Towards a comprehensive understanding of the structural dynamics of a bacterial diterpene synthase during catalysis

Terpenes constitute the largest and structurally most diverse natural product family. Most terpenoids exhibit a stereochemically complex macrocyclic core, which is generated by C–C bond forming of aliphatic oligo-prenyl precursors. This reaction is catalysed by terpene synthases (TPSs), which are capable of chaperoning highly reactive carbocation intermediates through an enzyme-specific reaction. Due to the instability of carbocation intermediates, the proteins’ structural dynamics and enzyme:substrate interactions during TPS catalysis remain elusive. Here, we present the structure of the diterpene synthase CotB2, in complex with an in crystallo cyclised abrupt reaction product and a substrate-derived diphosphate. We captured additional snapshots of the reaction to gain an overview of CotB2’s catalytic mechanism. To enhance insights into catalysis, structural information is augmented with multiscale molecular dynamic simulations. Our data represent fundamental TPS structure dynamics during catalysis, which ultimately enable rational engineering towards tailored terpene macrocycles that are inaccessible by conventional chemical synthesis.

T he intense structural diversity of terpenes is the molecular basis of diverse bioactivities, which render these compounds as development leads in the chemical, biotechnological, food and pharmaceutical industry 1,2 . Most terpenes constitute a macrocyclic core skeleton 3 , which is generated by cyclisation of aliphatic oligo-prenyl diphosphates 3 . This reaction is catalysed by terpene synthases (TPSs). By contrast, chemically mimicking TPS reactions to generate tailored terpene-type macrocycles can be considered in its infancy, although promising directions have been proposed [4][5][6][7][8][9] . During the cyclisation event, the acyclic precursor has to be brought into a defined conformation that positions the leaving diphosphate group and the reactive alkene entities in proximity to initiate C-C bond forming reactions. Moreover, the substrate environment has to stabilise several propagating carbocations, to adjust to conformational changes of the substrate as well as assist in intramolecular atom transfer reactions, such as hydride or proton transfers and carbon shifts. Such enzyme guidance is necessary to control the reaction intermediates, although the inherent reactivity of carbocations is also an important component in terpene biocatalysis 10 . In addition to the complex processes in the enzyme's active site, global effects such as dynamic structural changes of the protein scaffold and inter-protein interactions during catalysis are poorly understood. A detailed understanding of dynamic, biochemical and structural effects during catalysis would ultimately enable rational engineering of this enzyme family. This capacity would be a game changing advantage in structure-activity-relation studies of biologically relevant terpenes as even subtle structural modifications such as the additions of methyl groups to the native substrate (e.g., the "magic methyl effect") 11 or a simple H/D exchange may have a profound influence on the biological activity.
To elucidate the dynamic interactions of TPSs with their substrates during catalysis, we investigated the bacterial diterpene synthase CotB2 as a model system. This enzyme catalyses the cyclisation of the universal diterpene precursor E,E,E-geranylgeranyl diphosphate (GGDP) to cyclooctat-9-en-7-ol (Fig. 1a), which is subsequently elaborated to the bioactive compound cyclooctatin 12 . Cyclooctatin is a next-generation antiinflammatory drug targeting a lysophospholipase that is upregulated in eosinophilic leucocytes rather than a cyclooxygenase inhibited by ibuprofen or aspirin 13 . The CotB2 protein sequence exhibits a modified aspartate-rich DDXD motif 12 deviating from the conventional DDXXD motif and the NSE triad NDXXSXX(R, K)(E,D) as found in bacteria and fungi 3,14 , that is altered to a DTE triad in plants. Both motifs are involved in binding a conserved Mg 2+ triad that is indispensable for precise GGDP orientation in the active site. Upon substrate and Mg 2+ binding, TPSs undergo a discrete conformational change from an open, catalytically inactive one to a catalytically active closed conformation. Even though the active site cavity is already productshaped in the open conformation, closure, accompanied by translation and rotation of secondary structure elements, is essential for catalysis. Catalysis is initiated by lysis of the GGDP   15 . This initial carbocation propagates intramolecular cyclisation, which is guided by amino acids lining the active site, where π-cation and other electrostatic interactions drive the carbocation through an enzyme-specific reaction cycle. The cyclisation cascade is terminated by lining amino acids mediated deprotonation or addition of a water molecule to the final carbocation 3,16 .
