Enhanced genetic stability and expression control in growth-decoupled continuous two-stage E. coli fermentations using plasmid-dependent thyA auxotrophic selection Authors: Florian Simon, Patrick Stargardt, Natalia Danielewicz, Florian Weiß, Johanna Berein, Gerald Striedner, Juergen Mairhofer Published: 2025 ABSTRACT AbstractContinuous two-stage E. coli fermentations offer potential for high-efficiency bioprocesses but are often limited by plasmid instability, genetic mutations, and unintended expression during seed phases. In this study, we aimed to overcome these limitations by employing a plasmid-dependent auxotrophic selection system, specifically using a thymidine auxotrophy (thyA deletion), in conjunction with a series of plasmid modifications to enhance stability and expression control. We engineered the E. coli strain enGenes-eXpress V2 ΔthyA and constructed modified plasmids containing thyA along with regulatory elements to maintain plasmid stability and minimize basal expression. The modified strains were evaluated in continuous two-stage fermentations under carbon-limited conditions. Our results indicate significant reduction in plasmid loss, improved population homogeneity, and suppressed basal expression in non-induced phases. The addition of elements such as the cer (ColE1 resolution) site and a modified T7 promoter further enhanced plasmid stability and reduced basal expression levels. High-throughput screenings in microbioreactor setups confirmed that optimized constructs maintained a homogeneous producing population and suppressed non-producing cells over extended periods, which was validated by fed-batch cultivations and single-cell analyses. Finally, the two most promising constructs demonstrated high robustness in continuous two-stage chemostat fermentations lasting over 1000 h, maintaining stable GFP titers, plasmid concentrations, and cell dry mass throughout the process. Our findings demonstrate that the auxotrophic marker thyA-based selection system, combined with strategic plasmid modifications, can substantially improve the genetic stability and productivity of E. coli in continuous bioprocesses. This approach provides a robust platform for sustainable, antibiotic- free production in industrial biotechnology, highlighting its potential for scale-up in long-term continuous fermentations.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12934-025-02875-x.Keywords: Escherichia coli, enGenes-eXpress, Continuous biomanufacturing, Thymidine auxotrophy, Plasmid stability, Basal expression, T7 expression system Background Continuous bioprocessing has gained increasing interest in industrial biotechnology due to its potential to improve productivity, resource efficiency, and scalability compared to traditional batch and fed-batch processes [36]. Particularly in recombinant protein production, continuous systems offer substantial benefits by enabling stable, high-density cultivations and reducing downtime between production cycles. However, applying continuous fermentation strategies to Escherichia coli, one of the most widely used microbial production hosts, remains challenging due to genetic instability and plasmid loss, but also unintended gene expression during seed phases [58]. Genetic instability can occur at multiple levels and through diverse mechanisms, ultimately impairing expression of the protein of interest (POI) or overall product formation. In metabolite and enzyme production, even single base substitutions can reduce enzyme activity, while homologous recombination frequently undermines stability due to the reuse of regulatory sequences. In addition, transposable elements present in microbial chromosomes may spontaneously mobilize, leading to gene disruptions or altered transcription of adjacent genes [40]. Entire biosynthetic pathways can also be disrupted by structural rearrangements at various genomic loci caused by insertion sequences (ISs), which in most cases results in an increased population growth rate but a concomitant decline in product formation [39]. Heterologous expression systems are particularly prone to mutations, especially those imposing a high metabolic burden on the host such as the T7 expression system. Here, loss-of-function mutations frequently arise in the T7 RNA polymerase gene and/or in the T7 promoter [44]. These issues often lead to heterogeneous cell populations and reduced process efficiency, ultimately hindering robust and reliable production systems [58]. One of the major limitations in continuous single-stage chemostat processes is the rapid decline in productivity, often caused by the emergence of non-producing subpopulations. Striedner et al., [52] demonstrated that product formation in the T7-based E. coli HMS174(DE3) expression system ceased completely after only ten generations in a single chemostat. Because the T7 system drives very strong expression and imposes a substantial metabolic burden on the host, instability can arise even in the absence of induction due to leaky basal expression. This leaky expression is especially problematic when producing proteins that are toxic to the cells [10, 38]. One of the major consequences of such stressors, whether from leaky expression or metabolic burden, is an increased likelihood of plasmid loss and mutations [52]. To address these issues, two-stage chemostat systems have emerged as a promising alternative. The concept was first introduced in the mid-20th century, when Monod [28], described the growth of bacterial cultures and based on that Novick & Szilard [32], defined the basic principles of continuous culture in chemostats. By decoupling growth (seed reactor) and production (production reactor), this setup enables independent optimization of each phase. The first reactor is typically operated under conditions to maintain high biomass and low expression stress, while the second reactor is dedicated to protein production under induced conditions [43]. Thus, two-stage systems offer a clear advantage in this context, as they decouple growth and production, making the use of T7-based expression systems under continuous conditions feasible in the first place. The applicability of this system in two-stage continuous cultivations has already been demonstrated several times. Kittler et al., [19] were able to produce three soluble proteins with both BL21(DE3) and HMS174(DE3) over several days. Kopp, Kolkmann, et al., [21] achieved a similar result with the continuous production of inclusion bodies with BL21(DE3) over a longer period of time. This separation allows for reduced selective pressure and better control over expression kinetics, thereby extending process stability, increasing productivity, and minimizing adaptive evolution [20]. A comprehensive review on continuous processing with E. coli can be found elsewhere (Kopp, Slouka, et al., [22]). Beyond reactor configuration, the choice of expression system plays a crucial role in process stability. Host-RNA polymerase (host-RNAP)-based systems have shown promise due to their inherent genetic robustness. Mutations in the essential host RNAP are typically lethal, providing a natural form of protection against genetic drift [44]. However, these systems often suffer from limited expression strength. In contrast, T7-based systems offer significantly higher transcriptional activity [53] but are highly prone to instability [44]. Our approach aims to combine the strengths of both systems by introducing a “shielded” T7 RNA polymerase expression setup, inspired by the natural robustness of host-RNAP-based systems. Despite these improvements, several key challenges remain in long-term continuous fermentations: (i) plasmid loss due to insufficient selection pressure or multimer formation [2, 8], (ii) variation in plasmid copy number over time [8], and (iii) basal expression in non-induced phases [45], which can accelerate the emergence of non-producing subpopulations and genetic mutations. In this context, the carbon source plays a crucial role in regulating these expression systems. Glucose availability affects the activity of IPTG- and arabinose-inducible promoters through catabolite repression mediated by the cAMP - CRP complex. Under carbon-limited conditions, reduced glucose levels alleviate catabolite repression and enable inducer uptake, which is essential for proper induction during production. However, these same conditions can also increase promoter accessibility to the cAMP - CRP complex even in the absence of inducers, leading to elevated basal expression in non-induced phases [1, 42, 49]. This dual effect of carbon limitation represents a central challenge for maintaining tight transcriptional control in continuous fermentation systems. Plasmid instability represents, of the problems mentioned, a particularly critical one for robust bioprocessing. The literature offers various approaches to mitigate plasmid loss, including the optimization of plasmid copy number, engineering of stable host-plasmid systems, and the use of antibiotic-free selection mechanisms, such as toxin-antitoxin systems or auxotrophic complementation [23, 48]: i)Genome integration: One possibility to stabilize a continuous process is to integrate the GOI into the genome, instead of relying on plasmid-based expression. Although this enables stable cultivation over more than 40 generations in the non-induced state and induction at any time, evolutionary changes also lead to instability and thus to a decrease in productivity [52]. Genome integration: One possibility to stabilize a continuous process is to integrate the GOI into the genome, instead of relying on plasmid-based expression. Although this enables stable cultivation over more than 40 generations in the non-induced state and induction at any time, evolutionary changes also lead to instability and thus to a decrease in productivity [52]. ii)Growth-decoupled strains: The use of growth-decoupled E. coli strains, such as those employing T7 bacteriophage-derived Gp2, a 7 kDa protein, to inhibit host transcription, offers significant advantages, as the arrest of cell division not only prevents the formation of plasmid-free daughter cells but also substantially lowers the likelihood of mutation events [50]. This stabilization of the cells should also prevent the population collapse often mentioned in the literature [35]. Gp2 binds to the β′ subunit of the E. coli RNA polymerase and prevents the formation of transcriptionally active σ⁷⁰-dependent promoter complexes, effectively inhibiting host transcription while leaving elongating RNAP complexes unaffected [30, 46]. Controlled expression of Gp2 in the production phase enables inhibition of host transcription, thereby decoupling growth from recombinant protein synthesis and directing cellular resources toward product