Using yeast counter-selection for the improved development of ligand-stabilized biosensors
Matt Rich
Yeast and other microbes can be used to produce many small molecules, but optimization of the metabolic pathways involved is a non-trivial undertaking. Strains optimized to produce a small molecule of interest can be selected through the use of suitable biosensors, but few of these are currently available. As such, we must design new sensors for use in strain engineering contexts. We have previously shown that ligand-stabilized transcription factors are a viable, and likely general, biosensor platform. In this approach (Figure 1), an unstable transcription factor can induce expression of a reporter gene only in the presence of a stabilizing ligand. Optimization of this stabilizing interaction can be difficult, and would be aided by an in vivo selection and counter-selection strategy that enables the use of large populations of protein variants.
We have implemented such a strategy using the yeast CAN1 gene. CAN1 encodes an arginine transporter that also imports canavanine, a toxic arginine analog, into the cell. Because evidence has shown that low expression of CAN1 is not fully toxic in the presence of canavanine, we hypothesized that this gene could be used as a counter-selectable marker for ligand-stabilized transcription factor optimization. Indeed, increased biosensor stabilization causes a growth defect in the presence of canavanine (Figure 2). We are currently testing selection methods in which we initially positively select for HIS3 expression in the presence of ligand, followed by counter-selection in canavanine in the absence of ligand.
The use of CAN1 has drawbacks, though. For instance, the mutation rate at the CAN1 locus is relatively high, and loss-of-function mutations in CAN1 break the counter-selection. We are exploring the use of diploid reporter strains, which should diminish this problem. CAN1 seems to require a higher expression level for phenotypic effects when compared to the positive selectable marker, HIS3.
Genetically-encoded bacterial biosensors that function based on ligand-dependent stability
Ben Brandsen
Biosensors are genetically encoded proteins or nucleic acids that enable cells to sense and respond to stimuli. Biosensors can be used to control gene expression, to construct genetic circuits, and to detect metabolic products for genome engineering. Many bacterial biosensors are based on natural allosteric transcription factors, which change reporter gene expression after binding to ligand. Engineering allosteric transcription factors to bind new ligands is challenging, however, because engineered transcription factors often have broken allostery. A more general method to biosensor development would overcome the challenges associated with engineering allosteric transcription factors and enable rapid development of new biosensors.
In collaboration with George Church and David Baker, our group recently reported a modular design for biosensor development in eukaryotes. This biosensor design fuses a ligand-binding domain (LBD) to a DNA-binding domain (DBD) and transcription-activating domain (TAD). The biosensor is then engineered to have ligand-dependent stability, such that in the absence of ligand, the biosensor is unstable and rapidly degraded by cellular protein degradation machinery, but in the presence of ligand, the biosensor is stabilized by binding ligand. In this model, biosensor stability is linked to reporter gene expression, such that increased biosensor stability leads to increased levels of reporter gene expression. The primary advantage to this strategy is that existing LBDs, for which thousands have been structurally characterized, can be converted into new biosensors. I hope to use this strategy to develop new biosensors that function in E. coli.
In collaboration with George Church and David Baker, our group recently reported a modular design for biosensor development in eukaryotes. This biosensor design fuses a ligand-binding domain (LBD) to a DNA-binding domain (DBD) and transcription-activating domain (TAD). The biosensor is then engineered to have ligand-dependent stability, such that in the absence of ligand, the biosensor is unstable and rapidly degraded by cellular protein degradation machinery, but in the presence of ligand, the biosensor is stabilized by binding ligand. In this model, biosensor stability is linked to reporter gene expression, such that increased biosensor stability leads to increased levels of reporter gene expression. The primary advantage to this strategy is that existing LBDs, for which thousands have been structurally characterized, can be converted into new biosensors. I hope to use this strategy to develop new biosensors that function in E. coli.
As an initial experiment, we engineered the allosteric transcription factor LacI into a biosensor that functions based on ligand-dependent stabilization. A library of LacI variants was generated by error-prone PCR and cloned between the Zif268 DBD and RNA polymerase omega subunit TAD. Positive and negative selection was used to identify the LacI variant L5, which has six amino acid mutations. L5 drives HIS3 expression in a ligand-dependent manner, resulting in up to a 4.5-fold increase in growth rate in the presence of the IPTG ligand. I am now interested in developing E. coli biosensors to detect environmental hazards, including transition metals, endocrine-disrupting compounds, and herbicides.
Engineering natural product biosynthetic pathways with multiplexed mutagenesis and biosensor-based selection
Ben Brandsen
Antibiotic resistance is a serious and growing problem. In 2013, the U.S. Centers for Disease Control reported that antibiotic resistance is responsible for more than 23,000 deaths and 2 million infections each year, costing approximately $20 billion. The problem of increasing antibiotic resistance is compounded by the small number of new antibiotics approved by the FDA in the past decade and a dearth of new antibiotic drug candidates in clinical trials. One rich source of antibiotic lead compounds for several decades has been natural products, but isolation of new natural products has slowed, significantly hampered by the rediscovery of known compounds. If the biosynthetic pathways that synthesize therapeutically useful natural products can be efficiently engineered in a laboratory, then new, natural product-like compounds would be readily accessible.
