Engineering intelligent chassis cells via recombinase-based MEMORY circuits

The study addresses how to unify three foundational properties of intelligent biological systems—decision-making, memory, and intercellular communication—into a single engineered living chassis. Prior work provided separate technologies: transcription factor-based Boolean logic (e.g., Cello and Transcriptional Programming) for decision-making, quorum sensing for communication, and recombinase-based DNA state changes for memory. However, no single system has integrated all three. The authors hypothesized that coordinating multiple orthogonal, inducible serine integrases (recombinases) under discrete biosensing controls could enable inheritable DNA inversions, excisions, and insertions while remaining compatible with decision-making circuits. They aimed to build a genome-integrated platform with six orthogonal recombinases regulated by Marionette TFs, achieving near-digital switching and enabling programmable gain- and loss-of-function memory, with extensions to CRISPR-dCas9 protection and cross-species communication.

The paper builds on extensive work in synthetic gene circuits: CRISPR and TF-based logic for decision-making; multiple quorum-sensing systems enabling intercellular signaling; and recombinase-based memory for permanent DNA state changes. Prior recombinase circuits typically used up to three inducible recombinases, often on multicopy plasmids, limiting scalability and imposing resource burden. Large serine integrases (e.g., Bxb1, A118) are known to mediate site-specific inversion/excision/insertion, and their expression can be placed under inducible promoters. Cello and Transcriptional Programming (T-Pro) have enabled automated or programmable logic in cells. Landing pad strategies for genomic insertion (e.g., Bxb1 att sites) exist but often insert entire plasmids and rely on additional recombinases (e.g., FLP) to minimize inserts. Preliminary demonstrations of CRISPR-dCas9 blocking of integrase att sites existed in cell-free contexts, but not generalized or co-regulated with memory and decision-making in vivo. This work extends those foundations by doubling the number of independently inducible recombinases per cell, integrating them into the genome for stability and reduced burden, generalizing dCas9 protection to multiple integrases in vivo under programmable TF control, and demonstrating intercellular and cross-species signaling linked to memory.