formation [50]. Growth-decoupled strains: The use of growth-decoupled E. coli strains, such as those employing T7 bacteriophage-derived Gp2, a 7 kDa protein, to inhibit host transcription, offers significant advantages, as the arrest of cell division not only prevents the formation of plasmid-free daughter cells but also substantially lowers the likelihood of mutation events [50]. This stabilization of the cells should also prevent the population collapse often mentioned in the literature [35]. Gp2 binds to the β′ subunit of the E. coli RNA polymerase and prevents the formation of transcriptionally active σ⁷⁰-dependent promoter complexes, effectively inhibiting host transcription while leaving elongating RNAP complexes unaffected [30, 46]. Controlled expression of Gp2 in the production phase enables inhibition of host transcription, thereby decoupling growth from recombinant protein synthesis and directing cellular resources toward product formation [50]. iii)Externally regulated auxotrophic selection: Fu et al. [11]. , demonstrated overexpression of plasmid-encoded and enzymatically active β-lactamase for at least 50 days with the IPTG inducible tac promoter in a two-stage chemostat approach, as long as neomycin sulfate was supplemented to the media. This strategy represents a form of externally controlled, product-based auxotrophy for plasmid and production stability. The antibiotic exerts selective pressure on the population, but the strain itself lacks inherent resistance. Only cells actively expressing β-lactamase can degrade the antibiotic and thereby survive, ensuring that the population is continuously enriched for producers [11]. Externally regulated auxotrophic selection: Fu et al. [11]. , demonstrated overexpression of plasmid-encoded and enzymatically active β-lactamase for at least 50 days with the IPTG inducible tac promoter in a two-stage chemostat approach, as long as neomycin sulfate was supplemented to the media. This strategy represents a form of externally controlled, product-based auxotrophy for plasmid and production stability. The antibiotic exerts selective pressure on the population, but the strain itself lacks inherent resistance. Only cells actively expressing β-lactamase can degrade the antibiotic and thereby survive, ensuring that the population is continuously enriched for producers [11]. One way to prevent plasmid loss is thymidine auxotrophy. The thyA gene in E. coli encodes the enzyme thymidylate synthase. This enzyme plays a crucial role in the biosynthesis of thymidine monophosphate (dTMP), a precursor for thymidine triphosphate (dTTP), which is essential for DNA synthesis and repair. Thymidylate synthase catalyses the reductive methylation of 2′-deoxyuridine-5′-monophosphate (dUMP) to 2′-deoxythymidine-5′-monophosphate (dTMP), using N⁵,N¹⁰-methylenetetrahydrofolate as a donor of both the methylene and hydride groups. This reaction represents the primary pathway for de novo dTMP synthesis and is critical for maintaining balanced dNTP pools [31]. Deficiency in thymidylate synthase can lead to impaired DNA synthesis and cell division making this an effective, antibiotic-free selection mechanism [3, 37]. Unlike antibiotic-based systems, auxotrophic selection does not exert extrinsic toxic pressure and is GMP-compatible. In this work, we present a plasmid-stabilized, antibiotic-free system for continuous E. coli fermentations, combining (i) thyA auxotrophy for selection, (ii) rational plasmid design to modulate plasmid copy number, reduce basal expression and shield the T7RNAP against mutations, and (iii) growth decoupling to suppress cell division in the production phase. We evaluate multiple expression architectures, including monocistronic and polycistronic thyA-GOI arrangements, and assess their impact on long-term productivity, genetic stability, and cell population dynamics. Together, these modifications aim to provide a robust and scalable platform for stable, long-term recombinant protein production under continuous cultivation conditions with E. coli. Materials and methods Bacterial strains and plasmids Escherichia coli K-12 NEB5-α (fhuA2Δ(argF-lacZ)U169 phoA gln V44 Φ80 Δ(lacZ)M15 gyrA96 recA1 relA1 endA1 thi-1 hsdR17) was purchased from New England Biolabs (NEB, Ipswich, USA) and used for all cloning procedures. Experiments in this study were performed with E. coli strains enGenes-eXpress V2 [51] (F– ompT gal dcm lon hsdSB(rB–mB–) [malB+]K−12(λS) araB::T7RNAP-tetA ::attTN7 < pAraB-Gp2-CAT>) and enGenes-eXpress V2 ΔthyA (F– ompT gal dcm lon hsdSB(rB–mB–) [malB+]K−12(λS) araB::T7RNAP-tetA ::attTN7 < pAraB-Gp2-CAT >ΔthyA). enGenes-eXpress V2 ΔthyA was generated by homologous recombination of a linear double stranded (dsDNA) cartridge, containing a shorter non-functional version of the thyA gene according to Sharan et al., [47]. Since thyA deficient strains lack a functional folate pathway, they are resistant to trimethoprim, a dihydrofolate reductase (DHFR) inhibitor [37]. Supplementation of 100 µg mL− 1 Thymidine (THY, Sigma Aldrich #9250-5G) and 10 µg mL− 1 Trimethoprim (TRI, Sigma Aldrich #T7883-5G) was used for eradicating false positive clones, supplemented into liquid or agar plates of LB Veggitone. Selected clones were confirmed positive by PCR screening and Sanger sequencing. Plasmid pET30a(GFPmut3.1) was cloned as described elsewhere [50]. Unless otherwise described, all further modifications were inserted or removed by Golden Gate Assembly using Type IIS restriction enzyme BsaI-HF and T4 DNA ligase (New England Biolabs). All cloning procedures were performed by Q5® High-Fidelity DNA Polymerase according to the manufacturer’s instructions. Used plasmids, primers and important sequences in this study are listed in the supplementary material (Table S 1, Table S 2). Microbioreactor cultivations Simulated continuous two-phase cultivations were performed in 48-well flower plates (m2p-labs, Baesweiler, Germany) in a BioLector™ microfermentation system as described by Toeroek et al., [56]. C - limitation was set by the constant degradation of EnPump200 dextran (EnPresso GmbH, Germany) by addition of 0.6% v v− 1 of the enzyme amylase (EnPresso Reagent A, EnPresso GmbH, Germany). This corresponds to a glucose release rate of approximately 18 g L− 1 in 24 h, or 0.75 g L− 1 h− 1 (EnPump 200 Enpresso GmbH, Germany, product datasheet), which results in a growth curve that would correspond to a linear feed profile. Media composition is described elsewhere [12]. The working volume in each well was 800 µL. Inoculation of passage 0 was performed from an overnight culture in semi synthetic media (SSM) containing 50 µg mL− 1 kanamycin for constructs carrying a plasmid with the kanamycin resistance gene, and 100 mM thymidine to constructs of categories 2, 3, and to the empty enGenes-eXpress V2 ΔthyA. SSM is described elsewhere [9]. The initial optical density OD600 was set to 0.1. To mimic conditions in a real bioreactor, 1 g L− 1 glucose was added at passage 0, and 100 mM thymidine was added to constructs mentioned above. The inoculation density and the culture medium were designed to allow 5 generations per passage. At the end of passage 0, cells were washed with water to remove the remaining thymidine. One passage lasted 24 h, after which the cells were transferred to new wells. Passaging was done from non-induced wells. From passage 1 onwards, neither glucose was added to immediately create C-limiting conditions nor thymidine was added. This means that any growth of constructs based on thymidine selection can only be attributed to sufficient expression of ThyA. Recombinant gene expression was induced with 0.5 mM IPTG and 100 mM L-arabinose after 12 h of cultivation. GFPmut3.1 expression levels were monitored at 488 nm excitation and 520 nm emission wave length. Signals are given in relative fluorescence units (rfu). Cell dry mass (CDM) in g L− 1 was calculated from the light scattering signal using calibration settings obtained by linear regression analysis. The cycle time for all parameters was 10 min. At the end of each passage, 750 µL per well were sampled for subsequent plasmid analysis. Cells were centrifuged for 10 min at 10,000 rpm and the supernatant was removed. Fed-batch cultivations All clones were cultivated in a DASGIP parallel bioreactor system (Eppendorf AG), featuring four parallel vessels with a maximum working volume of 1.8 L. Each system included pH and dissolved oxygen (DO) probes (Hamilton Bonaduz AG, Switzerland) for online monitoring. Biomass yield was calculated using a substrate yield coefficient (Yx/s) of 0.33 g g− 1. The batch fermentation took place at 37 °C, with the temperature reduced to 30 °C during the feeding phase. pH was maintained at 7.0 ± 0.2 by adding a 12.5% ammonia solution, and dissolved oxygen was kept at 30% through a cascade control of stirrer speed, aeration rate, and gas composition. Foaming was controlled using the antifoam Struktol J673A (Schill + Seilacher GmbH, Germany). Inoculation was performed as described above with an ODini of 0.1 from an overnight culture in SSM medium, supplemented with 50 µg mL− 1 kanamycin and 100 mM thymidine in constructs where necessary. The fermentation itself was done without the addition of antibiotics, however, with 10 µg mL− 1 thymidine for category 2 und 3 constructs since those require C limitations for basal expression dependent growth. The batch medium was designed for a CDM concentration of 10 g L− 1. This was followed by an exponential feed phase with a µ of 0.114 h− 1 for 3 generations until a CDM of 42.6 g L− 1 was reached after 18 h. Subsequently, induction was performed with 1 mM IPTG and 100 mM L-arabinose and feeding was switched to a linear feed profile with a decreasing µ starting at 0.100 h− 1. This phase lasted 10 h or 1 generation until a theoretical CDM of 55.9 g L− 1 was reached. All constructs were cultivated non-induced and induced. Two-stage chemostat cultivations Continuous fermentations were carried out in the DASGIP parallel bioreactor system (Eppendorf AG) with different dilution rates (0.025–0.200 h− 1), but always with the same media composition. The batch medium for the seed reactor was designed for a CDM concentration of 20 g L− 1, that for the production reactor for 14 g L− 1. As in fed-batch cultivation, the batch phase was carried out at 37 °C and 30% DO, controlled by a cascade. After the batch phase, the system was immediately switched to continuous mode. The temperature was reduced to 30 °C in both reactors and the production reactor was induced with a pulse of IPTG and L-Arabinose to get a concentration of 1 mM and 50 mM respectively. For the remaining process time, the inducer concentration was kept constant at 1 mM and 50 mM via the feed. A constant antifoam feed of 0.3 mL h− 1 was applied in all reactors to