Programming Nutrient Detection with Modular Regulators for Dynamic Control of Microbial Biosynthesis

Abstract

Dynamic control of biosynthetic pathways improves bioproduction efficiency. One common approach is to use genetic sensors that control pathway expression in response to a nutrient molecule in the target feedstock. However, programming cellular response requires the engineering of numerous genetic parts, which poses a significant barrier to explore the use of different nutrients as cellular signals. Here we created a dynamic control platform based on a set of modular transcriptional regulators; these regulators control the same promoter for driving gene expression but each of them responds to a unique signal. We demonstrated that by only replacing the regulator, a different nutrient molecule can then be used for induction of the same genetic circuit. To show host versatility, we implemented this platform in both Escherichia coli and Pseudomonas putida. This platform was then used to program the induction of ethanol production by three nutrients, including fructose, cellobiose, and galactose, of which each molecule can be present in a different set of crops. These results suggest that our platform facilitates the use of different agricultural products for dynamic control of biosynthesis.

Graphical Abstract

INTRODUCTION

Many novel biosynthetic pathways have been created in microbial cells to generate a wide range of valuable chemicals. Nonetheless, activities of these heterologous pathways often lead to accumulation of toxic intermediates and overconsumption of energy and carbon sources, hampering cell growth and efficiency of bioproduction. One of the approaches to overcome this problem is dynamic pathway regulation, in which expression of pathway genes is driven by an inducible genetic system in response to a specific inducer ^{1, 2}. The inducer can either be an endogenous metabolite, such as a pathway intermediate, or an exogenous signal ³.

For the use of exogenous signals for dynamic control, an advantage is real-time regulation of pathway activities, which optimize the allocation of resources between cell growth and product generation, such that the bioproduction is a two-stage process—increase of cell population in stage 1 and generation of the desirable product in stage 2. Generally, microbes are grown to reach a high population and biomass, then the exogenous signal is added to induce pathway gene expression that upregulates the biosynthetic activities ⁴. Based upon cell density or alternative biological parameters, the exogenous inducer can be introduced at different time points to maximize yield and titer of the desirable product. Many studies have demonstrated the benefits of this two-stage approach with synthetic inducers, such as IPTG and ATc ⁵; however, the cost of these molecules is not economically feasible for industrial-scale production. Some existing solutions include the use of changes in temperature and pH value to induce metabolic gene expression ^6–8; nonetheless, these methods may not be broadly applicable to all biosynthetic processes as changes in these physical parameters may affect biological activities of microbial hosts, obstructing the bioproduction process.

An alternative strategy is to induce gene expression in response to a nutrient in the feedstock ^9–13—engineered cells first grow and accumulate biomass; after that, biosynthesis is triggered by switching the availability of an exogenous inducer by providing a different feedstock. For instance, Bothfeld et al. developed a genetic system that detects glucose levels for controlling metabolic gene expression ¹¹. Sugars are efficient inducers in bioproduction as they are also used by engineered microbes as sources of carbon and energy. The profile of sugars varies greatly among agricultural products; for example, sugarcane is abundant in sucrose, which can be hydrolyzed to fructose ¹⁴; in comparison, soybean contains a high level of galactose in its soluble sugars, including stachyose and raffinose ¹⁵; for many lignocellulosic crops, such as alfalfa and other types of hay, they have a high content of cellulose that can be robustly converted to cellobiose in industrial treatments, but not fructose or galactose ^{16, 17}. Genetic sensors for these sugars can provide a means to distinguish different agricultural products and harness them as inducers for triggering metabolic activities. Due to weather variables and other global factors, crop prices fluctuate drastically ¹⁸. Thus, it is economically favorable to switch the feedstock to different crops in response to these price fluctuations, which may require reprogramming of the dynamic control system to detect a different inducer. In this study, we aim to develop a dynamic control platform that can be seamlessly and robustly reprogrammed to detect different sugars for controlling metabolic gene expression.

To address these needs, in this study we have explored the development of a programmable dynamic control system by harnessing a set of hybrid allosterically regulated transcriptional regulators that we created with LacI family members ¹⁹. These regulators bind to a promoter to repress gene expression and the presence of an inducer triggers the regulator to dissociate from the DNA, inducing gene expression. These regulators comprise a DNA-binding module (DBM) and a ligand-binding module (LBM). A DBM binds to a specific DNA sequence within a promoter to blocks other genetic components to approach the promoter, repressing transcription; when an inducer molecule interacts with the LBM, a change in the protein confirmation causes the DBM to dissociate from the promoter, allowing genes driven by the promoter to be expressed ²⁰. We previously created a set of modular regulators by fusing DBMs and LBMs from different regulators, such that resulting hybrid regulators recognize the same promoter but each of them responds to a different inducer (Figure 1A). As a result, the inducing signal can be reprogrammed by only changing the modular regulator but not any other components in the genetic system. This approach differs from previous studies ^{1, 2}, which required switching the entire dynamic control system to detect a different inducer.

Figure 1. Programmable dynamic control system.

(A) Genetic design of the dynamic control platform for E. coli and P. putida. Three modular sensors were characterized, which all of them contained the ScrR DNA-binding module (DBM) to control the P_LscrO promoter and with a different ligand-binding module (LBM) for responding to the target nutrient. Five ribosomal binding sites (RBSs) were characterized for controlling the dynamic range of GFP expression. DNA sequences of these regulators and RBSs are listed in Table S1. Performance of the dynamic control circuit in (B) E. coli and (C) P. putida. Cells containing dynamic control platforms with native ScrR (left column), ScrR-CelR (middle), and ScrR-GalR (right) were exposed to the target inducer (6 hours for E. coli and 24 hours for P. putida) before measuring cellular levels of GFP (top row) and mCherry (bottom row) by flow cytometry. Each data point represents the mean ± S.D. of three biological replicates.