• Recombinase selection and regulation: Six serine integrases (A118, Bxb1, Int3, Int5, Int8, Int12) were chosen and each placed under a distinct inducible transcription factor from the Marionette array (PhlF/DAPG, TetR/aTc, AraC/L-arabinose, CymR/cuminic acid, VanR/vanillic acid, LuxR/3OC6 AHL). Expression cassettes were diversified via libraries (promoters, degenerate RBSs, start codons, degradation tags) and cloned on a single-copy BAC to emulate genomic expression.
• Reporter circuits and memory assay: Inversion-based gain-of-function (GOF) output plasmids (low-copy pSC101) placed an inverted strong promoter upstream of gfp flanked by anti-aligned att sites; additional architectures included inversion LOF, excision GOF, and excision LOF. A memory assay grew cells with/without inducer, then outgrew in inducer-free medium before flow cytometry to ensure readout reflects input history.
• Insulated genomic array: To prevent transcriptional readthrough between adjacent recombinase cassettes, strong terminators were placed upstream/downstream and transcriptional orientations alternated. Additional terminators were added based on cross-activation observations. The finalized insulated array was integrated into the E. coli MG1655 Marionette genome (EcMem), replacing unused TFs (retaining LacI for T-Pro demonstrations). Orthogonality and efficiencies were validated by cytometry.
• Kinetics, performance, and stability: Recombination time courses were measured in minimal medium, maintaining exponential growth with periodic dilutions. An 11-day genetic stability assay passaged EcMem harboring inversion GOF circuits every 12 h without inducers, with periodic inductions to assess retained function. Growth burden was assessed, including simultaneous induction of all six recombinases.
• Extrachromosomal programming and reset: Designed erasable pSC101-based extrachromosomal memory programs, where the origin of replication was flanked by aligned Int3 att sites to allow L-arabinose–induced excision (erase/reset) while preserving the genomic MEMORY platform. Demonstrated a 2-output fluorescent program (GFP via Int8 excision GOF; mKate via Bxb1 inversion GOF) and subsequent reset and reprogramming with a new circuit (Int5 inversion GOF).
• Programmed genomic insertion (MEMORY recorder): Engineered artificial genomic safe harbors (aGSHs) consisting of non-synonymous attP site pairs. Payload 1 (gfp, kanR) with nested att sites was integrated into aGSH1 via inducible Bxb1/Int8 recombination; the donor plasmid carried an origin eraser and sacB for counterselection. After insertion and plasmid erasure (Int3), integration was quantified by GFP and confirmed by colony PCR. Sequential insertion of Payload 2 (mkate, ampR) targeted a second aGSH, re-establishing aGSH1.
• CRISPR protection (CRISPRp): dCas9 was constitutively expressed (BAC), and sgRNAs were placed under inducible control (IPTG via LacI or synthetic TFs for T-Pro). sgRNAs targeted specific positions within or near att sites (natural or synthetic PAMs) to block integrase binding/action. Protection efficiency was quantified via the inversion GOF assay across all six integrases. Demonstrated orthogonal control of sgRNAs via T-Pro TFs (BUFFER operations) concurrent with MEMORY inputs.
• Next-generation recombinase-based state machine (ngRSM/GRSM): Built a 3-input register using Bxb1, Int3, and Int8 att architectures including odd-numbered att sets and CRISPRp to expand possible states. Designed a sequence-dependent unlocking of gfp requiring inputs in order (Int8 → Bxb1 → Int3); out-of-order induction triggers irreversible, nonfunctional states. All six input permutations were tested over three days and quantified by cytometry.
• Intercellular communication and autoinduction: Implemented biosynthetic pathways under LacI control to produce MEMORY inducers in vivo: phlACBD for DAPG (A118), luxI for 3OC6 AHL (Int12), and asbF+HsOMT for vanillic acid (Int8). Tuned promoter/RBS/degradation tags to achieve near-digital autoinduction that activates corresponding inversion GOF circuits. Validated sender–receiver cocultures (EcMem) and quantified recombination.
• Probiotic chassis and cross-species communication: Transferred MEMORY into E. coli Nissle 1917 (EcMemPro) and characterized all 24 circuits aerobically and anaerobically. Built a cross-species program where EcMemPro produces vanillic acid upon IPTG, which is sensed by a vanillic acid-inducible circuit in Bacteroides thetaiotaomicron driving Nanoluc; cocultures were assayed for luminescence.

• Six orthogonal, inducible recombinases integrated in the genome (EcMem) achieved near-digital switching: for inversion GOF targets, each showed >97% recombination upon induction with <3% in the uninduced state; complete recombination occurred in ~12 h under exponential growth for both inversion and excision GOF circuits.
• Expanded memory repertoire: 24 optimized circuits (inversion/excision; GOF/LOF) functioned robustly in EcMem, outperforming BAC-based versions and enabling flexible nesting of att sites and regulatory elements.
• Genetic stability and low burden: Over 11 days (~200 doublings), all six recombinases maintained functionality with negligible leak; every induction yielded >95% recombination. Simultaneous induction of all six recombinases had minimal impact on growth rate.
• Extrachromosomal programming and reset: A two-output plasmid program enabled independent activation of GFP (Int8 excision GOF) and mKate (Bxb1 inversion GOF). An Int3-based origin eraser reset cells by excising the pSC101 origin; >99.9% of cells lost the plasmid (by cytometry/CFU), preserving the genomic MEMORY platform. Reset cells were reprogrammed with a new Int5 inversion GOF circuit, showing correct inducibility and loss of prior responses.
• Programmed genomic integration: Using aGSHs and payloads, induced Bxb1/Int8 insertion of gfp/kanR achieved ~99% integration (GFP fluorescence before sacB counterselection), confirmed by colony PCR. A second sequential integration (mkate/ampR via Int5/Int12) reached ~90%, also PCR-confirmed. Plasmid erasing was verified in both rounds.
• CRISPR protection (CRISPRp): dCas9+sgRNA blocked recombination at targeted att sites with >95% efficiency for all six recombinases in vivo. CRISPRp control via synthetic TFs enabled orthogonal BUFFER operations running concurrently with MEMORY inputs, unifying decision-making and memory.
• Next-generation RSM: A 3-input ngGRSM with CRISPRp expanded the register to 9 discrete states; only the correct input order (Int8 → Bxb1 → Int3) unlocked gfp. All six permutations produced the predicted sequence-dependent state transitions with near-perfect correspondence to design.
• Intercellular communication: Autoinduction programs producing DAPG, 3OC6 AHL, or vanillic acid drove near-digital activation (>95% recombination; <5% leak) of corresponding inversion GOF circuits. Sender–receiver cocultures in EcMem achieved near-digital recombination in receivers.
• Cross-species exchange: EcMemPro producing vanillic acid upon IPTG induction triggered Nanoluc expression in B. thetaiotaomicron harboring a vanillic acid sensor, reaching luminescence comparable to exogenous vanillic acid controls (significant increases; P ≤ 2.0E-3 to ≤ 6.2E-5).

