Trends in Biotechnology

DNA language models offer a new paradigm for sequence design, yet their ability to generate functional genomic sequences remains underexplored. Plasmids act as a good testbed for evaluating DNA language model generation potential due to their simplicity and ease of construction. Here, we develop an end-to-end pipeline for generative design of Escherichia coli plasmid backbones, from large-scale data curation through fine-tuning, sampling, bioinformatic assessment, and candidate selection. A curated plasmid library was assembled from PlasmidScope and Addgene, and PlasmidGPT, a GPT-2-style DNA model, was fine-tuned on these corpora using circular-aware batching and random crops. Generations (1,000 per model) were produced under two prompting strategies: a minimal ATG seed to expose default tendencies, and a GFP cassette to enforce functional context. From 1000 generated synthetic plasmids, 16 candidates survived strict filtering and these were prioritised for wet-lab validation. Three shortlisted plasmids were synthesised and found to be functional, supporting growth, antibiotic resistance, and GFP expression in E. coli. These represent, to our knowledge, the first full AI-generated plasmids to be synthesised and validated in vivo. This work demonstrates that curated fine-tuning and prompt-aware generation enable DNA language models to progress from raw sequence sampling to experimentally testable plasmid designs. The approach offers a foundation for extending DNA design optimisation beyond E. coli, toward broader applications across engineering biology.

Genetic constructs meant for metabolic engineering in nonmodel microbes often use similar genetic parts to those familiar to E. coli work. The typical workflow is to clone these parts into plasmids in E. coli before they are transferred to the nonmodel host or its genome. In many cases, the metabolic burden of these constructs is stronger in the E. coli cloning phase of the workflow than in the eventual host, possibly resulting in mutation or other failure during cloning. Here, we apply generic knockdown of a range of popular expression systems, using CRISPR interference, by targeting guide RNAs to either promoters or RBSs that are commonly used in metabolic engineering. Generic targeting of a constitutive promoter series, combined with genome integration of CRISPR components, allows the use of only one or a few specific cloning strains to achieve strong knockdown of a wide range of constructs. Further, we present a recombinase-based workflow for easily adding guide RNAs with custom targets, so that users can knock down any desired promoter or ORF. Together, this group of strains comprises easy-to-use cloning strains meant for increasing success rates of difficult or burdensome cloning reactions, ultimately allowing more ambitious genetic constructs to reach their intended context.

Escherichia coli is a well-established model organism in molecular biology and biotechnology. Despite its long history as a laboratory workhorse, the efficient single-step chromosomal integration of large DNA fragments remains a challenge. Currently known methods are either simple but have limitations on insert size, or flexible but laborious requiring plasmid construction or multi-step procedures. Here, we present PhAGE (Phage-Assisted Genome Engineering), which enables the integration of [~]20 kb DNA fragments into E. coli genome within a single day. PhAGE method uses in vitro packaging of recombinant DNA into bacteriophage capsids, followed by general transduction to introduce pre-assembled DNA with flanking homology arms into recipient cells. This approach allows efficient and landing pad-free integration of large constructs into the target loci. We demonstrate its usefulness through rapid integration of multi-gene operons. PhAGE resolves the long-standing trade-off between simplicity and insert size in E. coli genome engineering, accelerating strain construction across a wide range of applications, from biosynthetic pathway engineering to genome-scale design.

Novel strategies for treating bacterial infections are needed to combat the growing threat of antibiotic resistance. Here we sought to engineer and produce phage-like particles to either harness the microbiome to secrete therapeutics or to hijack pathogenic bacteria for treatment and prevention of disease. For this, we used the P2/P4 system to design, produce and test P4 phage-mediated single- and dual-action antimicrobial prototypes. Upon successful completion of the in vitro proof of concept experiments, we focused on optimizing early-stage bioprocessing for in vivo studies, leading to 1011 plaque forming units (PFU) per mL and 0.25 endotoxin units (EU) per 109 PFU. We also challenged the P4 viral vector packaging limit by deleting the sid gene to package the payload into P2-sized capsids ([~]25.8 kb cargo capacity). Importantly, repressing the therapeutic payload during the production of particles improved viral titers about 2 logs, maintained viral payload sequence integrity and improved post-transduction functional activity. Altogether, this study demonstrates the potential of novel phage-based antimicrobials to go beyond elimination of bacteria. The in vitro optimized P2/P4 system constitutes a promising platform technology for in vivo evaluations of targeted antimicrobial candidates paving the way for future antimicrobial research in animal models of infection.

