Refactoring the nitrogen fixation gene cluster from Klebsiella oxytoca

Abstract

Bacterial genes associated with a single trait are often grouped in a contiguous unit of the genome known as a gene cluster. It is difficult to genetically manipulate many gene clusters because of complex, redundant, and integrated host regulation. We have developed a systematic approach to completely specify the genetics of a gene cluster by rebuilding it from the bottom up using only synthetic, well-characterized parts. This process removes all native regulation, including that which is undiscovered. First, all noncoding DNA, regulatory proteins, and nonessential genes are removed. The codons of essential genes are changed to create a DNA sequence as divergent as possible from the wild-type (WT) gene. Recoded genes are computationally scanned to eliminate internal regulation. They are organized into operons and placed under the control of synthetic parts (promoters, ribosome binding sites, and terminators) that are functionally separated by spacer parts. Finally, a controller consisting of genetic sensors and circuits regulates the conditions and dynamics of gene expression. We applied this approach to an agriculturally relevant gene cluster from Klebsiella oxytoca encoding the nitrogen fixation pathway for converting atmospheric N₂ to ammonia. The native gene cluster consists of 20 genes in seven operons and is encoded in 23.5 kb of DNA. We constructed a “refactored” gene cluster that shares little DNA sequence identity with WT and for which the function of every genetic part is defined. This work demonstrates the potential for synthetic biology tools to rewrite the genetics encoding complex biological functions to facilitate access, engineering, and transferability.

Many functions of interest for biotechnology are encoded in gene clusters, including metabolic pathways, nanomachines, nutrient scavenging mechanisms, and energy generators (1). Clusters typically contain internal regulation that is embedded in the global regulatory network of the organism. Promoters and 5′-UTRs are complex and integrate many regulatory inputs (2, 3). Regulation is highly redundant; for example, it contains embedded feedforward and feedback loops (4). Regulation can also be internal to genes, including promoters, pause sites, and small RNAs (5, 6). Further, genes often physically overlap, and regions of DNA can have multiple functions (7). The redundancy and extent of this regulation makes it difficult to manipulate a gene cluster to break its control by native environmental stimuli, optimize its function, or transfer it between organisms. As a consequence, many clusters are cryptic, meaning that laboratory conditions cannot be identified in which they are active (8).

Gene clusters have been controlled from the top down by manipulating the native regulation or adding synthetic regulation in an otherwise WT background (9). For example, knocking out a repressor or overexpressing an activator has turned on clusters encoding biosynthetic pathways (10–14). When the cluster is a single operon, it has been shown that a promoter can be inserted upstream to induce expression (15). The entire echinomycin biosynthetic cluster was transferred into Escherichia coli by placing each native gene under the control of a synthetic promoter (16).

In engineering, one approach to reduce the complexity of a system is to “refactor” it, a term borrowed from software development whereby the code underlying a program is rewritten to achieve some goal (e.g., stability) without changing functionality (17). This term was first applied to genetics to describe the top-down simplification of a phage genome by redesigning known genetic elements to be individually changeable by standard restriction digest (18). Here, we use it to refer to a comprehensive bottom-up process to systematically eliminate the native regulation of a gene cluster and replace it with synthetic genetic parts and circuits (Fig. 1). The end product is a version of the gene cluster whose DNA sequence has been rewritten, but it encodes the same function. The design process occurs on the computer, and then the resulting DNA sequence is constructed by using DNA synthesis (19). The first step of the process is to remove all noncoding DNA and regulatory genes. Next, each essential gene is recoded by selecting codons that produce a DNA sequence that is as distant as possible from the WT sequence. The intent is to introduce mutations throughout the gene to eliminate internal regulation (including that which is undiscovered), such as operators, promoters, mRNA secondary structure, pause sites, methylation sites, and codon regulation. Recoded sequences are further scanned by computational methods to identify putative functional sequences, which are then removed. The recoded genes are organized into artificial operons, and the expression levels are controlled by synthetic ribosome binding sites (RBSs), and spacer sequences physically separate the genes. The end result is a refactored gene cluster whose native regulation has been removed and has been organized into a set of discrete, well characterized genetic parts.

Fig. 1.

