Directed evolution of prenylated FMN-dependent Fdc supports efficient in vivo isobutene production

Isobutene is a high value gaseous alkene used as fuel additive and a chemical building block. As an alternative to fossil fuel derived isobutene, we here develop a modified mevalonate pathway for the production of isobutene from glucose in vivo. The final step in the pathway consists of the decarboxylation of 3-methylcrotonic acid, catalysed by an evolved ferulic acid decarboxylase (Fdc) enzyme. Fdc belongs to the prFMN-dependent UbiD enzyme family that catalyses reversible decarboxylation of (hetero)aromatic acids or acrylic acids with extended conjugation. Following a screen of an Fdc library for inherent 3-methylcrotonic acid decarboxylase activity, directed evolution yields variants with up to an 80-fold increase in activity. Crystal structures of the evolved variants reveal that changes in the substrate binding pocket are responsible for increased selectivity. Solution and computational studies suggest that isobutene cycloelimination is rate limiting and strictly dependent on presence of the 3-methyl group.

T he irrefutable harmful environmental effects and depleting reserves of fossil fuels have powered an extensive amount of research to seek sustainable alternatives for the production of petrochemicals, including the gaseous alkene isobutene [1][2][3] . Due to the favourable reactivity, isobutene is widely used as a building block for fuel additives, rubber, plastic and a broad range of fine chemicals. Over 10 million tons of isobutene are produced every year, primarily by steam cracking crude oil. Low levels of microbial production of isobutene were first detected in the 1980s [4][5][6] . More recently, isobutene production via a modified mevalonate (MVA) pathway using mevalonate diphosphate decarboxylase (MVD) to decarboxylate 3-hydroxyisovaleric acid was reported 7 and subsequently patented 8 (Fig. 1A). Further studies highlighted a more efficient route using mevalonate-3-kinase (M3K, Picrophilus torridus) that catalyses isobutene formation through an unstable phosphorylated intermediate 9,10 . The highest reported whole-cell isobutene production rate of 507 pmol min −1 g cells −1 was reached using E. coli engineered with M3K, however, this remains about 10 5 -fold lower than is economically viable 7,9 . The slow conversion could be surpassed by an alternative route, such as the more direct conversion of methylcrotonyl-CoA to isobutene through a combination of a thioesterase with a non-oxidative decarboxylase. The prenylated flavin (prFMN)-dependent ferulic acid decarboxylases (Fdc) catalyse reversible non-oxidative (de) carboxylation of a range of acrylic acids with extended conjugation [11][12][13][14] . Recently, a reversible 1,3-dipolar cycloaddition mechanism was conclusively shown to underpin catalysis in Aspergillus niger Fdc (AnFdc) 11,[15][16][17] . First, the cycloaddition of the substrate results in cycloadduct Int1 (Fig. 1B). Decarboxylation occurs concomitantly with ring opening to form Int2. Following the exchange of CO 2 with E282, protonation by E282 results in cycloadduct Int3 that releases the product through cycloelimination. Cycloadduct strain, mediated by a clash between the substrate R group and enzyme residues is key in ensuring reversible 1,3-dipolar cycloaddition 17 . Recent studies have shown rational engineering of AnFdc can expand substrate scope to include aromatic substrates such as naphthoic acid 18 . Crotonic acid was found to inhibit AnFdc by apparently irreversibly binding to the prFMN cofactor 17 . However, acrylic acid substrates lacking extended conjugation have rarely been reported in the wider UbiD enzyme family. Arguably, the natural UbiD substrate closest to 3-methylcrotonic acid is trans-anhydromevalonate 5-phosphate (tAHMP), which is decarboxylated by a UbiD-decarboxylase from a hyperthermophilic archaeon Aeropyrum pernix in an alternative mevalonate pathway 19 . Both 3-methylcrotonic acid and tAHMP contain a secondary beta carbon and lack extended conjugation, however, the phosphate group in tAHMP may facilitate strain manipulation in cycloadduct intermediates. A recent communication describes a pathway for the production of butadiene in E. coli where the decarboxylation of cis,cis-muconic acid is catalysed by Fdc from Saccharomyces cerevisiae enhanced by rational design 20 .
