A structural view of synthetic cofactor integration into [FeFe]-hydrogenases† †Electronic supplementary information (ESI) available: Tables listing and comparing the RMSD of the structures, distances and angles of the 2FeH-subclusters, the distances from 2FeH-subcluster atoms to selected amino acids

Crystal structures of semisynthetic [FeFe]-hydrogenases with variations in the [2Fe] cluster show little structural differences despite strong effects on activity.


Introduction
[FeFe]-hydrogenases are efficient natural catalysts for both the generation and oxidation of H 2 . 1 This reaction is accomplished by the H-cluster, a metal cofactor consisting of a cubane [4Fe4S] cluster (4Fe H ) connected via a cysteine to an unusual [2Fe] cluster (2Fe H ). The two iron atoms of the latter, termed distal (Fe d ) and proximal (Fe p ) iron are ligated by a total of three CO and two CN À molecules and an aza-dithiolato bridge. [2][3][4][5] Four structural characteristics seem to be important for the high activity of the H-cluster: 6,7 (a) The cyanide ligands besides having additional effects [8][9][10] shi the redox potential to more negative values, when compared to all-carbonyl complexes. 11 (b) The 4Fe H -cluster serves as an intramolecular redox partner. [12][13][14][15] (c) A proton donor is present in the dithiolato bridge. 16-19 (d) The ligand conformation at the 2Fe H subsite features a CO that bridges or semi-bridges the Fe atoms. This leads to an unoccupied coordination site on Fe d . 3,5,[20][21][22] For the hydrogenase HydA1 from Chlamydomonas reinhardtii three redox states are discussed as part of the catalytic cycle, which can be distinguished by EPR and FTIR spectroscopy. According to this hypothesis the H-cluster cycles from the H ox state 4Fe H 2+ -Fe(I)- The redox potentials for these transitions are À400 mV and À470 mV vs. SHE close to the H 2 /H + redox pair. 7 The crucial proton transfer to and from the active site seems to be accomplished by a proton transfer pathway through the protein towards the central atom of the dithiolato bridge in the 2Fe H -subcluster. 3,5,23 In nature, the 4Fe H -cluster and other FeS clusters of the enzyme not specic to [FeFe]-hydrogenases are synthesized by the widespread ISC or SUF systems for FeS cluster synthesis yielding inactive hydrogenases, which lack only the specic 2Fe H -subcluster. 24 For the sake of simplicity this pre-form will be referred to as apo-form of [FeFe]-hydrogenases throughout this text. The three maturase enzymes HydE, HydF and HydG are necessary for the synthesis of the 2Fe H -cluster and the assembly of the H-cluster within the protein. 25 In vitro, chemically synthesized [2Fe] complexes can be bound to the maturase HydF and transferred from there to apo-hydrogenases to form a complete H-cluster. 2 Notably, also in the absence of HydF or any other helper protein, an active H-cluster can be formed spontaneously by bringing together the inactive apohydrogenase and the chemically synthesized [2Fe] complex Fe 2 [m-(SCH 2 ) 2 NH](CN) 2 (CO) 4

2À
. 26 While the [2Fe] moiety alone is inactive under physiological conditions, the semisynthetic enzyme shows high catalytic activity, which demonstrates the importance of the protein environment.
[2Fe] complexes with variations in the dithiolato bridge and/or the other Fe ligands have recently been shown to integrate into HydA1 as well, but the enzymes were inactive or severely limited in their turnover rates especially if the dithiolato bridge was changed. 2,27 The central atom of the dithiolato-moiety seems to inuence the redox behavior of the H-cluster either directly or by interfering with the proton transfer to/from the active site. 28 Structures of the active bacterial [FeFe]-hydrogenases 3,29,30 and the inactive apo-form of HydA1 from Chlamydomonas reinhardtii 31  and its non-active derivatives CpI PDT (X ¼ CH 2 ), CpI ODT (X ¼ O) and CpI SDT (X ¼ S).

