Structural definition of hSP-D recognition of Salmonella enterica LPS inner core oligosaccharides reveals alternative binding modes for the same LPS

The crystal structures of a biologically and therapeutically active recombinant homotrimeric fragment of native human SP-D (hSP-D) complexed with the inner core oligosaccharide of the Salmonella enterica sv Minnesota rough strains R5 and R7 (rough mutant chemotypes Rc and Rd1) have been determined. The structures reveal that hSP-D specifically and preferentially targets the LPS inner core via the innermost conserved Hep-Kdo pair with the flexibility for alternative recognition when this preferred epitope is not available for binding. Hep-Kdo binding is achieved through calcium dependent recognition of the heptose dihydroxyethyl side chain coupled with specific interactions between the Kdo and the binding site flanking residues Arg343 and Asp325 with evidence for an extended binding site for LPS inner cores containing multiple Kdo residues. In one subunit of the R5-bound structure this preferred mode of binding is precluded by the crystal lattice and oligosaccharide is bound through the terminal inner core glucose. The structures presented here thus provide unique multiple insights into the recognition and binding of bacterial LPS by hSP-D. Not only is it demonstrated that hSP-D targets the highly conserved LPS proximal inner core Hep-Kdo motif, but also that hSP-D can recognise either terminal or non-terminal sugars and has the flexibility and versatility to adopt alternative strategies for bacterial recognition, utilising alternative LPS epitopes when the preferred inner core Hep-Kdo disaccharide is not available for binding.

Introduction While the H. influenzae Eagan 4A LPS [21] includes a truncated inner core with a single Kdo and a single Hep with a single Glc extension, the S. enterica Minnesota strains R5 and R7provide the opportunity to investigate, through crystal structure analysis, host recognition of more complex natural LPS structures with inner cores containing multiple Kdo and Hep alongside a terminal Glc in the R5 strain. Here we show that hSP-D recognition of Salmonella enterica sv Minnesota rough strains R5 and R7 (rough mutant chemotypes Rc and Rd1) LPS occurs preferentially through the inner core Hep-Kdo, with hSP-D having the flexibility and versatility to target the terminal glucose of R5 when the preferred epitope is not accessible for binding.

S. enterica Minnesota R5 and R7 oligosaccharide preparation
The oligosaccharide (delipidated) portions of Salmonella enterica sv. Minnesota R5 (Enzo Lifesciences-ALX-581-017-L002) and R7 (Enzo Lifesciences-ALX-581-018-L002) mutants were prepared under mild hydrolysis conditions with 2% acetic acid, in accordance with the method previously published by Masoud et al. [34] and our previously published procedures [21]. For the R5 LPS the molecular weight was calculated as % 3036 Da by the addition of all glycoconjugates and Lipid A. The lipid A portion accounted for % 1800 Da [35], making up 60% of the LPS. 2 mg of R5 LPS were initially hydrolysed, yielding %0.7 mg of the oligosaccharide representing 88% hydrolysis efficiency. The R5 oligosaccharide was dissolved in deionised water to give a KdoI-HepI-HepII-GlcI concentration of %16 mM (accounting for the beta elimination of KdoII-KdoII) for storage and for use in crystal soaking. The same procedure was followed for the R7 LPS providing a stock solution of %18mM for storage and for use in crystal soaking and cryoprotection preparation.

Generation and purification of a recombinant fragment of hSP-D
A recombinant homotrimeric fragment of human SP-D (rfhSP-D) was expressed and purified as described previously [5,6,14]. Each chain contains 177 amino acids (Gly179 to Phe355) comprising a short collagen region with eight Gly-X-Y repeats, the α-helical coiled coil neck region and the globular CRD. The contaminating level of endotoxin present in the rfhSP-D preparation was minimized [6] using a Polymyxin B column (Detoxi-Gel, Pierce, UK) to remove endotoxin followed by assaying with the QCL-1000 Limulus amoebocyte lysate system (BioWhittaker, Walkersville, MD) according to manufacturer's instructions. The assay was linear over a range of 0.1-1.0 EU/ml (10 EU = 1 ng of endotoxin).