Herein, we report on discovery and optimization through directed evolution of Fdc decarboxylation activity with 3-methylcrotonic acid to produce isobutene. We seek to understand how a substrate lacking extended conjugation and bulk can be decarboxylated by Fdc, especially in view of the fact crotonic acid acts as an inhibitor of prFMN. We discuss the structural basis for an increase in activity and selectivity in Trichoderma atroviride Fdc (TaFdc) evolved by directed evolution. Surprisingly, the optimized variants remain unable to decarboxylate crotonic acid, suggesting that in the case of the substrate 3-methylcrotonic acid the single additional methyl group plays a key role in the cycloelimination process. Computational studies are used to rationalize the effect of the 3-methyl substitution on product formation.

Results and discussion
Initial screening of Fdc homologues. Initial in vivo screening tested 15 UbiD homologues co-expressed with UbiX (E. coli K-12) in E. coli for conversion of 3-methylcrotonic acid into isobutene as detected by gas chromatography. TaFdc exhibited over twice the isobutene production compared to other homologues  7 and mevalonate-3-kinase (M3K) 9 produce isobutene via 3-hydroxyisovaleric acid. Fdc1, co-expressed with UbiX, catalyses the decarboxylation of 3-methylcrotonic acid to give isobutene. B Fdc decarboxylation reaction mechanism with 3-methylcrotonic acid in blue and common Fdc substrates with a conjugated R-group in red. First, the 1,3-dipolar cycloaddition of the substrate to prFMN iminium leads to the first pyrrolidine cycloadduct, Int1. Decarboxylation and ring-opening forms the noncyclic alkene adduct Int2. Protonation by a conserved glutamic acid residue yields the second pyrrolidine cycloadduct Int3 followed by cycloelimination to give the product.
Characterization of TaFdc and TaFdcV. TaFdc wild-type and TaFdcV with an N-terminal hexa-histidine tag were co-expressed with E. coli K-12 UbiX in E. coli and purified with Ni-NTA resin. UV-Vis spectra of both purified proteins exhibit a distinct peak at 380 nm, thought to correspond to the cofactor active form prFMN iminium (Supplementary Fig. 2A) 16 . ESI-MS confirmed the presence of prFMN iminium in both enzyme variants (Supplementary Fig. 2B and C). The shape of the 380 nm peak and cofactor content (assessed by the ratio of absorbances at 280 and 380 nm) varied from batch to batch.
TaFdc showed decarboxylation activity with cinnamic and sorbic acid, with rates k obs = 7.2 ± 0.3 and 3.2 ± 0.3 s −1 , respectively (reported for a batch with a 380:280 nm ratio of 0.067). These values are comparable to those reported for AnFdc 11,16 . In contrast, the TaFdcV variant showed compromised activity with sorbic acid (k obs = 0.33 ± 0.03 s −1 ) and no activity was detected with cinnamic acid. When exposed to light, TaFdc sorbic acid decarboxylation activity steadily deteriorates with a half-life of 1 h compared to enzyme stored in dark ( Supplementary Fig. 2H). This is consistent with Fdc light-sensitivity as described previously 17 . Upon irradiation with a 405 nm LED lamp, the characteristic 380 nm peak in the UV-visible absorbance spectra of TaFdc and TaFdcV irreversibly splits to peaks at 365 and 425 nm (Supplementary Fig. 2I and J).