prevent foam formation. The volume was kept constant using an immersion tube adjusted to the correct height of the liquid surface, ensuring that the bleed pump operated at a higher rate than the feed pump. This was done prior to the process at fixed agitation and aeration settings. Thus, in the continuous phase agitation was fixed to 1500 rpm and aeration to 100 sL h− 1. The feed medium for the seed reactor was designed for a CDM of 20 g L− 1, as in the batch medium. The feed medium for the production reactor was designed with a glucose concentration of 142 g L− 1. Through the in-line dilution with the efflux from the seed reactor the theoretical CDM in the production reactor amounted to 28 g L− 1. Assuming an immediate and complete decoupling of growth [50], a CDM of 14 g L− 1 would be set in the production reactor. For example, if the volume in the seed reactor was 400 mL, 40 mL h− 1 were fed at a D of 0.100 h− 1 and 40 mL h− 1 were transferred to the production reactor. The volume in the production reactor was 570 mL, so 17 mL h− 1 were fed here, which also results in a total D of 0.100 h− 1. Media composition for fed-batch and chemostat cultivations is listed in the supplementary (Table S 3). Offline analytics Offline measurement of GFP fluorescence Offline fluorescence measurements were performed with a Tecan Infinite 200Pro analyzer. For GFPmut3.1, an excitation wavelength of 485 nm and an emission wavelength of 520 nm were used. Quantification was achieved through calibration with an in-house purified target protein [17]. Single cell analysis by flow cytometry Flow cytometry analysis was conducted at each sampling point, using a CytoFLEX S flow cytometer (Beckman Coulter, Brea, USA) to monitor the formation of weakly induced or noninduced subpopulations during cultivation. The cell suspension was diluted in sterile-filtered PBS buffer at a 1:2025 ratio (achieved through a 1:45 dilution followed by a second dilution). The flow cytometer recorded forward scatter (indicating particle size), side scatter (indicating particle granularity), and fluorescence of GFPmut3.1 at 488/525 nm using the FL1 Channel (525BP40-A). MFI was calculated by geometric mean of FL1-A signal of induced cells. A total of 50,000 cells were analyzed per sample, and the data were processed using Kaluza Analysis software version 2.1 (Beckman Coulter). pDNA analysis For pDNA quantification, 750 µL from BioLector™ experiments or 7/OD samples from fed-batch and chemostat cultivations were used. These amounts provide sufficient material for HPLC analysis while keeping the bacterial culture below the recommended 15 OD units below 20 µg of pDNA, thereby avoiding column overload. The corresponding aliquots were centrifuged at 10,000 g for 10 min and the supernatant discarded. Cell lysis and plasmid extraction was performed using the Monarch® plasmid miniprep kit (New England Biolabs, Ipswich, USA). 600 µL of the neutralized lysate was loaded onto the spin column to ensure equal volumes for all samples. Final elution from the miniprep kit column was done in 100 µL preheated (50 °C) elution buffer. The analysis of the plasmid isoforms and the total plasmid content was performed with a Waters Arc HPLC (Waters Corporation) using a diode-array detector (DAD) (detector wavelength: 260 nm 4 nm− 1) on a CIMac™ pDNA − 0.3 monolith analytical column (Sartorius BIA Separations d.o.o.). 30 µL of the eluted plasmid samples were injected onto the HPLC column. Buffer A was a 50 mM Tris-HCl (pH 8.0), Buffer B a 50 mM Tris-HCl, 1 M NaCl (pH 8.0) solution. Plasmids were separated by a linear gradient with a slope of 1% B min− 1 (10 mM Cl− min− 1) at a flow rate of 1.0 mL min− 1. The column compartment temperature was set to 25 °C, the auto-sampler thermostat temperature to 4 °C. Results are given as the total integrated area of all isoforms per mg of analyzed sample. Results Evaluation of two-stage chemostat approach with growth decoupled E. coli We investigated the performance of continuous two-stage fermentations using the enGenes-eXpress V2 strain expressing the model protein GFPmut3.1 from a standard pET30a plasmid across a range of dilution rates (0.025–0.200 h⁻¹). The aim was to assess the maximum potential production duration under varying dilution rates and identify factors limiting the performance of continuous growth-decoupled E. coli processes. Across all tested conditions, the process performance rapidly declined due to three major issues: i.Basal expression in the seed reactor: We observed significant leaky expression of GFPmut3.1 even in the absence of inducer. This effect only occurred under carbon-limiting conditions and was more pronounced at lower dilution rates (Figure S 1, Figure S 2). During the batch phase the T7 system was tightly repressed, and no basal GFP expression could be observed. Basal expression increased upon starting the chemostat mode and its maximum was reached after ~ 154 h. At this point, cells in the seed reactor reached fluorescence levels comparable to the induced production reactor, indicating full induction (Figs. 1A, S1). Basal expression in the seed reactor: We observed significant leaky expression of GFPmut3.1 even in the absence of inducer. This effect only occurred under carbon-limiting conditions and was more pronounced at lower dilution rates (Figure S 1, Figure S 2). During the batch phase the T7 system was tightly repressed, and no basal GFP expression could be observed. Basal expression increased upon starting the chemostat mode and its maximum was reached after ~ 154 h. At this point, cells in the seed reactor reached fluorescence levels comparable to the induced production reactor, indicating full induction (Figs. 1A, S1). ii.Loss of population homogeneity and plasmid stability: Within just 12 h of chemostat operation, the cell population split into producing and non-producing subpopulations in both the seed and production phase. After 24 h (~ 1 generation) ~ 85% of the population in the production reactor still expressed GFP, but this dropped steadily, reaching nearly 0% after 276 h (~ 10 generations). At the same time, the proportion of non-producers in the production reactor (∼12% after 24 h) increased to almost 100% after 276 h (Fig. 1C). The increase of the non-producing population was indirectly proportional to the decrease in the plasmid copy number and the total amount of plasmid per sample, which is a strong indication of plasmid loss (Fig. 1B). In addition, plasmid multimerization and plasmid degradation seemed to enhance the problem of cultivation instabilities (Figure S 3). Loss of population homogeneity and plasmid stability: Within just 12 h of chemostat operation, the cell population split into producing and non-producing subpopulations in both the seed and production phase. After 24 h (~ 1 generation) ~ 85% of the population in the production reactor still expressed GFP, but this dropped steadily, reaching nearly 0% after 276 h (~ 10 generations). At the same time, the proportion of non-producers in the production reactor (∼12% after 24 h) increased to almost 100% after 276 h (Fig. 1C). The increase of the non-producing population was indirectly proportional to the decrease in the plasmid copy number and the total amount of plasmid per sample, which is a strong indication of plasmid loss (Fig. 1B). In addition, plasmid multimerization and plasmid degradation seemed to enhance the problem of cultivation instabilities (Figure S 3). iii.Genetic instability of the expression cassette: Sequencing of single colonies revealed that in roughly 2% of cases the T7 RNA polymerase expression cassette was completely deleted while the Gp2 gene remained intact, confirming the absence of contamination (Figure S 4). Genetic instability of the expression cassette: Sequencing of single colonies revealed that in roughly 2% of cases the T7 RNA polymerase expression cassette was completely deleted while the Gp2 gene remained intact, confirming the absence of contamination (Figure S 4). Challenges in achieving stable continuous production in E. coli and the proposed solution. Key limitations include high basal expression, illustrated by fluorescent biomass pellets under UV light in the seed reactor at D = 0.025 h⁻¹ (A). The image was taken at 154 h, when fluorescence intensity between seed and production reactors had already become nearly indistinguishable. Plasmid instability, represented by multimerization and plasmid loss, is shown in HPLC chromatograms of pDNA samples collected at different time points during continuous fermentation (B). The x-axis indicates retention time [min], the y-axis absorption units [AU], and the z-axis the corresponding cell generations at the sampling time. The chromatograms reveal distinct pDNA isoforms, with the largest peak corresponding to multimers. Strikingly, after only ~ 6 generations, most isoform peaks had already disappeared, reflecting severe plasmid loss. These instabilities ultimately lead to population heterogeneity in the production reactor (D = 0.100 h⁻¹), as evidenced by single-cell flow cytometry (C). Here, the x-axis shows GFP fluorescence intensity [rfu] and the y-axis cell counts. The right-hand peaks correspond to producing cells, while the left-hand peaks represent non-producers, illustrating the gradual shift in population composition over time. To overcome these limitations, we propose a combined stabilization strategy (D): growth decoupling via phage-derived Gp2 together with thymidine auxotrophy. The latter was implemented by deleting the chromosomal thyA gene, forcing cells to retain the plasmid carrying the functional thyA, which reduces basal expression and enhances plasmid stability, thereby enabling robust and sustained protein production Thus, the main goal of this study was to develop a system that overcomes the three main limitations of continuous fermentation using E. coli as host: (i) basal expression in the seed phase, (ii) plasmid loss, and (iii) genetic instability of the recombinant gene expression unit. Genetic engineering of host strain and plasmid optimization Since all three problems mentioned above strongly influence the performance of continuous fermentations, we decided to implement a thymidine auxotrophy in our originally used strain enGenes-eXpress V2 by deleting thyA, resulting in strain V2 ΔthyA. To couple plasmid retention with cell survival we reintroduced thyA into the original pET30a plasmid in various positions and configurations (Figs. 1D and 2). In addition, individual elements on the plasmid were modified, inserted, or removed to improve plasmid stability and to investigate their respective effects. These are: placIQ: point mutation in the promoter of the lac repressor protein LacI to achieve an overproduction and thus tighter repression of the GOI [13]. placIQ: point