Structural and catalytic characterization of Blastochloris viridis and Pseudomonas aeruginosa homospermidine synthases supports the essential role of cation–π interaction

The homospermidine synthases from P. aeruginosa and B. viridis, as well as their single-residue variants, are compared based on crystal structures and activity assays. A high structural similarity is demonstrated, suggesting the equivalent involvement of relevant residues in the reaction mechanism and catalytic dependence on cation–π interaction.


Introduction
Polyamines are involved in various processes in nearly all organisms in the three domains of life (Michael, 2016). The composition and roles of polyamines in prokaryotes are highly diverse and species-specific (Hamana et al., 2007;Hamana & Matsuzaki, 1992). Pseudomonas aeruginosa is a Gram-negative opportunistic pathogen that is frequently responsible for nosocomial infections (Walter et al., 2018) and is known to develop multi-drug-resistant or even pan-drug-resistant strains (Horcajada et al., 2019). Polyamines in P. aeruginosa have been demonstrated to be required for growth (Bitonti et al., 1982) and to affect biofilm formation (Cardile et al., 2017;Qu et al., 2016;Williams et al., 2010) and the susceptibility to antibiotics (Kwon & Lu, 2006a,b, 2007. Expression of the type III secretion system, a major virulence determinant in P. aeruginosa, was discovered to be influenced by the polyamine transporter SpuDEFGH (Anantharajah et al., 2016; Wu et al., 2012;Zhou et al., 2007). The polyamine-catabolizing ISSN 2059-7983 enzyme PauA2 was further shown to maintain the tolerance for polyamines, as disruption of this enzyme caused exogenous polyamines to exert bactericidal effects (Yao et al., 2012). While these studies suggest polyamine metabolism and the associated proteins to be potential antibiotic targets, neither homospermidine synthase (HSS; EC 2.5.1.45) nor its products have been addressed in P. aeruginosa so far.
Bacterial HSS is NAD + -dependent, active at neutral to basic pH with an optimum at pH $8.8, and shows increased activity in the presence of 50 mM potassium ion (Tait, 1979;Yamamoto et al., 1993). 1,3-Diaminopropane (DAP) acts as a strongly competitive inhibitor (Tait, 1979;Yamamoto et al., 1993), although it is converted with PUT into spermidine.
A profound structural characterization of the bacterial HSS from Blastochloris viridis (BvHSS) has previously been reported (Krossa et al., 2016). B. viridis is a Gram-negative, photoheterotrophic -proteobacterium with a simple photosynthetic system and thus is a commonly studied model organism for plant biochemistry (Konorty et al., 2009). Crystal structures of wild-type BvHSS and its single-residue variants in complex with various polyamines have been presented together with qualitative activity assays (Krossa et al., 2016). Based on these findings, the relevant residues involved in catalysis were identified and a detailed reaction mechanism was proposed. BvHSS forms a dimer, with each subunit containing a boot-shaped substrate-binding pocket between an NAD(P)-binding Rossmann-like domain and an HSS-like domain. The cofactor NAD + is bound as a prosthetic group in the binding pocket with its nicotinamide ring being part of the active site. An 'ionic slide' (residues Asp94 and Glu117) was proposed to lead positively charged amine substrates from the entrance to the binding pocket into the active site. The entrance tunnel is lined by a so-called 'track-and-trace' loop (residues 114-130). Krossa and coworkers identified three binding sites for the amino groups of reaction components within the substrate-binding pocket (Krossa et al., 2016). Assuming the amino groups of the reaction components to be protonated at neutral to slightly basic pH, the 'innermost' N atoms of PUT (N1) and of HSP (N01) (for numbering, see Fig. 2) mainly bind via a salt bridge to Glu210 and, in addition, via a hydrogen bond to Asn162 at the 'inner amino site'. At the 'outer amino site', the 'outer' amino group of HSP (N11) forms a salt bridge with Glu237 and a hydrogen bond to the O2D atom of NAD + . The N atoms PUT N2 and HSP N06 are located at the 'center amino site' between Asn162 and His296 as well as between the ring planes of the Trp229 indole ring and the nicotinamide ring. The residues constituting the 'center amino site' also form the active site by surrounding the C atoms PUT C4 and HSP C05. The nicotinamide ring of NAD + is assumed to serve as a hydride acceptor and donor (Bö ttcher et al., 1994), while His296 acts as a strong base by being arranged in a triad with the neighboring residues Glu237 and Glu298. Based on the interaction geometry between the Trp229 indole plane and the bound PUT and HSP molecules (PDB entry 4tvb; Krossa et al., 2016), the aromatic indole ring was suggested to stabilize positively charged amino groups and transiently charged C atoms of reaction components and intermediates by cation-interaction during catalysis (Krossa et al., 2016). Cation-interactions are noncovalent interactions (Dougherty, 1996;Sunner et al., 1981) and can approximately be explained by electrostatic models (Mecozzi et al., 1996a,b) as the attraction between the positive charge of a cation and the negative partial charge of the quadrupole moment of a system. Binding energies are in the range of those of hydrogen bonds and salt bridges in aqueous solution (Pletneva et al., 2001). The key steps of the previously proposed reaction mechanism (Krossa et al., 2016) are described in the following, assuming the amino groups of the reaction components to be protonated (Fig. 2, steps 1-8).
Initially, His296, as a strong base, probably generates a hydroxide anion, which in turn deprotonates the cation-stabilized PUT amino group at the 'center amino site' (steps 1-2). A hydride transfer from PUT C4 to NAD + C4N yields a protonated imine (steps 2-3). The positive charge of the imine might be transiently stabilized at C4 by a geometrically more favorable cation-interaction with C4 than with N2 ( Supplementary Fig. S2, step 4). This would further facilitate a nucleophilic attack by a water molecule, resulting in the formation of 4-aminobutanal by deamination (steps 3-4). After protonation of the aldehyde O atom by His296 (steps 4-5), a positive charge might again be transiently stabilized at the C4 atom (PUT nomenclature) of the aldehyde by cationinteraction to facilitate another nucleophilic attack (Supplementary Fig. S2, step 9). The amino group of a second PUT molecule is deprotonated by His296 (step 5) and performs a nucleophilic attack at the C4 atom (step 6), generating a protonated Schiff base (step 7). Hydride transfer from NADH ultimately results in the production of HSP (steps 7-8), which might accept a proton at its center amino group from the still protonated His296 before leaving the active site. Apart from the stabilization of transient carbocations, the Trp229 indole ring was also proposed to be involved in (geometrically less favored) cation-interactions with the positively charged nitrogen moieties of intermediates such as the protonated imine (step 3) and the protonated Schiff base (step 7). This reaction mechanism is supported by four inactive variants affecting the Glu-His-Glu triad (E237Q, H296S and E298Q) and the 'center amino site' (N162D) (Krossa et al., 2016). However, the roles of other residues, including the 'ionic slide' (Asp94 and Glu117), Glu210 at the 'inner amino site' and Trp229 as a potential cation-interaction partner, were not experimentally evaluated. In addition, the question arises as to whether these findings can be transferred to bacterial HSS from medically more relevant species such as P. aeruginosa.
To address these questions, we present the crystal structure of P. aeruginosa HSS (PaHSS; PDB entry 6y87) and confirm a high structural similarity between PaHSS and BvHSS, including the location of the catalytically relevant residues. We chose BvHSS as a model for further mechanistic characterization of PaHSS and of bacterial HSS in general because (i) a high sequence conservation of the catalytically relevant residues (BvHSS numbering: Glu117, Asn162, Glu210, Trp229, Glu237, His296 and Glu298) in bacterial HSS orthologs has previously been reported (Shaw et al., 2010), (ii) PaHSS and BvHSS consistently exhibit high structural similarity, as presented in this study, and (iii) BvHSS crystals provide superior X-ray diffraction data compared with PaHSS crystals. To substantiate the previously proposed roles of catalytically relevant residues, we have produced additional single-residue variants of BvHSS for characterization by X-ray crystallography and by quantitative activity assays. Our results confirm the important roles of the residues forming the 'ionic slide' and the 'inner amino site' and support the requirement for the active-site tryptophan as a potent cation-interaction partner. In the context of polyamine metabolism in P. aeruginosa being a potential target for antibiotic design, this study might provide possible approaches to interfere with HSP production in P. aeruginosa and other bacteria.

