Metabolic engineering of β-oxidation to leverage thioesterases for production of 2-heptanone, 2-nonanone and 2-undecanone

Abstract

Medium-chain length methyl ketones are potential blending fuels due to their cetane numbers and low melting temperatures. Biomanufacturing offers the potential to produce these molecules from renewable resources such as lignocellulosic biomass. In this work, we designed and tested metabolic pathways in Escherichia coli to specifically produce 2-heptanone, 2-nonanone and 2-undecanone. We achieved substantial production of each ketone by introducing chain-length specific acyl-ACP thioesterases, blocking the β-oxidation cycle at an advantageous reaction, and introducing active β-ketoacyl-CoA thioesterases. Using a bioprospecting approach, we identified fifteen homologs of E. coli β-ketoacyl-CoA thioesterase (FadM) and evaluated the in vivo activity of each against various chain length substrates. The FadM variant from Providencia sneebia produced the most 2-heptanone, 2-nonanone, and 2-undecanone, suggesting it has the highest activity on the corresponding β-ketoacyl-CoA substrates. We tested enzyme variants, including acyl-CoA oxidases, thiolases, and bi-functional 3-hydroxyacyl-CoA dehydratases to maximize conversion of fatty acids to β-keto acyl-CoAs for 2-heptanone, 2-nonanone, and 2-undecanone production. In order to address the issue of product loss during fermentation, we applied a 20% (v/v) dodecane layer in the bioreactor and built an external water cooling condenser connecting to the bioreactor heat-transferring condenser coupling to the condenser. Using these modifications, we were able to generate up to 4.4 g/L total medium-chain length methyl ketones.

Graphical Abstract

1. Introduction

Methyl ketones (MK) are naturally occurring compounds that have been shown to serve biological and industrial functions such as insecticides, pheromones, flavors, and fragrances, e.g. blue cheese flavor and fragrance comes from 2-heptanone and 2-nonanone produced by fungus in the maturation process (Collins et al., 2003). Similarly, 2-heptanone, 2-nonanone, 2-undecanone are found in plants, such as cinnamon, clove and coconut, where they contribute to the insecticide, flavor and fragrance profile (Forney and Markovetz, 1971; Kennedy, 2003; Zhu et al., 2018). Methyl ketones also have the potential to serve as renewable liquid transportation fuels. Medium-chain methyl-ketones, such as 2-nonanone and 2-undecanone, have high cetane numbers and low freezing points, making them compatible with diesel fuels (Table 1). In addition, these ketones and 2-heptanone can be condensed to yield long-chain, high cetane fuels (Frias et al., 2011; Harrison and Harvey, 2018).

Table 1.

Properties of methyl ketone and its derivatives as potential diesel fuel applications.

	Cetane Numbera	Melting Temperatureb (°C )	Citations
2-heptanone	30.0	−35.5	(Yanowitz et al., 2017)
2-nonanone	46.1	−15	(Yanowitz et al., 2017)
2-undecanone	56.5	15	(Goh et al., 2012)
2-tridecanone	60	30.5	(Harrison and Harvey, 2018)
dioxolanesc	81-91	−11	(Harrison and Harvey, 2018)
Palm oil derived biodieseld	61.9	13	(Hoekman et al., 2012)
No. 2 petroleum diesele	45	−18	(Hoekman et al., 2012)

^aLiterature reported cetane values that are either characterized experimentally or predicted by models; the minimum cetane number of diesel fuels in the U.S. is 40 according to ASTM standard D975.

^bMelting temperatures are obtained from a chemical information collection, PubChem (https://pubchem.ncbi.nlm.nih.gov/). In general, the freezing point trend is proportional to the melting temperature trend since both are measuring the required temperature of solid-liquid transition.

^cThe dioxolanes properties reported from the literature exemplified methyl ketones can be condensed to higher energy molecules.

^dpalm oil derived biodiesel is an example of commonly used biodiesel.

^eNo. 2 petroleum diesel is an example of commercial diesel fuels.

Methyl ketones are biologically produced from β-keto acids, unstable compounds that can spontaneously undergo decarboxylation under mild conditions (Kornberg et al., 1948). The medium chain β-keto acids, e.g. 3-oxooctanoate, are relatively less stable than shorter chain compounds, e.g. acetoacetate, possibly because the pentanyl moiety of 3-oxooctanoate is more favorable for attracting a proton to form a cyclic six-atom transition state of the decarboxylation process than methyl moiety of acetoacetate. This could explain why acetoacetate decarboxylase is an essential and rate-limiting step in acetone (Lan et al., 2013) or 2-butanone (Mehrer et al., 2019; Srirangan et al., 2016) biosynthesis while methyl ketone decarboxylase was shown to be unnecessary for 2-tridecanone biosynthesis (Goh et al., 2012) .