Here, we explored the use of these hybrid regulators to control an ethanol biosynthetic pathway. We first demonstrated that these hybrid regulators can efficiently detect target inducers and control gene expression in both Escherichia coli and Pseudomonas putida, showing the potential of this strategy for a broad spectrum of microbial hosts. By only switching the hybrid regulator gene and not any other genetic elements, ethanol production can be activated by three molecules that are abundant in different types of plant biomass, including fructose, cellobiose, and galactose, which demonstrates the ease of reprogramming feedstock detection in this platform. The use of monosaccharides and disaccharides such as fructose, cellobiose, and galactose as inducers represents a novel and valuable aspect of our study. Our approach leverages abundant plant-based nutrients commonly found in agricultural products and wastes—fructose from sugar cane ^{21, 22}, cellobiose from alfalfa and grasses ^23–25, and galactose from soybeans and legumes ^{26, 27}. Furthermore, we showed that all our dynamic control systems effectively maintained genetic stability compared to constitutive expression of pathway genes. These results suggest that our platform has potential for a wide range of biosynthetic applications with different feedstocks and microbial hosts.

RESULTS

Development of a programmable dynamic control platform in E. coli for responding to different signals

We first established a genetic platform on a plasmid to control gene expression in E. coli, as shown in Figure 1A; the sequence of this plasmid is in Data S1. This platform contains a modular transcriptional regulator gene driven by a P_L promoter ²⁸, which allows the regulator gene to express constitutively, serving as a genetic sensor. The regulator can be native ScrR, ScrR-CelR, or ScrR-GalR, in which all three regulators comprise a ScrR DNA-binding module that recognize the scrO operator sequence ¹⁹. Each regulator has a different ligand-binding module from ScrR, CelR, or GalR, rendering them to respond to different signals, including fructose, cellobiose, and galactose, respectively. We previously used the P_L promoter to develop a new promoter, P_LscrO, that can be regulated by our hybrid regulators. The P_L promoter has been robustly engineered to become controllable by a range of transcriptional regulators; this is achieved by incorporating the corresponding regulator’s operator sequence into the P_L promoter. For example, engineered P_L can be regulated by LacI, TetR, and AraC ²⁹.

In previous studies, we incorporated the scrO sequence into the P_L promoter to create a P_LscrO promoter that can be recognized by the ScrR DNA-binding module ^{19, 30}. All three regulators, native ScrR, ScrR-CelR, and ScrR-GalR, bind to the scrO site to repress gene expression and with the presence of their corresponding signal, they dissociate from the scrO site to restore the transcriptional activity of P_LscrO. To assess the performance of these modular sensors in this platform, we first used P_LscrO to drive the expression of a GFP gene with five ribosomal binding sites (RBSs), which were designed computationally to possess different translational strength ^31–33. The sequences of these RBSs are shown in Table S1 and their predicted strength descends from RBS 1 to RBS 5 (Figure 1A). Additionally, we introduced a mCherry fluorescence gene that is driven by a constitutive promoter, P_L, which served as a standard for comparing to GFP expression.

Each of the three dynamic control systems was tested with its corresponding signal, including 10 mg/mL of fructose for native ScrR, 5 mg/mL of cellobiose for ScrR-CelR, and 10 mg/mL of galactose for ScrR-GalR; these concentrations were selected based on our previous studies ¹⁹, where the corresponding regulator and inducible expression system were saturated by the inducer. GFP expression was induced upon cellular exposure of the signal (Figure 1A and Table S1). Cellular GFP levels were significantly different when its expression was driven by a RBS with different predicted strength, allowing to reach a broad range of expression levels (Figure 1B). For the ScrR-GalR system, GFP inductions were significantly higher than those in the other two systems; also, with the strongest RBS (RBS 1), the basal GFP expression is high. These results suggest that ScrR-GalR has a lower binding affinity towards the P_LscrO promoter relative to native ScrR and ScrR-CelR. For the constitutively expressed mCherry, its levels were not influenced by any of the signaling molecules.

Additionally, we evaluated the specificity of these three inducible expression systems (Figure S1). Each strain of engineered E. coli containing a dynamic control system was exposed to the three sugars, fructose, cellobiose, and galactose, and the levels of GFP expression in these exposed cells were measured. The results demonstrate that the three regulators respond specifically to their target inducer but not the other two signals, supporting that they can distinguish these three sugars as signals for triggering gene expression.

Implementation of the programmable dynamic control platform to P. putida

After the development of the programmable dynamic control platform in E. coli, we then explored its use in another microbe, P. putida, which is a soil bacterial species. We selected to implement our design to P. putida because recent studies suggest that it has high potential for metabolic engineering and biosynthesis applications ^34–36. The circuit design showed in Figure 1A was incorporated into a new plasmid (Data S2), which has a backbone with two origins of replication, one recognized by E. coli and the other one by P. putida. These features were required for us to construct the circuit by using E. coli for cloning and implementing the design in P. putida. For selection of promoters, we first explore the use of P_L and P_L-derivatives; this is because P_L is an effective promoter in P. putida ^{37, 38} and we have well-characterized these genetic elements in E. coli. Indeed, when a P_L promoter was used to drive the expression of a mCherry gene in P. putida, the engineered cells generated constitutive and high mCherry signal (Figure 1C). Therefore, we used P_L-based promoters for developing the dynamic control platform in P. putida, including P_L for the regulator and mCherry genes, and P_LscrO for the GFP gene, as illustrated in Figure 1A.

For characterization and optimization of the dynamic control platform in P. putida, the five versions of RBS as shown in Figure 1A and Table S1 were used to drive GFP expression for achieving a broad range of expression levels. For each regulator (native ScrR, ScrR-CelR, and ScrR-GalR), at least one version of RBS generated a GFP induction of above 10 folds (Figure 1C). Among these three regulators, the native ScrR required the strongest RBS (RBS 1) to express GFP upon induction with fructose and its dynamic range of induction is smaller than those from ScrR-CelR and ScrR-GalR. For ScrR-CelR, the fold of induction was correlated to the strength of RBS; and the GFP levels generated by all RBSs at uninduced state were similar to the cellular autofluorescence background. However, with ScrR-GalR as the regulator, the basal GFP levels were higher than the background (about 1000 a.u.) with strong RBSs (RBSs 1 and 2), showing that ScrR-GalR bound loosely to ScrO, which agrees to the results from the dynamic control platform in E. coli. These results imply that our design and approach on developing the dynamic control platform can be applied to this strain of soil bacteria.

