Combinatorial protein dimerization enables precise multi-input synthetic computations

Living cells interact with complex molecular environments and process information through genetically encoded, often binary, regulatory programs. Synthetic implementation of such decision-making in mammalian cells could enable applications in tissue engineering, stem cell differentiation and gene therapy. While electronic circuits use insulated components, cellular circuits must sense specific targets within complex mixtures, complicating higher-order logic. Prior mammalian transcription-based logic circuits built with multiple reporter regions or constitutive dimerization suffer from long induction times, low design flexibility and/or lack of inducibility, and the limited number of multiplexable inducible dimerization proteins constrains designs. In this study, the authors target one of the largest small-molecule-responsive protein families, bacterial helix-turn-helix (HTH) TFs, as modular building blocks for large digital circuits in mammalian cells. Many HTH regulators are orthogonal in mammalian cells and evolved to recognize diverse small molecules (sugars, amino acids, vitamins and metabolites). These TFs share a common three–alpha-helix DNA-binding domain (DBD). In mammalian cells, the key usable domains are the DBD and effector-binding domain (EBD). The authors discovered that truncated DBDs alone from some TFs can still bind cognate DNA strongly, and that full-length TFs can dimerize in response to effectors without requiring DNA interaction. Integrating these features with other chemically induced heterodimerization systems, they develop LOGIC, a framework for multi-input gates based on inducer-controlled cascades of protein fusions, enabling modular AND/OR logic up to 4 inputs and beyond, and conversion of native OFF-type switches into ON-type systems including band-pass behavior.

Design strategy: The LOGIC framework decouples DNA binding from effector binding by fusing either a full-length HTH TF (TF1) or only its DBD (TF1,DBD) to additional TFs or heterodimerization domains. Effector-induced dimerization colocalizes a transcriptional activator (TA; e.g., VPR or VP16) at the TF1 DNA-binding site to control a reporter, enabling n-input AND (serially chained dimer pairs) or OR (parallel fusions targeting the same DNA origin) logic gates.

Protein parts and constructs: HTH TFs tested included TetR, VanR (vanillic acid, VA), PIP (pristinamycin), AcoR (acetoin), LgnR (D-idonate), D-LIdR (D-lactate), TrpR (DBD), XylR (xylose), and ToxT (virstatin). Additional inducible heterodimerization domains used were FKBP/FRB (rapamycin), ABI/PYL1 (abscisic acid, ABA), and GAI/GID (gibberellic acid, GA). DNA-binding domains were modeled using SWISS-MODEL to design truncated HTH DBDs (TrpRDBD, AcoRDBD, VanRDBD, TetRDBD). Fusion proteins included combinations such as TetR–TF2, TF2–VPR, TrpRDBD–TF2, TF2–VPR, and serial chains (e.g., VanR–rTetR, FKBP–VanR, FRB–ABI, PYL1–VPR, etc.). VanR N-terminus truncations (deleting 5, 11, 16, or 21 residues) mapped functional requirements for dimerization versus DNA binding.

Cell culture and transfection: HEK293T cells were cultured in DMEM + 10% FBS + pen/strep. For transient assays, 10,000 cells/well (96-well) were transfected overnight with 150 ng total DNA using PEI (1:6 DNA:PEI). Media were exchanged next morning with or without inducers and incubated for 24 h before measuring secreted reporters. A stable HEK293T line carrying TetO7–SEAP (Sleeping Beauty transposon) was generated for genome-integrated reporter assays.

Inducers and concentrations: Dox (1 µM), VA (250 µM), rapamycin (50 nM), ABA (40 µM), GA (40 µM), D-lactate (25 mM), acetoin (10 mM), D-idonate (1 mM), pristinamycin (7.5–10 µM), virstatin (50 µM), xylose (5 mM). Dose–response curves were performed for virstatin and xylose.

Reporters and readouts: Primary reporter was secreted alkaline phosphatase (SEAP) driven by TetO or TrpR-responsive promoters with variable operator repeats (TetO2 vs TetO7). Nanoluciferase (NLuc; SV40 promoter) served as a constitutive control and as an output in specific OFF-switch configurations. A STAT3-responsive promoter reported GEMS receptor activation for extracellular dimerization tests.