The process of refactoring a gene cluster. Top: WT K. oxytoca nitrogen fixation gene cluster. The genes are colored by function: blue (nitrogenase), green (cofactor biosynthesis, shading corresponds to operons), yellow (e^- transport), and gray (unknown). The thin arrows show the length and orientation of the seven operons, and a horizontal bar indicates overlapping genes. The recoded genes are shown as dashed lines. The symbols used to define the refactored cluster and controller are defined in Figs. 4 and 5, respectively.

When the native regulation has been removed, synthetic regulation can be added back to control the dynamics and conditions under which the cluster is expressed. Constructing such regulation has been a major thrust of synthetic biology and involves the design of genetic sensors and circuits and the understanding of how to connect them to form programs (20). In our design, we genetically separate the sensing/circuitry from the refactored pathway by carrying them on different low-copy plasmids (Fig. 1). The plasmid containing the sensors and circuits is referred to as the controller, and the output of the circuits led to the expression of an engineered T7 RNA polymerase (T7* RNAP). The refactored cluster is under the control of T7 promoters. One advantage of this organization is that T7 RNA polymerase is orthogonal to native transcription and the T7 promoters are tightly off in the absence of the controller. In addition, changing the regulation is simplified to swapping the controller for one that contains different sensors and circuits, so long as the dynamic range of T7* RNAP is fixed.

As a demonstration, we have applied this process to refactor the gene cluster encoding nitrogen fixation in Klebsiella oxytoca (21). Nitrogen fixation is the conversion of atmospheric N₂ to ammonia (NH₃) so that it can enter metabolism (22). Industrial nitrogen fixation through the Haber–Bosch process is used to produce fertilizer. Many microorganisms fix nitrogen, and the necessary genes typically occur together in a gene cluster, including the nitrogenase subunits, the metallocluster biosynthetic enzymes and chaperones, e^- transport, and regulators (Fig. 2A) (23, 24). The gene cluster from K. oxytoca has been a model system for studying nitrogen fixation and consists of 20 genes encoded in 23.5 kb of DNA (Fig. 1, Top) (25). The biosynthesis of nitrogenase is tightly regulated by a two-layer transcriptional cascade in response to fixed nitrogen, oxygen, and temperature (26). The complete cluster has been transferred to E. coli, thus demonstrating that it has all the genes necessary for nitrogen fixation (27). The encoding of this function is complex, many of the genes overlap, the operons are oriented in opposite directions, and there are many putative hidden regulatory elements, including internal promoters and hairpins (25). The purpose of refactoring is to reorganize the cluster, simplify its regulation, and assign a concrete function to each genetic part.

Fig. 2.

The robustness of the nitrogen fixation pathway to changes in the expression of component proteins. (A) The pathway for nitrogenase maturation is shown and proteins are colored by function (Fig. 1). The metal clusters are synthesized by the biosynthetic pathway (23, 24). Nitrogen fixation catalyzed by the matured nitrogenase is shown with its in vivo electron transport chain. (B) The tolerance of nitrogenase activity to changes in the expression of component proteins are shown. Activity is measured via an acetylene reduction assay and the percentage compared with WT K. oxytoca is presented. WT operons are expressed from a P_tac promoter on a low-copy plasmid (SI Materials and Methods). The promoter activity is calculated as the output of the P_tac promoter at a given concentration of IPTG and compared with a constitutive promoter. The effect of not including NifY (-Y) and NifX (-X) are shown in red. (C) The comparison of the strength of WT (black) and synthetic (white) RBSs is shown. The RBSs were measured through an in-frame transcriptional fusion (−60 to +90) with mRFP. The strength is measured as the geometric average from a distribution of cells measured by flow cytometry (SI Materials and Methods). The synthetic RBSs of nifF and nifQ are not intended to match the WT measurement. Error bars represent the SD of at least three experiments performed on different days.

Results

Tolerance of Native Gene Cluster to Changes in Expression.

Before refactoring a cluster, a robustness analysis is performed to determine the tolerances of a gene or set of genes to changes in expression level (Fig. 2B). This informs the grouping of genes into operons and the selection of synthetic parts to obtain desired expression levels. In the WT background, genes are knocked out and complemented under inducible control. The tolerance is obtained by measuring nitrogenase activity as a function of the activity of the inducible promoter.

