Bailey et al.

Enzymatic control of cycloadduct conformation ensures reversible 1,3 dipolar cycloaddition in a

prFMN dependent decarboxylaseAtomic description of an enzyme reaction dependent on

reversible 1,3-dipolar cycloaddition

Samuel S. Bailey1, Karl A. P. Payne1, Annica Saaret1, Stephen A. Marshall1, Irina

Gostimskaya1, Iaroslav Kosov1, Karl Fisher1, Sam Hay1*, David Leys1*

1Manchester Institute of Biotechnology, School of Chemistry, University of Manchester, 131 Princess Street, M1 7DN Manchester, UK

* corresponding authors: sam.hay@manchester.ac.uk; david.leys@manchester.ac.uk

The UbiD enzyme plays an important role in bacterial ubiquinone (coenzyme Q)

biosynthesis. It belongs to a family of reversible decarboxylases that require

prenylated flavin (prFMN) to interconvert propenoic or aromatic acids with the

corresponding alkenes or /aromatic compounds using a prenylated flavin

(prFMN) cofactor. Thise azomethine ylide character of this cofactor is suggested

to support (de)carboxylation through a reversible 1,3-dipolar cycloaddition with

dipolarophile substrates, ultimately leadingprocess to (de)carboxylation. We

present Here we report an atomic-level description of the reaction of the UbiD

related ferulic acid decarboxylase with substituted propenoic and propiolic acids

(data ranging from 1.01 to 1.39 Å). The enzyme is only able to couple

(de)carboxylation of cinnamic acid-type compounds to reversible 1,3-dipolar

cycloaddition, while formation of dead-end prFMN cycloadducts occurs with

distinct propenoic and propiolic acids. The active site imposes considerable

strain on covalent intermediates formed with cinnamic and phenylpropiolic

acids. Strain reduction through mutagenesis negatively affects catalytic rates

with cinnamic acid, indicating a direct link between enzyme-induced strain and

catalysis that is supported by computational studies. Our data show Nature is

able to harness and control 1,3-dipolar cycloaddition chemistry by combination

of a unique cofactor with strict mechanochemical control.

1

Bailey et al.

Many enzymes make use of covalent catalysis to achieve substantial rate

enhancements, often by recruiting cofactors such as PLP1, TPP2 and flavins3. To

ensure high turnover, these enzymes are inherently required to ensure both rapid and

reversible cofactor-ligand adduct formation. In the case of the UbiD enzyme family,

reversible decarboxylation has been suggested to occur via a 1,3-dipolar cycloaddition

process between the substrate and the UbiD-cofactor, prenylated FMN (prFMN)4,

enabled by the azomethine ylide character of the latter. . While 1,3-dipolar

cycloaddition is commonly used in organic synthesis5, this is the first example of this

particular process in enzymatic catalysis. Diels-Alder reactions have also recently

been found to be catalysed by dedicated enzymes6, but these are not reported to be

reversible. Similarly, 1,3-dipolar cycloaddition reactions such as biocompatible

‘click’-chemistry are not considered reversible7, although the possibility of force-

induced cycloreversion is a recent and controversial topic8–11. In the case of the fungal

ferulic acid decarboxylase (Fdc) UbiD-enzyme12, 1,3-dipolar cycloaddition between

prFMNiminium and the cinnamic acid substrate (Fig 1) is proposed to lead to an initial

pyrrolidine cycloadduct (Int1) that supports decarboxylation concomitant with ring

Fig 1. A. niger Fdc proposed enzyme mechanism. The conversion of cinnamic acid (in red) by covalent catalysis using the prFMNiminium cofactor (in black) via a series of covalent intermediates is shown. The reaction with the alkyne substrate analogue phenylpropiolic acid is depicted by the additional unsaturated bond (shown in blue). Species for which crystal structures are determined here are underlined.

2

Bailey et al.

opening, resulting in formation of a distinct prFMN-alkene adduct (Int2). A

conserved Glu residue is proposed to donate a proton to the alkene moiety, leading to

a second pyrrolidine cycloadduct (Int3). Finally, the reaction cycle concludes with

cycloelimination of Int3, leading to alkene product formation and release. The

reaction is readily reversible at elevated [CO2] 12. There is no direct experimental

evidence that any of the proposed intermediates exist, although indirect evidence

supporting this mechanism has been reported13. Alternative mechanisms proposed to

occur with distinct forms of the cofactor have been discounted14. However, it remains

unclear how the enzyme is able to ensure reversibility of the 1,3-dipolar cycloaddition

steps. We aimed to determine whether detailed structural insights could be gained into

catalysis by A. niger Fdc1, an enzyme used as a model prFMN decarboxylase by

virtue of the fact it yields atomic resolution crystal structures12 .

Results

Structure of the AnFdc1 substrate complex

3

Bailey et al.

Here we sought to determine whether detailed structural insights could be gained into

catalysis by A. niger Fdc1, an enzyme used as a model prFMN decarboxylase by

virtue of the fact it yields atomic resolution crystal structures12. Catalytically

competent AnFdc1 crystals were made by avoiding illumination of the protein during

sample preparation and crystallisation, preventing photo-induced cofactor

isomerisation and inactivation14. Incubation of crystals with cinnamic acid for brief

periods of time (typically 10-15 seconds) led to rapid sample deterioration, likely due

to styrene accumulation. Electron density obtained with diffraction data from

cinnamic acid soaked crystals (prior to crystal dissolution) unambiguously revealed

the location of the substrate benzyl moiety above the prFMN iminium isoalloxazine ring

system, but revealed density corresponding to multiple conformations of the substrate

acrylic acid moiety (vide infra). In contrast, rapid soaking and flash-cooling of

crystals with alpha-fluorocinnamic acid revealed clear density for the substrate

positioned directly above the prFMNiminium azomethine ylide functionality (i.e. Sub

complex, to 1.1 Å resolution; Fig 2A2a). We could confirm that cinnamic acid binds

to the protein in the same conformation, by using inactive AnFdc1 crystals that

contained FMN rather than prFMNiminium (to 1.26 Å resolution; Supplementary Fig

S1aA). In the case of the alpha-fluorocinnamic acid prFMN complex, the substrate

Cα and Cβ carbons are located directly above the prFMNiminium C1’ and C4a (Fig 1), at

distances of 3.0 Å and 3.4 Å respectively, in a reactive conformation compatible with

the proposed cycloaddition.

The E282Q AnFdc1 variant allows structure determination of the Int2 covalent

adduct

4

Bailey et al.