Assessment on the feasibility of ethanol bioproduction with E. coli and P. putida

After the characterization of the dynamic control platform in E. coli and P. putida, we aimed to harness the platform for controlling the biosynthesis of ethanol. We first evaluated whether E. coli and P. putida are suitable to serve as the host for ethanol bioproduction, which includes characterizing their ethanol tolerance and the stability of ethanol in their cell cultures. To assess ethanol tolerance, cells were grown from lag phase to stationary phase in 0 mM and 50 mM ethanol; cell growth under these two conditions were compared by measuring colony forming units (Figure 2A). For both E. coli and P. putida, cell growth was highly similar in the two conditions, suggesting that cellular functions were not affected by this level of ethanol.

Figure 2. Cellular responses to ethanol.

(A) Determining the effect of ethanol on cell growth. E. coli (left) and P. putida (right) were grown in the presence of 0 and 50 mM ethanol. Population density in each culture was determined by counting colony forming units. (B) Assessing cellular efficiency in degrading ethanol. The culture medium was added with 25 mM ethanol, 100-fold diluted saturated cell culture, or both ethanol and cells. Ethanol levels in these samples were quantified with a GC-MS method after 24 hours under cell growth conditions (37 °C for E. coli and 30 °C for P. putida). In both panels, each data point represents the mean ± S.D. of three biological replicates.

Next, we characterized the stability of ethanol in cell cultures of E. coli and P. putida. Both E. coli and P. putida possess active ethanol metabolic pathways ^39–41 that are expected to diminish ethanol levels. Thus, we aimed to select bacterial strain with lower endogenous ethanol metabolic activities in our experiment conditions, such that it is more efficient in accumulating the bioproduction product. For this purpose, the two strains of bacteria were grown in culture medium with 25 mM ethanol. Additionally, two experimental controls were prepared by adding only bacterial cells or only ethanol to the medium. After a 24-hour growth, levels of ethanol in these samples were quantified by gas chromatography-mass spectrometry (GC-MS) as shown in Figure 2B and Figure S2.

After incubating those control samples with only 25 mM ethanol added, the ethanol concentrations of those samples in the E. coli (37 °C) and P. putida (30 °C) experiments were decreased to 10.7 ± 0.7 mM (mean ± S.D.) and 11.9 ± 0.6 mM, respectively. As these samples were incubated in a shaker with air ventilation to ensure an aerobic condition, the loss of ethanol could be due to evaporation. Additionally, ethanol could be oxidized to acetic acid. These results suggest that for the quantification of ethanol production, the measured ethanol levels are potentially lower than the actual amount generated by the bacteria.

After incubating ethanol with E. coli for 24 hours, ethanol concentrations only dropped slightly (8.2 ± 1.1 mM; mean ± S.D.) compared to the controls with only ethanol in the culture medium (10.7 ± 0.7 mM). Contrastingly, P. putida possessed significantly stronger ethanol metabolic activities, in which ethanol concentrations reached a background level (0.0021 ± 0.0014 mM) after incubating with the cell, as indicated by measurements from the cell culture controls (0.0020± 0.0010 mM). As P. putida was highly efficient in metabolizing ethanol under our conditions, we decided to not use it as the host of the ethanol synthetic pathway, but only harness E. coli for this following study.

Assessment on E. coli cell fitness with the dynamic control platform

Before implementing the ethanol biosynthetic pathway to our E. coli platform, we evaluated the influence of the platform on cell fitness. Potentially, those transcriptional regulators (native ScrR, ScrR-CelR, and ScrR-GalR for detecting fructose, cellobiose, and galactose, respectively) may interact with the bacterial genome and hamper the expression of endogenous cellular pathway genes. To investigate whether those regulators may bind to the E. coli genome, we performed a nucleotide blast search to identify genomic sequences that are similar to the scrO operator recognized by our regulators with the ScrR DNA-binding module (Figure S3A). The scrO contains 19 base pairs; in the E. coli genome, the top similar sequence only has 12 base pairs that are identical to the scrO, which should not be sufficient to interact with the ScrR DNA-binding module. Thus, it is unlikely that our dynamic control platform may reduce cell fitness.

To further assess the influence of the dynamic control system on cell fitness, we measured growth rates of cells containing the three versions of our platform (Figure S3B). Each strain of E. coli (described in Figures 1A and 1B) was grown overnight and then diluted 100 folds in fresh culture media. The growth of these cultures was measured by their optical density at 600 nm (OD600) for 5.5 hours. The growth activities of all these cultures were highly similar; OD600 increased exponentially between the 1- to 3.5-hour time points; after that, OD600 reached 1 and the growth rate reduced. The exponential growth rate of the wild-type strain was 0.34 h⁻¹ and those of the three engineered strains (containing native ScrR, ScrR-CelR, or ScrR-GalR) were between 0.33 and 0.35 h⁻¹. These results suggest that the constitutive expression of those three genetic sensors did not affect the cell growth of E. coli.

Performance of E. coli dynamic control systems at various extracellular temperature and pH

Bioproduction from each metabolic pathway often requires a different temperature ⁴² and pH ⁴³. We investigated the performance of the dynamic control systems in E. coli in a range of these two physical parameters (Figure S4). We used those strains of E. coli described in Figure 1A, in which GFP expression was regulated by native ScrR, ScrR-CelR, or ScrR-GalR. To characterize the effect of temperature changes (Figure S4A), each strain of cells was first cultured at standard conditions (37 °C). After reaching the early exponential growth phase, these cultures were grown at 30, 37, or 42 °C to perform induction by exposing them to the inducer for 4 hours. Our results show that these cells continued to grow under these conditions. For each dynamic control system, GFP expression induction was similar at these selected temperatures.