Delineating interactions of lining amino acids with the substrate along the entire reaction trajectory is fundamental to understand the complexity of TPS catalysis. This information can be reliably obtained by an interdisciplinary approach combining biochemical, structural biology, computational and biocatalytic studies.
In 2014, we reported the first crystal structure of CotB2 in its open, inactive conformation without substrate. CotB2 adopts an α-helical bundle fold that is conserved among class I TPSs (PDB-ID 4OMG 17 ). Single point mutations within the active site lead to pronounced structural product diversity (Supplementary Table 1 and Figs. 1, 2) [17][18][19] .
Here, we report cumulative data that provide significant insights into the chemical mechanisms that constitute the catalytic conversion of GGDP to cyclooctat-9-en-7-ol. Remarkably, even though structural information of TPSs bound to a product or different substrate analogues as well as Mg 2+ have been reported, a complete in crystallo cyclised product has not yet been observed, demonstrating that the previously reported substrate analogues were either not correctly positioned in the active site or catalytically inactive due to chemical modifications.

Results
Comparison of the open and closed structure of CotB2. In this study, we intended to probe the structural dynamics of the CotB2 reaction using X-ray crystallography of the catalytically relevant protein states in the presence of a substrate-analogue. Remarkably, we could capture the closed conformation of CotB2 by co-crystallisation with the substrate-analogue 2-fluoro GGDP (FGGDP) and determined the crystal structure at 1.8 Å (CotB2 wt •Mg  Table 2). FGGDP is functionalised with a fluorine atom at the C2 position of its hydrophobic tail (Fig. 1b) that interferes with the propagation of the generated carbocation (s). A direct comparison of the open and closed conformation is possible, since our crystals are isomorphous. The overall structure of the open and closed conformation reveals significant differences. Substrate binding induces a translation and rotation of α-helices B, C, D, F and H towards the active site (Fig. 1d) to accurately position the DDXD Mg 2+ binding motif and to bring the active site into product-shaped conformation. Upon binding of FGGDP the 12 C-terminal residues are structured to fold over the active site, resulting in the fully active and closed conformation (Fig. 1c,  d). Surprisingly, we did not observe the entire FGGDP molecule. Instead, we detected clearly resolved electron density for a single diphosphate moiety and for a ring-like structure positioned in the active site that we interpret as the reaction intermediate 2-fluoro-3,7,18-dolabellatriene (F-Dola; Figs. 1b-d, 2a and Supplementary  Fig. 3).
With the native substrate GGDP such in crystallo snapshots would not be feasible due to the highly unstable nature of the carbocation intermediates. Most likely the fluorinated substrate alters propagation of the carbocation and hence we do not observe the native product. The possibility of cyclisation reactions of fluorinated substrates has been demonstrated for 2fluorofarnesyl diphosphate 20,21 . To our knowledge such a crystal structure with a trapped intermediate has never been described before for other TPSs crystallised in presence of isoprene substrates. In these reported structures, the diphosphate is bound, but not cleaved off and consequently no cyclisation reaction has occurred (Supplementary Discussion).
The diphosphate is coordinated by three Mg 2+ -ions with Mg 2+ B bound to the NSE motif and Mg 2+ A and Mg 2+ C being coordinated by D110 of the non-canonical DDXD motif ( Fig. 2a and Supplementary Discussion and Supplementary  Fig. 11). Additionally, active site residues R227 and R294 make contact with the oxygen atoms of the diphosphate moiety to compensate its negative charge. Upon binding of Mg 2+ A and Mg 2+ C , D110 stretches out towards the active site, causing helix D movement thereby reducing the volume of the active site and trapping of the substrate. Moreover, this helix D movement positions the C-terminus for folding over the active site to shield it from bulk solvent. The closure by the Cterminus is initiated by formation of a salt-bridge between D111 and R294. In our previous studies we showed that D111E is an inactive mutant 17 . With our current data set, we can now explain this observation with the exchange at position 111 that disturbs the salt-bridge formation emphasising its catalytic relevance. Additionally, Y295 establishes hydrogen bonds to the diphosphate moiety derived from FGGDP as well as to N220 of the NSE motif. The latter RY-pair (R294/Y295) has been previously described for other TPSs 22 . Folding of the Cterminus is completed by formation of a short anti-parallel βstrand at the end of the C-terminus (Fig. 1c). Folding over of the C-terminus does not seem to be driven by strong ionic interactions, but is dominated by hydrogen bonds and van der Waals contacts.