β-keto acids are made via transthioesterification or hydrolysis reactions on β-ketoacyl thioester (-CoA, -ACP) substrates. For example, methyl ketone synthases act either as distinct enzymes (^ShMks1/2) or as an embedded domain in modular type I polyketide synthases (PKS) by hydrolyzing β-ketoacyl-acyl carrier protein (ACP) substrates (Yu et al., 2010; Yuzawa et al., 2017). Similarly, β-ketoacyl-CoA thioesterases produce β-keto acids via hydrolysis. The long-chain β-keto acids spontaneously decarboxylate to yield methyl ketones (Goh et al., 2014; Nie et al., 2008). Alternatively, β-ketoacyl-CoA transferase (PcalJ) transfers the CoA moiety from short-chain substrates to succinate to form succinyl-CoA and the corresponding β-keto acid. The latter is enzymatically converted to a methyl ketone by acetoacetate decarboxylase (Adc) (Lan et al., 2013).

Over the last decade, several groups have applied metabolic engineering strategies to bacteria and yeast with the goal of increasing flux to desired methyl ketones. For instance, Zhu et al., built a chimeric Saccharomyces cerevisiae type I fatty acid synthase (FAS) by embedding a ^ShMks2 next to an ACP domain to increase the cleavage activities. An engineered strain that co-expressed ^ShMks1 and the FAS genes produced ~20 μg/g DCW C₁₁-C₁₅ methyl ketones (Zhu et al., 2017). Using Escherichia coli as a host, Goh et al., compared pathway efficiencies between the acyl-CoA methyl ketone synthase ^EcFadM and the acyl-ACP methyl ketone synthase ^ShMKS1 and ^ShMKS2. The best strain (EGS895) overexpressed β-ketoacyl-CoA thioesterase (^EcFadM), acyl-CoA oxidase (Mlut_11700), acyl-ACP thioesterase (^Ec’TesA), and a bifunctional hydratase and dehydrogenase (^EcFadB). This strain produced significantly higher 2-tridecanone than that of the ^ShMKS2 overexpression strain (Goh et al., 2012). In a follow-up study, the authors increased the C₁₁-C₁₅ methyl ketone titer up to 3.4 g/L in a fed-batch fermentation by overexpressing native FadR and FadD, deleting key acetate forming pathways, and consolidating the pathway from two plasmids into one (Goh et al., 2014). In addition, the authors further improved the methyl ketone titer up to 5.4 g/L by overexpression of NADH-dependent β-ketoacyl-ACP reductase FabG from Acholeplasma laidlawii to mitigate the NADPH imbalance caused by the endogenous FabG (Goh et al., 2018). Thus far, the FadM pathway is the most frequently used pathway for methyl ketone production due to its high activity. Recent efforts have focused on transferring methyl ketone pathways into non-model organisms such as oleaginous yeast, Yarrowia lipolytica (Hanko et al., 2018), soil bacteria, Pseudomonas putida (Dong et al., 2019) and chemolithoautotrophs, Ralstonia eutropha (Muller et al., 2013). Other studies designed metabolic pathways to make shorter chain-length methyl ketones. For instance, Lan et al., constructed an E. coli strain that overexpressed Ralstonia eutropha thiolase BktB and Pseudomonas putida 3-oxoadipate CoA-succinyl transferase PcaIJ. The strain produced ~240 mg/L 2-pentanone from ~12 g/L glucose (Lan et al., 2013). Yuzawa et al., engineered a type I modular polyketide synthase (PKS) to produce short chain methyl ketones by inactivating the ketone reductase (KR) domain and switching to a malonyl-CoA preferred acyltransferase homologue domain. When the engineered PKS was overexpressed in Streptomyces albus, the culture generated ~140 mg/L 2-butanone and 2-pentanone (Yuzawa et al., 2018). Although various protein engineering strategies have been applied to control chain-length specificities for oleochemicals such as fatty acids (Yan and Pfleger, 2019), few studies have been reported for the production of 2-heptanone, 2-nonanone, or 2-undecanone (Park et al., 2012; Zhu et al., 2019).

In this manuscript, we describe efforts to engineer E. coli to produce medium-chain length methyl ketones. Our work leverages highly-active and selective thioesterases that target 8- and 12- carbon substrates as well as two routes of converting fatty acids to the β-keto acid precursors of our target products. In prior work, we isolated a C₈-specific Cuphea palustris FatB1 thioesterase variant, annotated as CpFatB1.2-M4-287, hereafter CpFatB1*, that exhibited a 15-fold increase in k_cat compared to the native variant while maintaining >90% selectivity towards producing octanoic acid (Hernández Lozada et al., 2018). An engineered E. coli strain that overexpressed the CpFatB1* from a single chromosomal copy produced 1.7 g/L octanoic acid from 20 g/L glycerol. In other studies, we leveraged the C₁₂-specific thioesterase from Umbellularia californica (BTE) to produce dodecanoic acid and other 12-carbon oleochemicals at ~g/L titers (Lennen et al., 2010; Agnew et al., 2012; Youngquist et al., 2013). Here, we used the selectivity of these thioesterases and permissive acyl-CoA synthetases to generate pools of acyl-CoAs with narrow chain-length distributions. We then compared a thiolase-mediated condensation and a β-oxidation pathway (Figure 1) for generating β-ketoacyl-CoAs with similar chain-length distributions. In order to maximize conversion to methyl ketones, we compared fifteen homologs of the E. coli methyl ketone synthase FadM in search of variants with higher activities to medium chain substrates. Lastly, we used a condenser and dodecane overlay to capture volatilized methyl ketones from a bioreactor, thereby achieving the highest reported 2-heptanone titer in a fed-batch fermentation.