To characterize the effect of pH values, these strains of E. coli were used to perform induction at pH 4, 7, and 9 (Figure S4B). Cultures were again grown to the early exponential growth phase. They were then centrifuged to collect cell pellets and resuspended in fresh culture media with the target pH values. They were then exposed to the corresponding inducer for 4 hours to induce GFP expression. The growth and GFP induction of all three strains were significantly impaired at pH 4; after incubating at this condition for 4.5 hours, OD600 of all cultures was still below 0.2 and induction of GFP expression was below 2 folds, suggesting that E. coli cells could not function at this pH. In contrast, these strains performed their biological functions normally at pH 7 and 9; growth and GFP expression behaviors were similar under these conditions. These characterization experiments enhance our understanding of the range of conditions for using the E. coli dynamic control platform.

Dynamic pathway control for ethanol biosynthesis in E. coli

After the characterization of the dynamic control platform with GFP expression, we then harnessed the platform for controlling the biosynthesis of ethanol. In this biosynthetic pathway ^{44, 45}, pyruvate decarboxylase (PDC) catalyzes the conversion of pyruvate to acetaldehyde and then alcohol dehydrogenase (AdhB) converts acetaldehyde to ethanol (Figure 3A). PDC activity causes cell stresses because the loss of pyruvate perturbs tricarboxylic acid cycle within the central of carbon metabolism ⁴⁶; additionally, the PDC reaction generates acetaldehyde, which is a DNA-damaging metabolite that obstructs cell division and production ^{47, 48}.

Figure 3. Programmable dynamic control in response to target nutrients for ethanol biosynthesis.

(A) Circuit design for controlling expression of PDC and AdhB in E. coli to generate ethanol from endogenous metabolite, pyruvate. (B) Ethanol production from different versions of the dynamic control system. Cells contained the system with different modular sensors (native ScrR, ScrR-CelR, or ScrR-GalR) and RBSs for controlling PDC expression. Each strain was exposed to the corresponding inducer for 1 day and ethanol levels in uninduced and induced cultures were quantified with a GC-MS method. (C) Influence of cell growth from activation of ethanol biosynthesis. Overnight cultures of the strains with the highest ethanol production for each inducible system (with RBS 2 for driving PDC expression) were grown in the uninduced and induced conditions. Cell population densities were assessed between 0 to 6 hours after inoculation by measuring optical density at 600 nm. Growth rates at the bottom table were calculated with data obtained at 2 and 4.5 hours. In panels B and C, each data point represents mean ± S.D. from three biological replicates; the error bars are not shown if they are shorter than the marker.

To implement the ethanol biosynthetic pathway in our dynamic control platform, we replaced GFP with PDC and mCherry with adhB (Figure 3A). In the resulting genetic circuit (Data S3), the PDC gene is driven by the P_LscrO promoter and its expression regulated by the modular genetic sensor (native ScrR, ScrR-CelR, or ScrR-GalR). Since adhB is not expected to be cytotoxic, it is constitutively expressed to ensure efficient conversion of acetaldehyde to the final product, ethanol, which reduces the toxicity from basal PDC expression.

We first characterized ethanol production in the native ScrR system, aiming to optimize the metabolic gene expression and the timing for collecting ethanol samples. After inducing ethanol biosynthetic activities with 10 mg/mL fructose, ethanol levels in culture media were assessed from 0 to 24 hours with the GC-MS method (Figure S5). The concentration of inducer was selected based on our previous studies, which saturated the inducible expression system ¹⁹. Similar to the optimization of inducible expression systems in Figure 1B, we used the five computationally designed RBSs (RBS 1 to RBS 5) to control PDC expression and then identified the RBS that led to maximal ethanol production.

Ethanol production was significantly higher with RBS 2 for driving PDC expression comparing to other strains. Intriguingly, while RBS 1 is expected to generate a higher level of PDC based upon GFP expression results in Figure 1B, the RBS 1 strain could not be generated in three cloning attempts; resulting plasmids contained mutations either in the PDC gene or in the P_LscrO promoter. On the other hand, RBSs 3, 4, and 5 did not lead to significant GFP expression in this platform (Figure 1B) but strains with these RBSs to drive PCD expression can still produce relatively low but measurable levels of ethanol. As the levels of ethanol were the highest at the 24-hour time point, we performed a 4-day ethanol production experiment with the optimal strain (with RBS 2 for driving PDC). However, a continuation of the bioproduction process did not increase ethanol levels in days 2 to 4 (Figure S5). Based on these results, we decided to induce all bacterial strains for 24 hours for ethanol production, which facilitates a direct comparison of production efficiency between different versions of the dynamic control platform.

Next, we characterized ethanol production by those strains with ScrR-CelR and ScrR-GalR (Figure 3B). Similar to results with the native ScrR, when incorporating RBS 1 to drive PCD expression, we have tried multiple times to create the system but the PCD gene was always mutated, suggesting that a combination of RBS 1 and PDC generated cytotoxicity. As shown in Figure 1A, ScrR-GalR with RBS 1 leads to a high basal level of GFP, implying that RBS 1 may also generate leaky PDC expression and reduce cellular fitness. Among other RBSs for ScrR-CelR and ScrR-GalR, the trend is similar to those for native ScrR, in which the strain with RBS 2 generated the highest level of ethanol upon induction. By implementing different hybrid regulators, a programmable dynamic control platform was established to control biosynthesis in response to different signaling molecules.

To demonstrate the necessity of regulating PDC expression to reduce cytotoxicity, we modified the dynamic control platform by removing the modular regulator gene by replacing it with a small DNA fragment (Figure S6), such that the PDC gene was expected to express constitutively. We then sequenced 5 resulting colonies and results suggest that PDC was silenced by DNA recombination, as a DNA fragment was incorporated into the P_LscrO promoter for driving PDC expression. These incorporated DNA fragments were originated from different endogenous genes, including rbbA, Phr, and yeaV (Figure S6). These results imply that PDC expression leads to a severe burden to E. coli cells and induce genetic instability. Thus, a dynamic control system is required for the ethanol bioproduction pathway.