Materials
Primers were obtained from Eurofins Genomics (Ebersberg, Germany). DNA purification kits were purchased from Macherey-Nagel (Dü ren, Germany). Pre-packed chromatography columns (HisTrap HP, HiTrap Q HP, GSTrap HP and HiTrap Desalting) for preparative protein purification were obtained from GE Healthcare (Chicago, Illinois, USA). Other reagents, unless stated otherwise, were reagent grade or better and were purchased from Carl Roth (Karlsruhe, Germany) or Sigma-Aldrich (Darmstadt, Germany).

Cloning of the PaHSS expression plasmid
The amino-acid sequence of PaHSS (UniProtKB Q6X2Y9) was back-translated into an Escherichia coli K12 codonoptimized DNA sequence and extended by terminal 5 0 -CC and TAATAACTCGAG-3 0 DNA sequences ( Supplementary  Fig. S3). The resulting DNA sequence was synthesized by General Biosystems Inc. (Durham, North Carolina, USA) and was provided as an insert in a pUC57 plasmid. This plasmid was transformed into E. coli XL1-Blue cells for plasmid production. The PaHSS-coding DNA sequence was transferred from the pUC57 plasmid into a pETM-14 plasmid (EMBL, Heidelberg, Germany) via the NcoI/XhoI restriction sites on both plasmids (endonucleases from New England Biolabs GmbH, Frankfurt am Main, Germany) and ligated with T4 DNA Ligase (Thermo Fisher Scientific, Waltham, Massachusetts, USA), yielding the PaHSS pETM-14 plasmid. The plasmid codes for an N-terminal 6ÂHis tag followed by a human rhinovirus 3C (HRV3C) protease cleavage site fused to the PaHSS sequence. The PaHSS pETM-14 plasmid was transformed into E. coli XL1-Blue cells for plasmid amplification and the correct sequence of the PaHSS insert was confirmed by DNA sequencing (Eurofins Genomics). For protein expression, PaHSS pETM-14 was transformed into E. coli BL21(DE3) Gold cells.

Cloning of expression plasmids for BvHSS variants
The expression plasmid BvHSS pETM-14 coding for wildtype BvHSS (UniProtKB O32323) was previously generated (Krossa et al., 2016). Expression of the plasmid yields wildtype BvHSS with an N-terminal 6ÂHis tag followed by an HRV3C protease cleavage site. For the generation of singleresidue BvHSS variants, the QuikChange site-directed mutagenesis protocol from Agilent was followed using Phusion high-fidelity polymerase (Thermo Fisher Scientific), BvHSS pETM-14 as the template and variant-specific primer pairs (Supplementary Table S1). After PCR, the template plasmid was digested with the endonuclease DpnI (Thermo Fisher Scientific), followed by the transformation of PCR products into E. coli XL1-Blue cells for clone screening and confirmation of the correct construct sequence by DNA sequencing. For protein expression, plasmids were transformed into E. coli BL21(DE3) Gold cells.

Protein expression and purification
Protein expression of wild-type BvHSS and BvHSS variants was performed as described previously (Krossa et al., 2016). The expression of PaHSS was similarly conducted with two modifications of the BvHSS expression protocol to increase the yield of soluble PaHSS: the induction of expression was performed with 0.1 mM isopropyl -d-1-thiogalactopyranoside (instead of 1 mM) and expression was continued for 20 h after induction (instead of 4 h).
For the purification of PaHSS, the cells were resuspended 1:10(w:v) in basis buffer [50 mM Tris-HCl pH 7.4, 25 mM KCl, 2 mM dithiothreitol (DTT)] supplied with 0.3 M NaCl, 25 mM imidazole, 2 mM NAD + and 1 mM phenylmethanesulfonyl fluoride. The cells were lysed by one passage through an Emulsiflex C3 homogenizer (Avestin, Mannheim, Germany) at pulses of approximately 1200 bar. After centrifugation (45 min, 75 600g, 283 K), PaHSS was purified from the clarified supernatant via chromatographic purification steps using an Ä KTA FPLC system (GE Healthcare) equipped with UV and conductivity detectors. The supernatant was first loaded onto an equilibrated [15 column volumes (CV) basis buffer] 5 ml HisTrap HP column. The column was then washed with 5 CV basis buffer supplied with 25 mM imidazole and elution was subsequently performed at $0.2 M imidazole by linearly increasing (25 mM imidazole per CV) the concentration of imidazole in basis buffer to 0.5 M. The buffer of the eluted protein was exchanged for fresh basis buffer supplied with 2 mM NAD + using 5 ml HiTrap Desalting columns. The N-terminal 6ÂHis tag was removed by incubating the PaHSS with 1:100(w:w) glutathione S-transferase (GST)-tagged HRV3C protease overnight at 277 K, followed by passage of the sample through connected 5 ml HisTrap HP and GSTrap HP columns in basis buffer. The flowthrough was loaded onto two connected 1 ml HiTrap Q HP columns equilibrated with basis buffer. PaHSS was eluted between $18 and 23 mS cm À1 by linearly increasing (20 mM NaCl per CV) the NaCl concentration in basis buffer to 0.3 M NaCl. The eluted sample was supplied with fresh basis buffer using HiTrap Desalting columns and was concentrated to $5 mg ml À1 using 10 kDa Amicon Ultra-15 centrifugal filter units (Merck Millipore, Darmstadt, Germany). After the addition of 2 mM NAD + , the PaHSS sample was flash-frozen in liquid nitrogen and stored at 193 K.
Purification of wild-type BvHSS and BvHSS variants was performed as described previously (Krossa et al., 2016), but research papers the standard buffer used in this study was composed of 50 mM bis-Tris propane-HCl pH 9, 25 mM KCl, 2 mM DTT.
Protein concentrations were estimated according to the Lambert-Beer law by measurement of the absorbance at 280 nm. The respective extinction coefficients were calculated with the ProtParam tool (Gasteiger et al., 2005). Reductive SDS-PAGE analysis revealed a purity of >99% for wild-type BvHSS and the BvHSS variants and >95% for PaHSS. The oligomerization states of the proteins were determined by SEC-MALS as described in Supplementary Section S1.