Figure 1.

Metabolic pathways designed for producing 2-heptanone, 2-nonanone and 2-undecanone in E. coli. Each pathway starts with a C₈- or C₁₂-specific thioesterases (CpFatB1* or BTE) which makes octanoate or dodecanoate from fatty acid biosynthesis. Free fatty acids are reactivated to octanoyl-CoA or dodecanoyl-CoA by fatty acid synthetases (^MtFadD6 or ^EcFadD), respectively. Octanoyl-CoA or dodecanoyl-CoA undergoes an oxidation process by an acyl-CoA dehydrogenase (^EcFadE), and a bi-functional enoyl-CoA hydratase/3-hydroxyacyl-CoA dehydrogenase ^EcFadBJ to generate β-keto-octanoyl-CoA or β-keto-dodecanoyl-CoA (pathways as marked in black). Alternatively, octanoyl-CoA and acetyl-CoA are condensed to generate β-ketodecanoyl-CoA by a thiolase ^EcFadA (marked in blue). The β-ketoacyl-CoAs are converted to β-keto acids by a methyl ketone synthase ^EcFadM (marked in red), then spontaneously decarboxylated to methyl ketones.

2. Materials and Methods

2.1. Bacterial strains, plasmids, oligonucleotides, and reagents

All bacterial strains used in this study are listed in Supplementary Table 1. Phusion DNA Polymerase was purchased from New England Biolabs (Ipswich, MA). Plasmid extractions were performed with QIAGEN (Valencia, CA) miniprep reagents. DNA, including oligonucleotide primers and short gBlocks, was synthesized by Integrated DNA Technologies (IDT), Inc. (San Diego, CA). Fatty acids and methyl ketone standards used in this study were purchased from Sigma-Aldrich (St. Louis, MO).

E. coli DH5α cells were used for cloning and assembling DNA molecules. Lysogeny broth (LB) was used to cultivate cells during the cloning and DNA assembly process. E. coli NHL17 (K12 MG1655 ΔaraBAD ΔfadD::trcCpFatB1*), E. coli ΔRABIJ (MG1655 ΔaraBAD ΔfadR ΔfadA ΔfadB ΔfadI ΔfadJ ) and E. coli TY34 (MG1655 ΔaraBAD ΔfadE::P_trc-BTE ΔfadAB::P_trc-BTE ΔackApta::P_trc-BTEΦ(P_trc-fadD)) were created as part of prior studies (Agnew et al., 2012; Hernández Lozada et al., 2018; Youngquist et al., 2013) and were used as base strains for production of 2-heptanone (NHL17), 2-nonanone (ΔRABIJ) and 2-undecanone (TY34). Pre-cultures were prepared in test tubes containing 5 mL LB and the appropriate antibiotics (final concentrations: carbenicillin, 100 μg/mL; chloramphenicol, 34 μg/mL; kanamycin 50 μg/mL).

2.2. Plasmid and strain construction

All plasmids used in this study are summarized in Supplementary Table 1. Plasmid maps are available as Supplementary Materials. Cloning of native E. coli genes into expression plasmids was performed by PCR amplification using the primers listed in the supplementary plasmid maps. Plasmids were constructed by annealing linearized DNAs using an isothermal assembly method (Gibson et al., 2009). Chromosomal expression cassettes were created using a combination of lambda red recombination and CRISPR/Cas9-mediated selection as described in prior work (Hernández Lozada et al., 2018; Mehrer et al., 2018). All cloned sequences, integration cassettes, and gene deletions were confirmed by Sanger sequencing performed by Functional Biosciences (Madison, WI).

Sequences of FadM homologs were acquired from the Thyme database (http://www.enzyme.cbirc.iastate.edu/) (Cantu et al., 2011). Protein similarities were performed using the BLAST alignment tool from NCBI. Terminal signal peptide sequences were identified using SignalP 3.0 (http://www.cbs.dtu.dk/services/SignalP-3.0/) (Dyrløv Bendtsen et al., 2004) and were subsequently truncated when genes were designed for synthesis. Constructs expressing FadA and FadB homologs were obtained from a prior study (Mehrer et al., 2018). Genes encoding FadM homologs, ^ReBktB and ^MlACO were codon-optimized and DNA sequences were chemically synthesized as gBlocks (IDT DNA).

2.3. Characterization of FadM activity in vivo

The relative activity of FadM homologs on various chain length substrates was measured by feeding free fatty acids to cultures of E. coli cells expressing a FadM homolog. These cultures were inoculated to starting OD₆₀₀ of 0.05 in 50 mL of LB supplemented with a defined amount of saturated fatty acid. All fatty acids were pre-dissolved in ethanol and supplemented to cultures at the following final concentrations: 1 g/L octanoic acid, 1 g/L decanoic acid, 200 mg/L dodecanoic acid, 100 mg/L tetradecanoic acid, or 100 mg/L hexadecenoic acid. Cultures were then grown at 37°C with shaking at 250 rpm until an OD₆₀₀ of approximately 0.2 was reached. At this point, cultures were supplemented with a final concentration of 1 mM IPTG and cultivated at 30°C with shaking at 250 rpm for an additional 24 h. After induction, samples were taken periodically to measure titers of fatty acids and methyl ketones.