The genus Mycobacterium is increasingly recognized as a major clinical concern due to diseases such as tuberculosis, along with the emergence of antimicrobial-resistant strains, underscoring the urgent need for advanced genetic tools to study mycobacterial biology and pathogenesis. Progress in this area relies heavily on the functional characterization of previously unannotated genes, which necessitates tightly regulated expression systems. Here, we report the development of an improved tetracycline-regulated vector platform, comprising the episomal pM(R)T2 and integrative pMI(R)T2 series, which builds upon the previously described pMT vector system. The T2 vector series incorporates a fine-tuned TetRO system for enhanced transcriptional control. The pMT2 vectors function as tetracycline-inducible systems, whereas the pMRT2 variants utilize a reverse tetracycline repressor (RevTetR) to enable tetracycline-repressible gene regulation. Additionally, the integrative variant, pMI(R)T2 switches the oriM element with the integrase and attP sites derived from mycobacteriophage L5, facilitating stable genomic integration and controlled expression of concentration-sensitive genes, including toxins. To expand the selection flexibility, the pAN(R)Tet series replaces the kanamycin resistance cassette with a hygromycin resistance cassette. Functional validation of gene regulation in M. smegmatis and M. bovis BCG shows that both TetR and RevTetR systems provide reliable inducible and repressible controls, respectively, upon anhydrotetracycline addition. Taken together, these vectors constitute a versatile, tightly regulated genetic toolkit with significant potential to accelerate research and therapeutic development in mycobacterial systems.

Cells are the fundamental unit of life. Yet there is no natural cell for which all its life-essential functions are understood. Here we demonstrate a complete cell cycle for a synthetic cell undergoing selection, with genome replication, growth, resource acquisition via feeding, and genetically encoded division. The cell is encoded via a 90kb genome that includes functions needed for resource uptake, transcription, translation, growth, genome replication, and division. The resulting synthetic cell is sufficiently encouraging to support routinization of synthetic cell engineering workflows, and will ultimately underlie diverse applications across all of biotechnology.

Genome synthesis is a major limitation in generative biology. Here, the half-sized genome of Escherichia coli was constructed by fleshing out an imperfect minimal genome through genome-scale debugging process. Our platform consists of integrated development environment (IDE) and runtime environment (RTE). The genome IDE supported the cell-free assembly of 200-300 kb plasmids and their in vivo fusion into a single 1.7 Mb plasmid. This imperfect genome was stably maintained in E. coli as a guest genome. The RTE relies on the restriction enzyme-mediated self-digestion of the host genome in the presence and absence of the RecA recombinase. The guest genome was tested, debugged, and partially replaced by the host genome to establish E. coli controlled by a 2.3-Mb genome. This is less than half in size of the wildtype and the smallest ever reported. Enfleshing a guest genome will facilitate genome printing that transforms AI-designed genomes into physical ones.

Efficient genome writing in mammalian cells requires robust methods for integrating large DNA payloads. The previously described method mammalian Switching Antibiotic resistance markers Progressively for Integration (mSwAP-In) enables iterative, biallelic genome rewriting in mammalian stem cells with DNA payloads exceeding 100 kb. However, the lack of standardized vectors and certain technical constraints have limited its broader adoption. Here we present an improved plasmid toolkit designed to streamline the implementation of mSwAP-In. The toolkit includes two core vectors. pLP-TK (pCTC174) is a landing-pad plasmid compatible with Golden Gate assembly of genomic homology arms and supports both mSwAP-In and the recombinase-mediated cassette exchange method Big-IN. mSwAP-In MC2v2 (pKBA135) is a versatile Big DNA assembly and delivery vector that supports Gibson-based assembly and incorporates positive, negative, and fluorescent selection markers, as well as a backbone counterselection cassette to minimize unwanted plasmid integration. The vector architecture also enables propagation in yeast and bacterial hosts, inducible plasmid copy-number amplification in standard E. coli strains, and CRISPR/Cas9-mediated payload release through preinstalled guide RNA target sites. We further characterize the FCU1/5-FC counterselection system in mouse embryonic stem cells and define conditions that minimize its bystander toxicity. Finally, we provide a set of Cas9-gRNA expression plasmids optimized for common mSwAP-In applications. Together, these reagents constitute a standardized and experimentally validated toolkit that simplifies large-scale genome writing using mSwAP-In.

