Rational Design of Thiolase Substrate Specificity for Metabolic Engineering Applications

Abstract

Metabolic engineering efforts require enzymes that are both highly active and specific towards the synthesis of a desired output product in order to be commercially feasible. The 3-hydroxyacid pathway (3HA), also known as the reverse β-oxidation (r-BOX) or coenzyme-A dependent chain elongation pathway, can allow for the synthesis of dozens of useful compounds of various chain lengths and functionalities. However, this pathway suffers from byproduct formation, which lowers yields of the desired longer chain products, as well as increases downstream separation costs. The thiolase enzyme catalyzes the first reaction in this pathway, and its substrate specificity at each of its two catalytic steps sets the chain length and composition of the chemical scaffold upon which the other downstream enzymes act. However, there have been few attempts reported in the literature to rationally engineer thiolase substrate specificity. In this work, we present a model-guided, rational design study of ordered substrate binding applied to two biosynthetic thiolases, with the goal of increasing the ratio of C6/C4 products formed by the 3HA pathway, 3-hydroxy-hexanoic acid (3HH) and 3-hydroxybutyric acid (3HB). We identify thiolase mutants that result in nearly ten-fold increases in C6/C4 selectivity. Our findings can extend to other pathways that employ the thiolase for chain elongation, as well as expand our knowledge of sequence-structure-function relationship for this important class of enzymes.

Introduction

Microbial fermentation affords many advantages for the synthesis of commodity and specialty chemicals over more traditional methods. These advantages include mild reaction conditions, avoidance of harsh and toxic chemicals, and the ability to utilize renewable feedstocks (Keasling, 2009). Advances in metabolic engineering and synthetic biology now allow for fast construction and manipulation of heterologous pathways in canonical production host strains (Lee et al., 2012). Although a wide variety of useful compounds have been synthesized using biological systems, few of these pathways have been commercialized. For a given pathway to be commercially viable, the process must produce the desired product in high yield, at a high titer and with high productivities.

The 3-hydroxyacid (3HA) pathway (Figure 1A), also referred to as the reverse β-oxidation (r-BOX) or CoA-dependent chain elongation pathway, can allow for the synthesis of dozens of useful compounds of various chain lengths and functionalities, including acids, alcohols, alkanes and aldehydes, with applications in the pharmaceutical, polymer and flavor and fragrance industries (Clomburg et al., 2015; Kim et al., 2015; Sheppard et al., 2014; Tseng & Prather, 2012; Blaisse et al., 2017). This is due to the promiscuous activities of pathway enzymes, which on the one hand makes the biological synthesis of these compounds possible, but on the other, always results in a mixture of products at the end of the fermentation (Cheong et al. 2016; Clomburg et al. 2012). Thus, it is imperative to select pathway enzymes with appropriate substrate specificities to maximize yields of the desired product and to minimize downstream separation costs. The lack of enzymes in the metabolic engineer’s toolbox that are both highly active and highly specific toward the manufacturing of a particular product is a major limitation in the construction of commercially feasible metabolic pathways.

Figure 1.

A Generalized 3HA pathway, which is also referred to as CoA-dependent chain elongation or reverse β-oxidation. This pathway consists of four core enzymes – a coenzyme-A (CoA) activating enzyme which converts a small acid precursor to a CoA thioester, a thiolase which brings about the condensation of the CoA activated acid and acetyl-CoA, a reductase which reduces the β-carbonyl of the resulting longer chain intermediate, and finally a thioesterase which cleaves the thioester bond of the 3-hydroxyacyl-CoA, releasing free CoASH and the free 3-hydroxyacid. A wide variety of other compounds can be produced by addition of other enzymes that can act on the 3-hydroxyacyl-CoA intermediates, such as enoyl-CoA dehydratases and reductases, and alcohol and aldehyde dehydrogenases. Biosynthesis of longer chain 3HAs and carboxylic acids, as well as ω-carboxylic acids, and longer chain alcohols has been demonstrated (Cheong, Clomburg, and Gonzalez 2016c; Sheppard et al. 2014). However, a mix of products of variable chain lengths always results.

B In this study we employ a four-enzyme pathway for the synthesis of poly-3HB-co-3HHx as a readout of thiolase selectivity. The cells are grown on glucose and supplied with butyrate. Activation of butyrate by the action of Pct (M. elsdenii), leads to butyryl-CoA which is then condensed with acetyl-CoA by a thiolase, either BktB (C. necator) or PhbA (Z. ramigera), to produce 3-oxohexanoyl-CoA. This intermediate is then reduced to 3HH-CoA by an acetoacetyl-CoA reductase PhaB (C. necator). The thiolase is also capable of condensing two acetyl-CoA molecules which leads to production of 3HB-CoA upon reduction by PhaB. The 3HA-CoA intermediates are then polymerized into PHAs by PhaC2 (R. aetherivorans I24).

C Reaction mechanism of the thiolase occurs by a biological Claisen condensation reaction though a sequential bi bi ping-pong mechanism. In addition to other thiolases, this mechanism is also similar to those utilized by acetyltransferase and ketosynthase domains of polyketide synthetases. Panel 2 corresponds to Bind 1 and Panel 5 corresponds to Bind 2 on which structure based design calculations were performed. Note that in this figure the residue numbering for PhbA is shown. Corresponding catalytic residue identifiers for BktB are (PhbA numbering in parenthesis): Cys90 (Cys89), Cys380 (Cys378), and His350 (His348).

D Atomic nomenclature used throughout the rest of the paper.

In the 3HA pathway, the thiolase enzyme sets the chain length upon which the downstream enzymes act. Our goal was to obtain a more selective thiolase with high catalytic activity towards the synthesis of longer chain products. Previously, we have used the 3HA pathway and demonstrated synthesis of both 3-hydroxy-valeric acid (3HV) and 3-hydroxy-hexanoic acid (3HH). While we achieved 100% conversion of the fed propionate precursor for the synthesis of 3HV, less than 1% of the fed butyrate was converted to 3HH indicating poor specificity of the pathway enzymes towards the longer chain substrates (Martin et al., 2013). In this work we focused on achieving selective production of the longer chain C6 product, 3-hydroxyhexanoyl-CoA (3HH-CoA), relative to the C4, 3-hydroxybutyryl-CoA (3HB-CoA). Formation of 3HH-CoA results from the thiolase catalyzed condensation of a priming butyryl-CoA and extending acetyl-CoA, and subsequent action of a reductase on the product, whereas 3HB-CoA is formed by the condensation of two acetyl-CoA substrates followed by reduction (Figure 1B). We sought to increase the thiolase selectivity ratio, which we define here as the ratio of C6 product formed relative to the C4 product. Ideally, this would mean the ratio of the C6 product to the C4 product at the end of the thiolase catalyzed reaction, i.e. 3-oxohexanoyl-CoA (abbreviated BA in the nomenclature of Figure 2) to acetoacetyl-CoA (abbreviated AA using the nomenclature of Figure 2), but the thermodynamics of this reaction require coupling to a downstream enzyme to enable product formation. We use formation of free 3HH and 3HB, as well as polyhydroxyalkanoates (PHAs) containing those monomers, which are derived from 3HH-CoA and 3HB-CoA, the condensation products after the reductase step, in calculating thiolase selectivity.