2.4. Methyl ketone production in shake flasks

To demonstrate methyl ketone production in shake flasks, overnight cultures of E. coli strains (Supplementary Table 1) were inoculated to a starting OD₆₀₀ of 0.05 in Clomburg medium (Clomburg et al., 2012) with 10% (v/v) dodecane and appropriate antibiotics and incubated at 37°C with shaking at 250 rpm. Once the cultures reached an OD₆₀₀ of approximately 0.2, cultures were supplemented to a final concentration of 1 mM IPTG and incubated for an additional 24 hours at 30°C with shaking at 250 rpm.

2.5. Extraction and Quantification of methyl ketones and fatty acids

Fatty acids and methyl ketones were extracted from culture according to an acid-based esterification method described previously (Grisewood et al., 2017; Hernández Lozada et al., 2018). Individual species were separated using a Shimadzu GC equipped with an Agilent RTX-5 column (Santa Clara, CA) and quantified by FID. Quantification of fatty acids was normalized by inclusion of 50 μL of 12.5 mg/mL nonanoic acid, and/or 1.25 mg/mL pentadecanoic acid internal standards dissolved in ethanol. Octanoic acid, decanoic acid and dodecanoic acid were quantified based on the nonanoic acid internal standard, and tetradecanoic acid and hexadecenoic acid were quantified based on the pentadecanoic acid. Quantification of methyl ketones was normalized by inclusion of 50 μL of 62.5 mg/mL 2-octanone and/or 62.5 mg/mL 2-dodecanone dissolved in ethanol. 2-heptanone and 2-nonanone were quantified by 2-octanone, and 2-undecanone and longer chain methyl ketones were quantified by 2-dodecanone. In order to evaluate methyl ketone concentration in the distinct organic or aqueous phases, 50 mL cell cultures were centrifuged at 4500 × g for 10 min and 0.5 mL samples from the dodecane layer and 2.5 mL samples aqueous phase were collected and evaluated separately.

To test for the rates of methyl ketone evaporation in our culturing apparatus, a final concentration of ~5.0 g/L 2-heptanone, 3.5 g/L 2-nonanone, and 6 g/L 2-undecanone was dissolved in 50 mL LB medium in 250 mL shake flasks. The concentration of each species was monitored for 48-hours during which the flasks were incubated at 30°C with shaking at 250 rpm. In order to evaluate the effect of dodecane for reducing methyl ketone loss, 10 - 20% (v/v) dodecane was added to flasks and the 2-heptanone, 2-nonanone and 2-undecanone titers were measured every 12 hrs. MK standards (manufacturer) were run at the following concentrations to generate a peak area-based standard curve: 1000 mg/L, 500 mg/L, 250 mg/L, 100 mg/L, 50 mg/L, 25 mg/L.

2.6. Fed-batch bioreactor fermentation

Fed-batch cultivation was performed using a 1-L Infors Multifors bioreactor. Overnight pre-cultures were inoculated to an initial OD₆₀₀ of 0.05 into a bioreactor containing 500 mL Clomburg medium with ~65 g/L glycerol. The bioreactor was operated at the following conditions: temperature was controlled at 30°C at post-induction, air flow was 1.5 L/min, stirrer rate was varied between 250 rpm and 1000 rpm to control dissolved oxygen at a value of 40%, pH was maintained at 7.0 using 3 M phosphoric acid and pure ammonia hydroxide. When the OD₆₀₀ reached 0.2, IPTG was added to achieve a final concentration of 1 mM and 100 mL dodecane was fed in the bioreactor. At 24 h of post-induction, ~500 g/L glycerol was one-time bolus-fed into the bioreactor and fermentation terminated 96 h post-induction. Measurements of methyl ketone, fatty acid, glycerol, optical density, and CO₂ evolution were recorded for 96 hrs total.

The bioreactor outlet gas stream was directed through a heat exchange system to condense methyl ketone vapors stripped from the culture broth. The condenser was constructed with coiled quarter-inch outer diameter stainless steel tubing to ensure sufficient residence time in a water-ice bath. A water chiller supplied 6°C to the outer shell of the condenser. The condensate was channeled to a bored-through union to facilitate the collection of the liquefied sample from the gaseous effluent.

2.7. Analysis of glycerol consumption

Glycerol was quantified on an HPLC (Shimadzu) equipped with an autosampler, quaternary pump, degasser and a refractive index detector. 1 mL fermentation broth samples were prepared by centrifuging at 15,000 × g, and filtering the supernatant through a 0.22 μm membrane filter. For each sample, 10 μL was injected and separated for 25 min on a Restek Organic Acids column with a mobile phase of 5 mM H₂SO₄ at a flow rate of 0.6 mL/min. Glycerol standards (manufacturer) were run at the following concentrations to generate a peak area-based standard curve: 100 g/L, 50 g/L, 25 g/L, 10 g/L, 5 g/L, 1 g/L.