Crystallisation, ligand soaking and cryoprotection
The endotoxin treated rfhSP-D at 8.8 mg/ml in PBS was used for crystallisation both directly (for the R7 studies), following addition of CaCl 2 to give 10mM, and after dialysis (for the R5 studies) into 50 mM tris(hydroxymethyl)aminomethane (Tris), 10 mM CaCl 2 , 150 mM NaCl, 0.02% NaN 3 at pH7.4. Protein concentrations used were 7.7-8.9 mg/ml. Native crystals of rfhSP-D were grown in sitting drops using screens around conditions previously reported [14]. Those for soaking with the delipidated R5 LPS were grown in precipitant buffer 0.1 M tris pH 8.0, 16% polyethylene glycol (PEG) 6000, and for the delipidated R7 LPS in 0.1 M Tris pH 7.0, 16% PEG 10000. Cryoprotectant buffers were prepared as previously described [21] to provide increasing concentrations of MPD (2,4-methylpentane diol) in precipitant buffer with ligand added to provide final concentrations of %8 mM and %13 mM in the crystallisation drop for R5 and R7 respectively. The crystals were soaked with ligand and prepared for cryocooling by successive addition of cryoprotectant containing increasing concentrations of MPD (5-15% for R7; 5-20% for R5) in 2 μl aliquots over three minutes followed by a final exchange of 6-8 μl of the drop with 6-8 μl of the final MPD cryoprotectant solution (15% MPD for R7; 20% MPD for R5). Crystals were flash-frozen at 100 K after a total of 24 minutes (R5) or 8 minutes (R7) soaking with ligand and cryoprotectant buffer.

Data collection, structure solution and refinement
Data for the rfhSP-D R5 and R7 complexes were collected at 100 K at Diamond Light Source, Oxfordshire, on beamlines I04-1 and I03, respectively, using Dectris Pilatus 6M hybrid photon counting detectors. Integrated intensities were calculated using the program MOSFLM [36]. The 1.6 Å native structure of rfhSP-D (PDB ID 1PW9 [14]) was used as a starting model for modelling both the rfhSP-D/R5 and rfhSP-D/R7 structures. Electron density maps were calculated using the CCP4 suite of programs including AIMLESS, TRUNCATE, SORTMTZ and, where necessary, REINDEX [37]. Model building was completed using COOT [38] and refined using maximum likelihood refinement in REFMAC5 [39], available as part of the CCP4 package, in alternating rounds. Ligand topology and parameter files were generated using ProDRG [40] following fitting into the density. Final refinement statistics are presented in Table 1 and the quality of the final model was confirmed using the MolProbity web server [41]. The coordinates and structure factors for the R5-bound (PDB ID 5OXS) and R7-bound (PDB ID 5OXR) structures have been deposited with the Protein Data Bank. All main and side-chain stereochemical parameters are within accepted limits or better, with more than 98% of residues in favoured regions of the Ramachandran plot with no outliers. Molecular figures were generated using CCP4mg molecular graphics software [42] and the PyMOL Molecular Graphics System Version 1.4 (Schrödinger, LLC, 2011).