In order to determine whether covalent adducts (either Int1 or Int2) might accumulate

in the absence of the Glu282 mediated protonation step, we sought to determine the

crystal structure of the inactive AnFdc1 E282Q variant14 in complex with cinnamic

acid and cinnamic acid derivatives. Indeed, UV-Vis spectroscopy of AnFdc1 E282Q

incubated with cinnamic acid leads to formation of a distinct spectral species,

indicative of adduct formation (Supplementary Fig S2aA). The E282Q crystal

structure obtained by co-crystallisation with cinnamic acid reveals a covalent prFMN-

substrate complex, corresponding to the proposed Int2 species following

decarboxylation (to 1.18 Å resolution; Fig 2B2b). The substrate alkene bond retains

the trans configuration but substantially deviates from ideal planar geometry, with a

torsional angle of = ~112° (as opposed to 180° for a planar trans configuration) and

an elongated Cα=Cβ bond length of 1.37 Å. Exchange of the bound cinnamic acid for

alpha-fluorocinnamic acid or pentafluorocinnamic acid readily occurs in crystallo,

Fig. 2. Crystal structures of A. niger Fdc with propionic acid substrates. A. Active site structure of Fdc in complex with alpha-fluorocinnamic acid. B. Omit electron density contoured at 4.5 igma and corresponding model of the cinnamic acid Int2 in the E282Q mutant. C. Overlay between the Int2 E282Q covalent adduct formed and the adduct obtained with phenylpyruvate (PDB 4ZA9) D. Omit electron density contoured at 3 igma and corresponding model of the cinnamic acid Int3 (populated at 30%) trapped by rapid flash-cooling of WT crystals following exposure to cinnamic acid (Fig. S1F shows the product complex for comparison). E. Omit electron density contoured at 4 igma corresponding to Int3crotonic. F. Overlay between the Int3 and Int3crotonic acid covalent adducts formed.

5

Bailey et al.

demonstrating a Sub ⇌ Int2 + CO2 equilibrium exists in E282Q (Supplementary

Fig S1b,cB,C). The alpha-fluorocinnamic acid Int2 structure (to 1.13 Å resolution)

directly demonstrates syn-pyramidalisation15 of the substrate alkene Cα

(pyramidalisation angle of ϕ ~20°), which is likely to be a result of the torsional strain

imposed. The observed alkene distortion is a likely consequence of the firm hold the

enzyme active site exerts on the substrate phenyl moiety, held in place by I187, I327

and F437 at a position directly above and stacked with the prFMN plane. For all

E282Q Int2 structures, the prFMN C1’ is located further out of the prFMN plane by

~0.4 Å when compared to other prFMN single bond adduct species (i.e. PDB 4ZA9;

Fig 2C2c). In the case of cinnamic acid E282Q Int2, this places the Cβ directly above

the prFMN C4a at a distance of 3.04 Å, while the Cα atom is in close proximity (~3.2

Å) of the location occupied by the Glu282 carboxylate moiety in the wild-type

enzyme structure.

Structure of cycloadducts Int1’ and Int3’ formed with alkyne compounds

The substrate and product alkyne analogues phenylpropiolic acid and phenylacetylene

were used to determine whether cycloadducts Int1 and/or Int3, respectively can be

observed in the wild-type enzyme. These compounds can also act as dipolarophiles,

but cycloaddition will lead to the double bond-containing 3-pyrroline Int 1’ rather

than the pyrrolidine cycloadducts (Fig 1). Upon addition of either alkyne compound, a

clear shift is observed in the enzyme UV-Vis spectrum indicating modification of the

prFMNiminium cofactor, with the formation of a species distinct from the E282Q Int2

adduct (Supplementary Fig S2bB). Stopped-flow experiments reveal the formation

of this alkyne intermediate occurs with kobs = 101.5 ± 0.4 s-1 (Supplementary Fig

S3A), but following prolonged incubation no product formation could be observed

(Supplementary Fig S2Cc). To investigate the phenylpropiolic acid complex prior to

reaction (i.e. Inhib complex), inactive crystals of AnFdc1 (with prFMN in a

semiquinone radical form) were used, revealing the inhibitor (to 1.29 Å resolution)

binds in a similar manner to the alpha-fluorocinnamic acid:prFMN or cinnamic acid

substrate:FMN complexes obtained. This positions the alkyne Cα and Cβ directly

above the prFMNradical C1’ and C4a at distances of 3.3 Å and 3.7 Å (Supplementary

Fig S1dD). In contrast, crystals of active AnFdc1 co-crystallised with

phenylpropiolic acid (to 1.01 Å resolution) or soaked with phenylacetylene (to 1.10 Å

6

Bailey et al.

resolution) revealed clear density for a 3-pyrroline cycloadduct (Int3’, analogous to

the proposed Int3) formed between the alkyne moiety and the prFMNiminium (Fig

3A3a). Hence, surprisingly, the phenylpropiolic acid derived adduct lacked density

for the carboxylate moiety, and was highly similar to that obtained with

phenylacetylene (Supplementary Fig S1eE). This suggests decarboxylation has

occurred over the prolonged time frame required for co-crystallisation (typically

exceeding 24 hours). Indeed, when using a rapid soaking procedure as an alternative

approach, a distinct cycloadduct (Int1’; to 1.21 Å resolution) is observed that retains

the carboxyl moiety attached to the prFMN C1’ group (and thus resembles the

proposed Int1; Fig 3bB). We used a mass spectrometry-based detection method to

demonstrate slow decarboxylation occurs within a 24 hour time period in solution,

confirming the conversion of Int1’ to Int3’ mimicking the natural activity of the

enzyme with cinnamic acid occurs (Supplementary Fig S4aA). W

Fig 3. Crystal structures of A. niger Fdc with propiolic acid substrate analogues. A. Omit electron density contoured at 5 igma and corresponding model of the phenylpropiolic acid derived Int3’ obtained through co-crystallisation (Calpha-C1’ 1.52 Å and Cbeta-C4a 1.66 Å bond distances). B. Omit electron density contoured at 4 igma and corresponding model of the

7

Bailey et al.

We then took advantage of the distinct spectral change associated with

cycloadduct formation to determine if cycloaddition with natural substrates can

also be monitored in solution. Indeed, rapid mixing of the enzyme with cinnamic

or sorbic acid substrates revealed spectral changes in the UV-Vis spectrum of

prFMN similar to those observed with the alkyne inhibitors (Supplementary Fig

S3b,cB,C). This suggest that, for these substrates, cycloadduct(s) accumulate

under multiple turnover conditions.