mutation in the promoter of the lac repressor protein LacI to achieve an overproduction and thus tighter repression of the GOI [13]. cer (ColE1 resolution) element: site-specific recombination sequence derived from the ColE1 plasmid, which plays a key role in the maintenance and stability of plasmid DNA by preventing multimerization and resolution of multimers. The recombination is catalysed by the XerC/XerD recombinases, thereby converting plasmid multimers back into monomers. The cer-mediated resolution acts post-replicatively and therefore does not directly influence plasmid replication control or the plasmid copy number [2]. Cer was inserted by replacing the f1 ori. cer (ColE1 resolution) element: site-specific recombination sequence derived from the ColE1 plasmid, which plays a key role in the maintenance and stability of plasmid DNA by preventing multimerization and resolution of multimers. The recombination is catalysed by the XerC/XerD recombinases, thereby converting plasmid multimers back into monomers. The cer-mediated resolution acts post-replicatively and therefore does not directly influence plasmid replication control or the plasmid copy number [2]. Cer was inserted by replacing the f1 ori. pT70.09: point mutation in the wild type T7 promoter to lower the transcription capacity to 9% [6, 15]. pT70.09: point mutation in the wild type T7 promoter to lower the transcription capacity to 9% [6, 15]. ΔlacO: removal of the lac operator sequence (only for thyA) to simplify the transcription of thyA from T7 promoters in the non-induced state. ΔlacO: removal of the lac operator sequence (only for thyA) to simplify the transcription of thyA from T7 promoters in the non-induced state. This resulted in 4 main construct categories, categories 1–3 still carry the kanamycin resistance gene, whereas category 4 is antibiotic marker-free: Modified pET30a plasmids: Here, only individual elements on the original plasmid or in the T7 promoter sequence were changed without inserting the thymidine selection. Modified pET30a plasmids: Here, only individual elements on the original plasmid or in the T7 promoter sequence were changed without inserting the thymidine selection. Polycistronic T7 expressed thyA variants: The model protein GFPmut3.1 and thyA are located under the same T7 promoter, separated by a spacer and another weaker RBS for ThyA. Polycistronic T7 expressed thyA variants: The model protein GFPmut3.1 and thyA are located under the same T7 promoter, separated by a spacer and another weaker RBS for ThyA. Monocistronic T7 expressed thyA variants: GFPmut3.1 and thyA are under separate T7 promoters, which are only separated from each other by a spacer. GFPmut3.1 has the wild-type T7 promoter, whereas thyA has a weaker T7 promoter (9% transcription rate compared to wt T7 promoter) to avoid overexpression of ThyA [7, 15]. Constructs from categories 2 and 3 are dependent on the basal expression of T7RNAP for the transcription of thyA and thus for the survival of the cells. This mainly occurs under C-limited conditions, which makes supplementation of thymidine in the batch phase necessary. However, they have the great advantage that they shield the entire T7 system from negative mutations, as otherwise the cells could not survive. Monocistronic T7 expressed thyA variants: GFPmut3.1 and thyA are under separate T7 promoters, which are only separated from each other by a spacer. GFPmut3.1 has the wild-type T7 promoter, whereas thyA has a weaker T7 promoter (9% transcription rate compared to wt T7 promoter) to avoid overexpression of ThyA [7, 15]. Constructs from categories 2 and 3 are dependent on the basal expression of T7RNAP for the transcription of thyA and thus for the survival of the cells. This mainly occurs under C-limited conditions, which makes supplementation of thymidine in the batch phase necessary. However, they have the great advantage that they shield the entire T7 system from negative mutations, as otherwise the cells could not survive. Constitutive expressed thyA variants: The kanamycin resistance cassette was removed and thyA with the constitutive bla promoter was inserted in the opposite direction instead. Since thyA is transcribed by a constitutive promoter, which is independent of glucose availability, there is no need to supplement thymidine. The idea here was to exploit the inefficiency of the T7 terminator and to improve the transcription of thyA via a possible readthrough of the T7RNAP [26]. This category allows antibiotic-free selection and cloning. Constitutive expressed thyA variants: The kanamycin resistance cassette was removed and thyA with the constitutive bla promoter was inserted in the opposite direction instead. Since thyA is transcribed by a constitutive promoter, which is independent of glucose availability, there is no need to supplement thymidine. The idea here was to exploit the inefficiency of the T7 terminator and to improve the transcription of thyA via a possible readthrough of the T7RNAP [26]. This category allows antibiotic-free selection and cloning. Visual representation of all clones used in this work. All constructs are based on the pET30a backbone and grouped into four categories. A: Modified pET30a plasmids: Individual elements of the original plasmid were altered. B: Polycistronic thyA variants: GFPmut3.1 and ThyA are co-expressed under a single T7 promoter. C: Monocistronic thyA variants: GFPmut3.1 and ThyA are each expressed under separate T7 promoters. D: Constitutive expressed thyA variants: ThyA is driven by the constitutive Bla promoter, inserted in reverse orientation in place of the kanamycin resistance gene A schematic overview of all cloned variants is shown in Fig. 2. Necessary DNA sequences are listed in the supplementary material (Table S 1, Table S 2). High throughput screening in simulated continuous microbioreactor cultivations To characterize growth behavior and basal expression levels under long-term conditions, we performed fed-batch like cultivations in 48-well microtiter plates in BioLector™. A complete list of all constructs, their growth behavior, and basal expression is summarized in Table 1. Growth and GFP curves are shown in the supplementary material (Figs. S 5 – S 8). Results were categorized as good/high (+), moderate (~), or bad/low (-) for growth and basal GFP expression. Desirable results are marked in green, undesirable in red, and moderate in black. Table 1Growth behaviour and basal expression analysis of constructs over four passages under non-producing conditionsThe study evaluated constructs from four categories: modified pET plasmids, polycistronic thyA variants, monocistronic thyA variants, and constitutive expressed thyA variants. Additionally, two control strains—enGenes-eXpress V2 and enGenes-eXpress V2 ΔthyA—were included. For constructs in categories 2 and 3, as well as the enGenes-eXpress V2 ΔthyA control strain, 100 mM thymidine was supplemented during the 0th passage since here thyA transcription is dependent on basal expression of the T7RNAP under carbon-limiting condition Growth behaviour and basal expression analysis of constructs over four passages under non-producing conditions The study evaluated constructs from four categories: modified pET plasmids, polycistronic thyA variants, monocistronic thyA variants, and constitutive expressed thyA variants. Additionally, two control strains—enGenes-eXpress V2 and enGenes-eXpress V2 ΔthyA—were included. For constructs in categories 2 and 3, as well as the enGenes-eXpress V2 ΔthyA control strain, 100 mM thymidine was supplemented during the 0th passage since here thyA transcription is dependent on basal expression of the T7RNAP under carbon-limiting condition Thresholds were defined based on the final CDM of the empty strains V2 (5.2 g L⁻¹) and V2 ΔthyA (4.9 g L⁻¹) in passage 0, chosen as benchmarks due to optimal growth conditions with minimal metabolic and selective burden. Basal expression was assessed relative to reference construct 1 (no modifications on plasmid or genome), with its mean of four passages serving as the benchmark (150 rfu g⁻¹ L⁻¹). Anything above this threshold was considered high basal expression. Categories were defined as follows: Growth: Good (CDM ≥ 3.4 g L⁻¹), moderate (1.7 g L⁻¹ ≤ CDM < 3.4 g L⁻¹), bad (CDM < 1.7 g L⁻¹). Growth: Good (CDM ≥ 3.4 g L⁻¹), moderate (1.7 g L⁻¹ ≤ CDM < 3.4 g L⁻¹), bad (CDM < 1.7 g L⁻¹). Basal Expression (BE): High (GFP ≥ 150 rfu g⁻¹ L), moderate (50 rfu g⁻¹ L ≤ GFP < 150 rfu g⁻¹ L), low (GFP < 50 rfu g⁻¹ L). Basal Expression (BE): High (GFP ≥ 150 rfu g⁻¹ L), moderate (50 rfu g⁻¹ L ≤ GFP < 150 rfu g⁻¹ L), low (GFP < 50 rfu g⁻¹ L). The two empty strains, V2 and V2 ΔthyA, exhibited similar growth curves and final CDM at passage 0. However, from passage 1 onward, V2 ΔthyA no longer grew, while V2 maintained a final CDM of approximately 4.7 g L⁻¹ over the following three passages. Basal expression was absent in both strains, with an average fluorescence of 22.4 rfu g⁻¹ L, indistinguishable from background noise. Category 1 constructs, which lacked thymidine selection, showed consistently good growth across all four passages. The only exception was construct 1.3 in passage 0, which failed to grow at the inoculated cell density. Reference construct 1 exhibited the highest basal expression, which gradually decreased over time, from 225.7 rfu g⁻¹ L in passage 0 to 122.9 rfu g⁻¹ L in passage 3. Individual elements contributed to basal expression reduction to varying degrees: the stronger placIQ promoter (construct 1.1) reduced basal expression by 37% (mean across four passages), while the cer element alone (construct 1.2) had only a minimal effect (1% reduction). The combination of placIQ and cer (construct 1.3) resulted in a 60% decrease, whereas the 9% T7 promoter (construct 6) had the strongest effect, reducing basal expression by 92%. When combined with the cer element (construct 6.2), the 9% T7 promoter led to a 77% reduction in basal expression. The polycistronic thyA variants (Category 2) exhibited poor to moderate growth, except in passage 0. By passage 3, all constructs except construct 2 had adapted and reached high final CDM. Construct 3 was the only one that maintained at least moderate growth in passages 1 and 2. Basal expression remained low across this category, with an overall reduction of more than 80%, except for constructs 2.2 and 3.3 in passage 3, where fluorescence reached 117.5 rfu g⁻¹ L and 57.7 rfu g⁻¹ L, respectively, placing them in the moderate expression range. The monocistronic thyA variants (Category 3), in which lacO was removed, exhibited excellent growth but also high basal expression, exceeding that of the reference construct by 166% (construct 4), 125% (construct 4.2), and 67% (construct 4.3). In constructs where lacO was retained, CDM was