Results and discussion
Only an ADT-bridged [2Fe] cluster induces H 2 evolution activity in CpI ) were synthesized following modied literature procedures 32-38 and used to prepare semisynthetic CpI as described before. 26 Specic hydrogen evolution activities with methylviologen as electron donor were 2874 AE 262 (mmol H 2 ) min À1 (mg protein) À1 for CpI with the ADT-bridged 2Fe H -cluster (CpI ADT ), which is in agreement with previously reported values. 10,26 Neither for apoCpI nor for the non-natural derivatives CpI PDT , CpI ODT or CpI SDT could any hydrogen evolution activity be detected above the detection limit of 0.02% of the activity of CpI ADT . The same [2Fe] complexes were recently integrated into HydA1. While the ODT-bridged and SDT-bridged complexes didn't induce H 2 evolution, 0.9 (mmol H 2 ) min À1 (mg protein) À1 were reportedly produced by HydA1 with the PDT-bridged 2Fe H -cluster. This equals 0.17% of the activity of the same enzyme with the naturelike ADT-bridged 2Fe H -cluster. 27 As HydA1 is smaller than CpI, a better accessibility of the active site from the protein surface might promote undirected proton transfer. This could enable slow H 2 production even though the directed proton transfer via the amine of the 2Fe H -cluster is disrupted.
All forms of CpI were crystallized under strictly anaerobic conditions and the crystal structures of both CpI ADT and apoCpI were solved with molecular replacement using the known structure of active, native CpI 3,30 as a search model and rened to 1.63Å and 1.60Å resolution respectively (Fig. 1, Table 1). CpI ADT was subsequently used as a search model during molecular replacement to determine the structures of CpI PDT , CpI ODT and CpI SDT at 1.82Å, 1.73Å and 1.93Å resolution respectively ( Fig. 1, Table 1). In contrast to already known structures of native CpI, 3,30,39 the space group of the crystals was P2 1 for all ve enzymes and the asymmetric units each contained two nearly identical molecules. Of these two molecules, one possesses a more exible N-terminal domain (residues 1-90), but at the same time a more rigid active site and thus yields a more reliable electron density in the important H-domain in all structures. This becomes evident through the slightly lower temperature factors around the active site when compared to the second molecule. Accordingly gures and values given in the text were taken from the former molecule (chain B) if not stated otherwise, while the complete values for both chains of all structures can be found in the ESI. † As in vitro maturation of apo-[FeFe]-hydrogenases with synthetic [2Fe] cofactors was described only recently, 2,26 we considered the exact structure of the 2Fe H -cluster and its environment in the semisynthetic enzyme to be of considerable interest. To minimize model bias of electron density in the active site cavity before starting to rene the 2Fe H -subcluster, at least two rounds of renement of each structure were performed without a 2Fe H -cluster in the models.
Subsequently, starting models of the 2Fe H -subcluster based on the structure of native CpI 30 with optimized geometry but adapted composition of the dithiolato moiety, 5 were used. Restraints were applied to all bond distances in the subcluster. We additionally restrained the angles dening the positions of the CO and CN À ligands. The position of the bridging CO was not restrained due to its reported exibility 5 depending on the redox state of the enzyme. Final models of the 2Fe H -subclusters were veried by inspection of composite omit maps.