Incubation of both TaFdc and TaFdcV with 3-methylcrotonic acid triggered a change in the protein UV-Vis spectrum to reveal peaks at 340 and 425 nm, suggestive of a covalent substrate:prFMN adduct accumulating under turnover conditions. Following a desalting step, the spectrum returns to the as-isolated 380 nm single feature, confirming that a long-lived, inhibitory covalent complex with 3-methylcrotonic acid is not formed (Supplementary Fig. 2D and E). Incubation of 80 μmol TaFdcV with 10 mM 3-methylcrotonic acid led to a complete shift in the corresponding UV-Vis spectrum. In contrast, the wild-type TaFdc required prolonged incubation with 50 mM 3-methylcrotonic acid to achieve full spectral conversion, suggesting a substantially higher K D and/or adduct formation rate for the wild-type enzyme. An ESI-MS spectrum of the desalted sample showed peaks corresponding to both prFMN iminium and a putative Int3 prFMN cycloadduct with 3-methylcrotonic acid ( Supplementary Fig. 3). This may be due to a small proportion of 3-methylcrotonic acid remains bound to prFMN as Int3, suggesting Int3 elimination is the rate-limiting step, or that a proportion of the Int3 species has irreversibly isomerized to a more stable conformation.
In order to assess the scope for activity with acrylic acids lacking extended conjugation, TaFdc and TaFdcV were incubated with trans-2-pentenoic and trans-2-hexenoic acid, compounds that have previously been reported to undergo some AnFdcmediated decarboxylation 11 . UV-Vis absorbance spectra indicated that TaFdc bound both acids ( Supplementary Fig. 2F), whereas the TaFdcV variant preferred the smaller pentenoic acid and required higher concentrations to fully bind hexenoic acid ( Supplementary Fig. 2G). In contrast to samples incubated with 3-methylcrotonic acid, the UV-Vis spectra of samples incubated with pentenoic or hexenoic acid were unaffected by a desalting step, indicating that pentenoic and hexenoic acid irreversibly binds to TaFdc/TaFdcV. Quantitative GC assay indicates that pentene production from hexenoic acid by AnFdc is limited to a single turnover ( Supplementary Fig. 4).
Crystal structures of TaFdc and TaFdcV reveal mutation impact on the substrate-binding pocket. In order to understand how TaFdcV mutations aid in isobutene production, crystal structures of TaFdc and TaFdcV were solved at a resolution of 1.74 and 1.89 Å, respectively. An overlay of the wild-type and the variant crystal structures shows that the key residues F447, Q200 and the catalytic network of E287-R183-E292 are unaffected by the mutations (Fig. 2A) 16 . The T405M mutation is located at the active site, extending towards the space above the prFMN uracil ring while the Q448W and F404Y mutations are situated in the second shell from the active site. The E292 residue side chain occupies 'up' and 'down' conformations, while weak electron density suggests a high degree of mobility for the L449 side chain. The mobile E292 and L449 gate access to the active site ( Fig. 2B) while the Q448W mutation in TaFdcV narrows the binding pocket (Fig. 2C). The T405M and Q448W mutations are likely to be responsible for the increased selectivity for 3-methylcrotonic acid in TaFdcV by enhancing the substrate/active site shape complementarity, blocking access to larger substrates (Supplementary Fig. 5). While comparison of TaFdc and TaFdcV crystal structures reveals the basis for increased selectivity in the evolved enzyme, it is not immediately clear why 3-methylcrotonic acid can yield isobutene from Int3.
Formation of stable cycloadducts with inhibitors. The effects of crotonic and 2-butynoic acid on TaFdcV were studied to determine whether the mutations that increase in 3-methylcrotonic acid turnover also affected activity with related compounds. Incubation of TaFdcV with 2-butynoic and crotonic acid led to the familiar split of the 380 nm prFMN peak in the UV-Vis spectrum ( Fig. 3A and D), similar to 3-methylcrotonic acid. However, as previously observed with pentenoic and hexenoic acid, the spectrum did not recover the following desalting, suggesting that a covalent inhibitory adduct is formed. Similar trends were observed with TaFdc, however, incubation at higher inhibitor concentration was required to drive changes in the UV-Vis spectrum.