DNA assembly is a cornerstone of synthetic biology, enabling the construction of bespoke genetic systems for applications ranging from metabolic engineering to DNA nanotechnology. Conventional Gibson Assembly (GA), the most widely used method, relies on 5' exonucleolytic resection and elevated temperatures ([~]50 {degrees}C), which together prevent the retention of 5' modifications and restrict compatibility with temperature-sensitive functionalities. Here, we report a DNA assembly strategy, 3 exonuclease-mediated low-temperature DNA assembly (3LTDA), which generates complementary 5' overhangs while preserving 5' end integrity. This approach enables the efficient assembly of blunt-ended, 5'-functionalised DNA fragments into both linear and circular constructs at ambient temperature (21 {degrees}C), with some assembly observed at temperatures as low as 4{degrees}C. We systematically optimise reaction conditions and demonstrate that this method supports efficient plasmid re-circularisation and multi-fragment assembly, including the construction of a [~]12.5 kbp plasmid from multiple DNA components. Comparative analysis across several DNA substrates shows that, under their respective optimal conditions, this approach matches or exceeds GA performance, improving assembly efficiency by up to 12.8%. Sequence analysis confirms high fidelity with no detectable base-pairing errors across assembled junctions. Crucially, this method preserves chemically functionalised 5' termini, enabling downstream conjugation and biochemical functionality. Retention of azide and biotin modifications was verified through fluorescence imaging, bead-based co-localisation, and enzymatic activity in ELISA-based assays. This is in contrast to GA-assembled controls, which showed complete loss of functionality under comparable conditions. We further assembled 5 kbp dsDNA using 3LTDA from four independent segments, three with different fluorescence reporters, and the fourth containing a biotin group for microparticle conjugation, each on the 5 end. Under fluorescence illumination, bead-bound DNA with all three fluorescence markers were detected. Conventional GA assembled constructs, on the other hand, failed to retain the reporter groups and the fluorescent images did not show the presence of any fluorescent markers. In addition to enhanced performance, the method could also reduce reagent cost and eliminate the need for elevated temperatures, simplifying workflows and expanding the applicability of multi-functionalised DNA constructs. Collectively, this work establishes 3LTDA as a robust, low-temperature alternative to conventional GA, with advantages for applications requiring precise chemical modification, temperature-sensitive components, or deployment outside conventional laboratory environments.

Microbial cultivation optimization remains labor-intensive and inefficient, requiring extensive experimental screening to identify suitable growth conditions. Traditional one-factor-at-a-time approaches are particularly ineffective for exploring complex, multidimensional nutrient parameter spaces. We present MicroGrowAgents, an AI-driven, agent-based system that automates the design of optimized growth media through integration of knowledge graphs, metabolic modeling, and optimal experimental design. The system employs 28 specialized agents and 50 skills that query structured biological knowledge (KG-Microbe: 864,363 validated species), mine literature evidence (245+ papers), perform genome-guided design (57 genomes, 667,000+ annotated features), and generate statistically optimal experimental designs using the MaxPro algorithm. We applied the approach to Methylorubrum extorquens AM1 by cultivating 70 designed conditions in quadruplicate and assessing three concurrent objectives: biomass (OD600 at 740 nm), redox activity (Abs590 Biolog proxy), and lanthanide uptake (residual Nd measured by arsenazo III). Monte-Carlo resampling of the replicate-level uncertainty (1000 iterations) identified a single stable Pareto-optimal medium, MPOB_058 (membership frequency 0.99), together with two borderline candidates and six rare appearers, providing a robust anchor set for subsequent rounds of design-build-test-learn. The integration of chemical similarity search (208,000+ embeddings), metabolic gap analysis, and multi-modal reasoning enables evidence-based hypothesis generation that reduces experimental burden while accelerating discovery of growth-promoting conditions. MicroGrowAgents provides complete provenance tracking with cryptographic checksums and 90.5% literature citation coverage, advancing reproducible, data-driven approaches to microbial cultivation. Author SummaryGrowing microbes in the laboratory is like figuring out the right recipe: too much or too little of any nutrient and they barely grow. Scientists have traditionally tested ingredients one at a time, an approach that is slow, expensive, and poorly suited to the dozens of interacting nutrients that real microbes need. We built MicroGrowAgents, an AI system that acts like a team of specialist scientists working together. It consults structured biological databases, reads the published literature, inspects microbial genomes, and uses statistical experimental design to recommend nutrient combinations worth testing in the laboratory. Applied to Methylorubrum extorquens AM1, a methanol-eating bacterium of interest for capturing rare-earth elements, the system designed 70 growth conditions and identified one robust winner that performed well across cell growth, metabolism, and lanthanide uptake. The software is free and open-source, helping any laboratory adopt these tools.