HSS activity assay
To quantify the conversion of PUT to HSP, 10 mM PUT dihydrochloride was mixed into three prewarmed reaction pools (independent triplicate measurements) consisting of 3.8 mM protein, 50 mM potassium phosphate pH 7.4 and 0.2 mM NAD + at 310 or 298 K. At different time points, 100 ml from the reaction pools was transferred into 50 ml 100%(w/v) trichloroacetic acid to stop the reaction. Precipitated protein was removed by centrifugation (18 000g for 10 min) and 1 ml supernatant was neutralized in 589 ml 0.2 M sodium borate-NaOH buffer pH 8.5. 25 ml of the neutralized sample was mixed with 10 ml 10 mM 6-aminoquinolyl-N-hydroxysuccinimidyl carbamate (Synchem UG & Co. KG, Felsberg, Germany) in dry acetonitrile to label the polyamines. After incubation for 5 min at room temperature in the dark, the sample was diluted with 35 ml 0.2 M sodium borate-NaOH buffer pH 8.5, and 50 ml was loaded onto a ProntoSIL HyperSorb ODS (C18) column (4.6 Â 250 mm, 120 Å pore size, 5 mm particle size; Bischoff Chromatography, Leonberg, Germany) connected to the HPLC system described in Section S1. The mobile phase was a gradient of 25 mM triethylamine (pH 4.8, adjusted with acetic acid), 80%(v/v) acetonitrile and methanol (Table 1) as described previously (Weiss et al., 1997).
The flow rate was 1.3 ml min À1 at 306 K and fluorescence was measured at ex = 248 nm and em = 398 nm at a photomultiplier gain of 11. Signal peaks were integrated for quantification using the ChemStation software version B.01.03 (Agilent). To confirm the retention times of the respective labeled compounds and to convert fluorescence peak areas to compound quantities, different amounts of PUT and HSP (Toronto Research Chemicals Inc., Ontario, Canada) were directly labeled and separated on the column (Supplementary Figs. 4a and 4b) to determine a standard curve (Supplementary Fig. 4c). The mean concentration of the produced HSP was plotted against the reaction time using GraphPad Prism version 5.00 (Supplementary Figs. 4d and 4e). The initial velocities (and their standard deviations) were calculated from the linear fit to all data points (not only the mean) of the initial, linear part of each progression curve, considering at least three time points.

Crystallization and data collection
Crystallization was performed by the hanging-drop vapordiffusion technique at 291 K. PaHSS was crystallized in drops consisting of 1 ml protein solution ($3.3 mg ml À1 PaHSS, 33 mM Tris-HCl pH 7.4, 17 mM KCl, 1.3 mM DTT, 1.3 mM NAD + , 0.2 M agmatine sulfate) and 1 ml reservoir solution [0.1 M HEPES-NaOH pH 7.5, 18-23%(w/v) poly(acrylic acid sodium salt) average molecular weight 5100 (Sigma-Aldrich), 10 mM adenosine triphosphate] by equilibration against 0.5 ml reservoir solution. The addition of agmatine sulfate was found to significantly improve the growth and diffraction quality of the crystals. Crystals appeared after $1-3 days and were disintegrated by vortexing with glass beads in reservoir solution. Crystal nuclei were streak-seeded into hanging drops of a fresh crystallization setup prepared as described above. Homogeneous, trigonal prism-shaped crystals grew within several days in a narrow concentration range of poly(acrylic acid sodium salt) and were briefly soaked with a mixture of 20%(v/v) glycerol and 80%(v/v) reservoir solution before vitrification in liquid nitrogen.

Structure determination and representation
Indexing and integration of diffraction data was performed with XDS (Kabsch, 2010). Using the CCP4 program suite (Evans, 2011;Potterton et al., 2003), the space group was estimated with POINTLESS. Scaling, data reduction and assignment of free R flags to 5% of the reflections were performed with the CCP4 program SCALA. Molecular replacement and automated refinement were performed using Phenix (Liebschner et al., 2019). Inspection of molecularreplacement results and manual model building between automated refinement cycles was performed by Coot (Emsley et al., 2010). For automated refinement by Phenix, the included refinement parameters were atomic positions (Cartesian and real space), individual B factors and occupancies. The 'maximum-likelihood' target function was applied and weight optimization was enabled for the X-ray target function and geometry restraints as well as B-factor restraints. Directly after molecular replacement, rigid-body refinement and Cartesian simulated annealing was included in the refinement. At later stages in the refinement, automatic addition/updating of H atoms and water molecules was performed. All structures presented in this study were solved by molecular replacement using the Phaser maximum-likelihood molecular-replacement tool implemented in Phenix.
For the BvHSS variants, the wild-type BvHSS structure (PDB entry 4plp; Krossa et al., 2016), excluding the NAD + molecules and any other ligands, served as a search model. After molecular replacement, the mutated residue of each variant was accordingly exchanged and the NAD + molecules were fitted into the binding pockets. After refinement including Cartesian simulated annealing, the model was iteratively optimized by manual model building and automated refinement.
For the PaHSS structure (PDB entry 6y87), a PaHSS homology model without the NAD + molecule or other ligands was calculated using the standard settings of the PRIMO server (Hatherley et al., 2016) and was used as a search model. For the homology model, the wild-type BvHSS model (PDB entry 4plp; Krossa et al., 2016) and the PaHSS amino-acid sequence (UniProtKB Q6X2Y9) served as input. Six molecules of PaHSS were estimated to fit into the asymmetric unit according to the Matthews coefficient calculator software within the CCP4 suite (Table 2). In the Phaser maximumlikelihood molecular-replacement tool, automated molecular replacement was performed with six PaHSS homology models as search models per asymmetric unit, enabling the option to try both enantiomeric space groups. After the correct placement of six PaHSS search models within the asymmetric unit, NAD + molecules were fitted into each active site and refinement including Cartesian simulated annealing was performed. The models were further improved by iterative manual and automated refinement.
Structural images were prepared with PyMOL (Schrödinger). Omit maps were calculated both with the composite omit map tool and the polder map tool implemented in Phenix (Liebschner et al., 2017). Using the composite omit map tool, only one specified part of the model was omitted, followed by refinement including Cartesian simulated annealing to remove model bias. Bulk solvent was allowed to occupy omit regions in the refinement step. For PUT composite omit maps, only the PUT molecule was omitted. For composite omit maps targeting the exchanged residue in each BvHSS variant, the mutated residue together with ten N-terminally and ten C-terminally adjacent residues were omitted. Polder maps were calculated by omitting the PUT molecules and exclusion of the bulk solvent within a distance of 5 Å around the ligands.
Superimposition of PaHSS molecule A onto the other PaHSS molecules (PDB entry 6y87; Supplementary Table S2) as well as superimposition of wild-type BvHSS onto BvHSS variants (Supplementary Table S3) was performed with PyMOL using the super algorithm over five cycles. Therefore, only protein atoms and NAD + atoms without hydrogens and only alternative conformations A were considered. Rootmean-square distance (r.m.s.d.) values were subsequently calculated with the command rms_cur. For the generation of a structure-based sequence alignment of BvHSS (PDB entry 4tvb chain B) with PaHSS (PDB entry 6y87 chain A; Supplementary Figs. S5 and S6) and of PaHSS (PDB entry 6y87 chain A) with L. pneumophila HSS (LpHSS; PDB entry 2ph5; Supplementary Fig. S7), the MatchMaker tool of UCSF Chimera 1.14 (Pettersen et al., 2004) was used. The initial sequence alignment was based on the Needleman-Wunsch algorithm, the BLOSUM62 substitution matrix and secondary-structure scoring. The subsequent superimposition of structures included iterative pruning of C -atom pairs which exceeded a distance of 2 Å . The resulting structure-based sequence alignment underlies a C -based residue-residue distance cutoff of 5 Å .
For the depiction of solvent-accessible surfaces of binding pockets, pockets were filled with dummy atoms (surface probe sphere radius, 4 Å ; dummy-atom radius, 1.4 Å ; grid spacing, 0.2 Å ) using HOLLOW 1.1 (Ho & Gruswitz, 2008). After the addition of water molecules present in the respective structures, the surface of these dummy atoms was depicted using PyMOL.