Figure 2.

Four different products can result from the condensation reaction of acetyl-CoA (A) and butyryl-CoA (B) catalyzed by the thiolase. The product formed depends on the order of addition of the acyl-CoAs into the active site of the enzyme. The priming acyl-CoA serves as an electrophile at the carbonyl carbon and forms an acyl-enzyme intermediate. The extending acyl-CoA in this case acts as a nucleophile after abstraction of an α proton and formation of a carbanion. Self-condensation of two acetyl-CoA molecules results in formation of acetoacetyl-CoA, which we term the AA condensation product, and subsequent reduction by PhaB leads to the formation of 3HB. Condensation with butyryl-CoA as the priming acyl-CoA and acetyl-CoA as the extending acyl-CoA forms 3-oxohexanoyl-CoA which we term the BA condensation product, and subsequent reduction by PhaB leads to the formation of 3HH. In this study we sought to increase the ratio of 3HH to 3HB by increasing the ratio of the BA condensation product relative to the AA condensation product.

Thiolases catalyze the condensation of a priming acyl-CoA and an extending acyl-CoA using a sequential bi bi ping-pong mechanism (Figure 1C). We were interested in the condensation of butyryl-CoA and acetyl-CoA to form 3-oxohexanoyl-CoA with high specificity; however, it is not possible to directly assay for this reaction for several reasons. First, for biosynthetic thiolases, such as BktB from C. necator and PhbA from Zoogloea ramigera, the condensation direction is thermodynamically unfavorable, requiring the condensation product to be reacted further in order to drive the reaction forward (Thompson et al., 1989). Here, the thiolase is coupled with a kinetically competent dehydrogenase enzyme. Reacting away CoASH, the other product of the condensation reaction, is insufficient to drive the reaction forward because it is released in the first half-step of the overall condensation reaction mechanism. In addition, the self-condensation of two acetyl-CoAs will always occur with some frequency, biasing any measured reaction rate. However, the low yields and high cost of synthesis of these acyl-CoAs precluded the development of a high-throughput activity screen.

For the above reasons, thiolase engineering has proved challenging, with few examples of such attempts. The first attempt at thiolase engineering described in the literature used directed evolution to arrive at a variant that exhibited robust acetoacetyl-CoA product formation and lower sensitivity to inhibition by CoASH (Mann and Lütke-Eversloh, 2012). Another effort to engineer the thiolase to accommodate α-substituted acyl-CoAs relied on intuition guided rational mutagenesis of just one residue in close proximity of the active site but employed coenzyme-A analogs (Fage et al., 2015). During the preparation of this manuscript, two additional studies were published which reported rational mutagenesis of S. cerevisiae Erg10 thiolase and two A. suum thiolases for increased selectivities towards α-substituted substrates (Torras-Salas et al., 2018 and Blaisse et al., 2018 respectively).

Limited by a low throughput in vivo assay, but armed with extensive crystallographic data, we followed a computationally driven, structure guided approach to engineer the biosynthetic thiolase for improved selectivity towards the synthesis of longer chain products. We apply a theoretical framework for the design of ordered binding to biosynthetic thiolase PhbA from Z. ramigera for which there is ample crystallographic data but which exhibits low activity towards longer chain substrates, such as butyryl-CoA, to identify mutants which we predict will exhibit higher selectivity ratios compared to wild type (WT). We then applied this same approach to the more active C. necator BktB thiolase. Mutants predicted to improve the selectivity ratio were screened in vivo within the context of two heterologous pathways, free HA production and PHA biosynthesis, which employ different downstream enzymes (thioesterase vs. PHA polymerase).

This process led to the identification of thiolase mutants with up to ten-fold increases in the selectivity ratio. In vitro characterization confirmed that one of the most selective mutants had 30-fold lower activity towards formation of the 3HB product, whereas the activity towards 3HH formation was comparable to WT. Thiolases represent a large conserved superfamily of enzymes central to many other biological pathways, and lessons learned from this study can help expand our understanding of the sequence-structure-function relationship for this important class of enzymes.

Materials and Methods

Chemicals and reagents

All chemicals were obtained from Sigma Aldrich unless stated otherwise. Protein purification reagents were purchased from BioRad Laboratories (Hercules, CA).

Strain and plasmid construction

Escherichia coli MG1655 K12 (DE3) was used as the host for all production experiments. pCDFDuet-pct-phaC2 was constructed by restriction enzyme cloning. First, pct from M. elsdenii was amplified using Q5 Polymerase (New England Biolabs, Ipswish, MA) from M. elsdenii gDNA. PhaC2 was synthesized as a codon optimized gBlock from Thermo Fisher and digested with the respective restriction enzymes. Construction of pETDuet-bktB-phaB is described in Martin et al. (2013). This plasmid served as the template for generating BktB mutants. Primer sequences can be found in Supplementary Table II.

Culture conditions and strain propagation

E. coli DH5α was used for construction and maintenance of all plasmids. For PHA production experiments, E. coli MG1655 K12 (DE3) was transformed by electroporation with pCDFDuet-pct-phaC2 and a pETDuet plasmid with a given thiolase variant and phaB. For every production experiment, three individual colonies were picked and grown overnight in LB medium containing carbenicilin (50 μg/mL) and streptomycin (50 μg/mL) at 30°C, 250 rpm. A 250-mL shake flask containing 50 mL of M9 minimal medium with 15 g/L glucose was used for production experiments and inoculated with 1% v/v of the overnight starter culture. Expression of heterologous genes was induced by addition of IPTG to 100 μM final concentration when OD₆₀₀ was 0.7–1.0. When required, butyrate was added to 15 mM final concentrations from a neutralized sterilized stock solution at induction. Cells were harvested by centrifugation and washed twice with water before freezing at −80°C and lyophilization for polymer extraction and derivatization. For analysis of free acids, cell-free culture supernatants were analyzed directly by HPLC.

Site specific mutagenesis

All point mutants were made using the QuikChange Lightning XL Kit from Agilent Technologies according to the manufacturer’s protocols (Agilent Technologies, Santa Clara, CA), except that DH5α cells were used for transformation of QuikChange products. The online primer design tool (http://www.genomics.agilent.com/primerDesignProgram.jsp) was used to generate the mutagenesis primers to be used in the thermal cycling reaction. Primer sequences can be found in the Supplementary Table II. Products of this reaction were used to transform chemically competent E. coli DH5α and plated on selective medium after recovery in SOC. Individual colonies were selected and mutations confirmed by sequencing (GeneWiz, Cambridge, MA).