Presumed maturation channel closed in apoCpI crystal structure
The crystal structure of apoCpI reported here is strikingly similar to structures of active CpI, both native and semisynthetic (Fig. 1). The overall RMSD of the backbone atoms of apoCpI and native CpI 30 is as low as 0.3Å, while apoCpI and CpI ADT display an RMSD of 0.4Å over all backbone atoms (Table S1 †). Signicant differences in side-chain orientation are mainly limited to surface exposed residues with V423 being a notable exception (Fig. 2). This residue in the central cavity is adapting a different rotamer, supposedly stabilized by one of the water molecules in the binding pocket for 2Fe H . As demonstrated earlier, 31 the structure of apoHydA1 from C. reinhardtii lacking the 2Fe Hsubcluster shows overall great similarity to the structure of the Hdomain of CpI 3 and DdH 29 with regard to the backbone geometry, but exhibits regions of pronounced differences. Amongst these differences is a channel from the surface to the site of the 2Fe H -  (Fig. S1A †). However, there has neither been a structure of a maturated [FeFe]-hydrogenase of the short chlorophyta type nor of an unmaturated bacterial type enzyme, which would have allowed for a direct comparison. The structure of apoCpI presented here shows the three regions 405-423 ("plug"), 437-453 ("lid") and 529-540 ("lock") clearly in a "closed" conformation nearly identical to active CpI (Fig. 1). Prominently, F417 in direct contact to the 2Fe H -subcluster shows minimal deviation in apoCpI when compared to CpI ADT (Fig. 2), while it is moved by 15Å in apoHydA1. Washed and subsequently dissolved crystals of apoCpI could be maturated with the synthetic ADT-bridged [2Fe] cluster to an activity of 1250 nmol H 2 /min/crystal, reassuring that the reported closed structure of apoCpI is not a dead-end conformation. This suggests an equilibrium between a "closed" and an "open" state in apoCpI, of which only the former readily crystallizes. Within the regions with striking deviation between apoHydA1 and apoCpI, several glycine residues can be identied, which are highly conserved in a recent sequence alignment of all known [FeFe]-hydrogenase sequences 40 (Table S2 †). These amino acids could function as hinges, as for some of them the "open" or "closed" conformation respectively would imply dihedral angle combinations commonly found only for glycine residues 41 (Table S2 † Rigid cavity in apoCpI forces the [2Fe] complex to move into its active conformation Being devoid of the 2Fe H -cluster, the active site binding pocket of apoCpI is occupied by seven water molecules and a chloride ion (Fig. 2) instead. Note that this leaves a water lled cavity of roughly 10Å diameter in the center of the protein. Nonetheless, residues which are assumed to interact with the cofactor in the active enzyme are shied only very slightly by 0.1-0.3Å towards a narrower cavity (Table S3, Table S4. † Numbering of amino acids as in the structure of native CpI.
identical structures with an RMSD of 0.3Å for the main chain atoms. Even with regard to the side chain atom conformations, signicant differences between CpI ADT and the native CpI can only be found in several surface exposed residues, which is surprising given the considerable differences in the crystal packing.
When comparing the important cofactor-peptide interactions in native CpI and the structure of CpI ADT , the distances between the atoms of 2Fe H and their respective interaction partners in the protein environment show a maximum deviation of 0.17/0.13Å and an average deviation of 0.06/0.05Å for chain A/chain B (Fig. 3, Table S4 †). This is well within the experimental error of crystal structure analysis at the given resolution. Moreover the synthetic 2Fe H -cluster itself in the structure of CpI ADT compares very well to the in vivo synthesized version in native CpI 30 (Fig. 3) 5,21,29 In our structure the CO ligand between the Fe atoms is positioned at an angle of 114 /132 (chain A/chain B) between Fe d -C-O (Table S5, † Fig. 4), which does not indicate terminal binding of the CO to Fe d as previously published for a structure of reduced DdH. 5 Thus we understand the here reported structure of CpI ADT to be mainly in the H ox state. While in earlier structures of CpI 3,30 a region of low but signicant electron density next to Fe d was assigned as a water molecule in this particular redox state, in the structure described here the H-cluster of both chains clearly features one coordination site on Fe d devoid of electron density (Fig. 4).