slightly lower but still within the good growth category, except for construct 5.3, which showed moderate growth in passages 1 and 2. In these constructs, basal expression was significantly reduced, with construct 5 showing a 66% reduction, construct 5.2 a 80% reduction, and construct 5.3 (placIQ + cer combination) achieving the strongest decrease (81%). The constitutive expressed thyA variants (Category 4) all exhibited good growth, with some constructs showing increased CDM in passage 0 due to higher inoculation density. The basal expression trends mirrored those of the kanamycin-resistant counterparts, with reductions achieved by modifying individual plasmid elements. For both the wild-type T7 promoter and the weaker 9% T7 promoter, the combination of placIQ and cer resulted in the greatest reduction, whereby pT70.09 alone contributed to the fact that all constructs were in the low basal expression range or just above. While construct 7 increased by 5%, all other constructs showed reductions of up to 91% (construct 7.3). The measured reductions were 22% for construct 7.1, 3% for construct 7.2, 73% for construct 8, 78% for construct 8.1, 74% for construct 8.2, and 80% for construct 8.3. Evaluation of thyA dependent plasmid stabilization under production conditions Yields are of crucial importance in industrially relevant processes. We aimed to assess the product formation kinetics of selected constructs under continuous production conditions, as well as the pDNA quantity and quality based on the isoforms. This time, two wells were inoculated with the same construct, one of which was induced with 0.5 mM IPTG and 100 mM L-arabinose after 12 h. After 24 h, two new wells were inoculated from the non-induced well. Thus, this setup allowed us to extend the cultivation over many generations, enabling the evaluation of plasmid stability and productivity under long-term conditions beyond those typically achieved in batch or fed-batch experiments. A graphical representation of the experimental setup is shown in Fig. 3A We decided to select one construct from each category with thymidine selection that had shown promising results. Construct 3 pT70.09-GFP-thyA was chosen from category 2, as it had the best growth behavior. From category 3, both constructs 4.2 pT7wt-GFP-pT70.09-ΔlacO-thyA_cer and 5.2 pT7wt-GFP-pT70.09-thyA_cer were selected. Although 4.2 had a very high basal expression, our hypothesis was that this would make induction and therefore production more efficient. 5.2 had the best performance in terms of growth and basal expression in its subcategory. There would have been several potential candidates in category 4. We opted for construct 8.3 pT70.09-GFP_cer_placIQ_pbla-thyA, as this construct showed excellent growth behavior, low basal expression and combines all elements. Therefore, we assumed it would provide the strongest contrast to the reference. As references, V2 ΔthyA empty was chosen to verify that cells are not viable without plasmid, as well as the original construct 1, without changes on the plasmid. Graphical representation of the experimental setup (A), growth curves (B), basal expression (C) and product formation kinetics (D) of selected constructs in simulated two-stage chemostat cultivations. The analysis included constructs 1, 3, 4.2, 5.2, and 8.3 over four passages, with the enGenes-eXpress V2 ΔthyA strain serving as a control. A two-stage chemostat was simulated by inoculating two wells with the same clone, one of which was induced after 12 h. After 24 h, two more wells were inoculated from the non-induced well (1:80 dilution). One passage corresponds to about 5 generations. In total, three further passages were made (A). The growth curves (B) represent the non-induced seed phases. For all constructs except 8.3, 100 mM thymidine was supplemented during the passage 0. Basal expression (C) shows the fluorescence signals per CDM in the non-induced seed phase. The product formation kinetics (D) illustrate the induced production phases, with induction performed using 100 mM L-arabinose and 1 mM IPTG. As the strains are growth-decoupled by induction (no growth), the growth curves of the induced samples are not shown in this graph The results could be replicated regarding growth and basal expression of previous non-induced cultivation experiments. Final CDM and results for basal expression were within a 5% variation to the mean for all constructs and passages (Fig. 3B). The only exception was construct 1 with slightly higher deviations in CDM and basal expression. As in the previous high throughput screening experiment, construct 4.2 had by far the highest basal expression. On average, this was 184% higher than in reference construct 1. All other constructs had such a consistently low basal expression that it could hardly be distinguished from noise. The reduction here was up to 88% (Fig. 3C). In the production phase, the expression of GFPmut3.1 was greatly improved with most constructs compared to the reference construct 1. With an increase of 37% over the mean value of the 4 passages, construct 5.2 was by far the best producing construct, followed by construct 4.2 (+ 18%) and construct 8.3 (+ 14%). In contrast, a decrease of -31% was observed for construct 3. In addition, 3 and 8.3 showed slower production kinetics and partially a decrease in the final yield in the last two passages (Fig. 3D). Plasmid analysis (Fig. 4) revealed major differences between constructs with and without cer. In the constructs with cer (4.2, 5.2 and 8.3), the specific total pDNA content was increased almost 10-fold, with 4.2 showing the highest increase of 810%. The proportion of the ccc fraction in constructs without cer (1 and 3) was only 3–14%, whereas the multimer fraction accounted for the largest proportion of the total pDNA content. In construct 3, the ccc content was even completely resolved by induction. The insertion of cer increased the ccc content up to 92%. In general, a decrease to almost complete loss of pDNA over the 4 passages was observed in the two non-cer constructs. In the cer constructs, however, this was kept constant or even increased over time. Constructs 4.2 and 5.2 had the highest pDNA content and also the highest ccc fraction. Induction increased the pDNA content in all variants. With a 103% increase, this was highest in 5.2 and lowest in 3 with 40%. The reported fold-changes in plasmid content are based on normalized values (pDNA peak area per mg CDM), obtained under standardized conditions, as detailed in the Materials and Methods section. Plasmid analysis of selected constructs in simulated two-stage chemostat cultivations over four passages in the non-induced (A) and induced (B) state. The analysis includes constructs 1, 3, 4.2, 5.2, and 8.3. The subfigures show the proportions of the four detectable pDNA isoforms OC (open circle), linear, CCC (covalently closed circular) and multimers, with the height of the bar representing the total amount of plasmid in the respective sample. The unit on the y-axis is represented as the integrated area under the curve of the chromatogram per biomass analysed. To simplify the comparison, the original number was divided by 10,000. Single cell evaluation and production kinetics in fed-batch cultivations To investigate the population distribution and production kinetics in more detail, a series of fed-batch cultivations were carried out on a benchtop scale in DASGIP® bioreactors. The production phase was chosen so that the cells were induced for 10 h, as this corresponds to the residence time in our standard continuous process at D = 0.1 h− 1. In addition to the selected constructs from the BioLector™ trials, 1.2 (pT7wt-GFPmut3.1_cer) and the constitutive expressed thyA variant 7.2 (pT7wt-GFPmut3.1_cer_pbla-thyA) were tested, as the results from the BioLector™ trials suggest that cer alone has a major influence on overall performance. Single cell analysis by flow cytometry of the 10 h sample showed an inhomogeneous population, both in the non-induced and in the induced cells of reference construct 1 as shown in Fig. 5A and B; Table 2. The larger producing population made up ≈ 83–86% of the total population. The smaller population (≈ 11%) was in a fluorescence range indistinguishable from the cell intrinsic fluorescence and was thus considered as plasmid-free cells. In contrast, all other constructs had a normally distributed homogeneous population with over 95% of the cells inside the chosen gate. Compared to the producing population of the reference construct 1, basal expression was greatly decreased in the other constructs (except 4.2), with the decrease in 5.2 being the largest with − 79%. As expected, basal expression was only strongly increased in 4.2 with + 289%. Although the mean fluorescence in the induced state was highest in the reference, a significantly higher dynamic range, which describes the ratio of fluorescence between induced to non-induced cells, was achieved for most constructs. This was particularly high for construct 5.2 with 142.9, construct 1.2 with 82.7 and construct 7.2 with 64.5. Although the basal expression of construct 8.3 was very low, only a slightly higher dynamic range could be achieved, as the fluorescence was also relatively low in the induced state. As none of the constructs reached a plateau within the 10 h production phase, it can be assumed that the capacity of the cells was not yet fully utilized. It is therefore even more important in a continuous process to use strains with the fastest possible production kinetics in order to achieve the best possible space time yield. The highest increase over time was observed for construct 7.2 with a final total yield of 260 mg g− 1. This means that the yield here was 33% higher than for the reference. The slowest increase and at the same time the lowest yields of 93.7 mg g− 1 (-52%) were measured for construct 8.3 (Fig. 5C). Flow cytometric analysis and production kinetics of selected constructs in 1 L fed-batch cultivations. The analysis includes constructs 1, 4.2, 5.2, 8.3, 1.2, and 7.2. The Graph presents flow cytometric data on population distribution and fluorescence intensity of single cells, with non-induced samples (A) and induced samples (B). Induction was performed with 100 mM L-arabinose and 1 mM IPTG, and data were collected at the end of fermentation after 10 h of production. Graph C illustrates the production curves, depicting the progression of total specific yields for the induced samples over the 10 h induction period. Table 2Single-cell evaluation and specific yields of selected constructs in 1 L fed-batch cultivations. The study assessed constructs 1, 4.2, 5.2, 8.3, 1.2, and 7.2 under both uninduced and induced conditions. Constructs 4.2 and 5.2 were supplemented with 10 µg mL-1 thymidine during the