Upon addition of crotonic acid, a gradual shift in UV-Vis spectrum occurs over minutes, allowing estimation of adduct formation rate (Supplementary Fig. 6A). The observed rate remains first order with respect to crotonic acid, with k obs = 0.34 ± 0.03 min −1 at the highest concentration tested (50 mM)  TaFdcII  Mutations  WT  T395M T395M  R435P P438W   WT  T405M  E25N N31G G305A D351R K377H P402V  F404Y T405M T429A V445P Q448W   F404Y T405M  V445P  ( Supplementary Fig. 6B). In contrast, a similar shift in UV-Vis spectrum upon addition of the substrate 3-methylcrotonic acid occurs rapidly within seconds, and at substantially lower 3-methylcrotonic acid concentrations. This suggests that crotonic acid adduct formation is hindered by a higher K D and/or slower rate of cycloaddition. ESI-MS and co-crystallization studies confirmed that the TaFdcV 2-butynoic acid adduct stalls as Int1, while the TaFdcV crotonic acid adduct undergoes decarboxylation to stall at the Int3 species (Fig. 3). Similar behaviour has been reported for AnFdc 17 . No decarboxylation of the Int1 with 2-butynoic acid was detected, even in 1-month-old crystals. In contrast, although only Int3 was observed in co-crystals with crotonic acid, ESI-MS also showed a peak for the corresponding Int1 (Fig. 3E). It is unclear whether Int1 can be detected in this case because decarboxylation of crotonic acid is slow, or because there is an equilibrium between Int1 and Int3 at ambient CO 2 levels.
AnFdcII with three point-mutations has an identical active site conformation to TaFdcV. To further understand how the architecture of the active site affects the decarboxylation of 3-methylcrotonic acid, corresponding key mutations from TaFdcV were introduced in AnFdc. AnFdc has been established as a model system due to the fact that it readily yields atomic resolution crystal structures 11,[16][17][18] . Two variants were studied: AnFdc T395M (AnFdcI) and the triple mutant AnFdc T395M R435P P438W (AnFdcII). Overlay of the AnFdc wild-type and TaFdcV crystal structures reveals a downward shift of the Y404 residue in TaFdcV in the secondary shell compared to the corresponding Y394 in AnFdc (Fig. 4A). The Y394 residue is unaffected in the AnFdcI variant compared to wild-type (Fig. 4B). In contrast, the active site of the AnFdcII variant matches that of TaFdcV in the conformation of Y394 and M395 (Fig. 4C).
As expected, neither AnFdcI nor AnFdcII were active with cinnamic acid, likely due to a clash between the substrate phenyl ring and M395. While binding of crotonic acid in AnFdc wildtype cannot be detected by the UV-Vis spectra over 2-h incubation, both mutants AnFdcI and AnFdcII readily bind the inhibitor, evident from UV-Vis spectra, demonstrating increased selectivity towards smaller substrates.
While AnFdc wild-type was included in the initial UbiD screen, the AnFdc wild-type was 90 times lower in activity in vivo compared to TaFdc. Hence, AnFdc was not selected for further directed evolution, despite having comparable in vitro activity to TaFdc. The disparate and lower activity in vivo might be attributed to AnFdc-specific inhibition by metabolites such as phenylacetaldehyde 11 . An initial comparison of in vitro isobutene production levels using crude cell lysate from cells expressing TaFdc variants with those expressing MVD and/or M3K reveals ã 50-fold increase is observed for TaFdcV compared to MVD/ M3K levels ( Supplementary Fig. 8). This demonstrates that the evolved TaFdcV is vastly superior in catalysing the decarboxylative step compared to the previously described enzyme systems.