Product analysis

Acidic methanolysis to analyze PHA composition was performed as described by Brandl et al. (1988) and is briefly described below. Cells were harvested by centrifugation and washed twice with water. The cells were then frozen at −80°C. Lyophilized cells were weighed to determine the CDW. Then, 5–20 mg of dried cells was used for methanolysis to determine PHA polymer composition by GC/MS. Hexanoic acid was added as an internal standard to a final concentration of 2.5 mM. In short, 1 mL chloroform, 0.85 mL methanol and 0.15 mL concentrated H₂SO₄ was added to each sample in a screw-capped tube with threads wrapped with PTFE tape. The samples were then boiled for 1.5–2 hours at 100°C on a heating block with intermittent manual mixing. After boiling, the tubes were cooled and placed on ice, followed by addition of 0.5 mL water and vortexing for 1 minute. Tubes were centrifuged to achieve phase separation. The bottom chloroform layer was then transferred into a glass vial, dried over MgSO₄, and filtered through a 0.45 μm PTFE filter into a GC vial. Derivatized 3HAs were analyzed on an Agilent 7890B/5977A GC/MS with a VF-WAX column (30 m × 250 μm × 0.5 μm). The following method parameters were used: inlet temperature of 220°C, initial oven temperature of 80°C and a linear ramp rate of 10°C/min until final oven temperature of 220°C, with a 10:1 split ratio. An FID detector was used for quantification of methyl-3HB and methyl-3HH. Quantification of free acids, 3HH and 3HB, was performed by HPLC. One mL of culture was harvested at induction and at 72 hours post induction and centrifuged at maximum speed for 6 minutes. A sample of the supernatant was then run on an Aminex HP-87x (BioRad, Hercules, CA) column on an Agilent 1200 HPLC instrument equipped with an RID detector. 5 mM sulfuric acid was used as the mobile phase at 0.6 mL/min with column temperature held at 35°C.

Protein purification

Thiolase variants were subcloned into a protein expression vector pTev5 with an N-terminal hexa-histidine tag using CPEC cloning with primers listed in the Supplementary Information. E. coli BL21(DE3) was used as the host for protein expression. One liter of culture was grown in TB medium with glycerol at 30°C and induced with 100 μM IPTG when OD₆₀₀ was ~ 0.5. Cells were harvested 15–18 hours post-induction by centrifugation and resuspended in 2.5× vol/wt buffer containing 50 mM Tris-HCl pH 8.0, 500 mM NaCl and 10% vol/vol glycerol. Lysozyme was added to 1 mg/mL final concentration and cells were lysed by sonication. Protein purification was then performed as described previously (McMahon & Prather, 2014). After purification, proteins were exchanged into storage buffer (50 mM Tris pH 8.0, 50 mM NaCl and 10% vol/vol glycerol), flash frozen in small aliquots and stored at −80°C. Protein concentration was determined by a Bradford assay using BSA as standard. PhaB reductase from C. necator, which was used as a coupling enzyme in condensation assays, was purified in the same manner as described above.

Enzyme assays

Thiolase variants were assayed in both condensation and thiolysis directions. The condensation assay was performed akin to that described previously (Bond-Watts et al., 2011), except at pH 7.0 and coupled to PhaB reductase (from C. necator). Each reaction contained 100 mM Tris pH 7 buffer, 100 μg/mL NADPH, and varying amounts of acetyl-CoA, and reaction progress was monitored by a decrease in A₃₄₀ nm corresponding to NADPH consumption on a Beckman-Coulter DU800 spectrophotometer. Thiolases were also assayed in the thermodynamically favored thiolysis direction with acetoacetyl-CoA and 3-oxohexanoyl-CoA. Each assay contained 100 mM Tris pH 7.0, 10 mM MgCl₂, 200 μM CoASH, an appropriate amount of enzyme, and varying substrate concentration. A decrease in A₃₀₃, corresponding to consumption of the Mg-keto-acyl-CoA complex was measured spectrophotometrically. The extinction coefficient for acetoacetyl-CoA was determined to be 4.22 μM⁻¹ cm⁻¹ under the enzymatic conditions. Concentrations of all enzymes used in the assays were such that the reaction rate was linear for at least 0.5 minutes. Enzymes were diluted in pH 7 dilution buffer (100 mM Tris pH 7, 50 mM NaCl and 10% vol/vol glycerol). Each substrate concentration was assayed at least in duplicate. Generated concentration vs. initial rate curves were fit to the Michaelis-Menten equation, from which catalytic parameters (k_cat and K_m) were determined using the nlinfit routine in MATLAB.

Synthesis of 3-oxohexanoyl-CoA

The generalized synthesis is outlined in Scheme I in Supplementary Information and was inspired from the synthesis of ethylmalonyl-CoA by Dunn et al. (2014) and adapted by Blaisse et al. (2018). 3-oxo-hexanoic acid methyl ester was purchased from Alfa Aesar. 1 mmol of the ester was allowed to react with 1.2 mmol aqueous NaOH at room temperature overnight. The reaction was then neutralized to pH 7.0 and extracted three times with ethyl acetate, dried over MgSO₄ and solvent evaporated. 3-oxo-hexanoic acid appeared as a white solid. This crude solid was used in subsequent thioesterification with 1.2 mmol of thiophenol, 1.5 mmol diisopropylcarbodiimide and 2 mg of dimethylaminopyridine in 10 mL of ethyl acetate. The reaction was carried out on ice for 2 hours, followed by 2 hours at room temperature, after which the white precipitate was filtered off and the filtrate extracted with saturated sodium bicarbonate. The organic layer was then dried over MgSO₄ and solvent evaporated. Crude thiophenol-coupled product was then re-dissolved in 200 μL acetonitrile and added to 1 mL of 0.5 M NaHCO₃ on an ice bath. 25 mg of CoASH was added and the reaction allowed to proceed for 1 hour on ice and then 1 hour at room temperature. The reaction was quenched with 50% formic acid, and extracted with diethyl ether. Final 3-oxohexanoyl-CoA product was purified by HPLC with 25 mM ammonium acetate pH 4.5 and 20% acetonitrile in water as the mobile phases using a linear gradient from 1% v/v acetonitrile to 20% over 25 minutes on an Agilent Zorbax Eclipse XDB C18 column. Identity of the compound was verified by mass spectrometry. Finally, the purified product was desalted on the same column but with only water and acetonitrile as mobile phases.

Starting X-ray Structures

For PhbA calculations with butyryl-CoA and acetyl-CoA bound with C89 unacylated (“Bind 1”), 1M3Z (C89A mutant with acetyl-CoA bound) was used as the starting crystal structure with the C89 built into the structure using the same dihedral angles as the C89 of the unliganded WT PhbA structure 1DLU (Kursula et al., 2002; Modis & Wierenga, 2000). For PhbA calculations with C89 acylated (“Bind 2”), 1DM3 (WT enzyme with acetyl-CoA bound and C89 acetylated) was used as the starting structure (Modis & Wierenga, 2000), and for all BktB calculations, 4NZS (WT enzyme, unliganded) was used as the starting crystal structure (Kim et al., 2014). All crystal structures were prepared for computer modeling with the CHARMM36 force field (Brooks et al., 2009) using the methodology outlined in Lippow et al. (2007).

