Metabolic engineering for the high-yield production of isoprenoid-based C5 alcohols in E. coli

Branched five carbon (C5) alcohols are attractive targets for microbial production due to their desirable fuel properties and importance as platform chemicals. In this study, we engineered a heterologous isoprenoid pathway in E. coli for the high-yield production of 3-methyl-3-buten-1-ol, 3-methyl-2-buten-1-ol, and 3-methyl-1-butanol, three C5 alcohols that serve as potential biofuels. We first constructed a pathway for 3-methyl-3-buten-1-ol, where metabolite profiling identified NudB, a promiscuous phosphatase, as a likely pathway bottleneck. We achieved a 60% increase in the yield of 3-methyl-3-buten-1-ol by engineering the Shine-Dalgarno sequence of nudB, which increased protein levels by 9-fold and reduced isopentenyl diphosphate (IPP) accumulation by 4-fold. To further optimize the pathway, we adjusted mevalonate kinase (MK) expression and investigated MK enzymes from alternative microbes such as Methanosarcina mazei. Next, we expressed a fusion protein of IPP isomerase and the phosphatase (Idi1~NudB) along with a reductase (NemA) to diversify production to 3-methyl-2-buten-1-ol and 3-methyl-1-butanol. Finally, we used an oleyl alcohol overlay to improve alcohol recovery, achieving final titers of 2.23 g/L of 3-methyl-3-buten-1-ol (~70% of pathway-dependent theoretical yield), 150 mg/L of 3-methyl-2-buten-1-ol, and 300 mg/L of 3-methyl-1-butanol.

There has been considerable interest in the biosynthetic production of C 2 -C 5 alcohols as commodity chemicals and potential biofuels 1,2 . Butanol, for instance, serves as a platform chemical for thousands of compounds and can be used as a biogasoline or fuel additive 3,4 . Some of the most promising work in the microbial synthesis of these compounds has used refactored amino acid biosynthesis pathways to produce a variety of C 3 -C 5 alcohols at high titers 5 . Using this method, valuable alcohols such as isobutanol (C 4 ) have been produced at impressively high yields and titers 6 .
Isoprenoid biosynthesis provides an additional route to energy-dense C 5 alcohols, namely isopentenol (3-methyl-3-and 3-methyl-2-buten-1-ol, also known as isoprenol and prenol, respectively) and isopentanol (3-methyl-1-butanol) 7 . These alcohols have octane numbers and combustion properties that make them potential gasoline replacements 8 . Isopentanol, for example, has been evaluated for use in homogenous charge compression ignition (HCCI) engines and significantly outperforms ethanol 9 . In addition, these alcohols were shown to function as ideal anti-knock additives in spark ignition engines 10 . Although isoprenoids can be produced from both the methylerythritol phosphate (MEP) and mevalonate (MVA)
Our engineering strategy was to first optimize 3-methyl-3-buten-1-ol production before moving towards the production of mixed C 5 alcohols. Rather than focusing on high-throughput strain generation or combinatorial pathway assembly, we applied robust metabolomics and proteomics methods to identify likely bottlenecks and rationally direct pathway engineering. Once we optimized the precursor pathway and improved 3-methyl-3-buten-1-ol production, we implemented additional engineering to produce 3-methyl-2-buten-1-ol and ultimately fully reduced 3-methyl-1-butanol.
Metabolite profiling of KG1 identifies IPP accumulation as a likely bottleneck. Strain KG1 produced ~1.2 g/L of 3-methyl-3-buten-1-ol on 1% glucose after 48 hours, equivalent to 36% of pathway-dependent theoretical yield 14 . While an improvement over the original strain 19 , this titer was well below levels necessary for economical, large-scale production. We hypothesized that one or more pathway bottlenecks were limiting product yields in strain KG1. To accurately identify these bottlenecks, we performed a comprehensive analysis of pathway metabolites and proteins during the fermentation time-course.