Polyketides are a class of natural products from which many important antibiotic compounds have been identified, including erythromycin, azithromycin, and clarithromycin. Polyketides are biosynthesized by polyketide synthase (PKS) enzymes, which consist of several modular protein domains that iteratively add small extender substrates together. One of the most well-studied polyketide biosynthetic pathways is that which produces the antibiotic erythromycin A. It consists of three synthases, named DEBS1-3. Each synthase contains two modules, each of which incorporates a single extender substrate. A thioesterase domain on DEBS3 cyclizes the polyketide to form 6-deoxyerythronolide B, and additional tailoring enzymes convert this cyclic intermediate to erythromycin A.
Polyketides are a class of natural products from which many important antibiotic compounds have been identified, including erythromycin, azithromycin, and clarithromycin. Polyketides are biosynthesized by polyketide synthase (PKS) enzymes, which consist of several modular protein domains that iteratively add small extender substrates together. One of the most well-studied polyketide biosynthetic pathways is that which produces the antibiotic erythromycin A. It consists of three synthases, named DEBS1-3. Each synthase contains two modules, each of which incorporates a single extender substrate. A thioesterase domain on DEBS3 cyclizes the polyketide to form 6-deoxyerythronolide B, and additional tailoring enzymes convert this cyclic intermediate to erythromycin A.
Because of the iterative manner by which polyketides are built and the modular nature of the PKS enzymes, engineering the biosynthetic pathways to produce new compounds is an appealing prospect. Efforts toward this prospect primarily focused on changing the order of protein domains, substituting exogenous enzyme domains, and supplying unique extender substrates. However, technical challenges, such as poor interaction between non-native protein domains and reduced enzymatic activity, have stalled these efforts.
I hope to overcome many of these challenges by using high-throughput DNA mutagenesis and biosensor-based selection to improve the activity of chimeric PKS enzymes. I will first develop E. coli biosensors that bind to polyketide macrolide compounds, which will be used to detect active polyketide pathway variants. I will then build a library of chimeric DEBS pathways that incorporates acyltransferase domains from other polyketide pathways. Acyltransferase domains control which substrates are incorporated into the polyketide, and therefore substituting the native AT domain for an AT domain with different extender substrate specificity will enable different extender substrates to be incorporated.
I hope to overcome many of these challenges by using high-throughput DNA mutagenesis and biosensor-based selection to improve the activity of chimeric PKS enzymes. I will first develop E. coli biosensors that bind to polyketide macrolide compounds, which will be used to detect active polyketide pathway variants. I will then build a library of chimeric DEBS pathways that incorporates acyltransferase domains from other polyketide pathways. Acyltransferase domains control which substrates are incorporated into the polyketide, and therefore substituting the native AT domain for an AT domain with different extender substrate specificity will enable different extender substrates to be incorporated.
Finally, I will use multiplexed genome engineering in E. coli to introduce mutations near the substitute AT domain, and select for active pathway variants using a biosensor to detect product formation. This strategy of high-throughput mutagenesis and biosensor-based selection could also be applied to other biosynthetic pathways that produce valuable small molecule or protein products.
Genetic sensors for the detection of biosynthetic products
Ben Jester
Model microorganisms such as E. coli and S. cerevisiae are valuable tools for the production of small molecules, such as therapeutics and biofuels, but the optimization of biosynthetic pathways in a new host is non-trivial. Identifying genetic modifications that enhance metabolite synthesis can be an exceptionally laborious process, particularly in the absence of a method to easily determine product yields. Genetically encoded biosensors that couple small molecule recognition to a readily measured output allow more rapid identification of cells with enhanced biosynthetic production or conditions that promote enhanced production. One of Nature’s most common mechanisms for detecting the presence of a small molecule – be it a metabolite or an environmental agent – has been to evolve biosensors that regulate the transcription of one or more genes upon binding of the relevant ligand. The goal of this project is to design a transcription factor (TF)-based strategy for biosensors that may provide a general solution to the problem of small molecule detection.
Our strategy is to use destabilizing mutations that impair functional expression of the biosensor until it binds its cognate ligand. It should be possible to design a protein-based biosensor for any ligand as long as a ligand-binding domain (LBD) exists – or can be designed – for the desired small molecule. By fusing an unstable LBD to a TF, we couple ligand-dependent stabilization to reporter gene expression. As a proof of principle, we chose to use a de novo designed LBD, DIG10.3, which binds digoxigenin, a steroid similar to drugs used to treat heart failure. By fusing this protein to a DNA-binding domain (DBD) and a transcriptional activation domain (TAD), we were able to generate a ligand-dependent TF, which we designated GDVP. From a library of mutants, FACS analysis allowed us to identify destabilizing mutations that improved sensor function by >10-fold (GDVP.1 and GDVP.2).