ARTICLE
To verify that the full closure of the active site and the folded C-terminus is not due to crystal contacts, we crystallised CotB2 with (4-amino-1-hydroxy-1-phosphonobutyl)phosphonate (alendronate (AHD); Supplementary Table 2), a compound that mimics the diphosphate group of GGDP. The CotB2 wt •Mg 2 + 3 •AHD structure has the same overall structure and active site architecture as described for CotB2 wt •Mg 2+ 3 •F-Dola, but crystallised in a different space group and clearly demonstrates that the folded C-terminus is not a crystallographic artefact caused by crystal contacts (Supplementary Fig. 4).
Importance of C-terminal region. In order to biochemically confirm the importance of the entire C-terminal region for catalysis, we generated a C-terminal truncation of CotB2 terminating at R294 (CotB2 ΔC ). CotB2 ΔC lacks the last 12 C-terminal residues, corresponding to the lid as observed in the structure of CotB2 wt •Mg 2+ 3 •F-Dola and CotB2 wt •Mg 2+ 3 •AHD. Indeed, CotB2 ΔC was an inactive mutant, not capable of catalysis as shown by GC/MS (Fig. 1e). Co-crystallisation of CotB2 ΔC with FGGDP resulted in the open-state conformation with no substrate bound (Supplementary Table 2). Additionally, this data provides further evidence that the salt-bridge between D111 and R294 is not sufficient to fully close the active site. Interestingly, there are two Mg 2+ -ions bound in the CotB2 ΔC structure. One is fully occupied Mg 2+ B and the other one is partially occupied Mg 2+ C , which can only be fully occupied if properly coordinated by the DDXD motif upon substrate binding and the accompanying conformational changes of the active site.
Closure of the active site seems to be an essential process in TPS catalysis. The implementation, however, appears to be very specific for each individual enzyme class. In several sesquiterpene synthases unstructured loop regions become ordered upon ligand binding, initiating active site closure 23,24 . In case of the plantderived bornyl diphosphate synthase, a monoterpene synthase, the N-terminus together with a loop segment close the active site, and the enzyme was co-crystallised with the final product bornyl diphosphate 25 . By contrast, in the structure of the sesquiterpene aristolochene synthase from a fungus it is merely a loop segment that closes the active site 26 . Interestingly, molecular dynamic (MD) simulations of the closed state of the plant derived taxadiene synthase indicated that the N-terminus might be involved in closure of the active site 27,28 . Therefore, the mechanism of active site closure may be very specific to the reaction type and taxonomic origin of the respective TPS.
Given the importance of the C-terminus of CotB2 for catalysis, we screened the PDB for other diterpene synthases for which structural information for the open and closed conformation is available and which are structurally most related according to a DALI search 29 (Supplementary Table 3). In the closed structure of the labdane-related diterpene synthase (PDB-ID 5A0K) its Cterminus is folded as well. The RY-pair ( Supplementary Fig. 5) is at the same position and engaging identical interactions with the diphosphate. Analysing protein sequences of other bacterial diterpene synthases for the presence of the RY-pair 22 and their flanking regions, we found a conserved tryptophan six amino acids upstream. Therefore, we suggest, that the WXXXXXRY ("X" any amino acid) motif is relevant for other diterpene synthases ( Supplementary Fig. 5). If the tryptophan of latter motif is exchanged in CotB2 for glycine (W288G), the product becomes 3,7,18-dolabellatriene (Supplementary Table 1 and Supplementary Fig. 2) 17 . In this study, we introduced the less stereochemcially demanding W288F mutation, which reduced CotB2's activity, but did not change the product profile. Thus, the aromatic character of the amino acid at position 288 in CotB2 is important for product formation and any drastic exchange to a non-aromatic side chain does interfere with the propagation of the carbocation. In the bacterial sesquiterpene synthase pentalenene synthase the corresponding mutation W308F leds to a product mixture 30 . However, in the plant-derived epi-aristolochene synthase mutations of W273 resulted in total loss of enzymatic function 23 . Therefore, tryptophan residues at this strategic position within the active site are of general importance in TPS to guide product formation.