3. Results and discussion

3.1. Bio-prospecting for methyl ketone synthases

E. coli FadM had significantly lower activity on C₈-C₁₂ saturated acyl-CoAs than C₁₄-C₁₆ acyl-CoAs in in vitro kinetic studies using saturated, unsaturated, and hydroxylated acyl-CoA substrates, (Nie et al., 2008). In our study, we targeted production of medium-chain methyl-ketones and therefore started with a bioprospecting study to identify FadM homologs with high activity towards substrates (acyl-CoA or acyl-ACP) with twelve or less carbons. We conducted a homology search with the Basic Local Alignment Search Tool to identify candidate protein sequences. The search generated a wide range of biological diversity that we sorted using the Enzyme Similarity Tool. We selected fifteen FadM homologs which had a protein sequence similarity range of 30-95% based on a pairwise comparison of each thioesterase, shown in Figure S1A. We were particularly interested in homologs from Salmonella enterica (SeFadM) and Pseudomonas aeruginosa (PaFadM) because these species were known to have strong β-oxidation activities (Chakrabarty et al., 1973; Iram and Cronan, 2006) and have been sources of active enzymes used in prior studies (Mehrer et al., 2018). Mycobacterium tuberculosis possessed several type III thioesterases that have been previously shown activities towards acyl-CoAs (Wang et al., 2007) and five different homologs (Mt1-Mt5 FadM) were found that shares 26-58% similarities to the ^EcFadM Figure S1B.

The activity of the FadM homologs was assayed in vivo by feeding fatty acids of varied chain length to E. coli ΔfadAIR + pTRC99a-FadD6-”FadM”, and monitoring fatty acid consumption and methyl ketone production, Figure 2. ^MtFadD6 was overexpressed to increase the acyl-CoA synthetase activity required to activate the exogenous fatty acids fed to cells. ^MtFadD6 has been shown to act on medium-chain free fatty acids with greater activity than native FadD from E. coli (Youngquist et al., 2013, Herneandez-Lozada et al., Under Review). The ΔfadR deletion was included to deregulate expression of FadE and FadB, which are responsible for the conversion of saturated acyl-CoA to the corresponding 3-ketoacyl-CoA. The ΔfadAI deletions were included to block the thiolase reaction that would otherwise lead to the shortening of the 3-ketoacyl-CoA. Given the metabolic engineering strategy used, any fatty acid consumption would be coupled to production of the corresponding methyl ketone. In Figure 2, we report both the fatty acid consumption rate (timecourses available in Figure S2) and the measured titer of methyl ketones because many medium-chain methyl ketones are volatile and lost from the headspace of the flask. In general, the trend of fatty acid consumption rate between enzyme variants tracked with the trend of methyl ketone production (Figure 2). We found that Providencia sneebia FadM showed the highest activities toward medium-chain substrates and produced ~50 mg/L 2-heptanone, ~43 mg/L 2-nonanone, and 50 mg/L 2-undecanone. In general, the P. sneebia FadM showed high activity against each chain length tested, within error of the best enzymes producing pentadecanone and heptadecanone (Figure 2).

Figure 2.

(A) Scheme describing bioconversion of free fatty acids to methyl ketones. FadM variants were expressed from pTRC99a-”FadM”-FadD6 in strain E. coli ΔRAI. (B) FadM activity was tested by feeding even chain-length saturated fatty acids to the culture and measuring the concentration of methyl ketones and free fatty acids over 24 hours. The MK titer is listed in the left column of each pair. A fatty acid consumption rate was calculated from the timecourses presented in Figure S2. Cells were fed 1 g/L octanoic acid, 1 g/L decanoic acid 200 mg/L dodecanoic acid, 100 mg/L tetradecanoic acid, or 100 mg/L hexadecanoic acid. The listed error is the standard deviation of measurements taken from triplicate flasks.

3.2. Metabolic engineering to improve 2-heptanone production

After identifying promising FadM variants, we next developed a strain to produce 2-heptanone utilizing the modified β-oxidation pathway outlined in Figure 3A and a previously characterized acyl-ACP thioesterase with strong activity towards octanoyl-ACP. The base strain for this study was E. coli NHL17 (MG1655 ΔaraBAD ΔfadD::trc-CpFatB1*) which has been shown to produce up to 1.7 g/L of octanoic acid (Hernández Lozada et al., 2018). To improve the flux towards 2-heptanone, fadA, fadI, and fadR were deleted from NHL17 to generate the strain TRS12. These deletions were included for the same reasons described above - blocking thiolase reactions and upregulating FadE and FadB. The FadM activity in TRS12 solely came from the chromosomal expression of the native ^EcFadM. This strain generated low titers, < 1 mg/L, of 2-heptanone (data not shown). To improve production further, we overexpressed ^EcFadM from the high copy plasmid, pTRC99a-FadD6-EcFadM. This strain generated a titer of 2.4 mg/L of 2-heptanone (Figure 3). In order to increase the flux from acyl-CoA to enoyl-CoA we overexpressed ^EcFadE. This modification led to a ten-fold increase in 2-heptanone concentration, to a final titer of 21.9 mg/L (Figure 3). We next replaced ^EcFadM with ^PsFadM, the most active variant from our bioprospecting study, to make the expression plasmid, pTRC99a-FadD6-PsFadM. When co-expressed with ^EcFadE, the strain generated a final titer of 78.4 mg/L 2-heptanone (Figure 3). Lastly, ^EcFadE was replaced with the acyl-CoA oxidase from M. luteus (Goh et al. 2012), and was co-expressed from the pACYC-Mlut_11700 plasmid alongside pTRC99a-FadD6-PsFadM. Replacing the membrane protein ^EcFadE with the soluble heterologous acyl-CoA oxidase increased the 2-heptanone titer to 215.2 mg/L (Figure 3) with 97% selectivity. We suspect that this titer is underestimated, because when we incubated a defined 5 g/L 2-heptanone/water solution with 10% dodecane (v/v) in the same shake flasks, we observed that ~70% of the 2-heptanone was lost (Figure S3A) after 24 hours.