Nitrogenase function is notably sensitive to expression changes, and each tolerance has a clear optimum (Fig. 2B). The chaperone NifY is required to achieve full activity and broadens the tolerance to changes in expression level. NifT did not have an effect on activity, as observed previously (28), and it is frequently absent from homologous clusters (29). The genes controlling electron transport (nifJ and nifF) need to be expressed at low levels, and activity decreases rapidly as expression increases. The optima for genes participating in the metal cluster biosynthetic pathways vary. The nifUSVWZM operon, which encodes proteins for early Fe-S cluster formation and proteins for component maturation, needs to be expressed at low levels, whereas nifBQ, encoding proteins for FeMo-co core synthesis and molybdenum integration, need to be expressed 10-fold higher. NifEN is tolerant to varied expression levels. However, activity is lost with the inclusion of nifX, which has been characterized as a negative regulator (30). The native cluster also includes the regulatory proteins NifL and NifA, which integrate environmental signals (26). The genes nifT, nifX, and nifLA are not included in the refactored cluster.

Complete Refactored Gene Cluster.

The nitrogenase activities of the refactored operons were measured as a function of the isopropyl-β-d-thiogalactopyranoside (IPTG)-inducible P_tac promoter (Fig. 3A). Each operon has a different optimum. To combine the operons, the P_tac promoters were replaced with T7 promoters that have a strength close to the measured optimum (Fig. 3B and SI Materials and Methods). The nitrogenase genes (nifHDK) are highly expressed in Klebsiella under fixing conditions (as much as 10% of cell protein) (31), so the strongest promoter was used to control this operon [P_T7.WT, 0.38 relative expression units (REUs)] (32). A long operon was built to include the nifEN and nifJ genes, whereby the lower expression required for nifJ was achieved through transcriptional attenuation. The nifF gene was encoded separately under the control of a medium strength promoter (P_T7.3, 0.045 REU). Finally, the nifUSVWZM and nifBQ operons were controlled by weak promoters (P_T7.2, 0.019 REU). Each of the individual refactored operons under the control of a T7 promoter was able to recover the activity observed from the P_tac promoter and corresponding optimal IPTG concentration (Fig. 3C).

Fig. 3.

Converting to T7* RNAP control. (A) Nitrogenase activity is shown as a function of promoter strength for each refactored operon in respective K. oxytoca KO strains (ΔnifHDKTY, ΔnifENX, ΔnifJ, ΔnifBQ, ΔnifF, and ΔnifUSVWZM). Vertical dashed lines indicate strength of the mutant T7 promoter that controls each operon in the complete refactored gene cluster. (B) A controller plasmid decouples operon expression from the inducible promoter. A T7 RNAP variant (T7* RNAP) was designed to reduce toxicity. A set of four mutated T7 promoters were used to control the expression of each operon (part numbers and sequences for mutants 1–4 are listed in SI Materials and Methods). P_tac activity under 1 mM IPTG induction is indicated by a dashed horizontal line. (C) Nitrogenase activity is compared for each refactored operon under the control of the P_tac promoter at the optimal IPTG concentration (black) and the controller with 1 mM IPTG and expression controlled by different T7 promoters (white). The T7 promoters used are P_T7.WT for operons HDKY, EN, and J; P_T7.2 for operons BQ and USVWZM; and P_T7.3 for F. Error bars represent the SD of at least three experiments performed on different days.

Transitioning the control to T7* RNAP and T7 promoters facilitates the assembly of the complete cluster from refactored operons. We first assembled half-clusters by using Gibson Assembly (33) and verified their function in strains with the corresponding genes deleted. The first half-cluster consisted of the nifHDKYENJ operon. The second half-cluster was assembled from the nifBQ, nifF, and nifUSVWZM operons. The half-clusters were able to recover 18% ± 0.7% and 26% ± 8.4% of WT activity, respectively. The full synthetic cluster was assembled from both half-clusters (Fig. 4), and its activity measured in a strain in which the full cluster is deleted. The synthetic gene cluster recovers nitrogenase activity at 7.4% ± 2.4% of the WT (Fig. 5A). Strains carrying the synthetic gene cluster used ambient N₂ as a nitrogen source, growing 3.5-fold slower than the WT strain (SI Materials and Methods) and incorporating ¹⁵N-labeled nitrogen into 24% ± 1.4% of their cellular nitrogen content, as measured by isotope ratio mass spectrometry (Fig. 5B).

