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

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 (Supplementary Table 1) and a directed evolution approach was taken to generate a variant of TaFdc with superior isobutene production activity and selectivity for 3-methylcrotonic acid over cinnamic acid (schematically presented in Supplementary Fig. 1). TaFdcI, with a T405M mutation, was the first variant with a considerable increase in isobutene. TaFdcV generated by 4 rounds of evolution has 11 mutations: E25N, N31G, G305A, D351R, K377H, P402V, F404Y, T405M, T429A, V445P and Q448W (Table 1).

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 previously¹⁷. 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 AnFdc-mediated decarboxylation¹¹. 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) (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 AnFdc¹⁷. 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 CO₂ 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 wild-type 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.

TaFdcII has comparable isobutene production activity to TaFdcV

Selected TaFdc variants (wild-type, TaFdcI i.e. T405M, TaFdcV, TaFdcII, see Supplementary Fig. 1) and AnFdc variants (wild-type, AnFdcI, AnFdcII) were purified and assayed for isobutene production. TaFdcII (F404Y, T405M, V445P, Q448W) was created by rational design based on the structural analysis of TaFdc wild-type, TaFdcV and AnFdcII. A comparison of the isobutene titre obtained following 2 and 4 h incubation revealed TaFdcI and AnFdcI produced 4–9 times the amount of isobutene compared to the wild-type enzymes. Additional mutations in AnFdcII and TaFdcII led to a substantial further increase of 18 and 81 fold, respectively, in isobutene production (Supplementary Fig. 7). Surprisingly, the in vitro titer obtained with TaFdcII was slightly higher than the corresponding TaFdcV levels obtained (Fig. 5). Thus, the 4 point mutations in TaFdcII (F404Y, T405M, V445P, Q448W) and 3 point mutations in AnFdcII (T395M, R435P, P438W), that create an active site architecture identical to TaFdcV (Fig. 4), appear largely responsible for the increased isobutene activity compared to wild-type TaFdc and AnFdc.

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 phenylacetaldehyde¹¹. 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 a ~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 compounds¹⁸. 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 prFMN-dependent 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 catalysis¹⁷. 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.