Computational Methodology

Computations for each of the two binding events sought to optimize $\mathrm{\Delta }\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B-\mathit{Bind}A}^{\mathit{Mut}-WT}=\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT}-\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}A}^{\mathit{Mut}-WT}$, where the subscript Bind B refers to the structure with a butyryl group in either the first or second binding event and the subscript Bind A refers to the corresponding structure with an acetyl group in either Bind 1 or Bind 2. Note that Bind B refers to the bound conformation leading to BA production in either step and Bind A refers to the bound conformation leading to AA production in either step. For example, the structure optimized for Bind B in the first binding event corresponds to step 2 of Figure 1C where R is a butyryl group, and the structure optimized for Bind A in the first binding event corresponds to step 2 of Figure 1C where R is an acetyl group.

Mutants were then sorted on $\mathrm{\Delta }\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B-\mathit{Bind}A}^{\mathit{Mut}-WT}$ and filtered with a fold cutoff of 15 kcal/mol in order to identify thiolase sequences with predicted differential selectivity as compared to wild type toward accommodating the butyryl group as opposed to the acetyl group. Sequences were then sorted on $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT}$ to allow identification of mutants for testing that were also predicted to accommodate the butyryl group more favorably than wild type, and not just resulting in improved differential specificity by accommodating both acetyl and butyryl groups more poorly than wild type, but with the acetyl worse than butyryl. Further details of the computational methodology are presented in the Supporting Information, including Supplementary Figures 3–4 and Supplementary Tables III–IV.

Results

Computational design of mutants predicted to exhibit increased selectivity in PhbA

The structure of Z. ramigera PhbA thiolase has been well studied, with crystal structures representing each step of the catalytic cycle (Meriläinen et al., 2009; Meriläinen et al., 2008; Modis & Wierenga, 2000; Modis & Wierenga, 1999; Kursula et al., 2002). Especially relevant to this study were the structures of the C89A mutant with acetyl-CoA bound (1M3Z), the wild-type thiolase with acetyl-CoA bound and C89 acetylated (1DM3), and unliganded wild-type thiolase with C89 butyrylated (1M4T), as these provided a basis for examining acyl group specificity at each binding event (Meriläinen et al., 2009; Meriläinen et al., 2008; Modis & Wierenga, 2000; Modis & Wierenga, 1999; Kursula et al., 2002). Due to this wealth of available crystallographic data, PhbA was chosen as the starting point for structure-based design calculations. No published BktB crystal structures existed at the start of this study.

In the case of the 3HA pathway, we were interested in improving the overall pathway selectivity ratio, i.e. the production of the longer chain C6, BA product relative to AA, using the nomenclature in Figure 2. At the thiolase level, this ratio could be improved either by increasing the formation of 3-oxohexanoyl-CoA (BA), by decreasing the formation of acetoacetyl-CoA (AA), or a combination of the two approaches. There are several possible steps in the thiolase catalytic cycle where BA production might be limited compared to AA production. For example, steric constraints might lead the thiolase active site to preferentially accommodate acetyl-CoA relative to butyryl-CoA during the priming CoA substrate binding step (“Bind 1” in Figure 1C). Similarly, steric constraints could also prevent the thiolase active site from accommodating butyrylated-C89 relative to acetylated-C89 in a conformation favorable to nucleophilic attack by the acetyl carbon of acetyl-CoA (“Bind 2” in Figure 1C). Effectively, the butyryl group must be accommodated in at least two orientations in the active site: on the bound priming butyryl-CoA, and on the butyrylated catalytic C89.

While it is possible that one of the catalytic steps (e.g. proton abstraction, breakdown of acyl-enzyme intermediate) might limit BA production, crystallographic studies by Kursula et al. (2002) suggest that butyrylation of C89 inhibits catalytically productive binding of the extending acetyl-CoA. Kursula et al. (2002) report that soaking experiments with butyryl-CoA and wild-type PhbA crystals result in butyrylation of C89 with no detectable CoA bound, indicating that butyryl-CoA is able to act as the priming acyl-CoA, but not as the extending acyl-CoA once the enzyme is butyrylated. We observe very low levels of 3HH formation in vivo with WT PhbA, suggesting poor PhbA activity with butyryl-CoA as the priming acyl-CoA and acetyl-CoA as the extending acyl-CoA. It should also be noted that on studies of ketosynthase domains in polyketide synthetases, which exhibit a similar bi bi ping-pong mechanism, it has been reported that the extending step is more often the bottleneck for acceptance of alternative substrates than the priming/acylation step (Jenner et al., 2015).

Superimposing the butyrylated C89 structure (which corresponds to step 4 of the catalytic cycle in Figure 1C; 1M4T) upon the acetylated structure with acetyl-CoA bound as the extending acyl-CoA (representing step 5 in Figure 1C; 1M3Z) reveals that the butyryl group of C89 lies directly over the sulfur atom of the bound extending acetyl-CoA, with the butyryl group pointing directly into the hydrophobic pocket formed by conserved active site residues M157 and M288, and the thioester oxygen pointing into the oxyanion hole formed by N(C89) and N(G380) (Kursula et al., 2002). Modis and Wierenga (1999) suggest that M157 and M288 in the PhbA active site prevent the accommodation of larger acyl-CoA substrates, although they did not test these observations experimentally (Modis & Wierenga, 1999). We sought to develop a model that would allow prediction of the ability of mutations at these positions, as well as additional positions to allow the butyrylated C89 to take on a conformation more favorable to catalytically productive binding of acetyl-CoA as the extending acyl-CoA.

Rather than building models of the transition state for each step of the condensation reaction and optimizing the active site binding of the transition state associated with each catalytic step leading to BA production, we chose to build on the published crystal structures of acetyl-CoA bound as the priming acyl-CoA (1M3Z) and acetyl-CoA bound as the extending acyl-CoA with C89 acetylated (1DM3). We assume these crystal structures represent catalytically productive binding modes at each step, and identify mutations that could accommodate a butyryl group in the appropriate place while keeping the rest of the crystal structure fixed outside of a defined radius (4.75 Å) of the residue to be mutated (see Computational Methodology).

Although poor binding affinity of the extending acetyl-CoA due to the native conformation of butyryl-C89 was likely the primary driver for poor BA production, it was nonetheless important to consider the effect of active site mutations on the ability to accommodate butyryl-CoA as the priming acyl-CoA. If a thiolase mutant was able to accommodate the butyryl group in Bind 2, but as a result of the mutation was unable to accommodate the butyryl group in Bind 1, then this would likely lead to poor BA production. Although they observe that butyryl-CoA is capable of acting as the priming acyl-CoA with WT PhbA, Kursula, et al. (2002) also report poor (mM) affinity of PhbA for butyryl-CoA. It was critical that designed mutants did not further decrease this affinity, or butyryl-CoA may no longer be capable of acting as the priming acyl-CoA.

We thus performed design calculations on conformations representing both Bind 1 and Bind 2. We focused structure-based design calculations on identifying mutations with the potential to improve the energy of bound conformations leading to BA relative to those leading to AA, at either the first or second Michaelis complex (steps 2 and 5 in Figure 1C, respectively). Calculations were performed as described in Methods, and Table I lists the PhbA mutants chosen for experimental testing along with their corresponding values of $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT},\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}A}^{\mathit{Mut}-WT}$, and $\mathrm{\Delta }\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B-\mathit{Bind}A}^{\mathit{Mut}-WT}$ for both Bind 1 and Bind 2.

Table I.