We further investigated the influence of ethanol biosynthetic activities on cell fitness; the results suggest that this pathway significantly reduced cell growth (Figure 3C). For each dynamic control system, including native ScrR, ScrR-CelR, and ScrR-GalR, the version with RBS 2 for driving PDC expression was selected for growth assessment because they generated the highest levels of ethanol (Figure 3B). Overnight cultures were diluted 100-fold in fresh media with and without the corresponding inducer. These fresh cultures were grown for 7.5 hours and measured for cell densities based on OD600. Data obtained at 2- and 4.5-hour were used to calculate growth rates, as all cultures showed exponential growth in the period. For cells containing native ScrR and ScrR-CelR, their uninduced cultures grew at 0.31 ± 0.02 and 0.35 ± 0.007 h⁻¹, respectively, which are similar to the growth rate of wild-type E. coli cells (0.34 ± 0.007; Figure S3B). However, uninduced cultures with ScrR-GalR had lower growth rates (0.24 ± 0.008 h⁻¹), which can be due to relatively high basal expression levels of PDC as demonstrated in the dynamic control characterization (Figure 1B). Nonetheless, cultures of all three strains reached OD600 above 1 at 7.5 hours. Contrarily, with induction, the growth of all three strains was dampened to about 0.14 h⁻¹ and OD600 only reached about 0.3. Together, these results demonstrate that our dynamic control platform provides a means to program the allocation of resources for cell growth and ethanol production.

DISCUSSION

The engineering of biosynthetic pathways has revolutionized the production of valuable chemicals, enabling the creation of novel compounds with diverse industrial applications. Recent studies have demonstrated significant progress in this field, from the production of polyketides in bacteria ⁴⁹ to the engineering of isoprenoid pathways in yeast ⁵⁰. However, these advancements are not without their limitations and challenges. One major challenge is the complexity of metabolic networks within microbial hosts, which can lead to metabolic burden and suboptimal production yields. Additionally, fine-tuning biosynthetic pathways for optimal performance often requires intricate genetic modifications and regulatory controls, posing technical and regulatory hurdles. As demonstrated in this study, constitutive activities of the ethanol biosynthetic pathway led to the loss of PDC expression by endogenous cellular mechanisms (Figure S6). PDC interferes with the central carbon metabolism by using pyruvate as a substrate ⁵¹. Additionally, the product of PDC, acetaldehyde, is a DNA damaging agent ^{47, 52}. Therefore, the ethanol biosynthetic pathway is required to be controlled temporally.

Addressing these challenges, our study focused on using two-stage bioproduction process with dynamic pathway control as a strategy to optimize bioproduction, where microbes first focused on cell growth to accumulate sufficient biomass before the biosynthetic pathway expression was triggered by an inducing condition. This strategy mitigates metabolic burden, and ensures precise regulation over biosynthetic activities. Previous studies demonstrate a diverse range of approaches and applications of dynamic regulatory systems in fine-tuning gene expression, optimizing metabolic flux, and responding to various stimuli ^{53, 54}. In our study, we emphasize the use of modular regulators for transcriptional control over a synthetic metabolic pathway, showcasing a unique feature of our dynamic control platform in tailoring towards different feedstocks for triggering target metabolic activities. The novelty and advantage of our approach is that different inducing signals can be flexibly connected to a target promoter for controlling metabolic activities, which supports the programming of signal detection that is desirable at different conditions and applications while the rest of the genetic pathway does not require additional alterations.

We demonstrated that by incorporating a different regulator in the genetic platform, ethanol production can be triggered using different sugars, including fructose, cellobiose, and galactose, which are nutrients commonly available in agricultural products and wastes. For example, fructose is abundant in sugar cane ^{21, 22}, cellobiose can be extracted from alfalfa and many types of grasses ^23–25, and galactose is at considerable levels in soybeans and some legumes, especially after they are fermented ^{26, 27}. Our dynamic control system can be programmed to respond to these plant-based materials as feedstocks. This ease in altering signal detection provides an advantage in microbial engineering for adapting to changes in bioproduction processes. Our study demonstrates the feasibility of this approach in facilitating alcohol production and it differs from previous studies, which typically required switching the entire dynamic control system to detect a different inducer. In contrast, our system enables flexibility and adaptability through its modular design, allowing seamless adjustments to different inducers within the same framework.

With this programmable dynamic control platform, different agricultural products and wastes can be mixed and matched to create the growth-bioproduction two-stage process. For example, in the case of using fructose as the signaling molecule, during the growth phase, E. coli would consume other feedstocks that do not contain fructose to build biomass and proliferate. Once the growth phase reaches an optimal cell density, the system would switch to the production phase, where the addition of a fructose-containing feedstock, such as biomass from sugar cane, would trigger the expression of genes responsible for ethanol production. Similarly, with cellobiose as an inducer, engineered microbes could utilize lignocellulosic feedstocks such as alfalfa or other types of hay, which are rich in this disaccharide ^23–25. This two-stage process showcases the flexibility of our system in utilizing different feedstocks, with each sugar acting as both a carbon source and a metabolic trigger for ethanol biosynthesis. The ability to switch between different inducers—depending on the availability of feedstocks—provides a highly adaptable approach, enabling efficient bioproduction from diverse plant-based materials.

As a proof of concept, the current study demonstrates the feasibility of using specific sugars (fructose, cellobiose, and galactose) as inducers, but the limitation lies in the restricted scope of the signal molecules, and future work could focus on expanding the platform to include a broader range of metabolic intermediates or other relevant feedstocks, further enhancing its versatility in biosynthetic applications. The three modular genetic sensors (native ScrR, ScrR-GalR, and ScrR-CelR) are from the LacI protein family, which contains over 70,000 identified protein sequences ¹⁹. While the inducers of most of these proteins are unknown, among family members that have been characterized, they can detect a range of saccharides, such as ribose ⁵⁵, allolactose ⁵⁶, fucose ⁵⁷, and trehalose ⁵⁸. It is expected that those uncharacterized LacI family members are genetic sensors of a wide range of sugars and nutrients that are naturally available. Following the increasing knowledge of these regulators and their use in constructing modular sensors ²⁰, our programmable dynamic control strategy has the potential to be extended to a broad spectrum of dynamic control applications.