Results and discussion
3.1. Organization of six highly similar PaHSS molecules in the asymmetric unit The PaHSS structure was solved in space group P3 2 12 using the Phaser tool implemented in Phenix and allowing both enantiomorphic space groups as possible solutions. Six PaHSS molecules were modeled into the asymmetric unit and refined to a maximum resolution of 2.42 Å ( Table 2). All PaHSS residues of molecule A could be modeled into the electrondensity distribution, including an N-terminal residue belonging to the remains of the cleaved affinity tag (residue 0).

research papers
Due to the local absence of interpretable electron density, a few residues belonging to the N-terminal loop and to a loop near the C-terminus were not modeled in the other PaHSS molecules (the missing residues were 1-3 in PaHSS molecule B, 1-5 and 449-458 in molecule C, 1-4 and 449-457 in molecule D, 1-3 and 451-456 in molecule E and 1-3 and 450-457 in molecule F). These loops are solvent-exposed on the protein surface without influence on the binding pocket. All six PaHSS molecules contain an NAD + molecule in their binding pockets, and molecules A, C and E additionally contain a PUT molecule in the active site. The degree of structural similarity between the PaHSS molecules was examined by superimposition of molecule A onto the other molecules. The resulting r.m.s.d. values considering all atoms (not only C ) were between $0.73 and $0.83 Å (Supplementary Table S2), demonstrating a high structural similarity. Determination of crystal contacts between the individual protein molecules (for PDB entry 6y87) was performed using the PDBsum web server (Laskowski et al., 2018). In addition to various smaller to moderate contact interface areas between the molecules, larger interface areas were detected between the pairwiseassociated molecules A and B (1381-1401 Å 2 ), C and D (1328-1332 Å 2 ), and E and F (1379-1370 Å 2 ). Although PaHSS was shown to be a monomer in solution by SEC-MALS (the theoretical molecular weight of the monomer and the estimated molecular weight according to HPLC-MALS were 52 kDa; data not shown), these interface areas indicate the formation of three crystalline dimers by the six protein molecules within the asymmetric unit. The monomeric state of PaHSS in solution is also supported by the shape complementarity ( Table 2 Data-collection and refinement statistics.
Values in parentheses are for the highest resolution shell.

Overall structure and binding pocket of PaHSS
PaHSS is composed of an NAD(P)-binding Rossmann-like domain (residues 1-155 and 389-419) and an HSS-like domain (residues 156-388 and 420-469), as confirmed by the CATH database (Knudsen & Wiuf, 2010) and as previously observed for the BvHSS structure (Krossa et al., 2016). The binding pocket is formed between the domains and accommodates an NAD + molecule (Fig. 3a).
The same residues as were previously discussed to be involved in catalysis by BvHSS (Krossa et al., 2016) are present in the binding pocket of PaHSS with the sequence numbering shifted by À4 (Fig. 3b). Transferring the proposed reaction mechanism of BvHSS to PaHSS, the following roles could be assigned to these PaHSS residues. Asp90 and Glu113 form the 'ionic slide' at the entrance tunnel of PaHSS. Tyr119 borders the entrance tunnel and is part of the so-called 'track-andtrace' loop (PaHSS residues 110-124). The equivalent BvHSS residue Tyr123 was assumed to regulate access to the binding pocket by modifying the pore dimensions (Krossa et al., 2016). The NAD + molecule in PaHSS lines the entrance tunnel with its pyrophosphate moiety, contributes to the 'outer amino site' with its O2D O atom and inserts its nicotinamide moiety into the active site. The nicotinamide ring plane, the indole ring plane of Trp225 and His292 surround the 'center amino site' and the active site. His292 thereby forms a triad with Glu233 and Glu294. The PUT molecule is located with its 'outer' amino group and C4 atom at the 'center amino site'/active site and with its 'inner' amino group near Glu206 at the 'inner amino site'. As in BvHSS, PaHSS forms a side pocket at the bottom of the binding pocket, which is confined by Asn131, among others.

PUT molecules and a potential NAD adduct in the active sites of PaHSS
The electron-density distribution of PaHSS molecules A, C and E revealed the binding of a PUT molecule in each active site (Fig. 4). PUT was not added at any stage of purification or crystallization. We assume it to originate from the expression host and to be co-purified within the active site of PaHSS. The possibilities of a contaminated agmatine solution or the catalytic conversion of agmatine by PaHSS was ruled out by separation of the utilized agmatine solution by reversed-phase HPLC before and after incubation with PaHSS. Neither contamination nor conversion of agmatine were detected.
Both the architecture of the active sites and the location of the PUT molecules are very similar in all three PaHSS molecules. The 'inner' and 'outer' amino groups of the PUT molecules form salt bridges with the carboxyl O atoms of Glu206 and Glu233, respectively, with distances of between 2.8 and 3.2 Å . The cation-interaction with the Trp225 indole ring is described later (Fig. 7). The conformation of the PUT carbon chains varies slightly between the respective PaHSS molecules. The exact tracing of the carbon chains might be complicated by their assumed flexibility and the moderate resolution of the structure. Omit maps and polder maps were calculated to evaluate the placement of the PUT molecules in the active sites (I, II and III in Fig. 4). Inspecting these omit  maps, the PUT molecule in PaHSS molecule E was most distinctly supported by the diffraction data. The electrondensity distribution at the active site of PaHSS molecule B could not be unambiguously interpreted (Fig. 5).
The electron density contouring the bond between the nicotinamide C4N and C5N atoms clearly merges into a longitudinal, uninterpreted density, which extends to the 'inner amino site' near Glu206. One explanation for this observation could be radiation damage caused by the X-ray diffraction experiment. This might have turned the nicotinamide ring into a radical, which in turn resulted in adduct formation between the nicotinamide ring and another molecule. Another potential explanation for this observation could be a covalent nicotinamide adduct occurring as an intermediate in the reaction mechanism, although additional experiments will be needed to confirm such an adduct. In the wild-type BvHSS structure (PDB entry 4tvb), close proximity of a PUT molecule and an HSP molecule to the nicotinamide ring was similarly observed and was suggested to reflect partial, transient hydride transfers between the respective molecules (Krossa et al., 2016). These observations might instead be caused by covalent NAD adducts occurring as intermediates in the bacterial HSS. In BvHSS (PDB entry 4tvb), the geometry of the potential bonds between (i) the PUT C4 and HSP C05 C atoms and (ii) the nicotinamide ring C4N atom would thereby satisfy an sp 3 hybridization of C4N. In the PaHSS molecule B presented here, the uninterpreted density contacts the nicotinamide ring in a rather side-on geometry, which perfectly satisfies neither sp 2 nor sp 3 hybridization of C4N in the context of an adduct structure. Also, the unin-  resolution, the presence and identity of such an adduct remains to be determined by additional experiments. In line with our assumption, NAD adducts have previously been described to be formed in various enzymes between the nicotinamide ring C2N or C4N atoms and both inhibitor compounds (Jacques et al., 2003;Rubach & Plapp, 2003;Benach et al., 1999;Becker & Roberts, 1984)  The electron-density distribution at the active site of PaHSS molecule D did not indicate the binding of a PUT molecule. Inspection of molecule F showed distinct electron density at the active site, which might be caused by a PUT molecule similarly located as in molecules A, C and E. However, the density was not sufficient to accommodate a PUT molecule. While the longitudinal density was quite defined near Glu206 (the inner amino site), it did not extend all the way to the site surrounded by Trp225, His292 and the nicotinamide ring. This might indicate low occupancy of a potentially bound PUT molecule or increased flexibility of the PUT carbon chain at the latter site.