Pre-catalytic states of CotB2. Very recently, Tomita et al. published a co-crystal structure of CotB2 bound to the inert substrate-analogue geranylgeranyl thiodiphosphate (GGSDP) (CotB2 wt •Mg 2+ B •GGSDP, PDB-ID 5GUE 19 ), which was described as representing the closed state of CotB2. Surprisingly, the diphosphate moiety of GGSDP is coordinated merely by a single magnesium-ion (Mg 2+ B ; Fig. 2c), in contradiction to the accepted principle of a trinuclear magnesium cluster required for catalysis. Notably, beside the two missing Mg 2+ , even though the saltbridge between D111 of the DDXD motif and R294 is present ( Supplementary Fig. 6b), the remaining C-terminus could not be resolved 19 Table 2). Interestingly, a single magnesium-ion (Mg 2+ B ) was bound to the active site, coordinated by the residues N220, S224 and E228 of the NSE motif and two water molecules ( Fig. 2c and Supplementary Fig. 6a). It has been proposed for the sesquiterpene synthase aristolochene synthase, that Mg 2+ B only binds the active site after the ligand has entered 31 . However, this does not seem to be the case for CotB2. Therefore, we suggest the binding of Mg 2+ B to be the first step in preparing CotB2 for substrate binding.
The position of the Mg 2+ B matches perfectly with that in the CotB2 wt •Mg 2+ B •GGSDP structure (Fig. 2c). Moreover, the conformation of the Mg 2+ B coordinating amino acids are oriented in the same manner as there are two water molecules in our structure that mimic the oxygens of the diphosphate group in CotB2 wt •Mg 2+ B •GGSDP ( Fig. 2b and Supplementary Fig. 6b). In the structure of CotB2 F107A •Mg 2+ B the salt-bridge of D111 and R294 has not yet been established (Supplementary Fig. 6a) and the remaining C-terminus could not be resolved. This suggests, that the accurate positioning of the diphosphate moiety is required for the formation of the latter salt-bridge. The overall structure of CotB2 wt •Mg 2+ B •GGSDP is most similar to CotB2 wt and CotB2 F107A •Mg 2+ B as indicated by a low root-mean-square deviation (Supplementary Table 4). Hence, the structures are likely different pre-catalytic states of CotB2.
CotB2 mechanism and structure-function dynamics. The in crystallo capture of the reaction intermediate F-Dola provides a static snapshot along the reaction coordinate in CotB2 (Fig. 3a). The mechanism in CotB2 has previously been delineated experimentally by Meguro et al. and Sato et al., and theoretically by Hong and Tantillo [32][33][34] . To explain mechanistic details of the entire catalytic cascade, we turned to in silico multiscale modelling. Modelling of the CotB2 reaction commenced with the bound 3,7,18-dolabellatriene (Figs. 3b, 4). Due to the high resolution of the crystal structure that captured CotB2 in a fully closed, catalytically competent state, QM/MM modelling yielded insight into the possible role of the enzyme in stabilising the reaction intermediates. We generated all-trans GGDP in a pre-folded conformation, ready for catalysis, as well as all intermediate steps until the final carbocation. In the initial pre-folded GGDP state, the C1-C11 distance is 3.9 Å and the C10-C14 distance is 3.3 Å (Fig. 4). Following initial C-O cleavage and presumably concerted C1-C11 and C10-C14 bond formations, dolabellatrienyl cation (A) is formed, with the cation at C15 being stabilised by π-cation interactions with W186 (distance 4.6 Å; Fig. 4). The cation dislocates to C8 via a 1,5-hydride transfer reaction to yield cation B, as shown by deuterium labelling 32 , with the cation being stabilised by an oxygen hole formed by N103 (OD1), N103 (O) and T106 (OG1) (Fig. 4).