Fig. 4.

Comprehensive schematic illustration for the complete refactored gene cluster and controller. Each of the 89 parts is represented according to the Synthetic Biology Open Language visual standard (www.sbolstandard.org), and the SynBERC Registry part number (registry.synberc.org) and part activity are shown. The full sequences of each plasmid have been deposited in GenBank (SBa_000534, JQ903614; SBa_000559, JQ903615; SBa_000560, JQ903616). The T7 promoter strengths are measured with monomeric red fluorescent protein and reported in REUs (Materials and Methods). Terminator strengths are measured in a reporter plasmid and reported as the fold reduction in monomeric red fluorescent protein (RFP) expression compared with a reporter without a terminator. The RBS strength is reported in as arbitrary units of expression from the induced P_tac promoter (1 mM IPTG) and a fusion gene between the first 90 nt of the gene and RFP. The nucleotide numbers for the plasmids containing the refactored cluster and controller are shown. The codon identity of each recoded gene compared with WT is shown as a percentage.

Fig. 5.

Regulation of the complete refactored gene cluster. (A) Nitrogenase activity for the three controllers are shown: IPTG-inducible, aTc-inducible, and IPTG ANDN aTc logic. The gas chromatography trace is shown for each, as well as the calculated percent of WT activity (7.4% ± 2.4%, 7.2% ± 1.7%, and 6.6% ± 1.7% respectively). SD is calculated by using data from at least two experiments performed on different days. (B) Incorporation of ¹⁵N into cell biomass is shown. Nitrogen fixation from N₂ gas by the refactored gene cluster was traced by using ¹⁵N₂ and measured by using isotope ratio mass spectrometry. Data are represented as the fraction of cellular nitrogen that is ¹⁵N. The SD represents two experiments performed on different days. (C) The effect of ammonia on regulation of nitrogenase expression is shown. Acetylene reduction traces shown with (red) and without (blue) addition of 17.5 mM ammonium acetate for WT cells (Left) and cells bearing synthetic nif system (Right). The synthetic system was induced by controller 1 using 1 mM IPTG and exhibited nitrogenase activity of 1.1% ± 0.5% and 6.1% ± 0.4% with and without ammonium acetate, respectively. (D) T7* RNAP expression of controller 1 corresponding to C is shown. Strains carrying controller 1 and an RFP reporter plasmid were characterized under 1 mM IPTG induction with or without addition of ammonium acetate.

The complete refactored cluster consists of 89 genetic parts, including a controller, and the function of each part is defined and characterized. Therefore, the genetics of the refactored system are complete and defined in the schematic illustration in Fig. 4. However, the process of simplification and modularization reduces activity (18). This is an expected outcome of refactoring a highly evolved system.

Swapping Controllers to Change Regulation.

The separation of the controller and the refactored cluster simplifies changing the regulation of the system. This can be achieved by transforming a different controller plasmid, as long as the dynamic range of the T7* RNAP expression is preserved. To demonstrate this, we constructed two additional controllers (Fig. 5A). Controller 2 changes the chemical that induces the system by placing the expression of T7* RNAP under the control of the aTc-inducible P_tet promoter. When induced, controller 2 produces nitrogenase activity identical to controller 1 (7.2% ± 1.7%). The controller can also serve as a platform to encode genetic circuits to control regulatory dynamics or to integrate multiple sensors. To this end, controller 3 contains two inducible systems (IPTG and aTc) and an ANDN gate (34, 35). In the presence of IPTG and the absence of aTc, nitrogen fixation is 6.6% ± 1.7% of WT activity. These controllers represent the simplicity by which the regulation of the refactored cluster can be changed.