A comparison of the crystal structures of the synthetic ADTbridged [2Fe] complex 37 before and aer integration into the protein environment as 2Fe H illustrates the distortions that the protein forces upon the [2Fe] complex (Fig. 4). An Fe-S-Fe bridge to the 4Fe H -cluster is formed and, as demonstrated earlier, one CO ligand is lost during the process of activation. 26 Another CO ligand shis into a bridging position between the two Fe atoms and the CO/CN À ligands move into an octahedral coordination at each Fe with nearly perpendicular equatorial planes (Fig. 4). This conformation has been attributed a crucial role in allowing the mixed Fe(I)-Fe(II) valency of the H ox state within the catalytic cycle, which is difficult to achieve in isolated [2Fe] clusters. 43 Additionally the new conformation features the open coordination site at Fe d trans to the bridging CO (Fig. 4). This promotes regioselectivity of H 2 binding or hydride formation close to the amine in the ADT-bridge, which is believed to be crucial for the mechanism. 16,44 2Fe H -subsite structure remains unaltered upon changes in the dithiolato bridge The structures of all three CpI derivatives with non-natural 2Fe H -subsites superpose very well with each other and the native CpI, apoCpI and CpI ADT with RMSD's for Ca atoms between 0.2Å and 0.5Å (Table S1 †). Comparison of the exact positions of amino acids supposedly involved in enzyme function, e.g. amino acids in the proton transfer pathway or around the active site, yielded little differences within the limits of exactness of macromolecular crystallography (Fig. 5). The average RMSDs of all atoms of selected amino acids were as low as 0.08-0.11Å when comparing the non-natural derivatives with CpI ADT . As signicant differences in the degree of maturation were observed for semisynthetic HydA1 with nonnatural 2Fe H clusters, 27 we allowed variation of the occupancies of the atoms of the 2Fe H -subclusters during renement. According to this rough estimate more than 90% of the molecules in the crystals contained the 2Fe H -subsite (Table 1). Even though the effects of partial occupancy and temperature factor are hardly discernible at the given resolution, we expect these results to be a good lower limit as the calculated temperature factors of the 2Fe H -clusters and the surrounding amino acids agree well ( Table 1). As apoCpI crystallizes in a nearly identical structure and an isomorphous unit cell we assume that the high occupancy of the 2Fe H -clusters does not result from For the ODT-bridged 2Fe H -subsite in HydA1 a less pronounced bridging character of CO b compared to other semisynthetic HydA1 enzymes was reported according to FTIR data of the "as isolated" state. 27 A very similar state represented a minor part of the mixed population of HydA1 with SDTbridged 2Fe H -subcluster in the same study. We found an angle of 145 between Fe d -C-O of the CO b ligand in CpI SDT which indicates a more terminal than bridging character, but the other structures including CpI ODT reveal angles suggesting a bridging CO (Fig. 6, Table S5 †).
Because of the above discussed effects of redox state changes on the CO b ligand, a potential dependent FTIR based investigation of the CpI enzyme derivatives would be needed to clarify if this is merely an effect of the redox state at the point of crystal mounting or inherent to the different dithiolato bridge. A detailed comparison of distances between the atoms of the 2Fe H subsite and the surrounding amino acids indicates a slightly different position of Fe d within the cavity for CpI PDT 0.1Å closer to Ala 230 and further away from Cys 299 (Table S4 †). Besides this, small differences in the dithiolato bridge can be observed. While the bridgehead atom is leaning about 0.2Å further away from Met 497 in the inactive CpI derivatives, the sulfur atom of Cys 299 is pushed back to keep roughly the van-der-Waals distance to the bridgehead atom of the dithiolato bridge in the three structures (Fig. 5B, Table S3 †). However, the position and geometry of the non-natural 2Fe H -clusters do not show any large differences ( Fig. 5 and 6, Table S5 †) when compared to native CpI or CpI ADT and thus do not offer a clear structural explanation for the impaired activity.
For CpI PDT this is in line with a recent ENDOR and HYSCORE study of HydA1 containing a PDT-bridged 2Fe H -subcluster, which showed very similar spectra in comparison to in vivo maturated DdH. 45 DFT calculations performed for ADT-bridged, PDT-bridged and ODT-bridged 2Fe H -subclusters in CpI also resulted in very similar geometries 30 (Table S5 †). Remarkably, there is no signicant electron density in our structures close to Fe d at the postulated site of H 2 binding in any of the 2Fe Hsubclusters (Fig. 6). Thus binding of an inhibitor to this open coordination site can be ruled out as cause for the quantitative loss of activity. For HydA1 with the PDT-bridged 2Fe H cluster no binding of CO to the active site was observed in a recent FTIR based study. 28 Our structure of CpI PDT rules out a rearrangement in the neighboring amino acids as explanation for this behavior. However, once an inhibitory CO is bound to Fe d , the distance between the central atom of the dithiolate bridge and the oxygen of CO was reported to be as close as $2.5Å. 5 While the single hydrogen of an amine bridgehead proposedly points towards C299 and thus away from Fe d , the PDT's central methyl group might considerably obstruct binding of CO to Fe d through its hydrogen atoms not visible in X-ray crystallography at the given resolution.  Table  S4. † Numbering of amino acids as in the structure of native CpI.
enzymes were subsequently cleaned from leover [2Fe] complex and buffered again into a 10 mM Tris-HCl buffer with pH 8.0 and 2 mM NaDT by use of a NAP™ 5 (GE Healthcare) size exclusion chromatography column. Enzyme preparations were concentrated using Amicon Ultra centrifugal lters 30 K (Millipore) under anaerobic conditions. Success of maturation and quality of puried protein samples of CpI ADT were determined by testing their H 2 production activity in vitro with methylviologen as electron donor using an established method. 49 To test for catalytic activity of the non-native semisynthetic enzymes, the same method was applied and additional measurements with 10 fold increased protein amount were conducted to lower the limit of detection.