Results
The trimeric rfhSP-D structures comprising three carbohydrate recognition domains (CRDs) and the α-helical coiled-coil neck region have been refined to 1.65 Å and 1.75 Å resolution for the R5-bound and R7-bound structures respectively. In all but one of the CRDs the structures reveal oligosaccharide ligand in the Ca1 binding pocket. No ligand is present in subunit A of the R7-bound structure while the mode of oligosaccharide recognition in subunit A of the R5-bound structure differs from the Hep-Kdo binding seen in subunits B and C of both structures.
The bound oligosaccharide is clearly defined in the electron density of both structures; Kdo-HepI-HepII for the R7 and Kdo-HepI-HepII-Glc1 for R5 (Fig 2). The Kdo is in the  (Fig 3) reported by Shrive and co-workers [21] in the rfhSP-D-H. influenzae Eagan 4A structure (pdb entry 4E52) but here the 5-membered ring and all substituents are fully defined in the electron density (Fig 4) except in subunit A of the R5-bound structure where LPS binding is not via HepI and the terminal anhydro Kdo is poorly defined. The Kdo electron density clearly shows the 2-oxobutanoic acid sidechain off C4 ( Fig 3B) in a beta conformation with respect to the glycosidic bond in both structures. KdoII and KdoIII (Fig 1) are not present in the electron density.
In subunits B and C, in both structures, recognition of the oligosaccharide by rfhSP-D ( Fig  5 and Table 2) is primarily via protein and Ca1 coordination of the 6'OH and 7'OH of the dihydroxyethyl side chain of HepI. Further HepI interactions with the binding pocket in these four subunits include those routinely associated with carbohydrate recognition by hSP-D [21].
In contrast to the Ca1-HepI binding in the B and C subunits of both structures, subunit A of the R5-bound structure reveals an alternative mode of LPS recognition with calcium-dependent binding of the terminal glucose GlcI of R5, with the remaining R5 oligosaccharide chain HepII-HepI-Kdo extending over Arg343 and the SP-A (surfactant protein A), SP-D conserved Arg349 and Glu333 (Fig 6). The relative positions of the LPS for the two alternative binding modes are shown in Fig 7A where least-squares fitting of the main chain CRD in the two cases reveals the completely different LPS orientations despite the conserved Ca1-OH interactions. The position of the Ca1-bound glucose GlcI and its interactions with Ca1 and the binding pocket closely mirror those reported in a variety of structures including the rfhSP-D maltose bound structure [14] as shown in Fig 7B. The Ca1 coordinates to the terminal glucose GlcI through 3'OH (2.38 Å) and 4'OH (2.36 Å). Protein coordination to both calcium (via Glu321, Asn323, Glu329, Asn341) and the calcium-bound Glc1 hydroxyls (via Glu321, Asn323,  Glu329, Asn341, Asp342) follows the established pattern for ligand-bound SP-D structures ( Table 2). Calcium Ca3, the more weakly coordinated of the three calcium ions Ca1, Ca2, Ca3 [15], is absent throughout the R5-bound structure. Ca2 is only weakly defined in subunit B of the R5-bound structure and is absent in subunit C where the calcium (Ca2 and Ca3) coordinating loop between Asp297 and Tyr304 is rearranged and a water molecule replaces Ca2.
In both structures and for both modes of binding (Ca1 coordination of HepI or GlcI) additional direct interactions between the protein and bound oligosaccharide are dominated by   the binding-pocket flanking residues Arg343 and Asp325 (Table 2). For both ligands where HepI is coordinated to Ca1 (subunits B and C, both structures) the Kdo (anhydro) interacts with both binding site flanking residues, forming hydrogen bonds with Asp325 through Kdo O6 (2.56-2.80 Å) and with Arg343 through the extended 2-oxobutanoic acid sidechain of the anhydro Kdo (Fig 5 and Table 2). In subunit A of the R5-bound structure where GlcI is coordinated to Ca1, additional direct interaction between protein and oligosaccharide again involves Arg343 (Fig 6) which forms weak hydrogen bonds to both GlcI 6'OH and O4' of HepI ( Table 2). The direct interactions between protein and oligosaccharide ligand in both structures and all subunits where oligosaccharide ligand is present are supplemented by extensive water molecule networks linking the ligand to the protein.