Assays: SEAP activity was quantified from supernatants via p-nitrophenyl phosphate turnover at 405 nm over 30 min. NLuc luminescence was measured for constitutive expression and toxicity checks. Live-cell imaging (phase-contrast) with confluency analysis (std-dev filter + Triangle threshold) assessed proliferation. RNA-seq (TruSeq stranded mRNA; NextSeq 500) evaluated transcriptomic impact of virstatin-responsive ToxT system; reads processed with trimmomatic, hisat2, samtools, featureCounts; statistics via edgeR and pathway analysis with GeneGo Metacore.

Circuit implementations: 2-input NIMPLY (B AND NOT A) gates used TetR fused to ligand-dimerizing TFs; 2-input AND used rTetR-based binding. Higher-order AND gates (3-, 4-, 5-input) were built by serially chaining heterodimerization domains, requiring all inducers for activation. Mixed AND/OR 4-input designs combined serial and parallel fusion architectures to realize (A AND B) AND (C AND D), (A OR B) OR (C OR D), (A AND B) AND (C OR D), (A OR B) OR (C AND D), (A OR B) AND (C OR D), and (A AND B) OR (C AND D). Sequential addition experiments tested assembly order independence. Promoter architecture effects were tested by comparing TetO2 vs TetO7 in 1–4 input gates.

Statistics: Data are mean ± SD of three biological replicates. Two-tailed unpaired Student’s t-tests or one-way ANOVA were used as appropriate, with significance thresholds as reported in figures.

• Decoupling DNA binding from effector binding: For several HTH TFs, their truncated DBDs (notably TrpRDBD and AcoRDBD) bind DNA strongly when fused to TAs, enabling dimerization-dependent activation rather than DNA-binding-dependent activation.
• OFF-to-ON conversion: Native OFF-type TF systems (e.g., VanR) were converted to ON-type switches by fusing TFs to DNA-binding modules and inducible dimerization partners.
• Screening of TF pairs: Using TetR as DNA anchor and TF2 fused to VPR, VanR-based LOGIC yielded up to 13.5-fold SEAP induction with VA; with TrpRDBD as the DNA anchor, VA-based induction improved to 23.1-fold.
• New mammalian gene switches: Built virstatin-responsive (ToxT) and xylose-responsive (XylR) switches in HEK293T cells. • ToxT: Virstatin induced SEAP ~16-fold; dose–response EC50 ~31.5 µM; Dox repressed TetR-mediated output even in presence of virstatin. • XylR: Xylose induced SEAP ~26-fold (5 mM); half-maximal expression at ~1.5 mM xylose; Dox abrogated xylose responsiveness.
• VanR functional dissection: N-terminal truncations showed that deletion of first 5 aa reduced DNA binding; deletion of 11 aa abolished DNA binding but preserved VA-induced dimerization; deletion of 16 or 21 aa abolished dimerization. Thus, distinct sequence requirements for DNA binding vs dimerization.
• Multi-compartment dimerization: VA-induced VanR dimerization was functional in cytosol (TEV protease reconstitution cleaving TrpRDBD–VP16 from a membrane-tethered construct) and extracellularly (GEMS receptor—VanR–EpoR–IL6 fusion activating STAT3), both significantly increasing SEAP.
• Band-pass filter: Combining VA-driven ON and OFF VanR systems with a ligand-independent dimer (D-LIdR) produced a compact transcriptional band-pass filter with >8-fold higher SEAP at intermediate VA vs OFF states.
• Multi-input logic: Built NIMPLY gates with several TFs; VanR–TetR NIMPLY achieved 24-fold ON/OFF. AND gates scaled to 2-, 3-, 4-, and 5-inputs: • 2-input AND (VanR–rTetR + VanR–VPR): ~42-fold ON over OFF. • 3- and 4-input AND (adding FKBP/FRB, ABI/PYL1): robust logic; NOT A AND B AND C and NOT A AND B AND C AND D variants had average ON/OFF ~47 and ~83, respectively (Extended Data). • 5-input AND (adding GA via GAI/GID): required all five small molecules; 8-fold change over highest OFF state; OFF-state leak decreased as complexity increased, though ON amplitude diminished.
• Promoter architecture: Higher-order gates (3-, 4-input) had higher outputs with longer operator arrays (TetO7 vs TetO2), suggesting larger binding regions facilitate assembly of larger fusion complexes.
• 4-input OR and mixed AND/OR gates: Constructed robust 4-input OR and combinations (e.g., (A OR B) OR (C OR D), (A AND B) OR (C AND D), etc.) using inputs A (Dox), B (VA), C (Rap), D (ABA). The 4-input OR gate averaged ~13-fold ON/OFF and remained functional across reporter plasmid titrations and with stably integrated reporters. Sequential input addition produced similar ON outputs, indicating order-independent assembly.
• Safety/physiology: Virstatin-responsive ToxT system showed minimal impact on viability, proliferation, protein production, and transcriptome (no clear separation by RNA-seq clustering).