The AnFdc1 active site constrains cycloadduct conformation

While both Int1’ and Int3’ structures confirm 1,3-dipolar cycloaddition can occur

with the prFMNiminium cofactor, several features indicate a significant degree of strain

in these cycloadduct(s). The phenylpropiolic acid Int1’ structure deviates from ideal

bond parameters both in the position of the inhibitor phenyl group (Cβ out-of-phenyl

plane angle is ~24° as opposed to ideal 0°), the deviation of the 3-pyrroline ring (C1’-

Cα-Cβ bond angle of ~99°, as compared to ideal 108°), prFMN tetrahedral C4a bond

angles (C2-C4a-Cβ bond angle of ~103° versus ideal 109.5°of sp3 C4α) as well as the

elongated C4a-Cβ bond length of 1.7 Å (Fig 4A4a). Following decarboxylation, the

position of the Int3’ 3-pyrroline ring shifts (Fig 3cC) to alleviate some of the phenyl

group strain (Cβ out-of-plane angle reduced to ~16°) as well as the deviation of the 3-

pyrroline ring (C1’-Cα-Cβ bond angle of ~111°; Cβ-C4a bond length of 1.65 Å). The

protein does not alter in conformation when comparing the Inhib/Int1’ and Int3’

structures, with the notable exception of Glu282, which reorients to occupy the

substrate/inhibitor carboxylate binding pocket in Int3’ (a motion similar to the CO2

for Glu282 exchange proposed to occur during catalysis, Fig 1). In the latter, Glu282

forms an interaction with the 3-pyrroline substrate-derived Cα (distance ~3.2 Å to

either E282 oxygen; Fig 3aA). Glu282 mediated proton donation or abstraction would

interconvert Int3’ with the putative prFMN C1’ – alkyne adduct Int2’ (the complex

analogous to Int2; Fig 1). However, an ‘unstrained’ linear C1’-alkyne adduct –with a

C1’-Cα≡Cβ angle of 180°– is sterically incompatible with the AnFdc1 active site,

suggesting a highly strained Int2’ (similar to strained cyloalkyne species15) connects

the Int1’ and Int3’ species, providing a likely explanation for the extremely slow rate

of Int1’ decarboxylation.

Mutagenesis alleviates cycloadduct strain and impacts catalysis

Fig 3. Crystal structures of A. niger Fdc with propiolic acid substrate analogues. A. Omit electron density contoured at 5 igma and corresponding model of the phenylpropiolic acid derived Int3’ obtained through co-crystallisation (Calpha-C1’ 1.52 Å and Cbeta-C4a 1.66 Å bond distances). B. Omit electron density contoured at 4 igma and corresponding model of the

8

Bailey et al.

As observed for the Int2 E282Q structures, the strain observed in the Int1’ and Int3’

structures appears to be a direct consequence of the inherent restrictions imposed by

the rigid AnFdc1 active site. The position of the substrate aromatic group appears

constrained to remain π-stacked with the prFMN isoalloxazine ring, by virtue of steric

interactions with the Phe437 sidechain located adjacent to the conserved Leu439. This

in turn imposes restrictions on the extent by which the prFMN isoalloxazine ring

system can deform upon formation of a C4a-Cβ bond with substrate, as the flavin

uracil ring is sandwiched between Ala172 on the si face and the substrate aromatic

group at the re face. We constructed the F437L and L439G variants and determined

the crystal structures in complex with phenylpropiolic acid, as well as the activity

with cinnamic acid. We found both mutations severely compromised enzyme activity

as judged from steady state turnover. In the case of L439G, a substantial increase of

KM was observed, while the F437L variant was only affected in kcat confirming a key

role for Phe437 in catalysis (Supplementary Fig S5b,dB,D). In the corresponding

phenylpropiolic acid adduct crystal structures for both variants, Int1’ decarboxylation

in crystallo was not observed, even following co-crystallisation and prolonged

incubation (up to several weeks) prior to X-ray exposure. This lack of

decarboxylation activity with Int1’ was confirmed in solution (Supplementary Fig

S4bB). In the case of the L439G Int1’ crystal structure (to 1.12 Å resolution), a

modest reorientation of the substrate carboxyl group along with a minor reduction of

the Cβ out-of-plane angle is observed, associated with a corresponding reorientation

of the F437 phenyl ring (Fig 3dD). The void created by the L439G mutation is filled

with a water molecule located within hydrogen bonding distance of the Int1’

carboxylate, reducing the hydrophobicity of the carboxylate binding pocket. In

addition, the L439G void was partially filled by the phenyl group of a second

phenylpropiolic acid molecule, bound at the entrance of the active site. This

observation offers a possible explanation for the drastic increase of the observed KM

for cinnamic acid in the L439G (as opposed to F437L) mutant, as similar binding of a

second cinnamic acid molecule could be required for decarboxylation by L439G. The

hydrophobic nature of the carboxylic acid binding pocket has been implicated in the

mechanism of other decarboxylases16,17. In marked contrast, the F437L mutation does

not affect the carboxylate binding or conformation, but subtly alters the position of the

substrate phenyl moiety (structure obtained to 1.08 Å resolution; Fig 3e,fE,F). This

reorientation, while retaining a strained conformation of the Cβ out-of-plane angle of

9

Bailey et al.

~25°, is associated with elongation of the Cβ-C4a bond to 1.8 Å. It thus appears the

Phe437 side chain counteracts the forces exerted on the prFMN 3-pyrroline ring

through enzyme interactions with the Int1’ carboxylate.

Smaller dipolarophile compounds form dead-end cycloadducts

Unfortunately, other mutations at the Phe437 position that could lead to further

reduction of strain imposed by the 437 side-chain led to variants that did not bind the

Fig. 4. The role of strain in the Fdc reaction. A. Schematic overview of the structural insights gained into the Fdc reaction cycle. The prFMN iminium cofactor and the substrate are shown for the various crystal structures of the Sub/Int1’ (as a model for Int1)/Int2 and Int3 states. The active site restricts the position of the substrate phenyl ring throughout the reaction (as indicated by black arrow labelled F437), leading to formation of strained intermediates Int1/Int2 and Int3. Substantial deviations from ideal bond lengths and angles are highlighted in black. For the Int1’/Int2 and Int3 species, DFT models of the corresponding prFMN adducts free in solution are overlaid (coloured in grey). The differences observed for Int1’ and Int3 with respective DFT models of the corresponding unconstrained prFMN adducts resembles those observed when comparing Int1’ and Int3 with Int1butynoic and Int3crotonic (Figs 3H and 2F).

10

Bailey et al.

prFMN cofactor. We therefore co-crystallised the WT enzyme with 2-butynoic acid,

avoiding the enzyme imposed adduct strain by interactions with the missing substrate

phenyl moiety. In this case, an Int1’butynoic structure was obtained to 1.02 Å resolution,

revealing no decarboxylation had occurred (Fig 3gG). Furthermore, the Int1’butynoic is

markedly different from the strained phenylpropiolic acid derived Int1’ structure (Fig

3hH). Little to no features associated with a strained Int1’ configuration can be

observed in the case of Int1’butynoic, a consequence of repositioning of the prFMNiminium

C4a and associated uracil plane. This repositioning leads to a near ideal conformation

of the Int1’butynoic 3-pyrroline ring, with a reduced Cβ-C4a bond length of 1.56 Å and

a C1’-Cα-Cβ bond angle of ~107°, but is incompatible with the presence of a

substrate phenyl group held within the context of either the WT or F437L variant

structures. As a consequence of the uracil plane repositioning, the prFMN O1 has

moved out of the oxyanion binding-pocket formed by Q191 and two water molecules.