Studies from other groups have developed modular regulators for sensing various biological molecules, including acidity ⁵⁹, metal ions ⁶⁰, and quorum sensing molecules ⁶¹. Both the repertoire of modular regulators and techniques for designing them are expanding rapidly ²⁰, which can be harnessed to elaborate our dynamic control platform for using different biomolecules as the trigger, tailoring for new biosynthetic applications.

Additionally, we demonstrated that the dynamic control platform can be applied to both E. coli and P. putida, suggesting its potential for use in different microbial hosts. The transfer of the platform is feasible because a range of genetic elements, such as promoters, are universal among many microbes ^{62, 63}. This versatility is important for metabolic engineering applications as each biosynthetic pathway may require a specific host. Taking the ethanol production as an example, our results suggest that P. putida had high activities in metabolizing ethanol under our experimental conditions (Figure 2B), rendering this strain to be unsuitable for hosting ethanol biosynthesis. Nevertheless, its inclusion highlights the adaptability and versatility of our dynamic control platform. We implemented our design in P. putida based on recent studies suggesting its high potential for metabolic engineering and biosynthesis applications ^34–36, making it a valuable test case despite its limitation in ethanol accumulation. We demonstrate that the platform can integrate into diverse microbial systems, and a suitable host can be selected for target bioproduction based on metabolic constraints. This adaptability is further emphasized by our ability to shift to E. coli as the primary host for ethanol production, leveraging the platform’s modularity without requiring extensive redesign. Addressing this limitation directly strengthens the argument for the platform’s broader applicability, as it underscores its potential to be tailored to different microbial hosts for a wide range of biosynthetic applications. While we only demonstrate the implementation of our dynamic control platform in two bacterial species, it is expected that it can be used in a wide range of microorganisms as LacI family regulators are found in many microbial genomes ⁶⁴ and these sugar inducers are common carbon sources that should get into most cells. Future efforts can build on this adaptability by optimizing the system for organisms with complementary metabolic traits, enabling more efficient production of ethanol and other target compounds.

Additionally, different strains from the same bacterial species may be required for different feedstocks to maximize the production yield. Our work used E. coli K-12 MG1655 as the host, which generated 24 to 75 mM of ethanol in the cultures with the three genetic sensors (native ScrR, ScR-CelR, and ScrR-GalR; Figure 3B). In a previous study, Alterthum et al. did not use the K-12 MG1655 strain but they implemented the PDC-AdhB pathway to a range of other E. coli strains ⁶⁵. When these strains were supplemented with glucose as a carbon source at 37 °C, the levels of ethanol in their cultures varied significantly, ranging from 0.9 to 4.8% (v/v), which is equivalent to 155 to 824 mM. They also demonstrated that the maximal level of ethanol was generated by a different strain when the cultures were supplemented with a different sugar, such as xylose and lactose.

Nonetheless, the adaptability of our platform to different microbes can be limited by undesirable interactions between genetic sensors and endogenous genetic elements. Allosterically regulated transcriptional regulators are relatively small cytosolic proteins that can be easily implemented in most microbes as genetic sensors. However, there are many examples that a heterogeneous regulator can bind to the host genome to affect essential gene expression, leading to cytotoxicity. For instance, overexpression of a regulator, ArsR, in E. coli caused severe reduction of cell growth ⁶⁶. In this study, we performed nucleotide blast and growth rate analysis to support that regulators with ScrR DNA-binding module had unnoticeable impact on cell fitness of our ethanol bioproduction host (Figure S1 and Figure 3C). Nevertheless, these regulators might be incompatible to another microbial hosts. This potential problem can be mitigated by using another set of modular regulators that recognize a different DNA sequence; in our previous study ¹⁹, we created a series of modular regulators that detect the same inducer but each binds to a different DNA operator. Among these modular regulators, it is promising to identify some with compatible DNA recognition properties for each microbial host.

The application of our dynamic control platform can potentially be broadened via several directions. It may be implemented to eukaryotes, such as yeast, to facilitate the use of these organisms for bioproduction. Gene expression in yeast can be controlled by the regulator, LacI ⁶⁷, which is both mechanistically and structurally similar to our modular regulators in this study. These studies imply a high possibility of success in engineering eukaryotic promoters to become compatible with our genetic sensors. Furthermore, the platform can be used to control the expression of additional genetic tools, such as CRISPRi, to repress endogenous metabolic activities, driving metabolic flux to favor the generation of target metabolites ^{68, 69}.

In conclusion, our study presents a promising solution through dynamic pathway regulation, offering a pathway to overcome technical complexities, enhance sustainability, and reduce costs in bioproduction processes. By highlighting the unique contributions and implications of our approach, we aim to advance the field of biosynthetic pathway engineering and pave the way for future innovations in sustainable bioproduction.

METHODS

Strains of bacteria.

Cloning was performed with commercial cloning strains of E. coli, including ig10β (Intact Genomics, Inc.; St. Louis, MO) and XL-1 Blue (Agilent Technologies, Inc.; Santa Clara, CA). Experiments were performed with E. coli K-12 MG1655 and P. putida KT2440 (ATCC; Manassas, VA).

Plasmid construction.

Plasmid construction was performed with standard cloning methods. DNA oligomers were purchased from Eurofins Genomics LLC (Louisville, KY) and cloning enzymes and reagents were from New England Biolabs (Ipswich, MA). For genetic elements that are not included in plasmid sequences in Supplementary Data 1, 2 and 3, they are listed in Table S1.