Discussion
The structures of rfhSP-D, a biologically active recombinant native fragment of hSP-D, bound to inner core structures of S. enterica Minnesota LPS from deep rough mutant strains reveal SP-D targeting of the proximal inner core Hep-Kdo motif through calcium-dependent binding of the (non-terminal) Hep supplemented by interaction of both Hep and Kdo with the binding site flanking residues Arg343 and Asp325. The structures also provide a definitive demonstration for the first time that hSP-D has the versatility to recognise alternative LPS epitopes when the Hep-Kdo inner core unit is inaccessible. This flexibility and versatility of LPS recognition is supplemented by evidence of an extended recognition surface that interacts with the LPS inner core oligosaccharide. The Kdo structure itself provides the first direct confirmation that mild hydrolysis removal of the lipid A from intact LPS exhibiting 4-KdoI substituted KdoII is accompanied by the elimination of KdoII from KdoI C4 [44,45] in a manner entirely equivalent to the β-elimination of phosphate from KdoI C4 through mild acid hydrolysis [21]. This is of some significance in the interpretation of mass spectrometry studies of bacterial LPS structure.
In all subunits where the carbohydrate binding site is openly accessible in the crystal (subunits B and C, both structures), HepI is coordinated to calcium Ca1 as reported for the H. influenzae Eagan 4A -rfhSP-D structure [21] with KdoI extending between the two bindingsite flanking residues, Asp325 and Arg343, both of which interact directly with the bound HepI-Kdo (anhydro) unit (Fig 5). Superposition of the anhydro Kdo furanose rings together with HepI from all these subunits reveals an essentially conserved position of both rings resulting in very similar Asp325 -anhydro Kdo O6 interactions across the structures (2.56-2.80 Å), but significant variations in the Arg343 -Kdo interactions (2.87-3.26 Å) ( Table 2). This suggests an important flexibility inherent in the Arg343 -Kdo interaction. Combined with the well documented ability of Arg343 to respond to different ligands through conformational change in the Arg343 side chain [15], this reinforces the importance of the Arg343-ligand interaction as a flexible determinant of specificity.
The R5 bound structure (Fig 6) reveals for the first time that hSP-D has the flexibility and versatility to recognise and bind alternative LPS epitopes if the HepI-Kdo core motif is not accessible or, in the physiological context, is shielded by extended oligosaccharides in the outer core and beyond [21]. In subunit A access to the carbohydrate binding site is restricted with the crystal packing precluding binding of either the R5 or R7 oligosaccharide via the HepI-Kdo pair. The crystal packing does however allow Ca1 binding of the terminal glucose of the longer R5 oligosaccharide, which binds to the SP-D subunit A binding pocket with the Recognition and binding of S. enterica Minnesota LPS by SP-D oligosaccharide ligand (Fig 6) oriented in a direction roughly orthogonal to the ligand when HepI is bound to Ca1 (Fig 7) as in subunits B and C of both structures (Fig 5). Interestingly, despite the completely different position of HepI with respect to the protein structure, Arg343 remains fundamental to the recognition of the LPS, stabilising the interaction through an Arg343-HepI hydrogen bond (Fig 6 and Table 2) and perhaps contributing to directing the glucose into the binding site [15,16]. While this novel structural demonstration that hSP-D can adopt alternative binding mechanisms for the same LPS is a fortuitous artefact of crystallisation, it clearly shows not only alternative recognition strategies for the same LPS but also the ability to recognise more than just the inner core heptose and Kdo residues in natural LPS ligands. This potentially provides an explanation for the extraordinary ability of hSP-D to recognise such a broad range of gram-negative bacteria by utilising Ca1-dependent binding of not just inner core HepI but also of multiple types of saccharide residue in the LPS [13,21]. The extent of the LPS, and the variety of different carbohydrate residues displayed on bacterial surfaces, provides the opportunity for versatility of recognition by the carbohydrate recognition domain.
The full S. enterica sv. Minnesota R5 and R7 LPS oligosaccharides (Fig 1) are known to be recognised by hSP-D [19]. Positioning the full R5 oligosaccharide, including KdoII, KdoIII and a full 6-membered KdoI ring, in the R5-bound structure (subunit C) by overlaying HepI reveals that KdoI may be oriented such that the dihydroxyethyl side chain (7'OH, 8'OH) off C6 interacts either with Asp325 or Arg343. In the orientation shown in Fig 8, KdoII sits between KdoI and the conformationally flexible side chain of Arg343 with both KdoII and KdoIII proximal to the SP-D conserved Glu347 and the SP-A, SP-D conserved Arg349 and Glu333. KdoI O7' interacts with Asp325, KdoII O1' with Arg343, and KdoIII O8' with Glu347. The lipid A linked to KdoI O2' extends out from the CRD and the trimer overall. Salmonella enterica mutants with more extended cores and which are recognised by SP-D [19] can also be accommodated in this orientation. For the Rb3 mutant for example (Fig 1), the sugars linked to HepII and Glc1 all extend away from the CRD. This strongly suggests an extended inner core binding site with recognition of the Hep-Kdo motif accompanied by additional binding of inner cores with multiple Kdo structures such as those of Pseudomonas aeruginosa, Escherichia coli and Klebsiella pneumoniae which are known to be recognised by rfhSP-D [46].
Previous studies of the hSP-D/LPS interaction have shown the importance of Asp325 and Arg343 as oligosaccharide ligand selection mediators, flanking the binding site and directing calcium recognition [13][14][15][16][17], alongside the targeting of the LPS proximal inner core. The structures presented here combine and extend these key elements, revealing the preferred inner core Hep-Kdo binding mechanism alongside a flexibility to recognise alternative epitopes, to provide unique insights into the recognition and binding of a wide variety of gram negative bacteria by hSP-D.