Cell-free expression systems (CFES) are increasingly used alongside conventional biotechnological approaches to accelerate early-stage prototyping and are particularly valuable in point-of-use settings. However, their broader adoption remains limited by time- and cost-intensive preparation, as well as stringent cryogenic storage requirements. To address this, several studies have explored lyophilization with protective additives to generate stable, solid-state CFES. These approaches had to balance the protection gained with a loss of activity due to the additives. In this study, we present a CFES that contains a tardigrade-derived Cytosolic-Abundant Heat-Soluble (CAHS) protein to protect the biosynthetic machinery in lysates from damages during drying. We show that the CAHS protein, without any other additives, preserves protein synthesis activity during low-cost room temperature desiccation, while unprotected lysates are affected in mRNA synthesis kinetics and translation yields. The diversity of tardigrade-derived protective proteins is a treasure trove for cell-free synthetic biology, in particular for making CFES more accessible and portable. Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=85 SRC="FIGDIR/small/715078v1_ufig1.gif" ALT="Figure 1"> View larger version (27K): org.highwire.dtl.DTLVardef@8ecc2eorg.highwire.dtl.DTLVardef@ff0432org.highwire.dtl.DTLVardef@6c940eorg.highwire.dtl.DTLVardef@6c5390_HPS_FORMAT_FIGEXP M_FIG C_FIG

Primary human hepatocytes (PHHs) are the gold standard for toxicology and drug metabolism studies in industry. However, their limited availability, substantial batch-to-batch variability, and high cost restrict their use. Here, we report a novel culture condition that reprograms PHHs into a proliferative state. These proliferating cells, termed precursors of chemically expanded hepatocytes (pre-cHep), expand over 106-fold within 30 days while retaining liver repopulation capacity comparable to PHHs. pre-cHep can further differentiate into chemically expanded hepatocytes (cHep) as three-dimensional (3D) spheroids within 7 days in vitro, exhibiting global gene expression profiles, albumin production, and cytochrome P450 (CYP) activities similar to 3D-cultured PHH spheroids (3D PHH). Efficient genetic manipulation of pre-cHep using CRISPR/Cas9 is also achievable. Together, pre-cHep and cHep represent a promising alternative to high-quality PHHs, providing a more affordable, reproducible, and scalable source of human hepatocytes for toxicology, drug metabolism studies, disease modelling, towards precision drug development.