batch phase to allow growth under non-C limited conditions. Induction was performed using 100 mM L-arabinose and 1 mM IPTG.ConstructInduction% Total of “producing” populationX-Mean [rfu]% Decrease/IncreaseDynamic rangefinal total yield [mg g− 1]1 pT7wt-GFPnon-induced83.318 319-51.8-induced85.5949 427-195.54.2 pT7wt-GFP-pT70.09-ΔlacO-thyA_cernon-induced97.871 223288.89.6-induced98.8680 622-28.3174.45.2 pT7wt-GFP-pT70.09-thyA_cernon-induced94.93 765-79.4142.9-induced97.9537 819-43.4147.48.3. pT70.09-GFP_cer_placIQ_pbla-thyAnon-induced95.05 769-68.557.4-induced98.1331 153-65.193.71.2 pT7wt-GFP_cernon-induced93.010 906-40.582.7-induced96.7902 034-5.0219.07.2 (pT7wt-GFP)_cer_pbla-thyAnon-induced95.813 751-24.964.5-induced98.2887 565-6.5260.0 Single-cell evaluation and specific yields of selected constructs in 1 L fed-batch cultivations. The study assessed constructs 1, 4.2, 5.2, 8.3, 1.2, and 7.2 under both uninduced and induced conditions. Constructs 4.2 and 5.2 were supplemented with 10 µg mL-1 thymidine during the batch phase to allow growth under non-C limited conditions. Induction was performed using 100 mM L-arabinose and 1 mM IPTG. Plasmid analysis data show a clear multimer formation in the reference construct 1. At around 70 to 80%, the multimer fraction accounted for the majority of the pDNA, both in the non-induced and in the induced samples (Fig. 6A and B). Although the amount of pDNA remained constant in the non-induced state, a significant decrease was observed upon induction. The far more stable and replication-competent CCC isoform dominated in all other constructs with around 95%. Other isoforms were negligible. Thus, the multimer formation could be almost completely prevented by the simple insertion of cer. In principle, an increase in pDNA was observed in all constructs compared to the reference. This was particularly pronounced in construct 4.2, which was to be expected. At hour 8, the increase in the non-induced state was even 5.6 times higher than in the reference. The average increase in pDNA over the entire cultivation of the non-induced samples compared to the reference is as follows: Construct 4.2: 4.1-fold, construct 5.2: 2.3-fold, construct 8.3: 2.3-fold, construct 1.2: 1.9-fold, construct 7.2: 2.6-fold. The induction further boosted plasmid replication, leading to an even higher amount of pDNA. The only exception was construct 8.3 where the mean pDNA amount was only about 75% compared to the non-induced sample. The average increase was particularly high for the T7 based constructs 4.2 and 5.2 at + 63% and + 64%, respectively. For constructs 4.2, 5.2 and 7.2, it also appeared that the amount of plasmid gradually increased due to induction. What is particularly striking, however, is that in none of the constructs, except for the reference, plasmid loss was detected, neither in the non-induced nor in the induced state. Chromatograms of the individual constructs, both non-induced and induced, are shown in the Supplementary material (Fig. S9). Plasmid analysis of selected constructs in 1 L fed-batch cultivations in the non-induced (A) and induced (B) state. The analysis includes constructs 1, 4.2, 5.2, 8.3, 1.2, and 7.2. The subfigures show the proportions of the four detectable pDNA isoforms OC (open circle), linear, CCC (covalently closed circular) and multimers, with the height of the bar representing the total amount of plasmid in the respective sample. The unit on the y-axis is represented as the integrated area under the curve of the chromatogram per biomass analysed. To simplify the comparison, the original number was divided by 10,000. Sustained cultivation for 1000 h in two-stage continuous mode The two most promising constructs at this stage were construct 5.2, due to its exceptionally low basal expression, excellent plasmid quality and shielded T7RNAP, and construct 7.2, due to the highest yields achieved, also good plasmid quality, and the dispensability of additional thymidine as a result of to the constitutive expression of ThyA. These were tested for their long-term stability in two-stage chemostat cultivations. The dilution rate was set at 0.1 h− 1. As in the fed-batch cultivations, 1 mM thymidine was added to the batch medium for construct 5.2, however, not in the feed medium. In principle, stable cultivation with constant CDM concentration, constant yields and no signs of plasmid loss for about 750 h (31 days) could be achieved with both constructs. After exactly one month, the robustness and flexibility of the system was tested by additionally pulsing 100 µg mL− 1 thymidine, or 100 µg mL− 1 thymidine and 50 µg mL− 1 kanamycin (only at 5.2) and reducing the dilution rate to 0.025 h− 1. The concentrations refer to the final concentration in the reactor. These triggers were applied in both reactors with the aim of resetting the cells to their initial state. Thymidine was added to revitalize cells affected by thymidine starvation. In the case of construct 5.2, kanamycin was introduced to selectively eliminate cells that were already irreparably damaged by the starvation. Additionally, the dilution rate was reduced to increase basal expression, thereby enhancing thyA supply. In this experimental phase, cultivation was continued for a further 300 h. In total, GFP could thus be continuously produced with both constructs for over 1000 h. Construct 5.2 showed a short adaptation phase at the beginning of cultivation. In the seed reactor (Fig. 7A), the CDM decreased constantly in the first 3 days after the batch phase and finally stabilized at about 22 g L− 1. This decrease of the CDM coincided with the increase and a subsequent stabilization of the pDNA at a lower level. Basal expression was de facto undetectable with the analysis method used. The first thymidine shot could not induce any measurable influence. Only the combination of thymidine and kanamycin, but especially the lower dilution rate, drastically reduced the CDM, pDNA suddenly shot up and GFP could be detected in small but significant amounts. The last thymidine and kanamycin shot allowed all three parameters to normalize again. The adaptation phase in the production reactor (Fig. 7B) was particularly evident in the GFP yield. After a spike in the titer at the beginning of cultivation with over 3 g L− 1, this decreased constantly over the first 200 h and stabilized at a moderately high but constant value of around 1 g L− 1. This could also be maintained until the first thymidine shot. The decrease of GFP correlated with the pDNA, which was also higher at the beginning than after the adaptation phase. The CDM, on the other hand, increased and stabilized unexpectedly at about 29 g L− 1. The other process changes led to similar effects as in the seed reactor. Thymidine plus kanamycin and a lower dilution rate briefly led to a spike in GFP yield (2.5 g L− 1) and in pDNA, the CDM decreased and was able to recover with the second thymidine and kanamycin shot. Two-stage chemostat cultivation with construct 5.2 (A, B) and construct 7.2 (C, D). The graphs show the three critical parameters CDM, GFP and pDNA over a feeding period of more than 1000 h. A and C show the non-induced seed reactors, B and D show the induced production reactors. The dilution rate in all 4 reactors was 0.1 h-1. The arrows show the time at which something was changed in the process. Thy = 100 µg mL-1 thymidine pulse, Kan = 50 µg mL-1 kanamycin pulse. The first pulse was after 744 h, the second pulse after 792 h, dilution rate decrease to 0.025 h-1 was after 864 h, third pulse after 984 h Construct 7.2 also showed an adaptation phase in the seed reactor (Fig. 7C). Similar to 5.2, this was reflected in a decrease in CDM and an increase in pDNA. However, the CDM then stabilized at a slightly lower value of around 17 g L− 1. The pDNA also decreased after the adaptation phase but then fluctuated at regular intervals. At the times when the pDNA was high, GFP was also measurable in small quantities. Since construct 7.2 has no antibiotic resistance gene, only thymidine was pulsed and the dilution rate was changed. The first two thymidine shots had no significant influence on the behavior of the cells. Only the low dilution rate caused the CDM to drop to 10 g L− 1. GFP and pDNA remained unaffected. The last thymidine shot was finally able to normalize the CDM again. The behavior in the production reactor (Fig. 7D) was very similar to that in construct 5.2. After an initial GFP spike at about 3 g L− 1, the yield leveled off at about 1 g L− 1and remained constant until the first thymidine shot. The CDM was significantly lower than in 5.2, averaging 20–22 g L− 1. The pDNA fluctuated quite strongly, as in the seed reactor, but without any signs of plasmid loss. The first two thymidine shots briefly led to a spike in GFP yields. The lower dilution rate again led to a decrease in CDM, which was recovered by another thymidine shot. In total, over 100 g GFP could be produced with both constructs over a period of 45 days in 1 L scale at a cell density of ~ 30 g L− 1. Interestingly, in both constructs we observed a continuous increase in a population that was significantly larger, based on forward and side scatter parameters, than the originally defined main population during cultivation. The main population (Gate A), located within the initially defined region based on forward and side scatter, is labelled as “inside gate” in Fig. 8. In contrast, the emerging larger population (Gate B), along with a minor fraction of smaller cell fragments (which are statistically negligible due to their low abundance) is labelled as “out of gate,” as it lies outside the original gating boundary. Figure S 10 shows an example of the forward scatter/side scatter data of construct 5.2 in the seed reactor at hours 24 and 528. Flow cytometric analysis of a two-stage chemostat cultivation with construct 5.2 (A, B) and construct 7.2 (C, D). The graphs display the respective percentages of cells that are located within or outside the initially defined population gate, based on forward and side scatter parameters over a feeding period of more than 1000 h. A and C show the non-induced seed reactors, B and D show the induced production reactors. The dilution rate in all 4 reactors was 0.1 h-1. The arrows show the time at which something was changed in the process. Thy = 100 µg mL-1 thymidine pulse, Kan = 50 µg mL-1 kanamycin pulse. The first pulse was after 744 h, the second pulse after 792 h, dilution rate decrease to 0.025 h-1 was after 864 h, third pulse after 984 h In principle, an increase in “large cells” was observed in all four reactors, although the effect was less pronounced in the production reactors compared to their corresponding seed reactors. The most substantial increase occurred in the seed reactor of construct 5.2, where the