Decarboxylation of Int1’, and by similarity Int1, is likely assisted by the relatively

hydrophobic nature of the carboxylate binding pocket, and ring opening of the

strained cycloadduct to Int2’/Int2. It is however unclear what drives the

cycloelimination step from Int3 to product release, a process not observed for Int3’.

Careful analysis of electron density associated with rapidly flash-cooled WT crystals

incubated with cinnamic acid reveals this can be modeled as a mixture of 30% Int3

and a 40% product complex (the remainder corresponding to the inactive

hydroxylated prFMN species), suggesting cycloelimination is the rate limiting step

under these conditions, as has previously been suggested13 (to 1.24 Å resolution; Fig

2dD). As observed with the Int3’ structure, the Int3 Cβ has an out-of-plane angle of

~13°, with additional strain observed for the Cα-Cβ-Cγ angle of ~122°, both

associated with the tight hold the enzyme exerts on the substrate phenyl moiety and

resulting in an C1’ endo envelope conformation of the pyrrolidine ring (as opposed to

the N endo conformation of the Int3’ enforced by the Cα-Cβ double bond (Fig 4aA).

Hence, to determine how Int3 strain might assist cycloelimination, we determined the

structure of an Int3crotonic species (to 1.39 Å resolution), obtained by co-crystallisation

with crotonic acid (2-butenoic acid; Fig 2eE). A comparison of Int3crotonic with the

cinnamic acid Int3 (Fig 2fF) reveals the Int3crotonic pyrrolidine ring adopts a distinct

envelope conformation with the N in endo position, that allows the prFMNiminium C1’

to adopt a more in-plane position, resembling the Int3’ geometry. The Int3crotonic

11

Bailey et al.

conformation cannot be adopted by the cinnamic acid Int3, as severe clashes between

the substrate phenyl ring and the active site occur. We suggest the lack of strain in the

Int3crotonic species is coupled to the fact no cycloelimination can be observed for this

species, while a strained conformation such as that observed for cinnamic acid Int3 is

required for progression along the catalytic cycle.

Combining the various crystal structures of Sub/Inhib, Int1’/Int2 and Int3’

complexes with our solution data leads to a comprehensive overview of the AnFdc1

catalytic cycle (Fig 4A). The rigid AnFdc1 active site brings the substrate into close

proximity of the prFMNiminium cofactor, aligning the Cα-Cβ double bond (or triple

bond in case of alkyne inhibitors) with the C1’-N5-C4a azomethine ylide to poise the

complex for cycloaddition. Formation of a pyrrolidine adduct between a similar

putative encounter complex of cinnamic acid and free prFMN iminium in solution would

be accompanied by repositioning of both the Cα-Cβ substituents (i.e.

carboxylate/benzyl group), as well as the prFMN uracil moiety. This is a direct

consequence of the ring formation and the accompanying sp2 to sp3 change for

substrate Cα/Cβ and prFMN C1’/C4a carbon atoms. However, within the confines of

the enzyme active site, the extent to which reorientation can occur is limited. In fact,

the enzyme active site appears evolved to have maximum complementarity with the

substrate and product complexes, and hence constrains both the substrate carboxylate

and phenyl moieties to adopt a position resembling the substrate complex throughout

catalysis. In the case of the Int1’butynoic and Int3crotonic adducts, the lack of a phenyl

moiety allows for a more relaxed conformation of the substrate/prFMNiminium adduct

within the confines of the active site. Crucially, neither Int1’butynoic decarboxylation

nor Int3crotonic cycloelimination are observed suggesting that strain contributes to

catalysis for both these steps.

Computational studies provide further insight into the AnFdc1 catalytic cycle.

We used computational studies to further our understanding of the AnFdc1 catalytic

cycle. An active ‘cluster’ site model comprising ~ 230 atoms was built from the

E282Q Int2 X-ray crystal structure. The positions of peripheral atoms were fixed to

maintain a crystal structure-like geometry, in addition to constraining the position of

the F437 and L439 side chains in most calculations to allow their contribution to

strain in the adducts to be investigated. Stable models of each species as identified in

12

Bailey et al.

Fig 1 were found for cinnamic acid, phenylpropiolic acid and crotonic acid. In the

case of cinnamic acid, cycloaddition (Sub → Int1) was found to occur as a 2-step

reaction, via a ring-open covalent adduct (Int1open), which possesses an unusual Cβ–

C4a interaction (2.60 Å) in addition to a Cα–C1’ bond (1.58 Å). The barrier to

formation of the ring-closed Int1 intermediate is small (Fig 4B5), so the reaction may

effectively occur as one asynchronous step. In contrast, no stable Int1open-type

intermediates were observed for phenylpropiolic or crotonic acid cycloaddition, which

also leads to significantly more stable Int1 cycloadducts.

Cinnamic acid decarboxylation occurs from the ring-open Int1open species and the

Int1open → Int2CO2 potential barrier is relatively low at ~ 16 kJ mol-1 (Fig 4B5). The

barriers for phenylpropiolic and crotonic acid decarboxylation were not determined,

but may be much higher than for cinnamic acid and/or occur via a different

mechanism, as these reactions cannot proceed via an Int1open intermediate and the

decarboxylation of these is then significantly endothermic. In all cases, the Int2CO2,

which contain noncovalently bound CO2, remain ring-open with broken (~3 Å) Cβ-

C4a bonds. The crotonic acid derived Int2CO2 is most stable as it can adopt an

unstrained ring-open conformation with near-ideal Cα–Cβ geometry. In contrast, both

cinnamic acid and phenylpropiolic Int2CO2 intermediates adopt strained conformations

with non-linear Cα–Cβ–Phenyl angles. Relaxation of the side chain constraints of

Phe437 and Leu439, as a proxy of the F437L and L439G variants, relieves ~14 kJ

mol-1 of strain from the cinnamic acid derived Int2CO2 adduct consistent with the role

of these residues in constraining the geometry of the phenyl moiety of the adducts.

Once the CO2 has vacated the active site, Glu282 mediated protonation of Cβ leads to

the formation of a ring-closed Int3 species for all compounds. The release of CO2 and

Cβ protonation were not studied in detail, as these reactions require substantial

rearrangement of Glu282, which would require much larger models. Cycloelimination

(Int3 → Prod) to form non-covalently bound product complexes is exothermic and

likely to be rate limiting in all three cases (Fig 4B5). Formation of styrene is the least

exothermic (38 kJ mol-1), while cycloelimination of phenylacetylene is highly

unfavorable, being 81 kJ mol-1 uphill and likely proceeding via a barrier > 100 kJ

mol-1 (not determined). Similarly, the barrier to propylene cycloelimination is

estimated at ~100 kJ mol-1. This is in agreement with our observation that turnover

13

Bailey et al.

does not occur with either crotonic or phenylpropiolic acid, with the reaction stalling

at the Int3 stage. Crucially, in the case of cinnamic acid, the cycloelimination barrier

was determined to be feasible, at 64 kJ mol-1 and 78.5 kJ mol-1 in the constrained and

F437/L439-relaxed models respectively. The 14.4 kJ mol-1 difference is at least

qualitatively consistent with the ~30 fold decrease in kcat observed in the F439L

variant (which retains some element of strain).