Extracellular vesicles (EVs) hold great promise as therapeutic delivery vehicles, leveraging their natural role as mediators of intercellular communication in all organisms studied. However, many barriers must be overcome to realize their full potential. Saccharomyces cerevisiae is an attractive chassis organism to explore solutions: It is used for drug biomanufacturing, it is amenable to complex genetic engineering, and their EVs can drive responses in human cells. To further develop this prospect, we sought to genetically modify S. cerevisiae EVs by devising a research framework amenable to iterative design, build, test, learn cycles - a core principle of synthetic biology. Using this approach, we focused on identifying new scaffolds - proteins that load cargoes into EVs - from a small pool of candidates. We first optimized a modular cloning strategy, called "EVclo", for plasmid and genome-integrated candidate gene expression. Candidate genes were fused to EGFP, and after confirming expression in cells, we showed that scaffold-EFGP proteins colocalized with mRuby2-tagged Nhx1, a biomarker of multivesicular bodies, presumed sites of EV biogenesis. We triggered release of EVs by heat stress, isolated these EVs by ultrafiltration and size exclusion chromatography, and confirmed the presence of exosome-sized EVs in all samples. We find that candidate scaffold proteins did not affect EV size, morphology or titers. Further analysis of these samples indicated that some EGFP-tagged scaffolds are present in EVs: Bro1, a yeast ortholog of ALIX, was most abundant and ExoSignal showed highest enrichment of the human candidates. In all, we conclude that Bro1 is a good scaffold for future engineering strategies, and that human proteins can be sorted into yeast EVs suggesting conservation of the sorting machinery and demonstrating that yeast EVs can be humanized. This synthetic biology-based, proof-of-concept study establishes S. cerevisiae as a platform to engineer and bioproduce designer EVs for many applications. GRAPHICAL ABSTRACT O_FIG O_LINKSMALLFIG WIDTH=167 HEIGHT=200 SRC="FIGDIR/small/710173v1_ufig1.gif" ALT="Figure 1"> View larger version (52K): org.highwire.dtl.DTLVardef@1cf61dcorg.highwire.dtl.DTLVardef@21f412org.highwire.dtl.DTLVardef@11ecde9org.highwire.dtl.DTLVardef@160b3f7_HPS_FORMAT_FIGEXP M_FIG C_FIG HIGHLIGHTS AND TOC BLURBO_LIsynthetic biology-based system was optimized to engineer EVs in S. cerevisiae C_LIO_LIEV scaffolds can be sorted to yeast EVs C_LIO_LIis an efficient scaffold to sort proteins into yeast EVs C_LIO_LIS. cerevisiae can be used to engineer designer EVs for drug delivery C_LI Extracellular vesicles (EVs) are a promising new modality for drug delivery. However, designer EVs must be engineered to broaden applications and improve efficacy. Here, Bouffard et al. optimize methods rooted in synthetic biology to genetically engineer EVs in S. cerevisiae, a yeast commonly used to manufacture biological drugs. They find that ectopically expressed human EV scaffolds (CD63, ExoSignal, PDGFR) can be sorted to yeast EVs, but Bro1 - the yeast ortholog of ALIX - was most efficient at sorting GFP into EVs. This proof-of-concept study demonstrates a single DBTL (design-build-test-learn) cycle that can be used to develop designer EVs for therapeutic applications.

Large language models and generative protein design promise to accelerate biotechnology, but it remains unclear whether they can engineer dynamic megasynth(et)ases whose activity depends on transient, context-specific domain interfaces. Non-ribosomal peptide synthetases (NRPSs) are an especially demanding target, yet a high-value one because they produce many clinically important natural products and offer a route to analogs that are often difficult or impractical to access by chemical synthesis. Here we integrate pretrained generative models (ESM3, ProteinMPNN and EvoDiff) with design-build-test-learn cycles and data-guided prioritization to generate 76 de novo thiolation (T) domains. We built and tested 578 recombinant NRPS variants in vivo spanning minimal, full-length and hybrid assembly lines. AI-designed T-domains supported product formation across architectures, enabled catalytically active hybrids at recombined junctions and increased yields by up to [~]3-fold relative to NRPSs carrying the native T-domain. A representative design showed improved soluble expression, refolding, and a 12 {degrees}C higher melting temperature, while molecular dynamics simulations indicated preserved global stability but reshaped, state-dependent interdomain contact networks. Together, these results establish generative design as an effective route to context-conditioned optimization and reprogramming of biosynthetic assembly lines.

Legitimacy screening, the process of verifying the identity and purpose of customers ordering synthetic nucleic acids, is a primary safeguard against the misuse of synthetic biology. However, the associated costs discourage the adoption of screening practices. To evaluate whether AI tools can facilitate this process, we tested five large language models on five verification tasks using customer profiles of life sciences researchers from around the world. We compared AI performance against an expert human baseline on flag accuracy, source quality, source fidelity, and cost. The best-performing model, Gemini 2.5 Pro aided by four bibliographic and sanctions APIs, achieved comparable flag accuracy to the human baseline (90% and 89%, respectively). Gemini 2.5 Pro outperformed the human baseline on source quality and fidelity, at roughly one-tenth of the cost ($1.18 vs. $14.04 per customer). For information-gathering tasks, which excluded the human review step, costs averaged $0.23 per customer, around 50 times cheaper than human screening. These results support piloting AI-assisted legitimacy screening at providers of synthetic nucleic acids and other dual-use biotechnology products, with AI systems handling information gathering and human reviewers retaining authority over order fulfillment decisions.

Statement for withdrawalThe authors have withdrawn their manuscript because it was submitted and made public without full consent of all authors. Therefore, the authors do not wish this work to be cited as reference for the project. If you have any questions, please contact the corresponding author.