Computational studies reveal a mechanistic basis for isobutene production. It is curious that a single methyl group difference, as occurs between crotonic acid and 3-methylcrotonic acid, determines whether the compound is a substrate or inhibitor for the evolved Fdc variants. The marked influence of the additional methyl group on Int3 cycloelimination suggests this step may proceed via a cationic or radical beta carbon stabilized through additional hyperconjugation effects. A density functional theory (DFT) active site 'cluster' model ( Supplementary Fig. 9) was used to investigate why 3-methylcrotonic acid is decarboxylated and eliminated by TaFdcV in contrast to crotonic acid. The potential energy surface for the cycloelimination of Int3 to the non-covalent product complex was computed for both crotonic acid and 3-methylcrotonic acid by varying the C α -C 1' and C β -C 4a bond lengths (Fig. 6). These suggest that 3-methylcrotonic acid undergoes a more asynchronous elimination, with the transition state C α -C 1' and C β -C 4a bond lengths of 1.96 and 2.97 Å, respectively, compared to 1.95 and 2.77 Å for crotonic acid,  respectively. This is linked to an increased charge separation occurring between the C β and prFMN for the 3-methylcrotonic acid compared to crotonic acid (Supplementary Tables 4 and 5), possibly affected by additional hyperconjugation in the case of 3-methylcrotonic acid. The release of propene from crotonic acid Int3 cycloadduct is more endothermic by~8 kJ mol −1 and has a higher energy barrier by 19 kJ mol −1 compared to the release of isobutene from Int3 with 3-methylcrotonic acid. If the activation entropy is similar for the two reactions then the transition state energy difference translates to a~2200 slower rate for the release of propene from Int3 at 293 K, explaining the lack of crotonic acid turnover under conditions tested.
The limit of prFMN-dependent (de)carboxylation by UbiD enzymes. Directed evolution of TaFdc to TaFdcV resulted in a marked increase in activity with 3-methylcrotonic acid. Surprisingly, the evolved mutant remained unable to convert crotonic acid to the corresponding propene. This contrasts with previous evolved studies aimed at expanding the AnFdc substrate repertoire to include (hetero)aromatic compounds 18 . In this case, the evolution of activity against heteroaromatic bicyclic compounds yielded a broad specificity variant able to convert even naphthoic acid. It is thus possible that 3-methylcrotonic acid represents a limit for bona fide UbiD-substrates, indicating that prFMNdependent catalysis requires more than an α,β-unsaturated acrylic acid (i.e. a secondary C β carbon) to yield reversible cycloelimination. Indeed, crotonic acid readily forms irreversible adducts with (evolved) Fdc that proceed to the last step prior to product formation. Detailed studies of the AnFdc mechanism revealed considerable enzyme-induced strain in substrate-cofactor adducts that avoid dead-end local energy minima during the covalent catalysis 17 . In the case of smaller substrates such as (3-methyl) crotonic acid, the scope for enzyme-induced strain as a tool to optimize the energy landscape is minimal. In this case, cycloelimination of isobutene appears feasible at ambient conditions whereas propene production is not. Computational studies provide a rationale behind these observations, suggesting a~2200 fold slower rate for the release of propene from Int3. Thus, further optimization of isobutene production and future evolution of propene producing Fdc variants will need to focus on the energetics of the hydrocarbon elimination step.

Methods
In vivo isobutene assay. In vivo screenings were carried out on a 96-well plate (DW96, 2.2 mL wells, sealed with a foil sheet). TaFdc and other UbiD homologues were co-expressed with UbiX (E. coli, K-12) in a petDuet vector (UbiD in MCS1 and UbiX in MCS2) in E. coli (BL21, DE3). Isobutene production from 0.4 mL reaction mix with 10 mM 3-methylcrotonic acid was detected from the headspace by gas chromatography. The GC method consisted of 100 µL of headspace with a split ratio of 10 injected to RTX-1 column (15 m, 0.32 mm internal diameter, 5 µm film thickness, from RESTEK 10178-111) using nitrogen as a carrier gas (1 mL/min flow rate). The oven temperature was held at 100°C and the injector and detector were maintained at 250°C. Isobutene was calibrated at 1000, 5000 and 10,000 ppm with standards from Messer.