Fig 4 B. Potential energy diagram of the Fdc reaction determined using DFT cluster models (Figs S6,S7), with potential energy normalised to the Sub/Inhib and Int3 energies. The relaxed cinnamic acid energies were determined for models where restraints on the side chains of Phe437 and Leu439 were removed, which alleviates some of the strain on the Cβ–Ph bond (Fig. S8). The two phenylpropiolic acid Int1 species interconvert via rotation of the Cα–CO2 bond and the conformation with the carboxyl group in the phenyl plane is the higher energy species and is not stable in the DFT model used here. The crotonic acid Int3 → Prod transition state actually represents a minimum energy crossing point (Fig. S9).

14

Bailey et al.

Discussion

In the case of the Fdc branch of the UbiD family, our data unambiguously establishes

for the first time that the enzyme relies on covalent catalysis by the unusual prFMN

cofactor through a reversible 1,3-dipolar cycloaddition process. The fact the active

site is highly complementary in shape to the substrate/prFMNiminium complex prior to

cycloaddition has the serendipitous consequence it does not yield to allow pyrrolidine

ring substituents to adopt the ideal conformation. This results in the formation of

strained intermediates, ensuring rapid progression through the catalytic cycle. While

substrate distortion/ground state destabilization has long been suggested to contribute

to enzyme catalysis18,19, it has only been conclusively demonstrated in a limited

number of cases20,21, and rarely includes atomic resolution insights22–25. Our study

reveals in AnFdc1 distortion occurs specifically for covalent substrate-cofactor

adducts only, and the strain is directly linked to progression along the reaction

coordinate. Hence, by targeted destabilization of intermediate species only, AnFdc1

harnesses 1,3-dipolar cycloaddition as a readily reversible reaction. This aligns with

the proposals from Albery and Knowles26 for evolution of an efficient enzyme by

control of the internal thermodynamics of the bound states. Interestingly, many

members of the UbiD family act directly on aromatic substrates, and can act in the

carboxylative direction as suggested for the anaerobic degradation of benzene27,28.

(De)carboxylation of aromatic substrates presents a large inherent barrier to

cycloaddition due to transient substrate dearomatisation, as well as scope for

formation of a highly stable Int2 adduct by re-aromatisation29. How these problems

are overcome, particularly in view of whether the prFMN-dependent catalysis in these

enzymes is similar to the Fdc branch, awaits a more detailed analysis of additional

members of the UbiD family. Finally, our data suggests mechanochemical reversion

of 1,3-dipolar cycloaddition reactions is feasible, providing further support to the

possibility of developing mechanoresponsive materials through reversible click-

chemistry8–11. Our data show Nature is able to harness and control 1,3-dipolar

cycloaddition chemistry by combination of a unique cofactor with strict

mechanochemical control.

Acknowledgements

15

Bailey et al.

This work was supported by the grants BBSRC BB/K017802 and ERC pre-FAB

695013. The atomic coordinates and structures factors have been deposited to the

Protein Data Bank (http://www.pdb.org). The authors acknowledge the assistance

given by the use of the Manchester Protein Structure Facility. We thank Diamond

Light Source for access (proposal numbers MX12788 and MX17773) that contributed

to the results presented here. D.L. is a Royal Society Wolfson Merit Award holder.

Correspondence and material requests should be directed to David Leys

(david.leys@manchester.ac.uk).

Author cContributions

SSB cloned, expressed and purified AnFdc1 and various variants. SSB determined all

crystal structures with guidance from DL. KAPP generated various AnFdc1 variants,

and collected solution data with cinnamic acid. AS assisted with crystallisation of the

Int3crotonic state. SAM assisted with AnFdc1 reconstitution. IG assisted with

crystallisation of FMN substitued AnFdc1 and the Int3butynoic state (with help from

IK). KF assisted with protein purification and solution data collection and analysis.

SH performed all computational studies. All authors discussed the results and

participated in writing of the manuscript. DL initiated and directed this research.

Competing financial interests

The authors declare no competing financial interests.

References

1. Toney, M. D. Controlling reaction specificity in pyridoxal phosphate enzymes.

Biochim. Biophys. Acta - Proteins Proteomics 1814, 1407–1418 (2011).

2. Kluger, R. & Tittmann, K. Thiamin diphosphate catalysis: Enzymic and

nonenzymic covalent intermediates. Chem. Rev. 108, 1797–1833 (2008).

3. Piano, V., Palfey, B. A. & Mattevi, A. Flavins as Covalent Catalysts: New

Mechanisms Emerge. Trends Biochem. Sci. 42, 457–469 (2017).

4. Marshall, S. A., Payne, K. A. P. & Leys, D. The UbiX-UbiD system: The

biosynthesis and use of prenylated flavin (prFMN). Arch. Biochem. Biophys.

632, 209–221 (2017).

5. Pellissier, H. Asymmetric 1,3-dipolar cycloadditions. Tetrahedron 63, 3235–

3285 (2007).

16

Bailey et al.

6. Jeon, B.-S. et al. Investigation of the mechanism of the SpnF-catalyzed [4+2]-

cycloaddition reaction in the biosynthesis of spinosyn A. Proc. Natl. Acad. Sci.

U. S. A. 114, 10408–10413 (2017).

7. Meldal, M. & Tornøe, C. W. Cu-Catalyzed Azide−Alkyne Cycloaddition.

Chem. Rev. 108, 2952–3015 (2008).

8. Jacobs, M. J., Schneider, G. & Blank, K. G. Mechanical Reversibility of Strain-

Promoted Azide-Alkyne Cycloaddition Reactions. Angew. Chemie Int. Ed. 55,

2899–2902 (2016).

9. Khanal, A., Long, F., Cao, B., Shahbazian-Yassar, R. & Fang, S. Evidence of

Splitting 1,2,3-Triazole into an Alkyne and Azide by Low Mechanical Force in

the Presence of Other Covalent Bonds. Chem. - A Eur. J. 22, 9760–9767

(2016).

10. Krupička, M., Dopieralski, P. & Marx, D. Unclicking the Click: Metal-Assisted

Mechanochemical Cycloreversion of Triazoles Is Possible. Angew. Chemie Int.

Ed. 56, 7745–7749 (2017).

11. Stauch, T. & Dreuw, A. Force-induced retro-click reaction of triazoles

competes with adjacent single-bond rupture. Chem. Sci. 8, 5567–5575 (2017).

12. Payne, K. a. P. et al. New cofactor supports α,β-unsaturated acid

decarboxylation via 1,3-dipolar cycloaddition. Nature 522, 497–501 (2015).