The study demonstrates a modular, protein–protein interaction-centric strategy to implement precise, multiplexed logic computation in mammalian cells. By exploiting the modular architecture of bacterial HTH TFs, especially the separable DBD and EBD, the authors achieve inducible, orthogonal, and compact logic elements. Decoupling DNA recognition from effector-controlled dimerization simplifies circuit design, enables conversion of native OFF switches into ON switches, and creates new small-molecule responsive systems (virstatin, xylose) without requiring native operator sequences. Compared to previous transcriptional cascades that suffer from long delays and limited dynamic range, the LOGIC approach multiplexes at the protein level to maintain high maximal expression while minimizing leakiness, supporting larger and faster networks. The framework generalizes across cellular compartments, as shown by cytosolic and extracellular dimerization, broadening application domains (e.g., receptor engineering). The band-pass filter showcases integration of ON and OFF modalities to emulate developmental patterning behaviors and could enable spatially precise gene expression in tissue engineering. Scaling to 5-input AND and multiple 4-input AND/OR combinations illustrates the flexibility and robustness of the approach; promoter architecture tuning (more operator repeats) further improves performance of higher-order assemblies. Orthogonality among small-molecule effectors and modularity of small DBDs expand the design space for complex Boolean logic gates in single cells.

This work introduces LOGIC, a generalizable framework that leverages inducible protein dimerization and truncated HTH DBDs to build compact, orthogonal, and tunable multi-input logic gates in mammalian cells. The approach converts OFF-type TF systems to ON-type, creates new small-molecule-responsive switches (virstatin, xylose), implements a high-performance transcriptional band-pass filter, and scales to multi-input AND and mixed AND/OR gates (up to five inputs). The methodology maintains strong ON/OFF ratios and low leak and operates across cellular compartments. Future directions include incorporating additional HTH TFs and heterodimerization systems to further increase input dimensionality, optimizing linker and promoter architectures to enhance ON-state amplitudes in higher-order gates, and translating these circuits to other cell types and in vivo applications for tissue engineering and therapeutic cellular computing.

• Reduced ON-state amplitude in the most complex gates: The 5-input AND gate showed weaker induction (~8-fold over the highest OFF state) compared to lower-order gates, likely due to reduced complex stability as fusion chains grow.
• Dependence on promoter architecture: Higher-order gates benefited from larger operator arrays (TetO7 vs TetO2), indicating that performance can be constrained by DNA-binding site design.
• Cell line and context: Experiments were performed in HEK293T cells; functional generalizability to other mammalian cell types and in vivo contexts was not assessed here.
• Finite set of orthogonal inducers: Although expanded by leveraging HTH TFs and heterodimerization domains, the available pool of fully orthogonal, non-toxic small-molecule effectors remains a practical constraint for very large multiplexing.
• Potential for homodimer formation: While tested ratios suggested minimal impact on performance, non-productive homodimers are an inherent consideration in fusion-based designs and may require tuning of expression ratios in some contexts.