The carbocation intermediate B then undergoes ring-closure via C2-C6 bond formation to yield cation C, with the cation located at C3 where it is stabilised by π-cation interactions, as it is sandwiched between F107 and F149 (Fig. 4). Further stabilisation is likely provided by additional electrostatic interactions, as the cation migrates towards the diphosphate moiety during the B→C transformation step. Subsequently, a proposed 1,3-hydride transfer takes place to generate cation E. However, Hong and Tantillo 34 suggested that a series of two 1,2-hydride transfers would be energetically favoured in the gas-phase. Based on the modelled intermediate states in this study, there seems to be significant steric hindrance to a direct 1,3-hydride transfer. Therefore, two sequential 1,2-hydride transfers to provide intermediate E appear plausible (Fig. 3b). Identical two 1,2hydride shifts were also experimentally reported for tsukubadiene synthase 35 . As suggested 34 , these different mechanistic possibilities can be differentiated via isotope labelling as shown by Sato  2 Å; Fig. 4). The following step involves another 1,5-hydride transfer to give carbocation G, as shown by deuterium labelling 32 , with the cation located at C10, which is partly stabilised by N285 (distance 5.4 Å) and W186 (C10-W186 distance 5.3 Å; Fig. 4). A subsequent ring-closure generates a propyl-ring in intermediate H, while moving the cation to C7, which is stabilised by N103 (OD1), N103 (O) and T106 (OG1) (distances 3.9, 5.6, and 5.0 Å respectively; Fig. 4). The subsequent step entails a 1,3-alkyl shift to yield I, as established by 13 C isotope labelling 32 , which moves the cation back to C10 (C10-N285 distance 4.7 Å, C10-W186 distance 5.8 Å), and shifts the cyclopropyl moiety along the central octa-cyclic ring (Fig. 4). Mutation of this residue to W186H results in formation of 3,7,18-dolabellatriene 19 . This is conceptually very similar to the taxadiene synthase W753H mutation, which yields (−)-(R)-cembrene A 27 . The approximate energetic contribution of the above-discussed interactions between active site moieties and the carbocation intermediates were quantified using densityfunctional theory cluster calculations (Supplementary Tables 6  and 7). Inspection of these interaction energies clearly show that the enzyme stabilises the evolving carbocations, with values between ca. −10 and −20 kcal/mol. The reaction concludes with attack by a water molecule at position C7, followed by deprotonation, to give the final product cyclocotat-9-en-7-ol. A potential hydroxylating water molecule was not observed in the crystal structure. But based on the current mechanistic model, it could be located between N103 (OD1), N103 (O) and T106 (OG1). The role of Asn in binding hydroxylating water molecules has been stressed in the context of 1,8-cineole synthase 36 . The above mechanistic model suggests that the CotB2 active site architecture has evolved to chaperone the changing cation along the entire reaction coordinate. Future QM/MM calculations will address the complete free energy surface for the CotB2 catalysed reaction. We note that the role of the enzyme environment in guiding the carbocations along the reaction coordinate is in line with our earlier work [37][38][39] and that of Peters and co-workers 40,41 .

Discussion
Co-crystallising CotB2 with FGGDP enabled us to capture an intermediate state of the reaction mechanism. In combination with structural and biochemical data we could point out the importance of the properly folded C-terminus for the enzymatic activity. Our findings allowed us to propose a complete catalytic mechanism of CotB2 at a structural level by combining crystallography, biochemical tools and theoretical modelling (Supplementary Fig. 7). The enzyme mechanistic cascade commences with Mg 2+ B binding to the NSE motif, although this binding event is not accompanied by any conformational changes of the main structural core of CotB2. Subsequently, the substrate binds with its diphosphate moiety to Mg 2+ B . This binding event induces the formation of the salt-bridge between D111 and R294.

Binding of the ions Mg 2+
A and Mg 2+ C to the diphosphate moiety of the substrate causes the largest conformational change in the enzyme, including rigid helix motion, which reduces the active site volume, and assures proper coordination of the substrate. At this point, the complete C-terminus can fold over the active site, thereby establishing a hydrogen network between the two catalytic motifs, and shields the active site from bulk solvent. Once all the pre-catalytic binding events and conformational changes are established, GGDP is correctly positioned to subsequently undergo cyclisation, which in the case of FGGDP results in premature quenching of the cyclisation reaction, as shown by the co-crystallisation of CotB2 with FGGDP. The closure of the C-terminus is the final trigger that initiates the chemical reaction cascade in CotB2, and possibly in additional class I TPSs.