The work demonstrates an integrated platform that unifies decision-making (via T-Pro TFs), permanent memory (via six orthogonal, genome-integrated recombinases), and intercellular communication (via biosynthesized inducers and sender–receiver systems). By moving recombinase expression to a single-copy genomic array and optimizing insulation, the platform achieves high orthogonality, near-digital switching, and low metabolic burden over extended culture, addressing prior limits of scalability and instability tied to multicopy plasmids. The suite of 24 fundamental memory operations provides modular building blocks for complex programs. CRISPRp extends memory capacity by post-translationally controlling integrase site usage with programmable specificity, and it interoperates with T-Pro decision layers, enabling sophisticated history-dependent state machines. The system’s capability to perform efficient, serial genomic insertions provides a path to rapidly engineer derivatives with durable traits. Finally, the probiotic deployment (EcMemPro) and cross-species communication to B. thetaiotaomicron validate the platform’s relevance for living therapeutics, where transient probiotics can program stable commensals. Collectively, these findings realize the proposed intelligent chassis cell paradigm and open broad applications in consortia-based therapies and advanced cellular programming.

This study establishes MEMORY chassis cells that integrate six orthogonal, inducible recombinases in the genome to provide robust, scalable, low-burden synthetic memory compatible with decision-making and communication. The platform delivers: (i) a library of near-digital GOF/LOF inversion/excision circuits; (ii) erasable/reprogrammable extrachromosomal programs; (iii) efficient, sequential genomic insertions; and (iv) CRISPRp for programmable, TF-controlled protection of att sites that expands memory capacity and enables next-generation, sequence-sensitive state machines. Intercellular and cross-species information exchange demonstrate the unification of intelligence tenets in living cells, including a probiotic chassis relevant to the GI tract. Future work could expand the number and diversity of inducible recombinases and TFs, integrate CRISPRp with additional logic layers (e.g., Interception and CRISPRi) on shared att sites, systematize design automation for MEMORY–T-Pro co-programming, and translate the platform to additional hosts and therapeutic contexts.

• Environmental dependence: In the probiotic EcMemPro strain, some circuits exhibited variable or unpredictable performance under anaerobic conditions, relevant to GI tract deployment.
• Residual cross-induction: Despite insulation, a low-level unintended recombination (~9% A118 sites in presence of 3OC6 AHL/Int12 induction) was observed, indicating room for further insulation/orthogonality improvements.
• Input/actuator limits: Although the platform doubles prior inducible recombinase count to six, overall capacity still scales with available orthogonal sensors and recombinases; complex programs may require extensive tuning.
• Dependence on small-molecule inducers and engineered pathways: Robustness in complex in vivo environments (e.g., metabolite availability, microbiome interactions) remains to be validated beyond lab conditions.
• Therapeutic translation: Nissle’s transient colonization necessitates repeated dosing or reliance on communication to colonizers; safety and regulatory considerations for genomic insertions and interspecies signaling will require further study.