proportion of large cells steadily rose and peaked at over 53% of the total population falling outside the main gate after 528 h. Notably, individual pulses or process alterations led, step by step, to a normalization of the population and the almost elimination of the large-cell subpopulation (Fig. 8A). A similar picture can be seen in the production reactor of construct 5.2, where the proportion of large cells amounted to a maximum of 35% (Fig. 8B). A clear increase in large cells was also observed in construct 7.2, with the population stabilizing at approximately 27% over time. The proportion of large cells peaked at 36% after 672 h. The first two thymidine pulses had little impact on normalizing the population distribution. Only the reduction of the dilution rate to 0.025 h⁻¹ resulted in a sudden decrease in the number of cells outside the main gate. The third thymidine pulse again showed no significant effect (Fig. 8C). In this case as well, the production reactor largely mirrored the dynamics of the seed reactor, although the proportion of large cells was consistently lower. The highest proportion of cells outside the main gate was observed after 360 h, reaching approximately 26% (Fig. 8D). Discussion Continuous bioprocesses are increasingly important in the context of climate change and sustainability due to reduced resource and energy consumption [18, 25, 27, 59]. However, maintaining stable long-term productivity in such continuous processes remains a challenge, particularly for E. coli which is known for fast adaptive evolution and its ability to escape production pressure, especially when using the T7 system for recombinant protein production [19, 44]. This study aimed to address the primary challenges associated with continuous E. coli fermentations, specifically basal expression in the seed phase, plasmid loss, and mutations in the T7 system. By implementing a thymidine auxotrophy and optimizing the plasmid design, we sought to improve the stability and efficiency of continuous fermentations. Our results suggest that thyA knockout is an excellent selection mechanism, as cells lacking the plasmid and thymidine supplementation failed to grow or adapt within 15 generations. This means that any growth can be attributed to sufficient expression of ThyA from the plasmid by the basal expression of the T7 system or by the host RNAP in the constitutive expressed thyA variants. The co-expression of GOI and ThyA from two separate T7 promoters (monocistronic variants) represents a particularly elegant and well-functioning variant, because here the basal expression of T7RNAP is utilized to ensure cell survival. Although mutations in the T7 system cannot be ruled out here either, the probability is greatly reduced. While the co-expression of different proteins or enzymes from two separate T7 promoters has been shown in literature [16, 24, 41], to our knowledge this has never been shown for auxotrophic genes. Especially the subgroup with two lacO sites, as for example in construct 5.2, is a promising variant as the basal expression was reduced to a minimum, but growth and also production was sufficiently good. With regard to basal expression, this finding is consistent with other works in literature. Schuller et al., [45]howed that the number, distance and symmetry of the lacO binding sites has a decisive influence on basal expression. Studies have demonstrated that the lac repressor can bind to multiple DNA sites simultaneously. This allows for cooperative interactions between distant lacO sites, facilitated by the lac repressor bridging them through the formation of a DNA loop structure [29]. This would also explain why basal expression was significantly higher in reference construct 1. Since there is only one lacO site on that plasmid, LacI can slow down transcription, but no DNA loop can form and therefore cannot completely prevent the T7RNAP from docking. What was very surprising was the unexpectedly high basal expression of the monocistronic variants with only one lacO, such as construct 4.2. Although we removed the lacO site for thyA to avoid a possible undersupply of thymidine, the basal expression of GFP was exorbitantly high instead. This is possibly due to an overlay effect. On the one hand, the phenomenon of missing DNA loop formation described above occurs. On the other hand, there is no element in the thyA cassette that would stop the T7RNAP from transcribing. This means that even in the non-induced state, the T7RNAP can bind and transcribe unhindered at the posterior T7 promoter. According to Mairhofer et al., [26] the termination efficiency of the native T7 terminator, which was also used in this study, is only about 79%. This means that there may be a readthrough over the entire plasmid, which in turn leads to further effects. Readthrough into the plasmid origin of replication may lead to an imbalance between RNA II and the negative regulator of it, RNA I [57]. Mairhofer et al., [26] described this phenomenon as Head-to-tail collision of T7RNAP with the E. coli RNAP, which then leads to runaway plasmid replication and thus to an increased pDNA content. This agrees well with our results, as the ΔlacO variants had by far the highest pDNA concentration. This effect is further enhanced by induction (Togna et al., [33]), which would explain why the pDNA content increased in the induced state in almost all constructs, but especially in the T7-based variants. In addition, it could be that the readthrough also impedes or prevents the transcription of lacI. This is transcribed in the opposite direction to GFP and thyA on the plasmids used, so T7RNAP and host RNA polymerase could collide with each other (head - to - head collision) and LacI could only be formed incompletely, too little or not at all. The stronger placIQ promoter was able to contain this problem to some extent. Within one group, basal expression was always lower in the lacIQ constructs compared to the constructs with normal promoter strength. This reinforces the hypothesis of insufficient LacI quantity. The poor overall performance of the polycistronic thyA variants can probably be attributed to the fact that the cells have too little ThyA expressed. This is a common phenomenon in prokaryotic operons, where several genes are connected in series under one promoter. In polycistronic cassettes genes located further downstream are often expressed less efficiently. This can happen at both the transcriptional and translational level. A study by Burkhardt et al., [5]emonstrated that the transcriptional distance between genes in an operon significantly influences gene expression levels, with expression decreasing for genes located further downstream. This is partly due to the structural organization of the mRNA, where the local degree of mRNA folding immediately downstream of the transcriptional start site correlates with the translation efficiency of the downstream gene. Secondary structures of the mRNA near the ribosomal binding site could obstruct the access of ribosomes and further hindering translation initiation. Another contributing factor is the progressive degradation of mRNA. This means that mRNA encoding downstream genes may be partially degraded before translation can occur, resulting in diminished expression levels [14].Bell-Pedersen et al., [4] showed that a transcription terminator is present in the thyA coding sequence. This serves as a terminator for the upstream genes, which, however, greatly reduces the amount of full-length thyA. On the other hand, ThyA is expressed monocistronic from its own mRNA as well as from polycistronic mRNA from upstream genes that extend into thyA. Thus, the amount of ThyA can be finely tuned by the cells. However, if the upstream genes are missing on the plasmids used, it could be that full length ThyA is not formed in sufficient quantities. All these factors together would mean that DNA synthesis is hindered by the lack of thymidylate synthase. The plasmid quantities in this category, such as in construct 3, would indicate that even pDNA synthesis is weaker compared to the other categories. This would also explain why the CDM concentrations in this category were also significantly lower. Finally, the cer element itself had a noticeably positive effect on the performance of the cells. The presence of an active recA gene in the strain used promotes homologous recombination, leading to plasmid multimer formation, which interferes with proper plasmid segregation and increases the risk of plasmid loss during cell division [54]. By inserting cer multimers were almost completely resolved, which also increased the amount of pDNA. As a result, there was no plasmid loss in fed-batch cultivations and no subpopulation formation. At this point, however, it must be mentioned that a 10 h fed-batch is maybe too short to facilitate reliable statements about the long-term stability of plasmids. In a study by Allen et al., [2], the introduction of cer into the plasmid backbone was shown to improve process performance in continuous cultivations. The cells were able to produce citramalate for a longer period of time than cells with non-cer plasmids. However, even with cer plasmids the decrease in productivity was drastic after approximately 60 h of feeding. This gives reason to believe that cer is indispensable for continuous cultivations but does not solve the problem of plasmid loss. We believe that the combination of cer and thymidine auxotrophy is an important step towards stable continuous recombinant protein production with E. coli. The two most promising constructs appeared to be 5.2 and 7.2. Construct 5.2 has the advantage that the T7RNAP is shielded from mutations. Construct 7.2 is expected to be partially protected from mutations due to transcriptional readthrough of the T7 RNA polymerase at the T7 terminator. An additional advantage is that thyA is transcribed constitutively, making its expression more independent and flexible with respect to the upstream GOI. It has been shown that a stable continuous process with constant yields over 31 days is possible. Interestingly, an adaptation phase was clearly recognizable for both constructs. In the seed reactors, this was evident from a decrease in CDM with a simultaneous increase in pDNA. We believe this effect is related to the transition from batch growth (without carbon limitation) to chemostat growth (under carbon-limited conditions). In chemostat operation, cells continuously lose precursors and metabolites due to the constant efflux of cells and nutrients. As a result, they are unable to accumulate or store essential compounds, which may affect their physiological state and robustness. This shifts the metabolism, which reduces the CDM. The cells presumably try to compensate for the lack of thymidine with more