13. Ferguson, K. L., Eschweiler, J. D., Ruotolo, B. T. & Marsh, E. N. G. Evidence

for a 1,3-Dipolar Cyclo-addition Mechanism in the Decarboxylation of

Phenylacrylic Acids Catalyzed by Ferulic Acid Decarboxylase. J. Am. Chem.

Soc. 139, 10972–10975 (2017).

14. Bailey, S. S. et al. The role of conserved residues in UbiD/Fdc decarboxylase

in oxidative maturation, isomerisation and catalysis of prenylated flavin

mononucleotide. J. Biol. Chem. 293, 2272–2287 (2018).

15. Borden, W. T. Pyramidalized Alkenes. Chem. Rev. 89, 1095–1109 (1989).

16. Okrasa, K. et al. Structure and Mechanism of an Unusual Malonate

Decarboxylase and Related Racemases. Chem. - A Eur. J. 14, 6609–6613

(2008).

17. Jez, J. M., Ferrer, J. L., Bowman, M. E., Dixon, R. A. & Noel, J. P. Dissection

of malonyl-coenzyme A decarboxylation from polyketide formation in the

reaction mechanism of a plant polyketide synthase. Biochemistry 39, 890–902

(2000).

17

Bailey et al.

18. Blake, C. C. et al. Crystallographic studies of the activity of hen egg-white

lysozyme. Proc. R. Soc. London. Ser. B, Biol. Sci. 167, 378–88 (1967).

19. Vocadlo, D. J., Davies, G. J., Laine, R. & Withers, S. G. Catalysis by hen egg-

white lysozyme proceeds via a covalent intermediate. Nature 412, 835–838

(2001).

20. Anderson, V. E. Quantifying energetic contributions to ground state

destabilization. Arch. Biochem. Biophys. 433, 27–33 (2005).

21. Zhang, Y. & Schramm, V. L. Ground-State Destabilization in Orotate

Phosphoribosyltransferases by Binding Isotope Effects. Biochemistry 50,

4813–4818 (2011).

22. Lehwess-Litzmann, A. et al. Twisted Schiff base intermediates and substrate

locale revise transaldolase mechanism. Nat. Chem. Biol. 7, 678–684 (2011).

23. Milić, D. et al. Crystallographic snapshots of tyrosine phenol-lyase show that

substrate strain plays a role in C-C bond cleavage. J. Am. Chem. Soc. 133,

16468–16476 (2011).

24. Vladimirova, A. et al. Substrate Distortion and the Catalytic Reaction

Mechanism of 5-Carboxyvanillate Decarboxylase. J. Am. Chem. Soc. 138,

826–836 (2016).

25. Lüdtke, S. et al. Sub-ångström-resolution crystallography reveals physical

distortions that enhance reactivity of a covalent enzymatic intermediate. Nat.

Chem. 5, 762–767 (2013).

26. Albery, W. J. & Knowles, J. R. Efficiency and Evolution of Enzyme Catalysis.

Angew. Chemie Int. Ed. English 16, 285–293 (1977).

27. Abu Laban, N., Selesi, D., Rattei, T., Tischler, P. & Meckenstock, R. U.

Identification of enzymes involved in anaerobic benzene degradation by a

strictly anaerobic iron-reducing enrichment culture. Environ. Microbiol. 12,

2783–2796 (2010).

28. Luo, F. et al. Metatranscriptome of an anaerobic benzene-degrading, nitrate-

reducing enrichment culture reveals involvement of carboxylation in benzene

ring activation. Appl. Environ. Microbiol. 80, 4095–4107 (2014).

29. Baunach, M. & Hertweck, C. Natural 1,3-dipolar cycloadditions. Angew.

Chemie - Int. Ed. 54, 12550–12552 (2015).

18

Bailey et al.

19

Bailey et al.

Figure Legends

Fig 1. A. niger Fdc proposed enzyme mechanism. The conversion decarboxylation

of cinnamic acid (in red) or the alkyne substrate analogue phenylpropiolic acid (in

blue) by covalent catalysis using the prFMNiminium cofactor (in black) via a series of

covalent intermediates is shown. Following the binding step, a 1,3-dipolar

cycloaddition between the dipolarophile ligand and the azomethine ylide cofactor

leads to the first cycloadduct species Int1. TThe reaction with the alkyne substrate

analogue phenylpropiolic acid is depicted by the additional unsaturated bond (shown

in blue highlights the reaction with the alkyne substrate analoguee). Decarboxylation

occurs concomitant with ring opening and formation of Int2. The bound CO2 leaves

the active site and is replaced by the E282 side chain (shown in green). Protonation

via E282 leads to formation of a second cycloadduct species Int3. A cycloelimination

process leads to formation of styrene and the entire reaction cycle is freely reversible.

The labels for sSpecies for which crystal structures are determined here are

underlined.

Fig. 2. Crystal structures of A. niger Fdc with propionic acid substrates. Aa.

Active site structure of Fdc in complex with alpha-fluorocinnamic acid. Bb. Omit

electron density contoured at 4.5 igma and corresponding model of the cinnamic

acid Int2 in the E282Q mutant. Cc. Overlay between the Int2 E282Q covalent adduct

formed and the adduct obtained with phenylpyruvate (PDB 4ZA9) Dd. Omit electron

density contoured at 3 igma and corresponding model of the cinnamic acid Int3.

(populated at 30%) trapped by rapid flash-cooling of WT crystals following exposure

to cinnamic acid (Supplementary Fig. S1F shows the product complex for

comparison). Ee. Omit electron density contoured at 4 igma corresponding to

Int3crotonic. Ff. Overlay between the Int3 and Int3crotonic acid covalent adducts formed.

Fig 3. Crystal structures of A. niger Fdc with propiolic acid substrate analogues.

Aa. Omit electron density contoured at 5 igma and corresponding model of the

phenylpropiolic acid derived Int3’ obtained through co-crystallisation (Calpha-C1’

1.52 Å and Cbeta-C4a 1.66 Å bond distances). Bb. Omit electron density contoured at

4 igma and corresponding model of the phenylpropiolic acid derived Int1’ obtained

through soaking and rapid flash-cooling (Calpha-C1’ 1.59 Å and Cbeta-C4a 1.70 Å

20

Bailey et al.

bond distances). Cc. Overlay between the phenylpropiolic acid derived Int1’ and

Int3’ covalent adducts formed. Dd. Omit electron density contoured at 4 igma and

corresponding model of the phenylpropiolic acid derived Int1’ for the L439G variant

(Calpha-C1’ 1.48 Å and Cbeta-C4a 1.86 Å bond distances). Ee. Omit electron density

contoured at 4 igma and corresponding model of the phenylpropiolic acid derived

Int1’ for the F437L variant (Calpha-C1’ 1.51 Å and Cbeta-C4a 1.79 Å bond

distances). Ff. Overlay between the Int1’ adduct obtained in the WT and the F437L

variant. Gg. Omit electron density contoured at 4 igma and corresponding model of

the 2-butynoic acid derived Int1’butynoic (Calpha-C1’ 1.52 Å and Cbeta-C4a 1.53 Å

bond distances). Hh. Overlay between the Int1’ and Int1’butynoic adduct.