In addition to making it possible to add new regulation, the process of refactoring eliminates the native regulation of the cluster. This is demonstrated through the decoupling of nitrogenase activity from the environmental signals that normally regulate its activity. For example, ammonia is a negative regulator that limits overproduction of fixed nitrogen (26). In the presence of 17.5 mM ammonia, no nitrogenase activity is observed for the WT cluster (Fig. 5C). In contrast, the refactored gene cluster maintains activity in the presence of ammonia (1.1% ± 0.5%). Interestingly, this sevenfold reduction of activity is not caused by residual regulation present in the system. Rather, it occurs because the addition of ammonia to the media reduces the output of the controller by 4.5-fold (Fig. 5C). In theory, this could be fixed by increasing the expression level of T7* RNAP, but it speaks to the need to create genetic circuits that are robust to environmental context.

Discussion

The objective of refactoring is to facilitate the forward engineering of multigene systems encoded by complex genetics. Native gene clusters are the product of evolutionary processes; thus, they exhibit high redundancy, efficiency of information coding, and layers of regulation that rely on different biochemical mechanisms (36–38). These characteristics inhibit the quantitative alteration of function by part substitution because the effect can become embedded in a web of interactions. Here, modularizing the cluster, physically separating and insulating the parts, and simplifying its regulation have guided the selection and analysis of part substitutions. The information gleaned from screening the permutations in a refactored system can be cleanly fed back into the design cycle.

The refactored cluster can also serve as a platform for addressing questions in basic biology. First, it allows for the impact of regulatory interactions to be quantified in isolation. For example, in the natural system, one feedback loop could be embedded in many other regulatory loops. Systematically removing such regulation provides a clean reference system (potentially less active than WT) from which improvements can be quantified as a result of adding back regulation. It also serves as a basis for comparison of radically different regulatory programs or organizational principles, for example, to determine the importance of temporal control of gene expression (4, 39) or the need for genes to be encoded with a particular operon structure (40, 41). Second, the process of reconstruction and debugging is a discovery mechanism that is likely to reveal novel genetics and regulatory modes.

Refactoring may enable the access of functions encoded in gene clusters that are identified within sequenced genomes. With advances in DNA synthesis technology, it is possible to construct complete gene clusters and specify every nucleotide in the design. This capability eliminates the reliance on the natural physical DNA for construction and enables the simultaneous specification of every part in the system. The systematic replacement of gene regulation will be required if the cluster is silent (unexpressed in laboratory conditions) or if it needs to be transferred into a heterologous host (42). This is particularly important of the source of the DNA is unknown, for example, from a metagenomic sample.

Two relevant challenges were encountered when refactoring the nitrogenase gene cluster. First, not all of the necessary genetics will be known or there will be errors in the sequence. To address this, debugging methods will have to be developed that do not require a deep, specific understanding of the pathway. This will be aided by high-throughput part assembly techniques where many designs can be evaluated simultaneously. Second, there is a need for context-independent parts to control expression and computational methods to scan genetic designs for interfering functions. Together, these approaches will enable the rapid mining of multi-gene cellular functions from sequence databases for industrial, agricultural, and pharmaceutical applications.

Materials and Methods

Strains and Media.

E. coli strain S17-1 was used for construction and propagation of all plasmids used in K. oxytoca KO mutant construction. K. oxytoca strain M5a1 (gift from Paul Ludden, University of California, Berkeley, CA) and mutants derived from M5al were used for nitrogen fixation experiments. Luria–Bertani/Lennox medium was used for strain propagation. All assays were carried out in minimal medium containing (per liter) 25 g Na₂HPO₄, 3g KH₂PO₄, 0.25g MgSO₄⋅7H₂O, 1g NaC1, 0.1g CaCl₂⋅2H₂O, 2.9 mg FeCl₃, 0.25 mg Na₂MoO₄⋅2H₂O, and 20 g sucrose. Growth medium is defined as minimal medium supplemented with 6 mL (per liter) of 22% (wt/vol) NH₄Ac. Derepression medium is defined as minimal medium supplemented with 1.5 mL (per liter) of 10% (wt/vol) serine. The antibiotics used were 34.4 μg⋅mL⁻¹ chloramphenicol, 100 μg⋅mL⁻¹ spectinomycin, 50 μg⋅mL⁻¹ kanamycin, and/or 100 μg⋅mL⁻¹ ampicillin.