Structural comparison of PaHSS and BvHSS
BvHSS was previously reported to form dimers (Krossa et al., 2016), which was equivalently confirmed by SEC-MALS in this study for the wild-type protein and all variants (the theoretical molecular weight of the wild-type and variant dimers was 106 kDa; the estimated molecular weight according to HPLC-MALS was 98-99 kDa; data not shown). While PaHSS was found to be a monomer in solution (the theoretical molecular weight of the monomer and the estimated molecular weight according to HPLC-MALS were 52 kDa; data not shown), six monomers in the asymmetric unit form three crystalline dimers with significant interface areas ($1330-1400 Å 2 , as determined by PDBsum; Laskowski et al., 2018). Superimposition of the crystalline PaHSS dimers onto the wild-type BvHSS dimer (PDB entry 4tvb) revealed a highly similar quaternary structure of the BvHSS and PaHSS dimers, demonstrating a similar dimerization interface. However, the interface area of BvHSS is larger ($1700 Å 2 according to PDBsum) than the interface areas of the crystalline PaHSS dimers, which could explain the different oligomerization states of PaHSS and BvHSS in solution.
The overall structural similarity between PaHSS (PDB entry 6y87 chain A) and BvHSS (PDB entry 4tvb chain B; Krossa et al., 2016) was examined by superimposition of both molecules and the subsequent calculation of a structure-based sequence alignment (Supplementary Fig. S5). This alignment shows a sequence identity of $44% and almost complete alignment of the structures. The few and short non-aligning gaps (C r.m.s.d. cutoff 5 Å ) are exclusively caused by solventexposed parts of both structures, which are not involved in formation of the binding pocket ( Supplementary Fig. S6).
Superimpositions of the structures exclusively based on a single domain (either the Rossmann-like domain or the HSSlike domain) effectively showed the same alignment as observed for superimposition of the complete structures. In conclusion, PaHSS and BvHSS share equivalent intramolecular domain orientations.
In addition, a structure-based sequence alignment of PaHSS (PDB entry 6y87 chain A) and L. pneumophila HSS (LpHSS; PDB entry 2ph5) was performed ( Supplementary  Fig. S7). The resulting sequence identity is $41% and the structures align with high similarity and with an equivalent intramolecular domain orientation. Non-aligning parts (C r.m.s.d. cutoff 5 Å ) are limited to a few solvent-exposed areas.
To further assess the similarity of the superimposed PaHSS and BvHSS structures, the architectures of their binding pockets were compared. Both binding pockets are 'bootshaped' and are mainly defined by the respective 'track-andtrace' loop, -helix J (PaHSS residues 226-236; BvHSS residues 230-240) and the NAD + molecule (Figs. 6a and 6b). The NAD + molecules similarly insert their nicotinamide rings into the active site of the binding pockets, albeit with a slight offset. In addition, the 'track-and-trace' loop and particularly -helix J of PaHSS are shifted in relation to BvHSS, providing more space for the pocket entrance compared with the BvHSS structure (Fig. 6b).
This spatial effect was more closely inspected by measuring the dimensions of the binding-pocket entrances of PaHSS and BvHSS ( Supplementary Fig. S8). The PaHSS pocket entrance is shaped like a funnel, with a large opening (4.7 Â 8.6 Å ) which narrows further to a maximum of 3.5 Å . Mostly caused by the constricting -helix J, BvHSS has a smaller pore-like entrance (3.2 Â 3.6 Å ). Still, both entrance tunnels effectively impose similar spatial limitations for the access of compounds to the active sites. The 'bottom' of the binding pockets, which includes the active sites, provides comparable space in both structures.
To determine whether the reaction mechanism proposed for BvHSS could also apply to PaHSS, the locations and conformations of relevant residues within the binding pockets were compared between the superimposed structures. In general, these residues align with good agreement, albeit not to the same degree (Figs. 6c and 6d). Referring to the PaHSS numbering, the side chains of Glu206, His292 and Glu294 align almost perfectly. Although PaHSS residue Glu233 is part of the shifted -helix J, its carboxyl group is still very similarly placed compared with the equivalent BvHSS residue Glu237. While Glu113 and Tyr119 belong to the shifted 'track-andtrace' loop, Asp90 is part of another slightly displaced loop. These residues exhibit a certain offset but retain highly similar conformations and/or placement of functional groups. Trp225 belongs to a short loop region directly preceding -helix J and is also affected by its shift in the PaHSS structure. Apart from a small offset comparing the tryptophan residues of PaHSS and BvHSS, the indole rings are equally oriented and parallel to each other. Similarly, the NAD + molecules are shifted with respect to one another but adopt highly similar conformations, including a coplanar alignment of their nicotinamide rings. In research papers summary, the architecture of both binding pockets and especially the orientation of the relevant side chains are highly similar in both the PaHSS and BvHSS structures. The most significant deviation regarding catalysis might be the shifted tryptophan residues. Still, BvHSS can be assumed to qualify as a mechanistic model for PaHSS.