Figure 3.

Scheme of genetic design to improve 2-heptanone titer (A), genetic modifications based on strain NHL17 were marked in red, resulted in strain TRS12 (NHL17, ΔfadAEIR) expressing pTRC99a-fadD6-fadM and pACYC-Mlut_11700. Stepwise improvement of 2-heptanone production from engineered E. coli NHL17 strains from 20 g/L glycerol in Clomburg medium (B). “WT” refers to native expression of the gene, “+” represents an exogenous gene is overexpressed in a plasmid, “++” represents an endogenous gene has a chromosomal copy and was overexpressed in a plasmid, “−” represents a heterologous gene that was not added, Δ represents a native gene that was removed from the chromosome.

3.3. Coupling bioreactor with condenser to reduce 2-heptanone loss

To benchmark methyl ketone production, we conducted fed-batch cultivations by periodically feeding glycerol boluses to cells grown in a stirred bioreactor. We applied two strategies to reduce the loss of 2-heptanone in the off-gas: adding 20% dodecane to the culture (Figure S3A) and coupling a condenser to the bioreactor to harvest vaporized 2-heptanone (Figure S4). After 72 h of induced cultivation, cells consumed 111.5 g/L of glycerol, reached a biomass density of OD₆₀₀ ~40, and produced 3.34 g/L methyl ketone titer (Figure 4A). When a condenser was added to the off-gas line, we observed a similar trend of glycerol consumption, biomass formation and methyl ketone production with modestly increased titers. After 72 h of induced cultivation, cells consumed 126 g/L glycerol, reached a biomass density of OD₆₀₀ ~50, produced 4.4 g/L methyl ketone at the maximum titer (Figure 4B). At longer times, the titer of methyl ketones decreased, likely due to the rate of evaporation exceeding the rate of generation, which one would expect to be low as the carbon source, glycerol, is exhausted. After 96 hours, we observed a small amount (~ 25 mL of 2.3 g/L) of a MK containing solution in the condensed phase, representing ~5% of total MK titer (Figure 4C). We suspect that we may either be failing to provide sufficient cooling duty or we may be stripping MK from the condensed phase with our current design (Figure S4). Further work will be needed to identify condenser configurations or other capturing strategies to recover volatile MK from culture. The volatility and ability to be stripped from culture may be advantageous when scaling up the process.

Figure 4.

Time course of concentration of glycerol, OD₆₀₀ and total methyl ketone in fed-batch fermentation from bioreactor (A) and bioreactor coupling a condenser (B) using E. coli TRS12 strain harbors pTRC99a-FadD6-PsFadM and pACYC-Mlut_11700 plasmids. (C) Evaluation of methyl ketone concentration from samples in dodecane layer, aqueous phase and condenser after 96 h fermentation.

3.4. Metabolic engineering to improve 2-nonanone production

Our initial experiments suggested that the oxidation of saturated acyl-CoAs was a limiting factor in converting free fatty acids to methyl-ketones (i.e. acyl-CoA oxidase improved production). Therefore, we considered an alternative pathway in which oxidation was replaced by thiolase-mediated extension of the saturated acyl-CoA to a β-ketoacyl-CoA with two extra carbons. This idea is modeled off of the reverse β-oxidation pathway that has been shown to produce long chain acyl-CoAs in E. coli fermentations (Dellomonaco et al., 2011; Mehrer et al., 2018). The thiolase route avoids the need to generate NADH when cells are actively metabolizing glycerol and already have a highly reduced ratio of NADH/NAD+. To test this approach, we cloned a family of seven thiolase enzymes into a pACYC vector and coexpressed each with pTRC99a-FadD6-PsfadM in E. coli ΔfadABIJR. The activity of the ^EcFadA homologs was assayed in vivo by feeding 0.5 g/L octanoic acid to E. coli ΔfadABIJR + pTRC99a-FadD6-PsFadM + pACYC-”FadA”, and monitoring octanoic acid consumption and 2-nonanone production, Figure 5A. In general, all seven thiolase enzymes conferred the ability to consume octanoic acid and produce 2-nonanone (Figure 5B and S5A). In terms of ranking, we found strains expressing ^ReBktB and ^SeFadA exhibited the highest consumption rates (over 80% octanoic acid consumed within 18 h) and 2-nonanone titers ( ~280 mg/L) (Figure 5B and S5B).