Codon Randomization.

Initial gene sequences were proposed by DNA2.0 to maximize the Hamming distance from the native sequence while seeking an optimal balance between K. oxytoca codon use and E. coli codon preferences experimentally determined by the company (43). Rare codons (<5% occurrence in K. oxytoca) were avoided, and mRNA structure in the translation initiation region was suppressed. Known sequence motifs, including restriction sites, transposon recognition sites, Shine–Dalgarno sequences, and transcriptional terminators, were removed by the DNA2.0 algorithm.

Elimination of Undesired Regulation.

Each synthetic operon was scanned before DNA synthesis to identify and remove undesired regulation. Multiple types of regulation were identified using publicly available software. The RBS Calculator was used (Reverse Engineering mode; 16S RNA:ACCTCCTTA) to identify RBSs throughout the proposed DNA sequence of the operon (44). The Prokaryotic Promoter Prediction server was used to identify putative σ70 promoter sites (e-value cutoff of 5, sigma.hmm database) (45). The PromScan algorithm was used to identify putative σ54 promoter sites using default options (46). TransTermHP software was used with default parameters to identify terminator sequences in both the forward and reverse directions (47). RBSs greater than 50 AU and all identified promoters and terminators were considered significant.

Nitrogenase Activity Assay.

In vivo nitrogenase activity is determined by acetylene reduction as previously described (48). For K. oxytoca whole-cell nitrogenase activity assay, cells harboring the appropriate plasmids were incubated in 5 mL of growth media (supplemented with antibiotics, 30 °C, 250 rpm, New Brunswick Scientific, Innova 44 incubator shaker) in 50-mL conical tubes for 14 h. The cultures were diluted into 2 mL derepression media (supplemented with antibiotics and inducers) to a final OD₆₀₀ of 0.5 in 14-mL bottles, and bottles were sealed with rubber stoppers (Z564702; Sigma). Headspace in the bottles was repeatedly evacuated and flushed with N₂ past a copper catalyst trap by using a vacuum manifold. After incubating the cultures for 5.5 h at 30 °C, 250 rpm in a New Brunswick Scientific, Innova 3100 water bath shaker, headspace was replaced by 1 atm Ar. Acetylene was generated from CaC₂ by using a Burris bottle, and 1 mL was injected into each bottle to start the reaction. Cultures were incubated for 1 h at 30 °C, 250 rpm in a New Brunswick Scientific, Innova 3100 water bath shaker, before the assay was stopped by injection of 300 μL of 4 M NaOH solution into each bottle. To quantify ethylene production, 50 μL of culture headspace was withdrawn through the rubber stopper with a gas tight syringe and manually injected into a HP 5890 gas chromatograph. Nitrogenase activity is reported as a percentage of WT activity. Briefly, ethylene production by strains was quantified by integrating area under the peak by using ChemStation software and dividing ethylene production of experimental strains by the ethylene production of a WT control included in each assay.

N₂-Dependent Growth and ¹⁵N₂ Incorporation Assay.

Nitrogen fixation by synthetic nif cluster in K. oxytoca is further demonstrated by N₂-dependent growth and ¹⁵N₂ incorporation. Cells are diluted as described in the acetylene reduction assay. The headspace of the bottles is replaced by normal N₂ gas or by stable isotope nitrogen, ¹⁵N₂ (¹⁵N atom 99.9%; catalog no. IN 5501; Icon Isotopes). After incubating the cultures for 36 h at 30 °C, 250 rpm in a New Brunswick Scientific, Innova 3100 water bath shaker, N₂-dependent growth of the cells is determined by measuring optical density at 600 nm (i.e., OD₆₀₀). To do the ¹⁵N₂ incorporation assay, the ¹⁵N-enriched cells with corresponding control cultures under normal nitrogen gas are collected by centrifugation, and the cell pellets are dried in a laboratory oven at 100 °C for 12 h. The dried pellets are analyzed for ¹⁵N/¹⁴N ratio at the Center for Stable Isotope Biogeochemistry at the University of California, Berkeley, by using the Finnigan MAT Delta plus Isotope Ratio Mass Spectrometer.