Fig. 4. The role of strain in the Fdc reaction. A. Schematic overview of the

structural insights gained into the Fdc reaction cycle. The prFMN iminium cofactor and

the substrate are shown for the various crystal structures of the Sub/Int1’ (as a model

for Int1)/Int2 and Int3 states. The active site restricts the position of the substrate

phenyl ring throughout the reaction (as indicated by black arrow labelled F437),

leading to formation of strained intermediates Int1/Int2 and Int3. Substantial

deviations from ideal bond lengths and angles are highlighted in black. For the

Int1’/Int2 and Int3 species, DFT models of the corresponding prFMN adducts free in

solution are overlaid (coloured in grey). The differences observed for Int1’ and Int3

with respective DFT models of the corresponding unconstrained prFMN adducts

resembles those observed when comparing Int1’ and Int3 with Int1butynoic and

Int3crotonic (Figs 3hH and 2fF).

Fig. 5. B. Potential energy diagram of the Fdc reaction. Potential energy diagram

determined using DFT cluster models (Supplementary Figss S6,S7), with potential

energy normalised to the Sub/Inhib and Int3 energies. The relaxed cinnamic acid

energies were determined for models where restraints on the side chains of Phe437

and Leu439 were removed, which alleviates some of the strain on the Cβ–Ph bond

(Supplementary Fig. S8). The two phenylpropiolic acid Int1 species interconvert via

rotation of the Cα–CO2 bond and the conformation with the carboxyl group in the

phenyl plane is the higher energy species and is not stable in the DFT model used

here. The crotonic acid Int3 → Prod transition state actually represents a minimum

energy crossing point (Supplementary Figs. S9, 10).

21

Bailey et al.

22

Bailey et al.

Online Methods

Cloning - The A. niger fdc gene was cloned as reported previously12,14.

Mutagenesis - Mutagenesis primers were designed using the NEBasechanger tool.

PCR, Dpn1 digest and ligation was performed using the Q5 Site-Directed

Mutagenesis kit (New England Biolabs). The PCR product was transformed into

E. coli NEB5 or Stellar (Clontec). Presence of the desired mutation wasα

confirmed by sequencing (Eurofins). The plasmid was then co-transformed with

the corresponding UbiX construct into E. coli BL21 (DE3).

Protein Expression - Protein was expressed in E. coli BL21 (DE3) grown in LB

media, supplemented with 50 g mLμ -1 ampicillin and 50 g mLμ -1 kanamycin at 37

°C, 180 rpm until mid log phase. The cultures were then cooled to 15 °C,

supplemented with 1 mM MnCl2, induced using 0.25 mM IPTG (Formedium) and

grown overnight. Cells were harvested by centrifugation (4 °C, 7,000 g, 10 min).

Protein Purification - Cell pellets were re-suspended in Buffer A (50 mM Tris, 200

mM NaCl pH 7.5 in Milli-Q water) supplemented with complete EDTA-free

protease inhibitor cocktail (Roche), lysozyme, DNase and RNase (Sigma). The

cells were lysed on ice using a Bandelin Sonoplus sonicator with a TT13/F2 tip,

set to 30% power, 20 s on 40 s off for 30 minutes. Cellular debris was removed

by ultra-centrifugation using a Ti50.2 rotor in a Beckman optima LE-80k

ultracentrifuge at 40k rpm for 1 hour at 4 °C. The supernatant was then passed

through a 0.45 m filter. The clarified supernatant was applied to a gravity flowμ

Ni-NTA agarose column (Qiagen). The column was washed with three column

volumes of Buffer A supplemented with 10 mM imidazole, then three column

volumes of Buffer A supplemented with 40 mM imidazole. Protein was eluted in

1 mL fractions using Buffer A supplemented with 250 mM imidazole. Fractions

containing the purified protein were buffer exchanged into Buffer B (20 mM Tris,

100 mM NaCl, pH 7.5 in Milli-Q water) using a 10DG column (BioRad). Protein

aliquots were flash frozen until required. Throughout purification light was

23

Bailey et al.

excluded as much as possible by using black plastic ware and foil wrapped

glassware.

UV-vis spectroscopy - UV-visible absorbance spectra were recorded using a Cary

50 Bio spectrophotometer (Varian). Protein concentrations were calculated

using ε280nm = 63,830 M-1 cm-1 for A. niger Fdc1 variants. All spectra have been

normalized for protein content.

UV-vis decarboxylation assays - The initial rate of decarboxylation was

determined by following consumption of substrate by UV-vis spectroscopy using

a Cary 50 Bio spectrophotometer (Varian). Assays were performed against

various concentrations of substrate in 350 L 50 mM KCl, 50 mM NaPi pH 6 in aμ

1 mm path length cuvette at 25 °C. The rate of cinnamic acid consumption was

measured at 270 nm. The extinction coefficient for cinnamic acid is ε270nm =

20000 M-1 cm-1, the extinction coefficient for styrene is ε270nm = 200 M-1 cm-1.

The rate of cinnamic acid consumption was calculated using Δε270nm. Protein

concentration was determined using A280, all kcat values are apparent due to

variations in prFMN content.

Stopped-flow – All stopped flow experiments were carried out at 4 °C using an

Applied Photophysics stopped flow instrument fitted with a diode array detector.

Spectra were recorded for 0.5 s to minimise light induced spectral effects. To

obtain the data for A. niger Fdc1 vs. phenylpropiolic acid a final concentration of

50 M Fdc1 and 50 mM phenylpropiolic acid was used. To obtain the data for μ A.

niger Fdc1 vs. cinnamic acid and A. niger Fdc1 vs. sorbic acid a final

concentration of 450 M Fdc1 and 1 mM of the corresponding acid was used. Allμ

data were reduced and analysed using Pro-KIV (Applied Photophysics.

Protein crystallization and structure determination - Crystallization was

performed by sitting drop vapor diffusion. An initial screening of 0.3 L 14mgμ

mL-1 A. niger Fdc1 in 100 mM NaCl, 25 mM Tris pH 7.5 and 0.3 L reservoirμ

solution at 4 °C resulted in a number of hits including PACT condition F4

24

Bailey et al.

(Molecular Dimensions). Seed stock produced from these crystals was used in an

optimization screen based around 0.2 M potassium thiocyanate, Bis-Tris propane

pH 6.5, 20% w/v PEG 3350 by mixing 0.05 L seed stock, 0.25 L proteinμ μ

solution and 0.3 L reservoir solution at 4 °C. Crystals were cryoprotected inμ

reservoir solution supplemented with 10% PEG 200 and flash cooled in liquid

nitrogen. All co-crystals were obtained by incubating protein with 1 mM ligand

prior to the crystallisation experiment, with the exception of the 2-butynoic acid

co-crystallisation which was carried out with 10 mM ligand. For ligand soaks the

cryoprotectant solution was supplemented with 100 mM ligand. Diffraction data

were collected at Diamond beamlines and processed using the CCP4 suite30.