Energetic calculations of Z. ramigera PhbA mutants selected for experimental testing

Mutant	Bind 1(a)			Bind 2(b)
	$\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT}$	$\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}A}^{\mathit{Mut}-WT}$	$\mathrm{\Delta }\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B-A}^{\mathit{Mut}-WT}$	$\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT}$	$\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}A}^{\mathit{Mut}-WT}$	$\mathrm{\Delta }\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B-A}^{\mathit{Mut}-WT}$
L88S	−0.09	0.09	−0.18	−1.05	0.09	−1.14
L88A	0.02	0.09	−0.07	−1.33	0.09	−1.42
L88G	0.53	0.15	0.38	−1.90	−0.26	−1.64
M157S	−18.09	0.83	−18.92	−3.98	1.32	−5.30
M157A	−17.36	1.38	−18.74	−2.76	1.88	−4.64
M157G	−16.34	2.20	−18.54	−1.34	2.37	−3.71
M288S	1.60	1.18	0.43	−0.03	0.48	−0.52
M288A	1.85	1.25	0.60	1.55	0.57	0.98
M288G	2.29	1.34	0.95	1.08	0.65	0.43
L377S	−0.65	0.02	−0.68	0.50	0.02	0.47
L377A	−1.01	0.07	−1.09	−0.30	0.08	−0.37
L377G	−1.14	0.11	−1.26	0.35	0.12	0.22

^(a)In the Bind 1 column $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT}$ is the difference in binding energies between mutant and wild type bound to butyryl-CoA with free C89, $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}A}^{\mathit{Mut}-WT}$ is the difference in binding energies between mutant and wild type bound to acetyl-CoA with free C89, and $\mathrm{\Delta }\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B-A}^{\mathit{Mut}-WT}$ is the difference between $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT}$ and $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}A}^{\mathit{Mut}-WT}$ corresponding to the differential specificities for mutant versus wild type for binding butyryl-CoA versus acetyl-CoA as the priming acyl-CoA. For reference, the distances from the C_α of each residue (from the Bind 1 configuration) to the Cδ of acetyl-CoA are M157 (6.49 Å), M288 (7.68 Å), L88 (7.76 Å), L377 (7.64 Å)

^(b)In the Bind 2 column $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT}$ is the difference in binding energies between mutant and wild type bound to acetyl-CoA with butyrylated C89, $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}A}^{\mathit{Mut}-WT}$ is the difference in binding energies between mutant and wild type bound to acetyl-CoA with C89 acetylated, and $\mathrm{\Delta }\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B-A}^{\mathit{Mut}-WT}$ is the difference between $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT}$ and $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}A}^{\mathit{Mut}-WT}$ corresponding to the differential specificities for mutant versus wild type for binding acetyl-coA as the extending CoA with C89 either butyrylated or acetylated. Negative energies are highlighted with a green background, while positive energies are highlighted with a red background. All energies are reported in kcal/mol.

Over 300 PhbA mutations were computationally screened, but only the 12 variants listed in Table I were deemed worthy of experimental testing based on the computed values of $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT},\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}A}^{\mathit{Mut}-WT}$, and $\mathrm{\Delta }\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B-\mathit{Bind}A}^{\mathit{Mut}-WT}$. The vast majority of mutations computationally screened had positive values of $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT}$ or $\mathrm{\Delta }\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B-\mathit{Bind}A}^{\mathit{Mut}-WT}$ for both Bind 1 and Bind 2, and/or did not satisfy the fold cutoff. In the selected set of PhbA variants, all mutants involve paring down of a bulky hydrophobic (L88, M157, M288, L377) residue to a smaller residue, such as serine, alanine or glycine. Note that all mutants except M288A and M288G have negative values of $\mathrm{\Delta }\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B-\mathit{Bind}A}^{\mathit{Mut}-WT}$ in Bind 1, Bind 2 or both Bind 1 and Bind 2. All mutants chosen for experimental testing also have negative values of $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT}$ in either Bind 1 or Bind 2. All mutants also have positive values of $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}A}^{\mathit{Mut}-WT}$ in both steps, indicating decreased binding preference for accommodating for the acetyl group in both binding events.

Of all positions, M157 was judged the most promising candidate due to its negative values of $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT}$ in both binding events, the high magnitude of $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}A}^{\mathit{Mut}-WT}$ relative to the other mutants, and the fact that a similar trend was exhibited for the similar mutations of M157S/A/G. Because according to energetic calculations and upon inspection M288S appeared to be a promising candidate for improving selectivity in Bind 2 but not M288A and M288G, M288A and M288G were also chosen for testing to account for the possibility that the model might not be able to accurately distinguish the small chemical differences between serine, alanine and glycine. This position was also included because previous crystallographic studies of PhbA hypothesized that the bulky hydrophobic group of M288 (along with M157) prevents the accommodation of larger acyl-CoA substrates (Modis & Wierenga, 1999). Figures 3A–D show the location of the residues chosen for PhbA mutagenesis relative to the active site catalytic residues in both the Bind 1 and Bind 2 orientations.

Figure 3.

A Structure of Z. ramigera PhbA thiolase active site during the first binding event (Bind 1, corresponding to step 2 in Figure 1C). The atoms colored black show the extra atoms of the butyryl group compared to the acetyl group that must be accommodated in order to preferentially produce 3-oxohexanoyl-CoA rather than acetoacetyl-CoA.

B Structure of Z. ramigera PhbA thiolase active site during first binding event with residues selected for mutation colored purple.

C Structure of Z. ramigera PhbA thiolase active site during second binding event (Bind 2, corresponding to step 5 in Figure 1C). Atoms colored black show the extra atoms that must be accommodated in order to preferentially produce 3-oxohexanoyl-CoA.

D Structure of Z. ramigera PhbA thiolase active site during second binding event (corresponding to step 5 in Figure 1C) with residues selected for mutation colored purple.

Initial screening of PhbA mutants identifies several improved enzyme variants

PhbA mutants were initially assayed in vivo in the context of a previously established pathway for 3-hydroxyalkanoic acid (3HA) production (Martin et al., 2013). This pathway consists of an activator enzyme (Pct, M. elsdenii), a thiolase (BktB from C. necator or PhbA from Z. ramigera), an NADPH dependent reductase (PhaB from C. necator), and a thioesterase (TesB from E. coli), which generates the final 3HA product. Specifically, when the cultures are supplied with butyrate and grown on glucose, the cells produce 3HB and 3HH. Examining the amount of 3HH produced relative to 3HB provides a measure of thiolase selectivity.