Figure 5.

(A) Metabolic pathway for synthesis of 2-nonanone. Strain E. coli ΔfadRABIJ expressing pACYC-”fadA” + pTRC99a-FadD6-PsFadM + pBTRCK-CpFatB1* was used for demonstrating 2-nonanone synthesis. Impact of acyl-CoA thiolase homologs on rates of octanoic consumption (B) and 2-nonanone production (C) were evaluated by in vivo feeding 500 mg/L octanoic acid to the culture. The negative control is the strain harbored pACYC empty plasmid. Shake flask batch fermentation for 2-nonanone production was conducted by grown E. coli ΔfadABIJR strain that harbors pACYC-ReBktB + pTRC99a-FadD6-PsFadM + pBTRCK-CpFatB1* plasmids in Clomburg medium containing 45 g/L glycerol. Time course of methyl ketone titer (C), glycerol titer and OD600 (D) were monitored every 4 - 12 hrs. (E) Time course of glycerol titer, OD₆₀₀ and methyl ketone titer in fed-batch fermentation in a bioreactor using the optimized strain. All error bars represent standard deviations calculated from at least three biological replicates.

Next, we evaluated 2-nonanone production from glycerol in the background strain ΔfadAIR pTRC99a-FadD6-PsFadM + pACYC-ReBktB + pBTRCK-CpFatB1*. After 24 h in a batch shake flask culture, strains produced ~1.15 g/L 2-nonanone representing 75% of total methyl ketones from the 45 g/L glycerol fed (Figure 5C and 5D). We also suspect that the titer is underestimated, because when we incubated a defined 3.3 g/L 2-nonanone/water solution in the same shake flasks, we observed that 90% of the 2-nonanone was lost within 24 h (Figure S3A).

To benchmark 2-nonanone production, we conducted fed-batch cultivations by bolus feeding glycerol to cells grown in a stirred bioreactor. In order to reduce the loss of 2-nonanone in the off-gas, we applied the two strategies: adding 20% dodecane to the culture and coupling a condenser to the bioreactor to harvest vaporized 2-nonanone (Figure S4). After 72 h post-induction, cells consumed 107.1 g/L of glycerol, reached a biomass density of OD600 ~52, and produced a 3.0 g/L methyl ketone titer (Figure 5E). When a condenser was applied to the off-gas line, we observed a similar trend of glycerol consumption, biomass formation and methyl ketone production (data not shown). After 96 h post-induction, we found ~23 mL liquid containing ~120 mg/L methyl ketones collected in the condenser, representing less than 1% total MK. Overall, our results showed the thiolase-mediated condensation route is capable of producing 2-nonanone, but further optimization is needed to avoid the loss of octanoic acid intermediates from the cell (Figure S5C).

3.5. Metabolic engineering to improve 2-undecanone production

The Umbellularia californica acyl-ACP thioesterase BTE has high activity towards C₁₂ acyl-ACPs. BTE has been used in metabolic pathways to produce dodecanoate, 1-dodecanol, and polyhydroxydodecanoate, etc (Agnew et al., 2012; Lennen et al., 2010; Youngquist et al., 2013). Therefore, we used BTE to generate dodecanoate as an intermediate to enzymatically produce 2-undecanone. To test this approach, we chose a base strain TY34 (ΔfadE::trcBTE ΔfadAB::trcBTE ΔackApta::trcBTEΦ(PTrc-fadD) from Youngquist et al., 2013, in which 1-dodecanol was produced in titers over 1 g/L. To this strain, we introduced Mlut_11700, ^PsFadM, and increased expression of ^EcFadD. Mlut_11700 was selected for its higher activity in catalyzing the dehydrogenation of acyl-CoA to 2-enoyl-CoA. ^PsFadM was selected from the 15 homologs tested in Figure 2 for its strong activity towards twelve carbon substrates. We selected ^EcFadD instead of ^MtFadD6 because ^EcFadD was previously shown higher activities converting dodecanoate to dodecanoyl-CoA (Youngquist et al., 2013). We chose to overexpress a FadB homolog because the endogenous ^EcfadAB operon in TY34 strain was replaced by one copy of BTE.

In prior work, we observed that the source of the FadAB complex affected the product profile from reverse β-oxidation cycles activities (Mehrer et al., 2018), indicating that there may be chain length preferences. We chose six FadB homologs from which have previously shown functional FadAB complexes when testing reversed β-oxidation pathways (Mehrer et al., 2018). We evaluated each by comparing 2-undecanone titers in TY34 strains harboring three plasmids: pACYC-”FadB”, pBTRCK-Mlu_11700, and pTRC99a-PsFadM (Figure 6A). While control strains harboring a pACYC empty vector could not produce 2-undecanone, strains expressing each FadB homolog showed substantial quantities of 2-undecanone (Figure 6B). The strain harboring FadB from Pseudomonas putida generated the highest titer (350 mg/L) of methyl ketones and the highest specificity (58%) towards 2-undecanone. This result is not surprising considering P. putida naturally accumulates medium-chain length polyhydroxyalkanoate (PHA) and likely maintains a relatively higher C₈-C₁₂ activity to provide sufficient 3-hydroxylacyl-CoA pools, the most common substrates for PHA synthases PhaC1/C2 (Hoffmann and Rehm, 2004, 2005).