Our strategy is to use destabilizing mutations that impair functional expression of the biosensor until it binds its cognate ligand. It should be possible to design a protein-based biosensor for any ligand as long as a ligand-binding domain (LBD) exists – or can be designed – for the desired small molecule. By fusing an unstable LBD to a TF, we couple ligand-dependent stabilization to reporter gene expression. As a proof of principle, we chose to use a de novo designed LBD, DIG10.3, which binds digoxigenin, a steroid similar to drugs used to treat heart failure. By fusing this protein to a DNA-binding domain (DBD) and a transcriptional activation domain (TAD), we were able to generate a ligand-dependent TF, which we designated GDVP. From a library of mutants, FACS analysis allowed us to identify destabilizing mutations that improved sensor function by >10-fold (GDVP.1 and GDVP.2).
In order to tune the sensor for selections using a HIS3 reporter, we fused the degron from the Mata protein to GDVP to increase degradation. The fusion generated a sensor that leads to growth in histidine-deficient media more than 3 orders of magnitude better when the ligand (digoxigenin) is present. Sensitivity to digoxigenin can be further improved by deleting yeast efflux pumps, like Pdr5.
While the biosynthetic pathway required to produce digoxigenin is not known, other steroids such as progesterone have been successfully produced by yeast. An additional round of mutagenesis and screening allowed us to create a sensor that is extremely selective for progesterone but not its biosynthetic precursor, pregnenolone. Collaborators have used this sensor to identify mutations in 3-beta-hydroxysteroid dehydrogenase that improve progesterone bioproduction in yeast. We are currently applying these methods to develop sensors for new small molecules.
While the biosynthetic pathway required to produce digoxigenin is not known, other steroids such as progesterone have been successfully produced by yeast. An additional round of mutagenesis and screening allowed us to create a sensor that is extremely selective for progesterone but not its biosynthetic precursor, pregnenolone. Collaborators have used this sensor to identify mutations in 3-beta-hydroxysteroid dehydrogenase that improve progesterone bioproduction in yeast. We are currently applying these methods to develop sensors for new small molecules.
Engineering ligand-gated dimers to control cellular functions
Ben Jester
Protein-protein interactions drive the overwhelming majority of cellular processes, such as signaling, differentiation, and proliferation. Interacting protein domains coordinate these critical functions by ensuring that functional domains act on the appropriate target. Native interacting domains can often be replaced with an alternative pair of interacting proteins that result in near wild-type function. For example, the protein pair of FKBP and FRB is a popular substitute for interacting domains because interaction between these two proteins is dependent upon the small molecule rapamycin. When fused to functional domains requiring co-localization for activity, they provide the researcher with temporal control over a cellular process. Currently, the toolkit for small molecule-gated protein-protein interactions is exceptionally small. As the field of synthetic biology continues to grow, so will the need for tools that provide temporal control over multiple cellular processes.
To approach this problem, we have taken an unstable, homodimeric ligand-binding domain (LBD) and converted it into a ligand-dependent dimer. The LBD used here, DIG10.3, is a small protein that is degraded by the cell until it binds to the small molecule digoxigenin. To measure LBD dimerization, we made two fusion proteins: a single monomer of the LBD fused to a DNA-binding domain (DBD), and a single monomer of the LBD fused to a transcriptional activation domain (TAD). In an assay conceptually similar to the yeast two-hybrid system, dimerization between these two fusion proteins should induce the expression of a reporter protein. Previous work with this protein has shown that mutations to the dimer interface of DIG10.3 can lead to destabilization of the protein. By testing pairs of proteins that contained mutations at the dimer interface or in the ligand-binding pocket, we were able to identify dimers that respond to either digoxigenin or progesterone, or that required both ligands to activate reporter expression.
To approach this problem, we have taken an unstable, homodimeric ligand-binding domain (LBD) and converted it into a ligand-dependent dimer. The LBD used here, DIG10.3, is a small protein that is degraded by the cell until it binds to the small molecule digoxigenin. To measure LBD dimerization, we made two fusion proteins: a single monomer of the LBD fused to a DNA-binding domain (DBD), and a single monomer of the LBD fused to a transcriptional activation domain (TAD). In an assay conceptually similar to the yeast two-hybrid system, dimerization between these two fusion proteins should induce the expression of a reporter protein. Previous work with this protein has shown that mutations to the dimer interface of DIG10.3 can lead to destabilization of the protein. By testing pairs of proteins that contained mutations at the dimer interface or in the ligand-binding pocket, we were able to identify dimers that respond to either digoxigenin or progesterone, or that required both ligands to activate reporter expression.
The use of dimeric proteins fused to two different components (i.e. a DNA-binding domain and a transcriptional activation domain) is limited by the possibility of non-functional dimers that can form between identical components. Using a crystal structure model of the DIG10.3 homodimer, we made several rational mutations to the dimer interface that were predicted to favor heterodimerization of the two proteins. Two redesigns based on either a “charge-swap” (1a and 1b) or “knob-in-hole” (2a and 2b) model generated a pair of orthogonal interfaces and improved inducible reporter activation by more than 10-fold. On-going work is focused on exploring how directed evolution and deep sequencing may allow us to identify additional unique interfaces and deploying multiple pairs of ligand-dependent dimers in the same cell.