Quantification of MVA pathway intermediates in strain KG1 over a 48 hour time-course (Fig. 2a) revealed that IPP concentrations exceeded those of any other observed MVA pathway intermediate by more than 8-fold. The comparatively low concentrations of acetyl-CoA, HMG-CoA, mevalonate, and mevalonate phosphate suggested that flux through the upstream pathway was unlikely to be limiting. The accumulation of IPP alone implied that NudB, the protein responsible for the conversion of IPP into 3-methyl-3-buten-1-ol, was the primary bottleneck in strain KG1. IPP has previously been shown to be toxic, resulting in growth inhibition and a reduction in glucose uptake at high concentrations. Although no obvious growth defects were apparent in strain KG1, more subtle effects such as feedback inhibition 20 could be deleterious to host and the pathway function. Even without exerting toxicity, the accumulation of a metabolite upstream of 3-methyl-3-buten-1-ol indicated suboptimal pathway performance. Consequently, reducing IPP accumulation was a priority in further engineering.
Quantification of pathway proteins in strain KG1 (Supplemental Figure S1) was conducted with targeted proteomics based on selected reaction monitoring (SRM) 21,22 to assess protein stability and relative concentrations over a 48-hour time-course. In general, pathway proteins appeared stable throughout the fermentation other than the expected increase following pathway induction (0 to 6 h). Although this Scientific RepoRts | 5:11128 | DOi: 10.1038/srep11128 SRM method reports relative peak areas for each protein rather than absolute concentrations, low signal intensity often indicates poor protein expression and potential pathway bottlenecks 14,22,23 . Peak area was the lowest for HMG-CoA reductase (HMGR), implying weak expression of this enzyme, but the low steady-state levels of HMG-CoA (the substrate of HMGR) and rapid accumulation of IPP (a downstream product after the HMGR-catalyzed reaction) suggested that HMGR may not be limiting under current The heterologous mevalonate pathway in E. coli consists of 7 reactions to convert acetyl-CoA into IPP and DMAPP. Dephosphorylation of these compounds by NudB, a promiscuous E. coli phosphatase, produces 3-methyl-3-buten-1-ol and 3-methyl-2-buten-1-ol, respectively. NemA, an endogenous reductase, is capable of reducing 3-methyl-2-buten-1-ol, but not 3-methyl-3-buten-1-ol, into fully reduced 3-methyl-1-butanol. (b) Plasmid architecture. A two plasmid system for 3-methyl-3-buten-1-ol production served as the initial engineering platform (strain KG1). Plasmid 1 contained genes from atoB to PMK with a medium copy p15A ori and weak lacUV5 promoter 16 . Plasmid 2 contained nudB and PMD with a high copy pBR322 ori and strong trc promoter. To produce mixed alcohols, a fusion protein and reductase were expressed on plasmid two (see Fig. 4). conditions. Protein areas of PMD and NudB, both of which were expressed from a high-copy plasmid, were an order of magnitude above the other pathway proteins. Although NudB is an E. coli native protein, the plasmid-borne copy was the primary source of quantifiable protein; background levels of NudB in DH1 wild type were more than 10-fold less than strain KG1 (data not shown).

3-buten-1-ol titers.
To prevent the accumulation of IPP, we could either slow down its formation through upstream pathway engineering or enhance its conversion to 3-methyl-3-buten-1-ol by improving NudB-catalyzed reaction efficiency. Since we did not want to decrease upstream pathway efficiency, we focused on improving flux to 3-methyl-3-buten-1-ol by increasing NudB expression. To increase NudB protein levels, we optimized the Shine-Dalgarno sequence of nudB using the RBS calculator 24 .