Figure 6.

(A) Designed a metabolic pathway for production of 2-undecanone. (B) Comparison of bi-functional 3-hydroxyacyl-CoA dehydratase. E. coli TY34 strain harbors pACYC-FadB + pTRC99a-PsfadM + pBTRCK-Mlut_11700 plasmids were grown in Clomburg medium containing 20 g/L glycerol for 24 h. (C) Time course of glycerol consumption and biomass formation, and (D) methyl ketone production by E. coli TY34 strain harbors pACYC-PpFadB + pTRC99a-PsfadM plasmids were grown in Clomburg medium containing 25 g/L glycerol for 72 h.

Next, we evaluated 2-undecanone synthesis in a time course experiment by growing strains in a shake flask in batch mode. We monitored cell growth, glycerol consumption, methyl ketone production, and fatty acid intermediates. As can be seen in Figure 6C and 6D, 25 g/L glycerol was depleted after 36 h. Cell growth reached a plateau after ~12 h, and the highest 2-undecanone concentration 335 mg/L was observed at ~48 h. The 2-undecanone titer represented ~62% of the total amount of methyl ketones. Overall, these data showed that we have successfully designed and optimized strains and metabolic enzymes to improve 2-undecanone production. Interestingly, the titers of 2-undecanone were significantly lower than those observed for 2-heptanone and 2-nonanone. This may be due to the difference in activity between the CpFatB1* and BTE thioesterases; residual levels of dodecanoic acid and tetradecanoic acid were substantially lower than residual octanoic acid (Figure S6).

4. Conclusion

In this work, we demonstrated metabolic engineering strategies for selective synthesis of 2-heptanone, 2-nonanone and 2-undecanone in E. coli. Through bioprospecting, we identified a FadM variant from Providencia sneebia that demonstrated higher activity on medium-chain substrates. When co-expressed with selective thioesterases it was capable of producing three different chain length MK products. In each case, the MK product profile was consistent with the selectivity of the thioesterase used. We demonstrated a novel thiolase-condensation route for converting acyl-CoAs to β-ketoacyl-CoA, the FadM substrate. When co-expressed with the CpFatB1*, a strain produced 2-nonanone with a titer up to 3.0 g/L; the highest reported titer to date (Table 2). Interestingly, this strain also excreted substantial amounts of octanoic acid, indicating that further optimization could further improve titers. In each of our studies, we observed substantial product loss due to evaporation and stripping. We coupled a rudimentary, water-cooling condenser to a bioreactor and found that it helped to increase titers over the timecourse, but failed to prevent product loss after metabolism ceased. While the evaporative loss is a challenge that leads to underestimation of product titers, it may be an advantage when scaling up MK production.

Table 2.

Summary of reported titer, yield and rate of methyl ketone production by microorganisms.

Product	Titers (g/L)	Yield (g/g)	Rate (g/L/h)	Cultivation Strategies	Species	References
2-heptanone	4.4	0.035 g/g consumed glycerol	0.061	bioreactor, fed-batch	E. coli	this study
2-nonanone	3.0	0.028 g/g consumed glycerol	0.042	bioreactor, fed-batch	E. coli	this study
2-undecanone	0.34	0.014 g/g consumed glycerol	0.007	shake flask, batch	E. coli	this study
C11-C15 MK	5.4	~0.05 g/g consumed glucose	0.05	bioreactor, fed-batch	E. coli	Goh et al., 2018
2-pentanone	0.24	0.024 g/g consumed glucose	0.0033	shake flask, batch	E. coli	Lan et al., 2013
C13-C23 MK	0.31	0.0074 g/g fed glucose	0.00155	bioreactor, fed-batch	Y. lipolytica	Hanko et al., 2018
C11-C15 MK	10 a	NR	NR	shake flask, batch	S. cerevisiae	Zhu et al., 2017
2-butanone 2-heptanone	0.24	0.0028 g/g fed hydrolysate	0.0011	bioreactor, batch	S. albus	Yuzawa et al., 2018
C13-C15 MK	1.1	0.169 g/g fed glucose	0.023	shake flask, batch	P. putida	Dong et al., 2019
C13-C15 MK	0.18	NR	0.0015	bioreactor, batch	R. eutropha	Muller et al., 2013

NR = Not reported

^aμg/DCW

^bYield is calculated as either g/g fed carbon source or g/g consumed carbon source, where former is quantified by dividing the reported titer (g/L) over the initial carbon concentration and the latter is directly reported in the original papers.

Highlights:

• β-oxidation and thiolase-mediated routes used to produce medium-chain methyl ketones

• Bio-prospecting identified FadM homolog from Providencia sneebia with higher activity

• We report production of highest medium-chain MKs titers, 4.4 g/L