Molecular replacement was undertaken in Phaser MR31 using 4ZA4 as a model;

further refinement was carried out using REFMAC532 and manual rebuilding in

COOT33. Ligand coordinates and definitions were generated using AceDRG34. All

crystallographic data collection and refinement statistics are given in Table S1.

To obtain the structure of Fdc1E282Q in complex with cinnamic acid cells

expressing the Fdc1E282Q gene were supplemented with 1 mM cinnamic acid 2

hours after induction. A total of 1 mM cinnamic acid was also present in the

buffers used during protein purification.

Production and crystallisation of FMN containing A. niger Fdc1 – Cells expressing

A. niger fdc1 but not E. coli ubiX were resuspended and lysed in Dulbecco’s

phosphate buffered saline (PBS) (Sigma) supplemented with complete EDTA-

free protease inhibitor cocktail (Roche), lysozyme, DNase and RNase (Sigma).

Cellular debris was removed as stated above and the clarified supernatant was

applied to a gravity flow Ni-NTA column. The column washed with PBS followed

by 2M KBr (2 column volumes), then 50 mM Hepes 500 mM KCl pH 7.0. A final

wash was carried out using 20 mM Hepes, 100 mM KCl. Protein was eluted in 20

mM Hepes, 100 mM KCl, 250 mM imidazole. The protein was then buffer-

exchanged using a 10DG column (Biorad) into 20 mM Hepes 100 mM KCl 1 mM

MnCl2 pH 7.0. FMN was added to a protein:FMN ratio of 1:4 and crystallised as

described above.

25

Bailey et al.

HPLC decarboxylation assays - Assays containing 10 mM propiolic acid, 50 mM

KCl, 50 mM NaPi pH 6, were incubated for 2 hours at 25 °C with 50 M enzyme.μ

100 L of sample was added to 900 L 50% v/v Hμ μ 2O/acetonitrile with 0.1%

trifluoroacetic acid (TFA). Sample analysis was performed using an Agilent 1100

Series HPLC equipped with a UV detector. The stationary phase was a Kinetex 5

m C18 column, 250 X 4.6 mm. The mobile phase was acetonitrile/ Hμ 2O (50/50),

with 0.1% TFA. Detection was performed at 254 nm.

Cofactor Mass Spectrometry - Samples were prepared by a desalting of proteins

into 100 mM ammonium acetate, pH 7.0 using a PD Minitrap G-25 column (GE

healthcare). A 1200 series Agilent LC was used to inject 5 μ L of sample into 5%

acetonitrile (0.1% formic acid) and desalted inline to release the cofactor from

the enzyme complex. This was eluted over 1 min by 95% acetonitrile. The

resulting ions were analyzed by an Agilent QTOF 6510 run in positive mode and

deconvoluted using Agilent Masshunter Software.

Protein Mass Spectrometry - All experiments were performed using nano-

electrospray ionisation (nESI) in positive ionisation mode. Mass spectra were

recorded on a (U)HMR Q-Exactive plus. The instrument parameters were set to :

Capillary voltage 1.2-1.5 kV; SID cone 0 V; S-Lens RF 200 V; Source Temperature

80 C; HCD cell pressure 7.0; Resolution 8750; Injection time 1 ms. HCD activation

energy was ranged from 0 V to 50 V and performed using nitrogen as collision

gas. Mass spectra were deconvoluted using UniDec software1. Errors in the

calculated experimental masses were obtained from the full-width half maxima

(FWHM).

DFT Calculations – Models of prFMN adducts in solution (Fig 4A) were modelled

at the B3LYP/6-311++G(d,p) level of theory with the D3 version of Grimme’s

dispersion with Becke-Johnson damping35 and implicit water solvation using the

Polarizable Continuum Model (PCM). Active site “cluster” models comprising ~

230 atoms were built from the E282Q X-ray crystal structure and modelled at the

B3LYP/6-31G(d,p) level of theory with the D3 version of Grimme’s dispersion

with Becke-Johnson damping35 and a generic polarizable continuum with = 5.7ε

26

Bailey et al.

using PCM. In all cases, calculations were performed using Gaussian 09 rev d36

and the models and additional details are given in Figs S6-S9. The positions of

peripheral atoms in the active site models were fixed to maintain a crystal

structure-like geometry and the positions of the side chains of Phe437 and

Leu439 were also restrained in most calculations for similar reasons. Three

substrates/inhibitors were studied: cinnamic acid, phenylpropiolic acid, and

crotonic acid and stable conformations of each species in Fig 4 were found for all

three reactions. Conformational sampling of the models was performed by

manually modifying unrestrained bond angles and dihedrals before performing

geometry optimization. This identified a number of local minima for many of the

presented species, with the lowest energy species presented here. Approximate

transition states were obtained for the cinnamic acid Sub → Int1open, Int1open →

Int 1, Int1open → Int2CO2 and Int3 → Prod steps by taking the highest-energy

species identified during relaxed potential energy scans (Fig S9). The crotonic

acid Int3crotonic → Prod step was similarly studied and no transition state was

identified. Instead the minimum energy crossing point is presented in Fig 4B. See

Supplementary Information for methods.

Data availability

The data generated and analysed in this study, including the computational

modeling data associated with all figures, are available from the corresponding

authors upon reasonable request. The diffraction data and corresponding

atomic models have been deposited in the PDB under accession codes 6R3G,

6R3F, 6R3I, 6R2Z, 6R3Q, 6R3O, 6R2T, 6R2R, 6R3N, 5R3L, 6R3J, 6R2P, 6R34,

6R33 and 6R32 (to be released prior to publication) .

27

Bailey et al.

Online method references

30. Winn, M. D. et al. Overview of the CCP4 suite and current developments.

Acta Crystallogr. Sect. D Biol. Crystallogr. 67, 235–242 (2011).

31. McCoy, A. J. et al. Phaser crystallographic software. J. Appl. Crystallogr. 40,

658–674 (2007).

32. Murshudov, G. N. et al. REFMAC5 for the refinement of macromolecular

crystal structures. Acta Crystallogr. Sect. D Biol. Crystallogr. 67, 355–367

(2011).

33. Emsley, P. & Cowtan, K. Coot: Model-building tools for molecular graphics.

Acta Crystallogr. Sect. D Biol. Crystallogr. 60, 2126–2132 (2004).

34. Long, F. et al. AceDRG: A stereochemical description generator for ligands.

Acta Crystallogr. Sect. D Struct. Biol. 73, 112–122 (2017).

35. S. Grimme, S. Ehrlich and L. Goerigk, “Effect of the damping function in

dispersion corrected density functional theory,” J. Comp. Chem. 32 (2011)

1456-65.

36. M. J. Frisch, et al. Gaussian, Inc., Wallingford CT, 2016.

28

Bailey et al.

Fig S1

29