Using theoretical modelling, we modelled the reaction using the natural substrate, GGDP. Importantly, based on in silico multiscale modelling we conclude that the three-dimensional active site architecture is such that the evolving carbocation is stabilised by precisely positioned amino acids and the bound diphosphate at each stage of the reaction cascade. Product release is possibly facilitated by the opening/unfolding of the C-terminus, which is entropically favourable, and might be the first stage in active site opening and release of the final product.
In summary, the cumulative experimental data, in conjunction with computational simulations, clearly demonstrate an active role of the protein scaffold in guiding the carbocation driven reaction towards a definitive end product. In line with our direct crystal structure evidence, there is mounting indirect evidence that the enzyme plays a crucial role in: binding the substrate, folding the substrate, selectively stabilising highly reactive carbocations to guide the reaction cascade towards the desired product and finally deprotonate or hydroxylate the final product by careful positioning of active site moieties. Therefore, the data delineated from the abrupt product trapped in the closed CotB2 structure provide fundamental advance in the understanding of TPS structural dynamics during catalysis.
Moreover, we identified universal structural features, which can be exploited for rational engineering of other bacterial TPS towards tailor-made terpene macrocycles. This information will pave the way for future semi-synthetic drug development strategies. The proof of concept for this strategy has already been well documented 42 . With respect to CotB2, site-directed mutagenesis of plasticity residues in the active site altered the enzymes product portfolio, which strongly argues for an active involvement of the protein scaffold in guiding the catalytic transformation of carbocations into distinct products 17 . The CotB2 mutation F107A afforded cembrene A as a product. The cembrene-type diterpenes display diverse bioactivities, including cytotoxic, insect deterrent and antimicrobial. Relevant work on insecticidal cembrenes has recently been reported 43 . Similarly, the single mutation of W288G alters to product to (1R,3E,7E,11S,12S)−3,7,18-dolabellatriene. Its structure has been reported to have potential antibiotic activity against multidrug-resistant Staphylococcus aureus. However, most of the targeted mutagenesis strategies were designed based on the open, non-catalytically active CotB2 structure. Therefore, the catalytic relevance of targeted amino acids in the active site could not be predicted, which often leads to generation of nonproductive mutants. Moreover, one could not predict the structure of the cyclisation product. On the basis of our current model with a catalytically relevant intermediate in the active site it is feasible to trace and predict productive mutations. Furthermore, future expansion of our data to other TPSs may also enable prediction of the resulting terpene cyclisation product when targeting a specific amino acid residue(s).

Methods
Cloning. DNA manipulations and cloning procedures were performed according to standard protocols (Supplementary Methods). Nucleotide sequences of all primers used in this study are summarised in Supplementary Table 5.
Protein expression and purification. CotB2 wt and CotB2 F107A fused to a Cterminal hexa-histidine-tag in a pET-24a vector were expressed in Escherichia coli Rosetta2 DE3 cells. Cells were harvested by centrifugation (6 min, 6000 rev min −1 at 4°C) and resuspended in buffer A (50 mM Tris/HCl pH 7.5, 500 mM NaCl, 5 mM MgCl 2 , 1 mM DTT). The cells were lysed by homogenisation and the lysate was cleared by centrifugation (1 h, 21,000 rev min −1 at 4°C). Purification included Ni 2+ -NTA affinity chromatography (elution in a linear gradient to buffer A containing 500 mM imidazole) and subsequent size-exclusion chromatography in buffer B (20 mM Tris/HCl pH 7.5, 150 mM NaCl, 5 mM MgCl 2 , 1 mM DTT) 17,18 . CotB2 ΔCter in pETM-11 was purified as CotB2 wt except for an additional proteolytic step after Ni 2+ -NTA affinity chromatography. The amino-terminal His 6 -tag of CotB2 ΔCter was cleaved overnight by tobacco etch virus protease while dialysing in buffer B.