In vitro directed evolution in synthetic microcompartments can generally support the evolution of genes with functions beyond affinity. The main challenge in the implementation of this strategy is the need to incorporate no more than a single DNA template molecule per microcompartment, thereby establishing a robust genotype-phenotype linkage, but which results in slow, inconsistent in vitro transcription and translation (IVTT) and poor DNA recovery after selection or screening. To address this challenge, we previously developed CADGE (Clonal Amplification-enhanceD Gene Expression) a strategy that allows the clonal amplification of linear gene-encoding DNA and coupled, in situ transcription-translation of the gene of interest. Here, we show that clonal amplification is highly sensitive to the cell-free systems composition and that robust, highly efficient cell-free DNA amplification via the CADGE platform can be achieved by replacing standard vendor-supplied energy mixes with DNA replication-optimized, homemade counterparts.

We have developed RepTiles, a system for generating traces of random DNA for synthetic genomics applications. RepTiles reduced the burden associated with the manual design of long DNA from standard biological parts. We are applying RepTiles to the construction of random DNA segments that will form part of a neochromosome in Physcomitrium patens. RepTiles has a base DNA collection of 52 1.7 kb chemically synthesised Phytobricks mini-chunks. We provide a user-friendly web application that facilitates the design of DNA assemblies based on the base collection. The system generates assembly plans for generating chunks using Golden Braid. The resulting chunks can then be assembled into megachunks using Transformation Associated Recombination cloning. It thus supports the generation of ultra-long synthetic DNA from phytobricks, contributing to the vision of synthetic plant genomics from modular parts to complete genomes.

The Modular Cloning (MoClo) and PhytoBrick standards have revolutionized plant synthetic biology by establishing a standardized, hierarchical assembly grammar. However, as the engineering of complex metabolic pathways, multi-trait stacks, and synthetic gene circuits expands, existing toolkits hit practical boundaries in assembly capacity and fixed grammars. To overcome these bottlenecks, we present MozClo, an expansion of the MoClo/PhytoBrick architecture. MozClo expands the standard Level 1 assembly framework to 10 positions using new L1 acceptors, end-linkers and dummy parts. We also identify and resolve a critical, sticky-end collision at L1 position 7 that has caused assembly failures during L2 cloning of large plasmids. To address commercial DNA synthesis length constraints and to lower cloning costs, we designed a universal 5-in-1 gene fragment multiplexing system. This architecture embeds up to five distinct parts flanked by orthogonal pairs of BpiI restriction sites into a single synthesized fragment, allowing them to sort independently into their respective L0 acceptor plasmids while maintaining complete modular flexibility of part types. Finally, we provide Level 2 cloning backbones with built in selection genes for common soybean transformation methods to facilitate downstream plant selection. Together, these advancements reduce DNA synthesis overhead and accelerate the construction of complex multigene payloads for plant biotechnology.

Compact tissue-specific promoters are highly desirable for gene therapy because viral vectors possess limited packaging capacity. However, existing promoter engineering strategies rely primarily on rational design or de novo sequence generation and lack efficient approaches for compressing long native promoters while preserving regulatory specificity. Although genome foundation models have substantially improved sequence-to-function prediction, they have not been effectively translated into computational platforms for promoter engineering. Here, we present VirEvo, a computational promoter engineering framework that integrates a virtual dual-luciferase assay (VirDLA), genome-foundation-model-guided genetic evolution, and an orthogonal Pan-Tissue Consistency Filter (PTCF). VirDLA introduces an internal-reference normalization strategy inspired by dual-luciferase reporter assays, enabling relative comparison of promoter activity across tissues without retraining AlphaGenome. Guided by these normalized activity scores, VirEvo iteratively optimizes promoter selectivity, off-target activity, and sequence length. Using the human p16INK4a promoter as a proof of concept, VirEvo evolved a compact synthetic promoter, SRP2M, of only 398 bp, representing an 85.9% reduction in sequence length. Experimental validation using dual-luciferase reporter assays in senescent IMR90 fibroblasts demonstrated that SRP2M retained 77% of wild-type senescence selectivity while reducing basal leakage to 52% of the wild-type level. Together, these results demonstrate the feasibility of genome-foundation-model-guided promoter engineering. VirEvo provides a generalizable framework for designing compact tissue-specific regulatory elements and extends the application of genome foundation models from functional prediction to synthetic regulatory engineering.