Of the twelve tested thiolase variants, several resulted in increased selectivity ratios in vivo (Figure 4A). This higher selectivity ratio is mostly due to decreased production of 3HB by the pathway, and not increased 3HH titers (Figure 4B). Specifically, five mutants: M157A/G and M288S/A/G resulted in an approximately 30-fold higher ratio of 3HH relative to the undesired 3HB by-product, with a roughly 80-fold decrease in their sum. Motivated by these results, we wanted to further characterize the most selective mutants. Because the extent to which the enzymes downstream of the thiolase could affect final product distribution was unknown, we also wanted to assay the mutants within the context of another pathway. Thioesterases exhibit varying levels of activity towards different acyl-CoA substrates, depending on the carbon chain length and functional group of the substrate (McMahon & Prather, 2014), suggesting that the ratio of 3HH to 3HB observed could reflect differences in TesB specificity rather than PhbA. The PHA biosynthesis pathway was thus subsequently used to screen the thiolase mutants because it is known that over 100 different 3HA monomers can be incorporated into PHAs, suggesting a broad substrate range for the PHA synthase (Agnew and Pfleger, 2013). We chose to use the PhaC2 polymerase enzyme from R. aetherivorans I24 because it has been previously employed to synthesize PHA polymers with large amounts of the longer chain C6 monomer, 3-hydroxyhexanoate, 3HHx (Budde et al., 2011). Using the polymerase as the terminal enzyme removes any possible limitation or specificity imposed by the thioesterase, providing further evidence for thiolase imposed selectivity on the distribution of observed products.

Figure 4.

A Initial screening of Z. ramigera PhbA thiolase variants as designed by our computational method, using the previously established 3HA pathway, which results in production of free 3HAs which are detected in the supernatant. Cultures were grown in M9 minimal medium supplemented with glucose and 15 mM butyrate. Products were analyzed from cell-free culture supernatants 72 hours post induction by HPLC and ratios calculated on a molar basis.

B Final concentrations of 3HB and 3HH acids 72 hours post induction as analyzed by HPLC.

C Z. ramigera mutant thiolases profiled within the context of PHA biosynthesis. Ratios represent the composition of the PHA polymer as measured by GC after methanolysis.

When the most selective PhbA thiolase variants were profiled using the PHA assay, M157 mutants resulted in an 18-fold higher 3HHx:3HB selectivity ratio (Figure 4C). The resulting PHA polymers synthesized by M157A/G/S PhbA mutants contained 83–85 mol% of the 3HHx monomer, as compared to WT, which only resulted in a 22 mol% of the 3HHx monomer (Table III). To eliminate the possibility that native E. coli thiolases or reductases could influence final PHA composition, we performed several control experiments; no PHA accumulation was observed without plasmid-based overexpression of all four genes of the pathway (data not shown). This assay was consistent with our previous observations and supports our initial hypothesis of these mutant thiolases exhibiting reduced activity towards the condensation of two acetyl-CoAs, while maintaining similar or better activity towards the condensation of butyryl-CoA and acetyl-CoA compared to the WT enzyme.

Table III.

Composition of PHAs extracted from engineered E. coli strains overexpressing different thiolases and grown on glucose with fed butyrate.

Thiolase	CDW (g/L)	PHA Content (wt%)	Mol% 3HHx
Z. ramigera PhbA WT	0.52 ± 0.022	4.3 ± 0.41	22.6 ± 1.81
	1.26 ± 0.073	0.41 ± 0.046	83.9 ± 2.7
	0.72 ± 0.34		85.6 ± 2.15
	0.96 ± 0.42	0.51 ± 0.17	84.2 ± 4.02
C. necator BktB WT	1.04 ± 0.13	27.2 ± 1.2	77.3 ± 2.3
C. necator BktB M158A	0.81 ± 0.16	29.3 ± 1.4	93.6 ± 0.42
C. necator BktB M158G	0.71 ± 0.17	18.9 ± 3.55	97.3 ± 0.43
C. necator BktB M158S	0.88 ± 0.16	22.1 ± 4.08	97.3 ± 0.22
C. necator BktB M290A	0.65 ± 0.23	1.47 ± 0.90	81.7 ± 1.4
C. necator BktB M290G	0.54 ± 0.04	0.90 ± 0.05	86.7 ± 0.46

Validated computational approach applied to more active BktB thiolase

Having successfully identified mutants with increased C6/C4 (BA/AA) selectivity in PhbA, we applied the newly validated modeling framework to identify mutants that might increase selectivity of the more active C. necator BktB thiolase. Although BktB only exhibits 51% sequence identity with PhbA, the active site is highly similar, with 86% of the residues within 10 Å of the PhbA acetyl-CoA carbonyl center conserved between PhbA and BktB (Supplementary Table I). Two unliganded crystal structures were available for BktB (Kim et al., 2014; Fage et al., 2015), and due to the active-site similarity, the Z. ramigera PhbA structures,1M3Z and 1DM3 were used as templates to build structures of BktB with acetyl-CoA and butyryl-CoA bound.

The results of the computational model applied to BktB (analogously as to PhbA described above) are shown in Table II. Given the active-site similarity, it was not surprising that two BktB residues with analogous PhbA positions (M157/M158, M288/M290) were also predicted to improve BA/AA selectivity. Additionally, a position unique to BktB was predicted, Y66, which is part of a loop that comprises the major structural difference between the PhbA and BktB active sites. The positions of the BktB residues chosen for mutagenesis relative to the Bind 1 and Bind 2 conformations are shown in Figures 5A–D.

Table II.

Energetic calculations of C. necator BktB mutants selected for experimental testing

Mutant	Bind 1			Bind 2
	$\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT}$	$\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}A}^{\mathit{Mut}-WT}$	$\mathrm{\Delta }\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B-A}^{\mathit{Mut}-WT}$	$\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT}$	$\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}A}^{\mathit{Mut}-WT}$	$\mathrm{\Delta }\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B-A}^{\mathit{Mut}-WT}$
M158S	−2.18	1.27	−3.45	−1.33	1.03	−2.36
M158A	−1.86	1.58	−3.44	−1.54	1.49	−3.03
M158G	−1.54	1.61	−3.15	−0.16	1.99	−2.15
M290S	−21.81	0.45	−22.26	0.13	0.31	−0.18
M290A	−22.03	0.55	−22.59	0.18	0.42	−0.23
M290G	−21.52	0.64	−22.16	0.29	0.51	−0.22
Y66Q	−0.36	0.12	−0.48	−2.52	0.08	−2.60
Y66N	0.10	0.09	0.01	−2.49	0.12	−2.61
Y66V	0.26	0.12	0.14	−2.49	0.11	−2.60
Y66T	0.09	0.12	−0.04	−2.46	0.12	−2.58
Y66S	−0.32	0.11	−0.43	−2.48	0.13	−2.61
Y66A	0.08	0.12	−0.04	−2.44	0.14	−2.58
Y66G	0.18	0.12	0.05	−2.45	0.15	−2.60

^(a)In the Bind 1 column, $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT}$ is the difference in binding energies between mutant and wild type bound to butyryl-CoA with free C90, $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}A}^{\mathit{Mut}-WT}$ is the difference in binding energies between mutant and wild type bound to acetyl-CoA with C90 unacylated, and $\mathrm{\Delta }\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B-A}^{\mathit{Mut}-WT}$ is the difference between $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT}$ and $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}A}^{\mathit{Mut}-WT}$ corresponding to the differential specificities for mutant versus wild type for binding butyryl-CoA versus acetyl-CoA as the priming acyl-CoA. For reference, the distances from the C_α of each residue to the Cδ of acetyl-CoA (from the Bind 1 configuration) are Y66 (12.61 Å), M158 (6.36 Å), M290 (7.54 Å).