For building the dynamic control platform in E. coli, a plasmid from our previous studies ⁷⁰, pTR, was used as the vector backbone, which contained a ColE1 origin of replication and a kanamycin-resistant gene. Three DNA fragments were cloned into the vector in the following order: 1) the P_L-mCherry fragment, 2) the P_LscrO-GFP-T1 terminator fragment, and 3) the T0 terminator-P_L-(scrR-celR)-T0 terminator fragment. The resulting plasmid contains the dynamic control platform that responded to cellobiose as an inducer, as shown in Figure 1A. The sequence of this plasmid (pPDC_Ec) is in Supplementary Data 1. The scrR-celR gene was replaced with native scrR and scrR-galR (Table S1) to generate the systems that responded to fructose and galactose, respectively. To change the ribosomal binding site of GFP (Figure 1B), a DNA hybridization method was used ¹⁹. Briefly, two complementing single-stranded DNA were mixed to reach a final concentration of 50 μM; the mixture was heated to 100 °C and then cooled down under room temperature for hybridization. The double-stranded DNA possessed sticky 5’- and 3’-ends that mimic digested restriction sites. These RBS fragments (Table S1) were incorporated into the plasmid, via restriction sites NheI and SacI. The strength of these RBSs was predicted with an online software (https://salislab.net/software/).

The plasmids for P. putida experiments were constructed with pSEVA-gRic6T ⁷¹ from Addgene (Watertown, MA) as the vector. This plasmid was restriction digested to obtain a backbone with a gentamycin-resistant gene, a pRO1600 origin of replication for P. putida, and a pUC origin of replication for E. coli. This backbone was ligated to a fragment with multiple cloning sites to generate pPpVK. From pPDC_Ec described above, a fragment was cloned into pPpVK, which contained the fragment with the GFP and mCherry genes and their promoters and terminators. Following that, a fragment, containing the scrR-celR gene and its promoter and terminator, was cloned into the plasmid to generate pPDC_Pp (plasmid sequence shown in Supplementary Data 2). Similar to the process for modifying pPDC_Ec as described above, the regulator gene was replaced with native scrR and scrR-galR, and the RBS of GFP was substituted with the list of RBSs in Table S1. Resulting plasmids were used to generate results in Figure 1C.

To generate plasmids for ethanol biosynthesis, pPDC_Ec and its derivatives were modified by replacing GFP with PDC and mCherry with adhB, creating the plasmid with the ethanol biosynthetic pathway (pPDC_Ec-ethanol). The genes PDC and adhB were cloned from the plasmid pLOI297 ^{44, 65}. The sequence of a representative plasmid with scrR-celR to control ethanol production is shown in Supplementary Data 3. These plasmids were used to generate results in Figure 3B.

To test the genetic effects from the constitutive expression of ethanol biosynthetic pathway (Figure S1), a DNA fragment was used to replace the modular sensor gene with the DNA hybridization method ¹⁹ described above. The DNA fragment sequence is shown in Table S1.

Culturing conditions.

All culturing reagents were purchased from VWR (Radnor, PA) and Fisher Scientific (Waltham, MA). Both E. coli and P. putida were grown in LB broth at 200 rpm, with E. coli at 37 °C and P. putida at 30 °C. The following reagents were added to culture media when required: 50 μg/mL kanamycin, 20 μg/mL gentamycin, 10 mg/mL fructose, 5 mg/mL cellobiose, and 10 mg/mL galactose.

Characterization of GFP expression with flow cytometry.

For results in Figure 1 and Figure S1, E. coli and P. putida were grown in LB with kanamycin and gentamycin, respectively. Overnight cultures were diluted 100 folds in fresh LB with the same antibiotic. After growing for 2 hours, appropriate inducers were added to these cultures, including fructose for cells containing plasmids with native scrR, cellobiose for scrR-celR, and galactose for scrR-galR. Cells were collected for GFP expression analysis after 6 hours for E. coli and 24 hours for P. putida.

For characterizing the E. coli dynamic control platform at different physical conditions (Figure S4), engineered E. coli cells were grown in LB with kanamycin overnight. Saturated cultures were diluted 100 folds in fresh LB and grown with the same conditions. After reaching an OD600 of 0.1, these cultures were used to assess the influence of temperatures (Figure S4A) and pH values (Figure S4B). For the temperature experiments, each culture was transferred to two wells in each 96-well plates (200 μL in each well); each plate was incubated at 30, 37, or 42 °C for 0.5 hours. For pH experiments, each culture was centrifuged and the collected pellets were resuspended in the same volume of LB with kanamycin at different pH (4, 7, and 9; the pH was adjusted by adding hydrochloric acid and sodium hydroxide); each resuspended culture was transferred to two wells in a 96-well plate and was incubated at 37 °C for 0.5 hours. After these cultures were transferred to these new physical conditions, inducer was added to one of the two equivalent wells to serve as the induced sample and the other well was the uninduced sample. Fructose, cellobiose, and galactose were added to cultures containing native ScrR, ScrR-CelR, and ScrR-GalR, respectively. After 4 hours of induction, OD600 of these cultures were measured with a Synergy H1 microplate reader (BioTek Instruments, Inc.) and their GFP expression was measured with a flow cytometric method.

NovoCyte 3000VYB flow cytometer (Agilent Technologies, Inc.; Santa Clara, CA) was used to analyze fluorescence levels in bacterial cells to generate data in Figures 1B, 1C, and S1. Data analysis was performed with NovoExpress software (Agilent Technologies, Inc.). Flow cytometry data were gated by forward and side scatter to eliminate multi-cell aggregates. The PE-Texas Red and FITC channels were used to measure mCherry and GFP fluorescent signals, respectively. The geometric means of fluorescence level distributions were calculated with the software. At least 10,000 events were collected for each measurement.

Characterization of E. coli and P. putida cell growth in response to ethanol.

For data in Figure 2A, we measured the colony forming units (CFU) of the wild-type strain of E. coli and P. putida cultures from lag phase to stationary phase to assess their ethanol sensitivity. Each overnight culture was diluted 100 folds and then grown in LB medium with 0 mM and 50 mM ethanol. Cells were grown at standard conditions described above. For E. coli cultures, after 0, 0.5, 1, 2, 3, 4, 5, 6 and 8 hours, each culture was diluted in PBS for 10 folds; the 10-fold diluted culture was serially diluted with PBS to reach 10²-, 10³-, 10⁴-, 10⁵-, 10⁶-, 10⁷-, 10⁸-, 10⁹-, and 10¹⁰-fold dilutions. A volume of 5 μL of each diluent was spotted on a LB agar plate and incubated at 37 °C (for E. coli) and 30 °C (for P. putida) for colony formation to determine CFU.