Box-like protein crystals of apoCpI and the semisynthetic hydrogenases were obtained with PEG 3000 or PEG 4000 as precipitant using the hanging drop or sitting drop vapor diffusion method at 277 K under anaerobic conditions within 2-4 days when mixing reservoir solution 1 : 1 with protein solution (10 mg ml À1 ). The crystallization conditions for the selected crystals of apoCpI were 12% PEG 3000, 0.1 M MES pH 6.5, 0.2 M MgCl 2 in a sitting drop vapor diffusion experiment and cryo-protection was achieved with a nal concentration of 15% glycerol in 15% PEG 3000, pH 6.5, 0.2 M MgCl 2 . CpI ADT crystals selected for diffraction experiments were grown in 11% PEG 4000, 0.1 M MES pH 7.0, 0.2 M MgCl 2 in a hanging drop experiment and protected against formation of ice crystals with paraffin oil. Crystals of the non-native semisynthetic enzymes were grown by hanging drop vapor diffusion using 0.1 M MES pH 6.0, 0.4 M MgCl 2 and a total of 40% v/v of PEG4000 and glycerol to avoid the need of additional cryo-protection during crystal mounting. In detail the reservoir solutions contained 15% PEG 4000, 25% glycerol for CpI PDT , 19% PEG 4000, 21% glycerol for CpI ODT and 21% PEG 4000, 19% glycerol for CpI SDT .
Maturation capability of crystallized apoCpI was tested by washing a crystal in three fresh drops of its reservoir solution followed by dissolution of the crystal in cold 0.1 M K 2 HPO 4 / KH 2 PO 4 buffer at pH 6.8 with 2 mM NaDT under strictly anaerobic conditions. Maturation was started by addition of 1.5 pmol Fe 2 [m-(SCH 2 ) 2 NH](CN) 2 (CO) 4 [Et 4 N] 2 in 0.1 M K 2 HPO 4 / KH 2 PO 4 buffer, pH 6.8, and allowed to proceed for 1 h at 4 C. Subsequently the mixture was transferred completely into a solution of methylviologen, NaDT and phosphate buffer as for standard tests for H 2 evolution activity and treated accordingly. 49 Mounting of protein crystals into CryoLoops™ (Hampton Research) and subsequent ash-freezing in liquid N 2 was performed under strictly anaerobic conditions at 298 K. Diffraction data were collected at 100 K at beamline BL44-XU at SPring-8 (Hyogo, Japan) and beamline PXII at the SLS (Villigen, Switzerland) and the data were processed using the soware package HKL2000 (ref. 50) and XDS 51 for apoCpI and the semisynthetic hydrogenases, respectively. Molecular replacement and structure optimization were performed with the soware packages CCP4 (ref. 52) (apoCpI and CpI ADT ) and PHENIX 53 (CpI PDT , CpI ODT and CpI SDT ) and Coot. 54 At least two nal renement runs were conducted with PHENIX on all structures to improve comparability of the nal models. In order to estimate the occupancy of the 2Fe H -cluster in the structures of CpI ADT and other derivatives, we applied a partial occupancy renement at the nal stage of PHENIX renement. Simulated annealing omit maps were calculated with PHENIX, omitting the H-cluster with the bridging cysteine residue and the residues around the central cavity as well as all atoms within the central cavity for the semisynthetic [FeFe]-hydrogenases and apoCpI, respectively.

Conflict of interest
The authors declare no competing nancial interest.

Accession numbers
The coordinates and structure factors for all structures were deposited with the Protein Data Bank under the following accession numbers: apoCpI: 4XDD, CpI ADT : 4XDC, CpI PDT : 5BYR, CpI ODT : 5BYQ, CpI SDT : 5BYS.