Starting with the RBS sequence in strain KG1 (NudB RBSo ) as an input, 10 sequences were generated that were predicted to improve NudB protein expression (Fig. 2b). We designed the 10 RBS sequences (RBS1 -RBS10) to achieve a range of expression to allow for the titration of NudB levels in subsequent engineering (Supplemental Table 1). After cloning these variants into plasmid 2, we co-transformed each of them into E. coli DH1 with pJBEI-6829 to generate strains KG1 R1 (strain KG1 containing NudB RBS1 ) through KG1 R10 (strain KG1 containing NudB RBS10 ) ( Table 1). We performed a production test with these strains and collected protein samples to quantify levels of NudB (Fig. 2c). Protein levels of NudB were improved in all RBS sequence variants with the notable exception of NudB RBS2 . The best-performing RBS sequences, RBS5, RBS9, and RBS10, yielded NudB protein levels 9-fold higher than the original sequence (Fig. 2c). As anticipated, strains with improved NudB expression also produced more 3-methyl-3-buten-1-ol (Fig. 2c). Surprisingly, the variability in 3-methyl-3-buten-1-ol titer was minor compared to the large changes in NudB protein level; even the slight increase in NudB protein in strain KG1 R7 yielded a 3-methyl-3-buten-1-ol titer comparable to strain KG1 R10 , which produced 8-fold more NudB. This observation suggested that NudB expression was no longer limiting in the highest-performing strains, which produced ~1.6 g/L of isopentenol at 36 hours. Indeed, expression of additional NudB on a supplemental plasmid yielded no additional improvement in titer (Supplemental Figure S2). Still, we chose KG1 R10 for further study, reasoning that this strain provided room for additional improvement in the precursor pathway and increased flux to IPP.
Metabolite analysis of KG1 R10 confirmed that improved NudB expression relieved IPP accumulation. Over a 48-hour time-course, IPP levels in strain KG1 R10 were reduced by 4-fold relative to KG1 (Fig. 2d). At 48 hours, the 3-methyl-3-buten-1-ol titer in strain KG1 R10 reached 1.94 g/L, a 60% increase over strain KG1. Given the significant reduction in IPP level observed in strain KG1 R10 , we suspected that the pathway bottleneck might have shifted from NudB reaction to the upstream precursor pathway. If this were the case, further increases in IPP production could yield additional increases in 3-methyl-3-buten-1-ol titer.

Optimization of MK yields minor improvements in 3-methyl-3-buten-1-ol titer.
Although the 3-methyl-3-buten-1-ol pathway consists of seven genes, MK is the primary determinant of flux to IPP 14 . Sufficient flux to IPP is required for high product yields, but excessive MK expression can lead to IPP accumulation and a reduction in growth and 3-methyl-3-buten-1-ol titer. Since we increased the "pull" on IPP using NudB RBS10 (Fig. 2d), we decided to increase the "push" to IPP through additional MK engineering. We employed two approaches: 1) pairing improved NudB expression with increased expression of MK, and 2) introducing MK homologs with desirable properties from other organisms. In each case, the MK gene was positioned downstream of PMK as this gene order previously increased MK protein levels 14 .
MK expression was enhanced by inserting a trc promoter upstream of MK (pJBEI-6832), which was shown to increase MK protein levels by ~4-fold 14 . We paired this plasmid with the previously characterized plasmids containing NudB RBSo (creating strain KG2), NudB RBS3 (strain KG2 R3 ), and NudB RBS10 (strain KG2 R10 ) to titrate low, medium, and high levels of NudB protein, respectively (Fig. 3). We compared strain KG2 with the control strain KG3, which contained an identical version of the pathway with the same gene order but no supplemental promoter (Table 1). 3-Methyl-3-buten-1-ol titer in KG2 was 268 mg/L, a 4-fold reduction compared to KG3. If this low titer was due to IPP accumulation, we anticipated that elevated NudB levels would dramatically increase 3-methyl-3-buten-1-ol titer. In accordance with our expectation, strain KG2 R3 , which expressed a higher level of NudB, produced 4-fold more 3-methyl-3-buten-1-ol than strain KG2. Even with this improvement, however, titers were about 20% less than the comparable control (strain KG3 R3 ). With NudB RBS10 (i.e. strain KG2 R10 ), we recorded an additional increase in titer to 1550 mg/L of 3-methyl-3-buten-1-ol, matching the performance of the control KG3 R10 (Fig. 3). Attempts to further increase NudB levels with an additional plasmid (pBbB8k-NudB, pJBEI-6835) were unsuccessful (Supplemental Figure S2), suggesting that IPP accumulation was no longer problematic in strain KG2 R10 . Although elevated MK expression appeared to increase flux to IPP, the strong performance of the control strain suggested it was not necessary for peak pathway efficiency. Consequently, we focused on changing MK identity rather than expression level for further optimization.