pDNA in order to provide more template for transcription. The apparently optimal pDNA content was quickly established at 5.2 after the adaptation phase. At 7.2, the cells appeared to find the optimal amount of pDNA iteratively, as the pDNA concentration fluctuated periodically. A possible reason for this could be that the host RNAP is used here for the transcription of thyA, which has a much lower transcription rate than the T7RNAP and also has more interference with its actual task. This means that the cells have to counteract this with the PCN. The same applies to the two production reactors, where, as expected, the pDNA content due to induction was far higher than in the respective seed reactors. The decrease in GFP yields correlates with the decrease in pDNA. This is probably also the reason why significantly more was produced at the beginning than after the adaptation phase. The transition from the adaptation phase to the constant phase is known in the literature and has already been described by Kittler et al., [20]. To shorten or even contain the adaptation phase, for example, a higher temperature could be used in the seed reactor to accelerate metabolism, or complex components could be added to the feed medium. Interestingly, no population collapse was observed in our continuous cultivations, despite the use of a synthetic minimal medium. This contrasts with the findings of Rai et al., [35], who reported a reproducible collapse of E. coli populations in chemostat cultures. In their study, cell density dropped sharply after an initial steady phase, accompanied by extreme morphological changes, before the population recovered through adaptive mutations, a phenomenon termed “adaptive rescue.” The absence of such collapse in our system suggests that factors such as medium composition, but especially a genetically stabilized strain [50] may play a key role in stabilizing population dynamics during long-term continuous cultivation. The significantly increased CDM in both production reactors, but especially in construct 5.2, was unexpected. This is probably partly due to the fact that growth decoupling, as described by Stargardt et al., [50], does not occur immediately and the cells do not stop growing completely under the conditions tested. Above all, however, we believe that this is related to the effect of thymidine starvation. E. coli mutants suffering from thymidine deficiency normally show a resistance phase in which damage to the DNA first accumulates before it becomes irreparable. This resistance phase is facilitated by an internal pool of dTDP sugars (mainly dTDP-glucose) that serves as a temporary source of thymine. During this phase, cells can become extremely enlarged (up to 6-fold). This can cause the cells to lose their cytoplasm and even their genomic DNA, which ultimately leads to cell lysis, known as thymineless death [34]. We also detected a strong increase in cell size, with average values approximately one log level higher than those of the healthy main population. In some cases, the size increase exceeded even two log levels. At peak levels, large cells comprised more than 50% of the total population. This could be triggered by thymidine starvation, and which would explain the increased CDM. This was also the reason why we pulsed thymidine, or thymidine and kanamycin in combination into the reactors after 31 days to test whether the cells could be reset to the initial state. Thymidine pulse alone had no measurable effect. Additional kanamycin caused the CDM to drop sharply and the pDNA to rise. We assume that cells with thymidine deficiency have lost the ability to deactivate kanamycin due to a possible loss of genomic DNA. As a result, they were eliminated in the reactor. By reducing dilution rates, we hoped to increase basal expression to boost ThyA synthesis in the seed reactors. The further decrease in CDM (fewer large cells) and the increase in GFP in the seed reactor confirmed our assumption that ThyA expression was too low. Whether the low ThyA levels were caused by inefficient transcription or plasmid loss could not be fully determined. Further optimization and fine balancing of the optimal amount of ThyA could be achieved with a stronger T7 promoter or a stronger RBS for ThyA. Taken together, these findings highlight the potential of combining growth-decoupling, rational plasmid design, and thyA-based selection to overcome major limitations of continuous E. coli production. Based on these insights, the most important conclusions for future applications are summarized below. Conclusions The results of this study highlight the significant progress achieved by implementing thymidine auxotrophy-based selection systems in continuous fermentation of E. coli. Among the constructs tested, 5.2 (Monocistronic T7 expressed thyA variant) and 7.2 (Constitutive expressed thyA variant) proved to be particularly promising candidates for application in two-stage continuous cultivation processes due to their superior performance. Construct 5.2, using a T7-based system, showed little basal expression and remained tightly regulated under non-inducing conditions. At the same time, it exhibited robust growth, excellent plasmid stability, and high productivity after induction. The survival of construct 5.2 was dependent on basal expression of the T7RNAP, which shield the T7 system and thus improve genetic stability during long-term cultivation. These properties make it an attractive platform for tightly regulated and efficient production in continuous processes. Although construct 7.2 is not explicitly dependent on the T7 system, the T7 RNA polymerase is expected to be partially protected from mutations due to the alignment of the thyA cassette on the plasmid and transcriptional readthrough of the T7RNAP. Furthermore, this construct offers the advantage that cell survival is largely independent of the upstream GOI, making it more flexible and robust across different bioprocess operation modes and compatible with various POIs. The construct achieved high growth rates, maintained excellent plasmid stability, and exhibited efficient inducible controlled expression, providing a sustainable and compliant alternative for industrial applications. Despite these successes, the study highlights the importance of construct-specific and process-specific evaluations, particularly for T7-based systems. Factors such as mRNA stability, transcription kinetics, and plasmid structure can significantly impact performance. While the constructs work well with the model protein GFP, extrapolation to other proteins requires careful validation to account for differences in expression requirements and potential metabolic burden. Additionally, the role of the cer element in improving plasmid stability requires further investigation. In multiple constructs, the inclusion of cer resulted in a significant reduction in plasmid loss and multimer formation, resulting in a significant improvement in process stability. The exact contribution of cer to these results compared to other plasmid modifications should be investigated in more detail to optimize its use in future designs. In summary, constructs 5.2 and 7.2 provide compelling evidence for the potential of thymidine auxotrophy-based selection systems to alter continuous E. coli bioprocessing. These systems combine stability, compliance, and high productivity to address key challenges in long-term fermentations. Achieving more than 1000 h of stable continuous fermentation without plasmid loss or productivity decline underscores the exceptional robustness of this system. To our knowledge, this represents one of the longest-running E. coli-based processes without the addition of antibiotics reported to date. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1. Acknowledgements We sincerely thank Karoline Reznar, Moritz Dielacher, and Corinna Schödl for their invaluable support in analytics. We also express our gratitude to Florian Strobl for his technical expertise and assistance with the DASGIP reactors. Furthermore, we acknowledge the financial support from the Austrian Research Promotion Agency (FFG) as well as the scientific support from the Doctoral School Bioprocess Engineering (BioproEng). Open access funding was provided by BOKU University, Vienna. Graphical illustrations were created with BioRender. Declaration of generative AI and AI-assisted technologies in the writing process During the preparation of this work, the author used ChatGPT (OpenAI) to assist in drafting and polishing text for clarity and style. After using this tool, the author reviewed and edited the content as needed and takes full responsibility for the content of the publication. Abbreviations Cell dry mass Escherichia coli Green fluorescent protein Gene of interest Gene product 2 Protein of interest RNA polymerase Plasmid deoxyribonucleic acid Plasmid copy number Isopropyl-β-D-thiogalactopyranosid Deoxythymidine monophosphate Deoxythymidine triphosphate Dihydrofolate reductase Polymerase chain reaction Semi synthetic media Optical density Relative fluorescence units Dissolved oxygen Dilution rate Diode-array detector ColE1 resolution Ribosome binding site Basal Expression Covalently closed circular Author contributions Conceptualization: PS, JM. Methodology: FS, PS, ND, JM. Investigation: FS, ND, PS, FW, JB. Visualisation: FS, PS.Funding acquisition: JM. Project administration: PS, JM. Resources: JM. Supervision: JM, PS. Writing – original draft: FS, PS, JM. Writing – review & editing: FS, PS, JM, GS. Corresponding author: PS. Funding This research was funded by the Austrian Research Promotion Agency (FFG) as part of the project “ECOnti - Accelerated, low ecological footprint, manufacturing platform for continuous production of biotechnological products”. Data availability All data generated or analyzed during this study are included in this article and its supplementary information. Additional raw datasets (e.g., full flow cytometry files) are available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate Not applicable. Consent for publication All authors agree to publishing this manuscript with Microbial Cell Factories. Competing interests The authors declare the following competing interest: Two patent applications entitled “UNCOUPLING GROWTH AND PROTEIN PRODUCTION” and “CONTINUOUS BIOPRODUCTION BY DECOUPLING GROWTH AND PRODUCTION” were filed by enGenes Biotech with application number PCT/EP2016/059597 and PCT/EP2020/087354. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1. Data Availability Statement All data generated or analyzed during this study are included in this article and its supplementary information. Additional raw datasets (e.g., full flow cytometry files) are available from the corresponding author on reasonable request.