Cation-p interaction geometry in PaHSS and BvHSS
Cation-interaction was previously proposed to be part of the reaction mechanism of BvHSS (Krossa et al., 2016). The negative electrostatic potential of Trp229 was suggested to interact with the positively charged amino group of substrates and intermediates, resulting in efficient positioning of the compounds in the active site. In addition, a more favorable cation-interaction geometry was shown between the side chain of Trp229 and the C4 atom of the bound PUT molecule. Since the HSS reaction mechanism is based on nucleophilic attacks at C4 (Fig. 2, steps 3 and 6), cation-interaction was suggested to stabilize a transient positive charge at C4, with a higher interaction energy compared with N2, to catalyze the reaction. In the following, cation-interaction strengths are comparatively estimated for the PaHSS structure (PDB entry 6y87 chain E) and the BvHSS structure (PDB entry 4tvb chain B; Krossa et al., 2016). Estimations are based on the interaction geometry between the two PUT atoms N2 and C4 and   Krossa et al., 2016). Protein models are depicted in cartoon representation with selected side chains and NAD + molecules in ball-and-stick representation. (a, b) Depiction of the solvent-accessible binding pocket surfaces of PaHSS (red, transparent) and BvHSS (gray, opaque). In (b), the binding-pocket surfaces were clipped and the NAD + molecules were hidden for clarity. (c, d) Depiction of selected side chains (numbered according to the PaHSS sequence) in the superimposed binding pockets. In (d), the NAD + molecules are hidden for clarity. the benzene moiety of the indole rings (Fig. 7). The PUT molecule in PaHSS molecule E was chosen for the following comparison because it is most distinctly supported by all electron-density maps (Fig. 4) and provides the most favorable interaction geometry with regard to the C4 atom (an explanation is given below).
The geometry of cation-interactions has already been statistically evaluated based on protein structures and correlated to binding energies by quantum-chemical calculations (Pinheiro et al., 2017;Rapp et al., 2014;Marshall et al., 2009;Minoux & Chipot, 1999;Chipot et al., 1996;Burley & Petsko, 1986). As rule of thumb, the binding energy of a cationinteraction between the respective PUT atoms and the benzene moiety of tryptophan should increase with (i) decreasing centroid-cation distance, (ii) decreasing angle (with a maximum at 0 ) and (iii) increasing ' angle (with a maximum at 30 ; angles are explained in Fig. 7). However, the ' angle should play a minor role at the relatively small angles occurring in this study (Marshall et al., 2009).
The binding of the substrate PUT is assumed to be stabilized at the 'center amino site' by cation-interaction of its positively charged 'outer' ammonium moiety (atom N2). In PaHSS molecule E, the geometry between PUT N2 and the benzene moiety exhibits an N2-centroid distance of 4.5 Å at a angle of 34.7 and a ' angle of 21.1 . The PUT N2 atoms are similarly located in the three PaHSS molecules (A, C and E) and the interaction geometry between N2 and the tryptophan benzene moiety is effectively the same for the three bound PUT molecules. In BvHSS, the distance amounts to 4.2 Å , with a angle of 24.8 and a ' angle of 11.9 . The cation-interaction in BvHSS should consistently be stronger due to a smaller cation-centroid distance and a smaller angle.
In addition, reaction intermediates with a transient positive charge at the C4 atom were proposed to be stabilized by cation-interaction to facilitate a subsequent nucleophilic attack at C4 (Krossa et al., 2016). These intermediates could include the protonated imine and the protonated 4-aminobutanal (Fig. 2, steps 3 and 5), in which the positive charge could transiently be shifted to the C4 atom, where it might subsequently be stabilized by cation-interaction ( Supplementary Fig. S2, steps 4 and 9, respectively). In PaHSS molecule E, the PUT C4-centroid distance is 3.6 Å , with a angle of 18.5 and a ' angle of 15.5 (Fig. 7). The interaction geometries in the other PaHSS molecules with bound PUT molecules are less favorable but still qualify as a cation-interaction (PaHSS molecule A, C4-centroid distance of 4.2 Å , angle of 26.4 ; PaHSS molecule C, C4-centroid distance of 4.8 Å , angle of 16.6 ). In BvHSS the distance is 3.5 Å , the angle is 5.5 and the ' angle is 29.5 , representing a near-optimal cation-interaction geometry. While the C4-centroid distance in PaHSS molecule E is very similar, the angle in particular indicates a less favorable geometry compared with BvHSS. It should be noted that the PUT molecules detected in the crystal structures are not necessarily intermediates with a positive charge at C4 but are more probably regular substrates 'waiting' for further conversion. In particular, the interaction geometry in PaHSS should therefore be governed by cationinteraction involving PUT N2 rather than C4. This could also explain the assumed multitude of conformations of PUT molecules, resulting in complicated and different tracing of the carbon chains in the respective PaHSS molecules. Consistently, the stabilized amino  Geometry of cation-interaction between PUT and tryptophan in PaHSS and BvHSS. Interaction geometry between PUT atoms C4 and N2 and the tryptophan benzene moiety in (a, b) PaHSS (PDB entry 6y87 chain E) and (c, d) BvHSS (PDB entry 4tvb chain B; Krossa et al., 2016). Structures are shown in ball-and-stick representation, with distances as yellow dashed lines, angle legs as gray lines and angles as gray transparent triangles [not visible for ' in (c)]. The orthogonal projections of C4 and N2 onto the ring planes are shown as black spheres (C4 0 and N2 0 ). All angle legs originate from the centroid of the benzene moiety, including the dashed distance vectors (centroid to C4 and N2). The angle is spanned by the normal of the ring plane (gray, infinitely pointing upwards) and the C4 or N2 distance vector (yellow, dashed). The angle ' is between the vector pointing to C4 0 or N2 0 and the vector pointing to the CH2 ring carbon.
groups (N2 atoms) of the bound PUT molecules are similarly located within the active sites (with equivalent cationinteraction geometries), while the location of the presumedly unstabilized C4 atoms varies in relation to the tryptophan indole ring (with the most favorable cation-interaction geometry in PaHSS molecule E). However, a transient positive charge at PUT C4, occurring in the course of the catalytic reaction, would be stabilized with higher cation-interaction energy according to the interaction geometry than a positive charge at N2. In BvHSS, the closer proximity of PUT C4 to the nicotinamide ring was interpreted as hydride transfer from PUT C4 to NAD + C4N (Krossa et al., 2016). This potential situation leaves more room for speculation regarding the oxidation state of PUT C4 and its resulting involvement in cation-interactions. In summary, the interaction geometry between the indole ring and both PUT N2 and PUT C4 in PaHSS suggests significant cation-interaction energies (Pinheiro et al., 2017;Rapp et al., 2014;Marshall et al., 2009). However, the geometry in PaHSS is less favorable and the interaction energies are thus assumed to be weaker than in BvHSS. In conclusion, this inferior cation-interaction geometry in PaHSS might be the reason for its reduced activity compared with BvHSS (see below).

Activity of PaHSS, BvHSS and BvHSS variants
To assess the effect of each single-residue mutation on BvHSS catalysis and to compare the activities of wild-type BvHSS and PaHSS, the initial reaction velocities (V 0 ) of the enzymes were determined under identical reaction conditions. BvHSS has previously been reported to have a catalytic optimum at pH $8.8 (Tait, 1979;Ober et al., 1996). To better reflect the physiological reaction conditions and considering the instability of PaHSS at basic pH (thermal shift assays, not shown), assays were performed at neutral pH 7.4 and mostly at 310 K (Table 3, Supplementary Fig. S4).
As the only exception, comparison of wild-type BvHSS and PaHSS was performed at a reaction temperature of 298 K due to the inferior stability of PaHSS at elevated temperatures (thermal shift assays, not shown). Wild-type BvHSS was determined to exhibit a V 0 of 0.34 AE 0.05 mM min À1 at 310 K and 0.108 AE 0.009 mM min À1 at 298 K. For PaHSS, a V 0 of 0.037 AE 0.004 mM min À1 at 298 K was determined, amounting to roughly 33% of the BvHSS activity at 298 K. The reason for the lower activity of PaHSS cannot be resolved at this point; the less favorable cation-interaction geometry could be one explanation (see above). Another explanation could be the inferior stability of PaHSS compared with BvHSS in terms of pH and temperature, which might cause a certain fraction of the purified PaHSS to be inactive/less active. However, protein aggregation was ruled out by the detection of a monodisperse PaHSS elution peak corresponding to a monomer by HPLC-MALS analysis. Furthermore, the high structural similarity of the six PaHSS molecules in the asymmetric unit of the crystal structure (PDB entry 6y87) did not indicate the presence of structurally impaired protein molecules. The temperature-dependence of BvHSS is in line with previous studies, describing an increase in activity up to 318 K . Activity assays were performed at a substrate concentration of 10 mM PUT dihydrochloride, which is sufficient for the oversaturation of wild-type BvHSS (BvHSS K m = 0.26 mM; Bö ttcher et al., 1994). The presented V 0 value of BvHSS at 310 K can accordingly be converted to a k cat value of 1.5 s À1 . This k cat value lies between the previously reported V max value of 22 nkat mg À1 (Bö ttcher et al., 1994) (corresponding to a k cat value of 1.2 s À1 ) and a specific activity of 83 nkat mg À1 (corresponding to a k cat value of 4.4 s À1 ) . The reason for the distinct difference between the two reported values remains elusive. Also, the reported values were determined near the catalytic optimum at pH 8.5 and are thus expected to be higher than the value determined at pH 7.4 in this study. In any case, our assay provides comparable data and primarily serves for a relative catalytic comparison of the different BvHSS variants in this study. More extensive activity assays will be required in the future to determine K m and k cat values for PaHSS and the BvHSS variants. However, our approach allowed the confirmation of significant effects of the single-residue mutations on HSP production. Considering the activity of wild-type BvHSS at 310 K as a reference (0.34 mM min À1 = 100%), the D94N (61%) and D94K (52%) variants showed moderately reduced activity. The W229F (2%), E117Q (1.5%) and W229Y (1%) variants were still able to produce HSP from PUT, albeit with drastically decreased activity. The E117K, E210A, E210K, E210Q, W229A, W229E, W229H and W229K variants were completely inactive and did not show quantifiable production of HSP over 5 h (V 0 < 9 Â 10 À5 mM min À1 ). These results are discussed below in the context of the structural evaluation of the variants.

Overall structures of BvHSS variants
All BvHSS variants were solved in space group P22 1 2 1 with similar unit-cell parameters and mostly refined to a resolution research papers Acta Cryst. (2021). D77, 1317-1335 Helfrich and Scheidig Homospermidine synthases 1329 Table 3 Catalytic activity of PaHSS, BvHSS and BvHSS variants.
The time-resolved production of HSP was quantified after the addition of 10 mM PUT dihydrochloride. Initial reaction velocities were determined from the linear fit to the initial data points of the progression curves.

