every two days by caliper measurements. Two weeks after inoculation, when the tumors reached an average volume of approximately 120 mm³ (calculated using the formula: 1/2 × length × width²), the mice were randomly assigned to four treatment groups: saline control, free siRSur, siR@tFNA, and siR@MAJ. Each group received 100 μL of treatment via tail vein injection. The siRNA doses for the siRSur, siR@tFNA, and MAJ@siR groups were normalized to the body weight at 1 mg/kg. Treatments were administered every two days for a total of 10 doses. Finally, all the mice were euthanized, and the tumors were harvested for further analysis. Statistical analysis All the experimental data were statistically analyzed via GraphPad Prism software. The flow cytometry data were analyzed via FlowJo. Statistical significance between groups was determined using one-way analysis of variance (ANOVA) or Student’s t test. The quantitative data are presented as the means ± standard deviations (n ≥ 3). Statistical analysis, *p < 0.05, **p < 0.01, ***p < 0.001. Supplementary information Supplementary Information - clean version.docx Acknowledgements This study was supported by the National Natural Science Foundation of China (82501232, 82401145, 82322015, and 82171006), the China Postdoctoral Science Foundation (2024M762249), the Postdoctoral Fellowship Program of CPSF (GZC20241123 and GZC20231790), the Sichuan Province Science Foundation for Youths (2025ZNSFSC1595 and 2024NSFSC1621), the Sichuan Science and Technology Program (2022NSFSC0002), the Sichuan Province Youth Science and Technology Innovation Team (2022JDTD0021), the Research and Development Program, West China Hospital of Stomatology Sichuan University (RCDWJS2024-3), and the Health Commission of Sichuan Province Medical Science and Technology Program. Data availability The data that support the findings of this study are available in the manuscript or the Supplementary Information. Source data are provided with this paper. Any additional specific datasets are available on request. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Yang Gao, Yun Wang, Yunlong Qin Contributor Information Itamar Willner, Email: itamar.willner@mail.huji.ac.il. Taoran Tian, Email: Taoran.tian@scu.edu.cn. Xiaoxiao Cai, Email: xcai@scu.edu.cn. Supplementary information The online version contains supplementary material available at 10.1038/s41392-026-02767-5. References Associated Data Supplementary Materials Supplementary Information - clean version.docx Data Availability Statement The data that support the findings of this study are available in the manuscript or the Supplementary Information. Source data are provided with this paper. Any additional specific datasets are available on request.