Although most work on MVA pathway optimization has focused on S. cerevisiae-derived MK (MK SC ), MK enzymes from alternative hosts such as Staphylococcus aureus (MK SA ) and the archaea Scientific RepoRts | 5:11128 | DOi: 10.1038/srep11128 Methanosarcina mazei (MK MM ) have promising characteristics. MK MM , for example, was shown to resist feedback inhibition from IPP, GPP, and FPP 20 . We tested these MK variations with a MVA pathway "top" portion that provided high flux to mevalonate (MevTco) 14 , reasoning that this context could better reveal differences in MK activity. Strains KG4, KG5, and KG6 contained this high flux pathway paired with MK enzymes derived from S. cerevisiae, M. mazei, and S. aureus, respectively ( Table 1). As practiced above, we titrated each pathway with low (NudB RBSo ), medium (NudB RBS3 ), and high (NudB RBS10 ) levels of NudB.
Since the preliminary results for MK MM were promising, we investigated this homolog further by pairing the enzyme with a different MVA pathway "top" portion, MevT SA , to create strain KG7 (Table 1). In this pathway context, a maximum 3-methyl-3-buten-1-ol titer of 1730 mg/L was recorded with NudB RBS10 (Fig. 3). Although MK MM did appear to improve flux to IPP relative to MK SC , similar maximum titers were reached with each enzyme. This apparent "plateau" in 3-methyl-3-buten-1-ol titer may suggest  Table  S1) were generated using RBS calculator and cloned into plasmid 2, forming JPUB-004498 to JPUB-004507. (c) 3-Methyl-3-buten-1-ol titer and NudB protein level in KG1 R1 -KG1 R10 . Strains containing nudB RBS1 through nudB RBS10 (KG1 R1 -KG1 R10 ) were assayed for 3-methyl-3-buten-1-ol and NudB protein level. Bars represent 3-methyl-3-buten-1-ol titer after 36 hours. Squares show NudB protein area at 24 hours relative to that measured in strain KG1 (normalized to a value of 1). Error bars represent standard deviation (n = 3). (d) IPP accumulation and 3-methyl-3-buten-1-ol production in strains KG1 and KG1 R10 . IPP concentrations in KG1 R10 were reduced by 4-fold relative to KG1, while 3-methyl-3-buten-1-ol titer increased by 60%. Error bars represent standard deviation (n = 3). that further gains from MVA pathway engineering will be difficult to achieve. Although NudB protein expression is no longer a bottleneck, the poor kinetics of the enzyme 12 may play a key role in limiting 3-methyl-3-buten-1-ol production. It is possible that an improvement in NudB kinetics is required to facilitate additional titer increases in the current pathway context.
Compared to strain KG1, total C 5 alcohol content in KG8 decreased by over 40%. We hypothesized that competition with endogenous pathways was responsible for this reduction in overall titer. We suspected that expression of Idi1-required for DMAPP and 3-methyl-2-buten-1-ol production-would divert carbon towards the formation of geranyl diphosphate (GPP) and farnesyl diphosphate (FPP), both of which are used in a variety of cellular processes (Fig. 5a). Consistent with this hypothesis, GPP and FPP were observed in strain KG8, but below detection in the 3-methyl-3-buten-1-ol production strain KG1 (Fig. 5b). A 10-fold decrease in the concentration of IPP/DMAPP was also observed in KG8, likely reflecting the multiple routes of IPP/DMAPP depletion in this strain. The detection of GPP and FPP in strain KG8 makes E. coli FPP synthase (ispA) an attractive engineering target; it is likely that reducing or eliminating IspA activity will yield increased production of 3-methyl-2-buten-1-ol and subsequently 3-methyl-1-butanol. Though levels of IPP, GPP, and FPP varied significantly between KG8 and KG1, mevalonate concentrations were similar in both strains. This suggested that carbon flow through the upstream mevalonate pathway was unaffected by the implementation of downstream engineering.