Mutagenesis. Point mutations (TaFdcI, TaFdcII, AnFdcI, AnFdcII, Supplementary Table 6) were generated with a Q5 mutagenesis kit from New England Biolabs. Primers were designed with NEBaseChanger (New England Biolabs). The presence of the point mutation was confirmed by sequencing (Eurofins).
Protein expression. A pETDuet-1 vector containing genes for T. atroviride Fdc (with an N-terminal 6-histidine affinity tag) and UbiX (E. coli, K-12) was transformed into BL21(DE3) competent cells following the manufacturer's protocol (Novagen). A colony was inoculated into Lysogeny Broth (supplemented with 100 µg/mL ampicillin) and incubated by shaking overnight at 37°C. 5 mL of LB culture was inoculated into 1 L of Terrific Broth (TB, Formedium), supplemented with 100 µg/mL ampicillin. The culture was incubated by shaking at 37°C until the optical density of 0.6-0. 8 Supplementary Fig. 7. Source data are provided as a Source Data file.
Crystallization and X-ray structure determination. Crystallization was performed by sitting-drop vapour diffusion. Screening of 0. In vitro isobutene assay comparing TaFdc, ScMVD and PtM3K. An equal amount of E. coli cells containing either empty pETDuet (as control) or one of the following plasmids: pETDuet TaFdc_UbiX, pETDuet TaFdcV_UbiX, pETDuet PtM3K (P. torridus mevalonate 3-kinase, Uniprot: Q6KZB1), pETDuet ScMVD (S. cerevisiae MVD, Uniprot: P32377) or pETDuet PtM3K-ScMVD, were lysed in 50 mM Tris-HCl pH 7.5, 20 mM KCl, 2 mM MgCl 2 , 1 g/L lysozyme, 0.03 g/L DNAse for 1 h at 37°C. A total of 150 μL of lysate was transferred to a 2 mL GCvial and MgCl 2 (10 mM final concentration) was added. Substrates were added to 50 mM final concentration and 200 μL total volume, and consisted of either 3hydroxyisovalerate/ATP, 3-phosphonooxy-isovalerate/ADP or 3-methylcrotonate. Following 4 h of incubation at either 37 or 50°C, the reaction mixture was inactivated by incubation at 90°C for 5 min. GC analysis of the gas phase was carried out as described above to determine isobutene levels produced. All reactions were carried out in duplicates.
DFT calculations. TaFdcV active site cluster model with crotonic (365 atoms) and 3-methylcrotonic acid (368 atoms) was built based on the TaFdcV crystal structure with crotonic acid bound as Int3 adduct (Supplementary Fig. 9) and modelled at the B3LYP/6-31 G(d,p) level of theory with the D3 version of Grimme's dispersion with Becke-Johnson damping and a generic polarizable continuum with ε = 5.7 using the polarizable continuum model 25 . C α -C 1' and C β -C 4a bonds were both fixed for any single DFT optimization and substrate release was modelled using Gaussian 09 revision D.01. by lengthening one bond by 0.05 Å at a time, resulting in a 3D energy landscape consisting of~900 DFT optimized models (for 3-methylcrotonic acid).
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Data supporting the findings of this work are available within the paper and its Supplementary Information files. A reporting summary for this Article is available as a Supplementary Information file. Crystal structure data that support the findings of this    Tables 2 and 3) scheme for 3-methylcrotonic (red) and crotonic acid (blue) with the Int3 set as 0 and the projected approximate transition state denoted with double daggers. D overlay of the DFT optimized transition states between Int3 and product for 3-methylcrotonic (pink, Cα-C 1' and C β -C 4a bond lengths of 1.96 and 2.97 Å, respectively) and crotonic acid (blue, Cα-C 1' and C β -C 4a bond lengths of 1.95 and 2.77 Å, respectively). Source data underlying A-C are provided as a Source Data file.