^(b)In the Bind 2 column

$\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT}$ is the difference in binding energies between mutant and wild type bound to acetyl-CoA with butyrylated C90, $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}A}^{\mathit{Mut}-WT}$ is the difference in binding energies between mutant and wild type bound to acetyl-CoA with acetylated C90, and $\mathrm{\Delta }\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B-A}^{\mathit{Mut}-WT}$ is the difference between $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT}$ and $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}A}^{\mathit{Mut}-WT}$ corresponding to the differential specificities for mutant versus wild type for binding acetyl-coA as the extending CoA with C90 either butyrylated or acetylated. Negative energies are highlighted with a green background, while positive energies are highlighted with a red background. All energies are reported in kcal/mol.

Figure 5.

A Structure of C. necator BktB thiolase active site during the first binding event (corresponding to step 2 in figure 1C). The atoms colored black show the extra atoms of the butyryl group that must be accommodated compared to the acetyl group in order to preferentially produce 3-oxohexanoyl-CoA rather than acetoacetyl-CoA.

B Structure of C. necator BktB thiolase active site during first binding event with residues selected for mutation colored purple.

C Structure of C. necator BktB thiolase active site during the second binding event (corresponding to step 5 in figure 1C). Atoms colored black show the extra atoms that must be accommodated in order to preferentially produce 3-oxohexanoyl-CoA.

D Structure of C. necator BktB thiolase active site during second binding event (corresponding to step 5 in figure 1C) with residues selected for mutation colored purple.

BktB thiolases enable synthesis of PHAs enriched in 3HHx

BktB has been previously used by us and other groups to achieve synthesis of longer (>C4) and branched chain acids, aldehydes and alcohols by the same CoA dependent pathway (Dhamankar et al., 2014; Tseng et al., 2009; Kim et al., 2015; Cheong et al., 2016). Of the M158, M290 and Y66 mutants assayed, the M158 mutants resulted in the highest selectivity ratios, with M158G and M158S exhibiting selectivity ratios 10-fold greater than WT for 3HHx in PHAs (Figure 6A). Based on previous reports of their activities, it was not surprising that WT baseline selectivity was higher for BktB at 3.45 compared to PhbA at 0.292 (Slater et al., 1998). The PHA polymers isolated from E. coli strains expressing these mutants varied from 77 to 97 mol% 3HHx (Table III), with BktB M158A mutant resulting in the highest yields of 3HHx as a percentage of the CDW (Figure 6B). Protein gels of lysates of strains expressing WT vs. mutant enzymes showed no significant difference in the soluble expression level of the thiolase enzymes, pointing to differences in the activities of these enzymes (Supplementary Figure 1). Surprisingly, the M290 mutants resulted in very low yields of PHAs in vivo. Although it is possible that BktB expression or solubility was affected as a result of this mutation, soluble expression was detected via a protein gel.

Figure 6.

A Computationally predicted mutant BktB thiolases assayed within the context of PHA biosynthesis. Cultures were grown in M9 minimal medium supplemented with glucose and 15 mM butyrate. Ratios represent the composition of the PHA polymer as measured by GC after methanolysis. P-values from a two-tailed t test were <0.0003 for M158G/S and M158A, M158A/G/S, M290A, and Y66Q relative to wild type enzyme. P-values for M290A, Y66Q relative to wild type, were 0.21<p<0.73, indicating no statistical significance.

B PHA content as a weight percentage of the CDW of E. coli overexpressing a given BktB thiolase variant. For 3HHx production, p-values from a two-tailed t test were <0.0001 for M158A, M290A, M158A, M158G/S, M158A and Y66Q relative to wild type. P-values for M158G and M158S relative to wild type were p=0.16, and p-values for the rest of the mutants relative to wild type were >0.22. For 3HB production, p<0.0001 for all mutants relative to wild type except M158G/S and M290A, for which 0.10<p<0.54, indicating no significant differences between those mutants for 3HB production.

C Overexpression of a trans-enoyl-CoA reductase (ter_Td) and reductase (PhaJ4b_Cn), in addition to a thiolase, and acetoacetyl-CoA reductase and PHA polymerase, allows for synthesis of C6 products solely from glucose. In this case, cultures were grown in M9 minimal medium without butyrate supplementation. Increased selectivity for longer chain products is achieved with BktB M158A in place of wild type BktB. A two tailed t-test was performed on 3HHx, and 3HB content for the two strains. For both 3HHx and 3HB, p=0.03; and p=0.04 for the total weight percent comparison. These samples are different at a significance level of 0.05.

In vitro characterization of BktB mutants with highest selectivity ratios

Having achieved increased selectivity ratios of the 3HHx:3HB in the PHA polymers with our computationally designed mutants, we next studied the effects of the M158 mutations on thiolase activity. Our in vivo data suggested that we were able to obtain increased selectivity ratios due to decreased activity of these mutant thiolases for the formation of the AA condensation product (and subsequently 3HB), and not due to increased activity towards the condensation of butyryl-CoA with acetyl-CoA, which results in the formation of 3-oxohexanoyl-CoA (and 3HH-CoA upon reduction). We thus sought to purify and assay both the mutant and WT BktB thiolases in vitro to remove many of the confounding variables present in vivo. For example, differences in stability of the enzymes as well as fluctuating pools of substrates and coenzymes could influence thiolase activity. Further, activities of the downstream enzymes could also influence the final product distribution. Each WT and mutant enzyme was purified as a His-tagged fusion protein to homogeneity and assayed in the condensation direction with acetyl-CoA, and thiolysis directions with acetoacetyl-CoA (AA) and 3-oxohexanoyl-CoA (BA). In vitro characterization of the BktB WT and M158A enzymes reveal a 10-fold lower catalytic activity of the mutant towards the condensation of two acetyl-CoA molecules (Table IV). This result is consistent with in vivo observations of reduced 3HB product formation which arises from the condensation of two acetyl-CoAs. From the in vitro kinetic parameters it can be concluded that the M158 mutants do indeed have lower catalytic efficiencies towards the formation and degradation of AA, the C4 product (1.52 × 10⁴ vs. 1.46 × 10³ M⁻¹sec⁻¹, WT vs mutant), whereas the thiolysis k_cat/K_m value towards the degradation of BA, the C6 product, is 3-fold higher as compared to WT (2.67×10⁵ vs. 9.82×10⁵, WT vs mutant, Table IV). In all, there is an 80-fold improvement in the selectivity ratio of the M158A thiolase as compared to WT. Further, activity measurements of the BktB M290A mutant revealed a very low k_cat for the condensation of two acetyl-CoAs consistent with in vivo observations (data not shown).

Table IV.

In vitro kinetic characterization of thiolase variants

Reaction	kcat (sec−1)	Km (μM)	kcat/Km (M−1sec−1)	C6/C4 Selectivity
BktB WT C4 Condensation	14.1	919	1.52 × 104	N/A
BktB M158A C4 Condensation	1.33	913	1.46 × 103
BktB WT C4 Thiolysis	148	17.5	8.45 × 106	0.032
BktB WT C6 Thiolysis	4.06	15.2	2.67 × 105
BktB M158A C4 Thiolysis	4.63	14.1	3.28 × 105	2.99
BktB M158A C6 Thiolysis	16.7	17	9.82 × 105

In vitro characterization of C. necator BktB wild type and mutant thiolases in the forward direction (condensation) with C4, and reverse direction (thiolysis) with both C4 and C6 substrates. Catalytic parameters were computed from fits to the Michaelis-Menten equation (Supplementary Figure 2).