Protein
T (K) Initial velocity V 0 (mM min À1 ) near or below 2 Å (  Table S3) when only considering 'track-andtrace' loop residues 120-130 (not calculated for the E117Q variant due to missing residues). These r.m.s.d. values do not suggest significant perturbations of the BvHSS variant structures. The increased structural flexibility and therefore deviation of the 'track-and-trace' loop was previously reported (Krossa et al., 2016) and thus was expected. Furthermore, local inspection of the catalytically relevant residues in the binding pockets of the variants showed a high structural similarity to the wild-type residues where not stated otherwise in the following.
3.8. Requirement for negatively charged residues at the 'ionic slide' and the 'inner amino site' For the attraction of protonated and thus positively charged amine substrates into the binding pocket of BvHSS, Asp94 and Glu117 in particular were suggested to form an 'ionic slide'. After reaching the active site, the 'inner' amino group of the reaction components is stabilized by Glu10. The importance of the 'ionic slide' was analyzed by exchange of the acidic residues for neutral and basic residues. The activity of these BvHSS variants in relation to wild-type BvHSS correlated to the negative electrostatic potential of the exchanged residue in the order D94N (61%) > D94K (52%) >>> E117Q (1.5%) > E117K (inactive). The deeper lying residue Glu117 consistently had far more influence on the activity than Asp94 at the pore opening. This effect might also be explained by an assumed lower dielectric constant, considering the protein environment of Glu117 compared with the more solventexposed environment of Asp94. Glu117 would consistently provide higher salt-bridge interaction energies for the attraction of positively charged substrates. The predominant role of Glu117 compared with Asp94 is also reflected by a sequence alignment of bacterial HSS orthologs, which revealed that only $40% of the enzymes contain an acidic residue at the position equivalent to Asp94 in BvHSS (Shaw et al., 2010). Only the BvHSS variant E117Q (PDB entry 6s6g) provided sufficiently diffracting crystals for structure solution (Fig. 8a).
While the architecture of the binding pocket in the variant closely resembles the wild-type pocket, a few minor structural deviations were detected. Large parts of the 'track-and-trace' loop are not traceable in the electron-density distribution (residues 123-127 and 121-127 are missing in subunits A and B, respectively), which is probably caused by the previously described flexibility (Krossa et al., 2016) of this loop. Delocalization of the loops causes a shift of Gln117 compared with the wild-type residue Glu117 (Fig. 8b). The shifted side chain of Gln117 is thereby moved away from the entrance tunnel and does not cause steric occlusion. In addition, the electrondensity distribution of the NAD + molecule was interpreted as alternative ribose conformers of the nicotinamide riboside moiety. While the ribose moiety in the wild-type structure is exclusively present as a C2 0 -endo conformer, both C2 0 -endo and C3 0 -endo conformers seem to be present in the BvHSS variant E117Q. However, the extent of the above-described structural deviations is rather small and should not exert significant effects on the activity of the variant. Instead, the drastically reduced activity of the BvHSS variant E117Q is assumed to be caused by the lack of negative charge at position 117. In line with this, the introduction of a basic residue in the BvHSS variant E117K completely inactivated the enzyme. However, steric repulsion caused by the long lysine side chain or other structural disturbances cannot be ruled out without a structure of this variant. A simulated model of the BvHSS variant E117K indicates that there are many possible conformations of Lys117. Therefore, the passage of educts and products through the entrance tunnel might be sterically hindered considering the length of the lysine side chain and the different possible conformations (see Supplementary Figs. S9a and S9b).
Following a similar approach to probe the role of BvHSS residue Glu210, the E210A, E210Q and E210K variants were examined and were found to be inactive. The crystal structures of the E210Q (PDB entry 6s3x; Fig. 8c) and E210A (PDB entry 6s49; Fig. 8d) variants were solved and showed clear contouring of the exchanged residues by electron density. Superimposition of the variant structures onto the wild-type structure revealed nearly identical locations of all residues in the binding pocket, including the wild-type and variant residues Glu210 and Gln210. These findings convincingly demonstrate the indispensable need for a negative charge at position 210 to stabilize the positively charged amino groups of the reaction component at the 'inner amino site'.

Strong evidence for cation-p interaction between PUT and tryptophan at the active site
The BvHSS residue Trp229 has been proposed to play a central role in the reaction mechanism by stabilizing reaction components and intermediates by cation-interaction. Until now, this assumption was exclusively based on evaluation of the interaction geometry between both the PUT and HSP molecules and the indole ring of Trp229 (Krossa et al., 2016).