We increased the level of NudB protein to optimize the pathway given its positive effect in strain KG1 (Fig. 2). Since NudB in strain KG8 was part of a protein fusion with Idi1, we increased NudB levels by using a third plasmid, pBbB8k-NudB (pJBEI-6835) 14 . We designated this 3-plasmid strain as KG9 (Table 1). Compared to strain KG8, total mixed alcohol content at 48 hours increased more than 20% to 880 mg/L (Fig. 6). However, 3-methyl-3-buten-1-ol accounted for the entirety of this increase-levels of 3-methyl-2-buten-1-ol remained constant while those of 3-methyl-1-butanol slightly decreased. More engineering is clearly required to direct carbon flow from DMAPP towards 3-methyl-2-buten-1-ol and 3-methyl-1-butanol and away from GPP and FPP. One promising approach would be the development or discovery of a phosphatase specific for DMAPP. Although NudB is effective in producing high titers of C 5 alcohol, the enzyme is promiscuous and does not appear to discriminate between IPP or DMAPP, both of which are non-native substrates 25 . More broadly, carbon loss to GPP and FPP should be mitigated through a selective knockdown of IspA activity. This may prove challenging, however, since the formation of FPP and thus ispA is essential for E. coli growth 26 . An oleyl alcohol overlay improves yields of each C 5 alcohol. Since short-chain alcohols and terpenes are often volatile, calculated yields are commonly underestimated during fermentation. In the production of 3-methyl-3-buten-1-ol, an apparent decrease in recoverable alcohol observed from 48 to 72 hours (Supplemental Figure S4) suggested that evaporation was occurring. To quantify the amount of 3-methyl-3-buten-1-ol, 3-methyl-2-buten-1-ol, and 3-methyl-1-butanol lost during production assays, we spiked each alcohol into culture tubes containing growth medium at various concentrations and monitored recovery over time at 30°C (Supplemental Figure S5). After 48 hours, losses of 20%, 10%, and 40% were observed for 3-methyl-3-buten-1-ol, 3-methyl-2-buten-1-ol, and 3-methyl-1-butanol, respectively. When the same tubes were incubated at 4°C, 100% recovery of each alcohol was observed after 48 hours (data not shown).
Hydrophobic overlays such as decane or dodecane are often used to prevent evaporation during microbial fermentations, particularly in the case of fatty acids or longer chain isoprenoids 27,28 . For the recovery of alcohols such as 1-butanol or 3-methyl-1-butanol, however, oleyl alcohol has proven a better choice in both Clostridia 29 and E. coli 30 . We assessed the efficacy of an oleyl alcohol overlay with KG1 R10 and KG9, high producers of 3-methyl-3-buten-1-ol and mixed alcohols, respectively. Using a 20% oleyl alcohol overlay (10 mL overlay added to 50 mL of culture) with strain KG1 R10 , a 3-methyl-3-buten-1-ol titer of 2.23 g/L (70% theoretical) was recorded after 48 hours, a 20% increase over the same strain without an overlay (Fig. 7). With strain KG9, an overlay led to a 26% increase in total C 5 alcohol content (Fig. 7). The largest increase was in 3-methyl-1-butanol, where titers more than doubled to ~300 mg/L. The increased volatility of 3-methyl-1-butanol relative to 3-methyl-3-and 3-methyl-2-buten-1-ol (Supplemental Figure  S5) and differential partitioning of 3-methyl-1-butanol into oleyl alcohol (Supplemental Figure S6) provide potential explanations for the improved recovery of this alcohol. We suspected this was due to the increased volatility (Supplemental Figure S5) and partitioning (Supplemental Figure S6) of this alcohol relative to 3-methyl-3-and 3-methyl-2-buten-1-ol.
Scientific RepoRts | 5:11128 | DOi: 10.1038/srep11128 production. Although increased NudB expression resulted in high yields in our most productive strains, significant improvements in pathway productivity will be required for fermentation scale-up. Future work to develop a more catalytically active phosphatase should be undertaken to achieve this goal.