Using C. necator BktB M158A mutant allows for biosynthesis of PHAs enriched in 3HHx

Having demonstrated increased selectivity for the BktB M158A mutant in vivo while supplying both butyrate and glucose, we next wanted to determine if this mutant could allow for more selective synthesis of longer chain products using glucose as the sole source of carbon. We sought to use the same model system as before, except that now we had to overexpress additional enzymes that would allow for conversion of 3HB-CoA to butyryl-CoA. Trans-enoyl-CoA reductase, Ter from Treponema denticola was cloned into the first MCS of pCDFDuet and enoyl-CoA hydratase, PhaJ4b from C. necator was cloned into an operon with PhaC2 generating pCDFDuet(ter_Td)-(phaC2-phaJ4). This vector, along with pETDuet(BktB WT or M158A)-(phaB) was used to transform E. coli MG1655(DE3) and the strain was grown in M9 medium with glucose as a sole carbon source. Figure 6C shows that the residual activity of BktB M158A towards the condensation of two acetyl-CoAs, perhaps combined with the native activity of E. coli AtoB, was sufficient to allow for formation of butyryl-CoA and subsequently 3HB -CoA. Using the BktB M158A mutant led to an almost 2-fold increase in selectivity for the 3HHx monomer as compared to using wild type BktB, though the overall yield of PHAs was low. However, we used an almost WT E. coli for all production experiments in this work, and it is likely that strain engineering to increase precursor supply and elimination of competing pathways will lead to increased product yields.

Discussion

In this work we present a rational design framework for increasing the thiolase selectivity ratio, defined as the ratio of C6 to C4 condensation products. We then apply this framework to two related biosynthetic thiolases, PhbA from Z. ramigera and BktB from C. necator. In vivo, we observe the synthesis of PHAs that are highly enriched for 3HHx (C6) when our rationally selected mutants are employed. In vitro characterization of one of the most selective mutants (M158A) revealed a 10-fold reduction in activity for formation and breakdown of the C4 product with uncompromised thiolysis activity toward the C6 substrate as compared to the wild type enzyme.

Although designed thiolase mutants exhibited nearly 10-fold improvements in the selectivity ratio, this increase was primarily driven by the reduced ability of the thiolase to synthesize AA (acetoacetyl-CoA, C4) and not improved ability to synthesize BA (3-oxohexanoyl-CoA, C6). The decreased synthesis of C4 products by the mutants we tested is consistent with in silico predictions, as all mutants tested exhibited positive computed values of $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}A}^{\mathit{Mut}-WT}$ for both Bind 1 and Bind 2. From the in vitro kinetic characterization of the BktB M158A mutant, the reduced rate of condensation of two acetyl-CoA substrates is consistent with reduced 3HB production within both pathway contexts (3HA and PHA).

The fact that all mutants tested failed to significantly increase 3HH titers was not consistent with in silico predictions however. With the exception of M288A/G, all mutants tested had negative computed values of $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathit{Bind}B}^{\mathit{Mut}-WT}$ for either Bind 1 or Bind 2, meaning that each of the tested mutants were expected to preferentially accommodate the butyryl group in either the first (PhbA L377S/G and BktB M290S/A/G), second (PhbA L88A/G, PhbA M288S, BktB Y66N/V/T/A/G) or both (PhbA L88S, PhbA M157S/A/G, PhbA L377A, BktB M158S/A/G, and Y66Q/S) binding events.

It remains possible that activities of downstream enzymes on the longer chain substrates limited 3HH production in vivo. Activities of all downstream enzymes (3-ketoacyl-CoA reductase and/or thioesterase and PHA polymerase) with the pathway acyl-CoA intermediates must be examined to rule this out. This is challenging due to the commercial unavailability of required substrates as well as lack of robust assays as is the case with PHA polymerase, in which the observed in vitro and in vivo substrate specificities differ (Stubbe & Tian, 2003; Yuan et al., 2001).

In vitro, we were unable to directly assay the rate of condensation between acetyl-CoA and butyryl-CoA to test whether mutants exhibited increased production of 3-oxohexanoyl-CoA (and subsequently 3-HH-CoA) in the absence of any potential confounding factors in vivo. While we can assay the thiolase in the thermodynamically favored direction, thiolysis, and observe higher activity of the M158A mutant with the C6 substrate, we cannot conclude that a similar rate enhancement results for the forward condensation direction. One might expect the observed increase in the C6 thiolysis rate to lead to decreased overall titers in the biosynthetic direction; however, one must keep in mind that the reductase which is present in both in vivo and in vitro contexts is necessary to allow for formation and detection of condensation products. For example, when PHB synthesis was modeled in vitro, inclusion of the reductase enzyme was necessary to observe accumulation of the 3-ketoacyl-CoA condensation product (Burns et al. 2007). Put another way, both in vivo and in vitro, the thiolase must always be coupled to the reductase and the substrate specificity and activity of the reductase enzyme will influence the behavior of the overall system. Indeed, a similar approach has been used to model the kinetics of in vivo PHB accumulation (Leaf and Srienc, 1998; van Wegen et al, 2001). For this reason, the system must necessarily be examined while considering, at a minimum, the thiolase and reductase enzymes in combination.

During preparation of this manuscript, we note that Torres-Salas et al. (2018) reported a structure-guided design of thiolase from S. cerevisiae towards the production of branched chain products. Our findings in this report support their results, as several of the mutations they test were also predicted by our computational methods. One difference between our approaches is that we focused on the two distinct binding events of the thiolase reaction, and optimized both towards the formation of linear, rather than branched products. In addition, the mutation M158A was not identified in their report. Blaisse et al. (2018), have recently solved two crystal structures of A. suum thiolases, which share several active site residues with the thiolases described in this study. In particular, the structural studies of Blaisse et al. (2018) report that M158 exhibits high fluxionality and plays a key role in the “covering loop” governing accommodation of α-substituted substrates in the thiolase active site. Together, these recent reports should provide other research groups access to greater chemical diversity in the field of metabolic engineering and synthetic biology.

From a metabolic engineering standpoint, the thiolase mutants identified in this study should be useful in other pathways where the condensation of acetyl-CoA and different acyl-CoA species is required (Sheppard et al., 2014; Cheong, Clomburg, and Gonzalez, 2016). In addition, we have shown that the thiolase can be used to modulate PHA polymer composition, resulting in PHAs that are highly enriched for medium-chain length monomers. Typically, PHA composition is modulated by process engineering such as novel feeding strategies and choice of feedstock, as well as various strain engineering strategies to remove endogenous competing enzymes from native PHA synthesizing microbes. Using the thiolase to control PHA monomer composition opens up a new avenue for achieving the synthesis of PHAs with specific, desired properties for diverse applications.