Comparison of cell fitness among engineered strains of E. coli.

To investigate the potential of undesirable binding of hybrid regulators to E. coli genome (Figure S3A), a nucleotide blast search was performed (https://blast.ncbi.nlm.nih.gov/Blast.cgi). We searched for the scrO sequence (5’-ttattaaaccggtttagca-3’) in the genome of E. coli K-12 MG1655 (taxid:511145) under the Core nucleotide BLAST database (core_nt).

To evaluate the influence of the dynamic control platform (Figure S3B) and the ethanol biosynthetic pathway (Figure 3B) on cell fitness, we measured the growth rates of engineered cells. Each engineered strain of cells described in these figures was first inoculated into LB culture media with kanamycin; only the wild-type strain was inoculated in LB without antibiotics. Overnight cultures were diluted 100 folds in fresh media. For assessing the influence of ethanol biosynthetic pathway (Figure 3B), those strains were diluted in media with or without the corresponding inducer. Optical density at 600 nm (OD600) was measured at 0, 0.5, 1, 1.5, 2, 2.5, 3, 3.5, 4, 4.5, 5, and 5.5 hours after the dilution; 200 μL of each culture was transferred to a flat-bottom 96-well plate for measurement with a Synergy H1 microplate reader (BioTek Instruments, Inc.). Growth rates were calculated with the following equation:

$$Growthrate=\frac{\log \left(\mathrm{O}\mathrm{D}{600}_f\right)-\mathrm{l}\mathrm{o}\mathrm{g}\left(\mathrm{O}\mathrm{D}{600}_i\right)}{{\mathrm{T}\mathrm{i}\mathrm{m}\mathrm{e}}_f-{\mathrm{T}\mathrm{i}\mathrm{m}\mathrm{e}}_i}$$

, where $\mathrm{O}\mathrm{D}{600}_f$ and ${\mathrm{T}\mathrm{i}\mathrm{m}\mathrm{e}}_f$ are from the 3.5-hour time point data and $\mathrm{O}\mathrm{D}{600}_i$ and ${\mathrm{T}\mathrm{i}\mathrm{m}\mathrm{e}}_i$ are from the 1-hour time point data. These time points are selected because results show exponential growth in this period.

Quantification of metabolites on ethanol bioproduction pathway by gas chromatography-mass spectrometry (GC-MS).

For characterizing the metabolism of ethanol by E. coli and P. putida (Figure 2B), a volume of 50 mL of LB culture medium in a 250-mL conical flask was used in each experimental condition. For those conditions with ethanol, 73 μL of pure 200 proof ethanol was added to the 50 mL LB culture medium. For those conditions with bacterial cells, a volume of 50 μL of an overnight culture of a wild-type strain (either E. coli or P. putida) was added. These flasks were incubated for 24 hours at standard conditions (200 rpm; 37 °C for E. coli and 30 °C for P. putida). Samples were then collected for GC-MS analysis.

For quantifying ethanol production from engineered E. coli (Figure 3B), cells containing a pPDC-ethanol plasmid were grown and induced similar to the process in flow cytometric analysis described above. Overnight cultures of each strain were 1:100 diluted into 50 mL of LB culture medium with kanamycin in a 250-mL conical flask and incubated at 37 °C and 200 rpm. After OD600 reached 0.2, the appropriate inducer (fructose, cellobiose, or galactose) was added. For experiments to characterize the native ScrR-containing dynamic control system (Figure S4), samples were collected at time points 0, 1, 2, 3, 6, 24, 48, 72, and 96 hours after induction. As the system generated the highest levels of ethanol at the 24-hr time point, characterization of other dynamic control systems (Figure 3B) were also performed with samples collected at 24 hours after induction.

For sample collection, 6 mL of a culture was centrifuged to pellet the cells and 5 mL of supernatant was transferred to a 20-mL vial to mix with 2 g of sodium chloride. Samples were stored at −20 °C before GC-MS analysis.

For GC-MS analysis, ethanol was detected and quantified using a Trace1310 gas chromatography system coupled to an ISQ single quadrupole mass spectrometer from Thermo Fisher Scientific. Each sample (5 mL of supernatant with 2 g of sodium chloride) in a 20-mL vial was spiked with 5 μL of 0.5% (v/v) acetonitrile in water (internal standard). It was then incubated at 80 °C for 10 min with 10 seconds shaking/10 seconds pause cycle using a Triplus RSH autosampler. Then, 500 μL of extracted sample in the gas phase was injected to a split/splitless injector whose temperature was maintained at 225 °C. A split ratio of 5:1 was applied to the extract. Samples were resolved using a DB-624 UI (30 m x 0.25 mm x 1.4 μm) column under a constant flow of helium setup at 1.6 mL/min. The GC conditions were as follows: initial temperature ramp was set to 32 °C with a 2 min hold, first temperature ramp was 3 °C/min up to 40 °C with no hold, second temperature ramp was 50 °C/min up to 230 °C with 1 min hold, and the third temperature ramp was 50 °C/min up to 250 °C with a 5 min hold. The total GC-MS run was 15 min.

For the MS analysis, samples were ionized using electron impact (EI) ionization in positive ion mode. The temperatures of the transfer line and the EI source were 200 and 230 °C, respectively. A mixture of external standards (38.3 mM of acetonitrile and 34.3 μM of ethanol) was utilized to record the retention time of each ionized metabolite that were scanned over a mass range of 20–200 amu. Once the compounds had assigned retention time, the most abundant product ion for each analyte was selected to perform single ion monitoring (SIM). In this study, product ions for acetonitrile and ethanol were 41 and 45 amu, respectively. We also generated a set of external standards with different concentrations of ethanol to demonstrate the linear relationship between the ethanol level and signal intensity (Figure S2); the resulting standard curve was also used for determining ethanol concentrations in biological samples. Data acquisition was performed with Xcalibur version 2.2. Data were processed with FreeStyle 1.7.