Production titers of 3-methyl-2-buten-1-ol and 3-methyl-1-butanol were improved by more than 10-fold compared to previous work, where titers on 0.2% glucose were <10 mg/L 12 . Furthermore, we demonstrated that an oleyl alcohol overlay minimizes product loss due to evaporation and is effective for C 5 alcohol fermentations. Still, additional engineering is required to improve total C 5 alcohol content and optimize for the production of 3-methyl-2-buten-1-ol and 3-methyl-1-butanol. The detection of GPP and FPP, metabolites of endogenous isoprenoid metabolism, makes E. coli FPP synthase (ispA) an   Table 1) in strain KG9 yielded a 20% increase in total alcohol content at 48 hours post induction. Error bars show standard deviation (n = 3). attractive engineering target. A reduction in IspA expression or activity should reduce carbon loss to FPP and improve flux to 3-methyl-2-buten-1-ol. Although NemA successfully catalyzes the conversion of 3-methyl-2-buten-1-ol to fully reduced 3-methyl-1-butanol, it does so at a low efficiency 12 . Protein engineering to improve the kinetics of this reaction should result in the complete conversion to 3-methyl-1-butanol. As titers of C 5 alcohols increase, product toxicity is likely to become an issue 31 . Although use of oleyl alcohol may reduce toxicity, host engineering for increased tolerance 32 may eventually be required to reach production goals.

Methods
Chemicals, solvents and media components were purchased and used without modification from Sigma-Aldrich (St. Louis, MO), Fisher Scientific (Pittsburgh, PA), or VWR (West Chester, PA) unless otherwise noted. E. coli strains DH10B (Invitrogen, Carlsbad, CA) and DH1 (ATCC) were used for plasmid construction and production experiments, respectively. For targeted proteomics experiments, mass spectrometric-grade trypsin was obtained from Sigma-Aldrich and prepared according to manufacturer's instructions.
Plasmid and strain construction. E. coli DH10B was used as the host for all cloning and plasmid manipulations. The BglBrick standard 33 was used to assemble all plasmids as previously described. With the exception of pTrc99A 17 , plasmids were derived from the BglBrick plasmid library 16 . E. coli DH1 was used as the host for all production assays. Plasmids used in this study are available on the JBEI public registry and listed in Table 1 along with a brief description of production strains.
Production assays. Starter cultures of E. coli DH1 harboring production plasmids were grown overnight in LB medium containing appropriate antibiotics at 37 °C and shaken at 200 rpm in rotary shakers. Chloramphenicol, ampicillin, and kanamycin were provided at final concentrations of 25 mg/L, 100 mg/L, and 25 mg/L, respectively. Production assays were performed in triplicate in EZ-Rich defined medium (Teknova) containing 1% glucose. Briefly, starter cultures were used to inoculate 5 mL of production media in a culture tube or 50 mL of production media in a 250 mL Erlenmeyer flask to an OD 600 of 0.1. Production cultures were grown in rotary shakers (200 rpm) at 37 °C to an OD 600 of 0.4-0.6 and induced with 500 μ M isopropyl β -D-1-thiogalactopyranoside (IPTG). Strains harboring pJBEI-6835 were also induced with 20 mM arabinose. Following induction, cultures were moved to 30°C for the duration of the assay. At set times, samples were taken for C 5 alcohol quantification analysis by GC-FID as described previously 12 . For samples containing oleyl alcohol, the entire culture volume was extracted with ethyl acetate after 48 hours. Samples were analyzed by GC-FID as previously described 30 . Metabolite quantification. Glucose and organic acids were quantified in filter-sterilized supernatant by high performance liquid chromatography (HPLC) at set time points using an Agilent 1200 Series HPLC system. Intracellular concentrations of mevalonate and IPP were measured by liquid chromatography mass spectrometry (LC-MS). Please see references 14 and 34 for complete protocols. Volatility assays. To assess the evaporation of each C 5 alcohol, 3-methyl-3-buten-1-ol, 3-methyl-2-buten-1-ol, and 3-methyl-1-butanol were spiked into 5 mL of EZ-Rich media in culture tubes (triplicate) at various concentrations and placed at 30 °C while shaking (200 rpm) for 48 hours. One set of tubes was placed at 4 °C to act as a control. Samples were collected for alcohol quantification by GC-FID at times 0, 24, and 48 hours.
Targeted proteomics analysis. At 24 hours, 1.5 mL of production culture was collected and pelleted by centrifugation at 8000 × g (4 °C). After the supernatant was decanted, cell pellets were frozen in liquid nitrogen and stored at − 80 °C. Sample preparation and protein extraction was performed as described previously 14,22 .