research papers
To experimentally substantiate this interaction, the tryptophan residue was exchanged for residues with reduced aromatic character (W229F, W229Y and W229H), for both acidic (W229E) and basic (W229K) residues and for alanine (W229A). The W229E variant was generated to scrutinize the possible stabilization of positively charged reaction components by a salt bridge instead of a cation-interaction. The W229K variant served as 'negative control', which should repel positively charged reaction components. Except for the W229F and W229Y variants, which retained very low activity ($2% and 1% of the wild-type activity, respectively), all other variants were inactive. The crystal structures of the W229A, W229F and W229E variants could be solved and allowed several conclusions to be drawn in combination with the activity data. Apart from the exchanged residues, all residues in the binding pockets of these variants superimpose almost perfectly onto the wild-type residues without considerable deviations. In conclusion, the decreased or abolished activity should be caused by the exchange of Trp229 and not by structural perturbation of the binding pocket. In the W229F variant, the location of the Phe229 phenyl ring is distinctly defined by the electron-density distribution (Fig. 9a).
Superimposition of the variant and wild-type structures shows coplanar alignment of the Phe229 phenyl ring and the wild-type Trp229 pyrrole moiety (Fig. 9e). Consequently, the reduced activity of the variant is not caused by obstruction of  . Atomic models are depicted in ball-and-stick representation and electron-density maps as mesh. The large images (a, c, d) show the 2mF o À DF c maps (black, 1) and mF o À DF c maps (green/red, AE4) of the complete models. Models lacking the respective exchanged residue together with ten N-terminally and ten C-terminally adjacent residues were used to calculate (I) 2mF o À DF c omit maps (black, 1) and (II) mF o À DF c omit maps (green/ red, AE4) after refinement including simulated annealing. the active site due to an inappropriately orientated phenyl ring. Without a bound PUT molecule in the active site of the structure, the exact location of the substrate cannot be determined. Catalytically unfavorable binding of the PUT molecule, also considering the smaller, less space-occupying phenyl ring, might therefore cause the decreased activity. However, the active site of BvHSS is considerably narrow and both the W229F and W229Y variants provide bulky side chains to maintain the spatial dimensions. In the structure of the W229F variant, the side chain of the variant residue Phe229 sterically confines the active site to similar spatial dimensions as in wild-type BvHSS. A substantially different binding of the PUT molecule in the narrow active site of the W229F variant therefore seems unlikely. Still, the activities of the W229F and W229Y variants are drastically reduced. In conclusion, we suggest the weaker cation-interaction in the W229F and W229Y variants to be the reason for their reduced activities as explained in the following. Firstly, the Phe229 benzene moiety is shifted in relation to the Trp229 benzene moiety and therefore also in relation to the 'center amino site'. The resulting cation-interaction geometry in the W229F variant would consistently be more off-centered and therefore energetically less favored (Marshall et al., 2009). Secondly, the phenyl ring (and similarly the phenol ring in the W229Y variant) exhibits less negative electrostatic potential and a smaller size of its -electron system compared with the indole ring, resulting in a smaller cation-interaction energy (Mecozzi et al., 1996b;Reddy & Sastry, 2005). The structure of the W229H variant could not be determined and correct positioning of the His229 side chain and the PUT molecule cannot be verified. Simulation of the possible side-chain orientations of His229 did not reveal steric hindrance or hydrogen-bond problems (Supplementary Figs. S9e and S9f ). However, notable cation-interaction can be doubted even assuming optimal interaction geometry since the system of the imidazole ring in the W229H variant provides even less negative electrostatic potential than the benzene and phenol rings (Mecozzi et al., 1996b;Reddy & Sastry, 2005). In line with these assumptions and considering the already low activity of the W229F and W229Y variants, the W229H variant was found to be completely inactive.
More evidence for cation-interaction as a requirement for efficient catalysis might be inferred from the inactive variant W229A. Soaking the variant crystals with 0.2 M PUT for 5 min allowed the identification of a PUT molecule in the active site of subunit A (Figs. 9c and 9d). While stable binding of the substrate in the active site was possible in the variant, catalytic conversion was completely disrupted. Superimposition of this variant subunit onto the wild-type subunit also containing a PUT molecule (PDB entry 4tvb chain B; Krossa et al., 2016) revealed different orientations of the PUT molecules (Fig. 9f ). The 'inner' amino group of both PUT molecules is similarly placed near the carboxyl group of Glu210. The 'outer' amino group in the W229A variant, however, is shifted towards or attracted by the carboxyl group of Glu237 (distance 3.3 Å ) compared with the wild-type protein (distance 4.4 Å ). In context of the oxidation of PUT C4 by NAD + as part of the proposed reaction mechanism (Krossa et al., 2016), the distance between PUT C4 and NAD + C4N is 4.4 Å in the variant and 2.2 Å in the wild-type structure. An explanation for the inactivity of the W229A variant could therefore be the missing cation-interaction, which prevents appropriate stabilization of the 'outer' amino group of PUT at the 'center amino site' near the nicotinamide ring. This could ultimately impede hydride transfer between PUT C4 and the nicotinamide ring as a prerequisite for imine formation (Fig. 2, steps 2-3). Another explanation would be the missing stabilization of a partial positive charge at PUT C4 to facilitate the nucleophilic attack of a water molecule ( Supplementary Fig. S2, step 4), resulting in oxidative deamination (Krossa et al., 2016). However, the lack of a bulky, confining side chain at position 229 broadens the active site and might thereby allow increased flexibility of the PUT molecule. This could also contribute to the inappropriate positioning of the PUT molecule in this variant, preventing further conversion of the substrate.
Regarding the structure of the inactive variant W229E, the location of the carboxyl group of Glu229 is probably not completely fixed. The carboxyl group is rotated between the two subunits of the variant and the omit maps reveal a lack of electron density clearly supporting the location of the carboxyl group (Fig. 9b). A reduced occupancy of the carboxyl group at a defined location due to flexibility of the side chain or radiation damage affecting the carboxyl group should therefore be assumed. In any case, the side chain of Glu229 is able to protrude into the active site, as modeled in the variant structure, which would sterically interfere with substrate binding. The same principle might apply to the inactive variant W229K (for which no structure is available). Simulation of the W229K variant suggests a very diverse ensemble of side-chain rotamers, which interfere with the 'inner amino site' for positioning of the substrate (see Supplementary Fig. S9g). In conclusion, the inactivity of both variants W229E and W229K is not necessarily caused by their inability to provide cationinteraction but might rather result from steric issues or an enlarged space enabling increased flexibility of the substrate. Still, both variants are inactive as expected for a cation-dependent reaction mechanism. A future Asp229 variant might allow better evaluation of a salt bridge as potential compensation for a cation-interaction without steric issues. However, the short side chain of an aspartate would require a distinct shift of PUT N2 from the active site towards Asp229 to form an energetically significant salt bridge (distance of 4 Å ). Such a displacement would probably reduce or prevent efficient catalysis as proposed by the reaction mechanism.
In summary, the combination of activity assays and structural data convincingly suggests an indispensable catalytic need for strong cation-interaction in BvHSS and thus probably also in PaHSS. The drastically decreased activity of the W229F and W229Y variants is particularly remarkable and emphasizes the need for (i) the more negative electrostatic potential and/or (ii) the more favorable interaction geometry provided by the indole ring. Exchange of Trp229 for nonaromatic residues generated inactive variants, which might be caused by the lack of cation-interaction. However, other effects caused by the introduced residues might equivalently provide an explanation, including steric repulsion of the substrate or increased flexibility of the substrate in the enlarged space of the active site. As previously proposed by Krossa and coworkers, efficient catalysis might require the stabilization of both positively charged amino groups and/or carbons at the active site by cation-interaction (Krossa et al., 2016). In this context, the influence of cation-interactions in catalysis for the stabilization of carbocations has been confirmed for aristolochene synthase (Faraldos et al., 2011) and triterpenoid synthases (Hoshino, 2017;Chang et al., 2013) by similar experimental approaches as in this study. The role of critical aromatic residues was also examined by sitedirected mutagenesis and activity assays of the resulting variants, although structural data were limited to the wild-type enzyme or a homology model, respectively. To more precisely assess the requirement for cation-interaction in bacterial HSS, finer modification of the interaction energy and thus of the resulting activity could possibly be achieved by the 'fluorination strategy' (Davis & Dougherty, 2015). This approach entails the replacement of relevant aromatic residues with fluorinated derivatives, which exhibit decreased cation-interaction energies but very similar van der Waals radii compared with their native counterparts. Steric effects on catalysis, which cannot completely be ruled out in the case of exchange for native, sterically different amino acids, are negligible when using these derivatives. Also, the decline in activity would be expected to quantitatively correlate with the extent of fluorination of the introduced derivative. Simulation of BvHSS variants with 5-fluorotryptophan or 6-fluorotryptophan introduced for Trp229 indicates equivalent orientations of the fluorinated indole rings and the wild-type indole ring without steric issues ( Supplementary Fig. S10). These simulations suggest the feasibility of the corresponding experiments.

Conclusions
Polyamines contribute to medically relevant processes in bacteria, including pathogens such as P. aeruginosa. The bacterial HSS is part of polyamine metabolism and exhibits high sequence conservation of catalytically relevant residues in different bacteria. Despite the potential medical significance of HSS in P. aeruginosa and other pathogenic bacteria, its role remains unexplored so far. Apart from an unpublished crystal structure of L. pneumophila HSS with moderate resolution (PDB entry 2ph5), structural as well as catalytic characterization is limited to the HSS from B. viridis. This study presents the crystal structure of PaHSS in complex with its main substrate PUT. The architecture of the substratebinding pocket and the conformation of catalytically relevant residues are highly similar to those of BvHSS. Choosing BvHSS as mechanistic model for PaHSS (and for bacterial HSS in general), single-residue variants of BvHSS have been characterized by structure determination and activity assays. The results emphasize the importance of attraction and binding of substrate amino groups by acidic residues via salt bridges at the entrance tunnel ('ionic slide') and at the 'inner amino site'. Strong evidence for a catalytic dependence on cation-interaction provided by the tryptophan residue at the active site is demonstrated. Exchange of this tryptophan for phenylalanine or tyrosine but not for histidine or nonaromatic residues can maintain catalysis, albeit with drastically reduced activity. These findings substantiate the previously proposed roles of the residues involved in the reaction mechanism of bacterial HSS and might provide guidelines for the design of HSS inhibitors.

Related literature
The following reference is cited in the supporting information for this article: Webb & Sali (2016).