Crystallisation. CotB2 F107A was concentrated to 28 mg/ml as measured by the absorbance at 280 nm. Crystals were obtained by the sitting-drop vapour-diffusion method at 18°C with a reservoir solution composed of 17.5% (v/v) polyethylene glycol 4000, 100 mM Tris/HCl at pH 8.5 and 100 mM MgCl 2 . For co-crystallisation experiments, CotB2 wt was concentrated to 20 mg/ml and incubated in a fivefold molar excess of FGGDP for 30 min on ice. Crystals were obtained by the sittingdrop vapour-diffusion method at 18°C with a reservoir solution composed of 30% (v/v) polyethylene glycol 400, 100 mM HEPES/NaOH at pH 7.5 and 200 mM MgCl 2 . CotB2 ΔCter was concentrated to 27.9 mg/ml and incubated in a 15-fold molar excess of FGGDP for 30 min on ice. Crystals were obtained by the sittingdrop vapour-diffusion method at 18°C with a reservoir solution composed of 16% (v/v) polyethylene glycol 4000, 100 mM Tris/HCl at pH 8.5 and 100 mM MgCl 2 . For co-crystallisation experiments with alendronate (Alfa Aesar, Germany), CotB2 wt was incubated in a twofold molar excess for 30 min on ice. Crystals were obtained by the sitting-drop vapour-diffusion method at 18°C with a reservoir solution composed of 26% (v/v) polyethylene glycol 4000, 100 mM Tris/HCl at pH 8.5 and 150 mM MgCl 2 . All crystals were cryo-protected with 25% (v/v) 2-methyl-2,4-pentanediol supplemented to the reservoir resolution and subsequently flashcooled in liquid nitrogen.  Table 2). The structures were determined by molecular replacement with the coordinates of CotB2 wt (PDB-ID: 4OMG 17 ) as search model using PHASER 45 . The structure was refined by maximum-likelihood restrained refinement in PHENIX 46,47 . Model building and water picking was performed with COOT. Geometrical restraints, used in the refinement of the fluorinated intermediate, were generated by using the PRODRG Web Server or ELBOW 48  Computational modelling. The modelling commenced with construction of cation A based on 2-F-3,7,18-dolabellatriene trapped during crystallisation of FGGDP in CotB2 (monomer A of PDB-ID 6GGI). The substrate GGDP was then constructed by performing the reverse reaction in silico 16 . Hydrogen atoms were added using the HBUILD facility of CHARMM. Subsequently, the enzyme was soaked in a preequilibrated TIP3P water 52 box of size about 80 × 80 × 80 Å 3 , and the effect of ionic buffer environment and system charge neutralisation was accounted for by addition of 29 Na + and 20 Cl − ions 53 . Subsequently, we relaxed the system by performing step-wise energy minimisation, followed by MD simulations. The MD simulations entailed a 25 ps heating period up to 298 K followed by 5 ns of equilibration. During the first 1 ns of the equilibration simulations, weak harmonic restraints (1 kcal/mol Å 2 ) were imposed on the backbone C α atoms (residues 24-290), as well as nuclear Overhauser effect (NOE) restraints on selected hydrogen bonds in the protein and between the protein and cofactor diphosphate and Mg 2+ ions. The MD heating was performed in the NVT ensemble, while equilibration was performed in the NPT ensemble 54 . Long-range electrostatics were treated via particle-mesh Ewald summation. Following MD equilibration, we performed hybrid QM/MM energy minimisations on the complete system, with the QM region defined as GGDP and the three Mg 2+ ions. Employing a combination of NOE restraints and general distance restraints, we generated the following cations along the reaction pathway 34 : A, B, C, E, G, H and I (Figs. 3b, 4). All MM calculations used the CHARMM protein 34 and nucleic acid force fields 55 , in conjunction with parameters for GGDP 56 , while QM/MM calculations 57 employed the SCCDFTB method 58 for the QM region. All simulations used the CHARMM simulations programme 59 .
Synthesis of isprenoid ligand. Detailed experimental procedures are provided in the Supplementary Methods. Intermediate reactions steps and the final product were characterised by 1 H-NMR (Supplementary Table 8A), 13 C-NMR (Supplementary Table 8B), 19 F-NMR (Supplementary Table 9A) and 31 P-NMR (Supplementary Table 9B). The precise molecular weight was confirmed by mass spectrometry (Supplementary Table 10).