Impact of branching on the conformational heterogeneity of the lipopolysaccharide from Klebsiella pneumoniae: Implications for vaccine design

Resistance of Klebsiella pneumoniae (KP) to antibiotics has motivated the development of an efficacious KP human vaccine that would not be subject to antibiotic resistance. Klebsiella lipopolysaccharide (LPS) associated O polysaccharide (OPS) types have provoked broad interest as a vaccine antigen as there are only 4 that predominate worldwide (O1, O2a, O3, O5). Klebsiella O1 and O2 OPS are polygalactans that share a common D-Gal-I structure, for which a variant D-Gal-III was recently discovered. To understand the potential impact of this variability on antigenicity, a detailed molecular picture of the conformational differences associated with the addition of the D-Gal-III (1→4)Gal branch is presented using enhanced-sampling molecular dynamics simulations. In D-Gal-I two major conformational states are observed while the presence of the 1→4 branch in D-Gal-III resulted in only a single dominant extended state. Stabilization of the more folded states in D-Gal-I is due to a O4-H…O2 hydrogen bond in the linear backbone that cannot occur in D-Gal-III as the O4 is in the Galp(1→4)Galp glycosidic linkage. The impact of branching in D-Gal-III also significantly decreases the accessibility of the monosaccharides in the linear backbone region of D-Gal-I, while the accessibility of the terminal D-Gal-II region of the OPS is not substantially altered. The present results suggest that a vaccine that targets both the D-Gal-I and D-Gal-III LPS can be developed by using D-Gal-III as the antigen combined with cross-reactivity experiments using the Gal-II polysaccharide to assure that this region of the LPS is the primary epitope of the antigen. Author Summary Klebsiella pneumoniae (KP) is a bacterial pathogen commonly associated with hospital acquired infections to antibiotics and is of increasing concern due to the development of resistance to antibiotics including those of last resort. Development of an efficacious KP human vaccine would not be subject to the mechanisms governing antibiotic resistance and, is thus, a public health priority. The present study applies computer simulations to understand the effects of branching on the antigenic D-Gal-I and D-Gal-III O polysaccharide (OPS) species of KP. Results show that branching leads to a single dominate extended conformation, though that conformation occurs in both species. In addition, the terminal D-Gal-II region of both OPS sample similar, exposed conformations while regions of the central repeating units of the OPS are potential hindered from interactions with antibodies in the presence of the branched D-Gal-III OPS. The results suggest a strategy for vaccine development based on D-Gal-III as the antigen combined with cross reactivity experiments to assure that the D-Gal-II region is the primary epitope of the antigen.


Introduction
Klebsiella pneumoniae is a gram negative encapsulated bacterial pathogen, common in the environment, and is a major cause of hospital acquired infections, for which the recent development of widespread antimicrobial resistance among clinical isolates has become an urgent threat. [1] Individuals with impaired host defenses due to chronic illness or immunosenescence are generally at highest risk. [2][3][4][5][6] With the limited pipeline of new and novel antibiotics, development of an efficacious Klebsiella vaccine is urgently needed. K. pneumoniae typically expresses both lipopolysaccharide (LPS), comprised of a conserved core polysaccharide (CP) linked to lipid A and a repeating polymer of O polysaccharide (OPS), and a capsular polysaccharide (CPS, K-antigen), both of which contribute to virulence. While there are greater than 80 different Klebsiella capsule serotypes for which no single type predominates, there are only 8 recognized OPS serotypes of which 4 (O1, O2a, O3, O5) account for most human disease globally. [2,7,8] Thus, the development of vaccines based on OPS is preferable due to the lower valency requirement to enable broad coverage. However, variability in the composition of OPS subtypes may impact the utility of OPS as a K. pneumoniae vaccine antigen. [2][3][4][9][10][11] Klebsiella O1, a poly-galactan, is formed by a short stretch of a D-Galactan-I (D-Gal-I) comprised of repeats of →3)-β-Galf-(1→3)-α-Galp-(1→ , that is linked at the non-reducing end to a longer stretch of D-galactan-II repeat units (D-Gal-II) that is generated by →3)-β-Galp-(1→3)-α-Galp-(1→. The D-Gal-I structure, when uncapped with D-Gal-II, forms the O2a serotype. [4] A recent report documented a variant of D-Gal-I, that contains a (1→4)-α-Galp branch, which has been designated as D-Gal-III. [12] In this work, we investigate the conformational properties associated with the addition of this branch to the K. pneumoniae O1/O2a OPS (Scheme 1). [4,12] To investigate the conformational properties we apply computational molecular dynamics (MD) simulations with enhanced sampling via Hamiltonian replica exchange, a technique that has been successfully used to elucidate the conformational properties of other polysaccharides. [3,4,[13][14][15][16][17][18][19][20] Enhanced sampling was achieved through the use of the Solute Tempering 2 method (HREST) in conjunction with the use of biasing potentials in the context of the correction map (CMAP) [21,22] approach, termed HREST-bpCMAP. [18,20,[23][24][25][26] This approach is applied in the present study to elucidate the changes in the conformational properties associated with the addition of 1→4 branches to the D-Gal-I repeat units (RU) yielding D-Gal-III. [27] The mechanism by which branching on the K. pneumoniae serotypes O1 and O2a that may impact antigenicity is explored in terms of both the conformational properties and accessibility of the monosaccharide components of the LPS.
Scheme 1) Sequences of the D-Gal-I, D-Gal-III, D-Gal-II repeating subunits, the core polysaccharide (CP) and of the full D-Gal-I and D-Gal-III polysaccharides considered in the present manuscript where n indicates the number of repeating units (n = 3, 4 or 5). [4,28]

Results and Discussion
The goal of the present study was to investigate the impact of the 1→4-Gal branch on the conformational properties and accessibilities of the K. pneumoniae O1 and O2a OPS using explicit solvent MD simulations ( Table 1) properties. [27] The change in the number of RUs was found to have minimal impact on the overall conformational properties.  Gal-II  Gal-I  Gal-III  CP   3RU  3  3  1  30330   4RU  3  4  1  36138  D-Gal-I   5RU  3  5  1  49797   3RU  3  3  1  30303   4RU  3  4  1  36168  D-al-III   5RU  3  5 1 49713       Fig 7) and α-D-Galp (Cx in Fig 7) monosaccharides that are common to both D-Gal-I and D-Gal-III is substantially decreased in the presence of the D-Galp (Dx in Fig 7) branched monosaccharides. This decreased exposure is facilitated by the O6(H) to O6 hydrogen bonds associated with the branch in D-Gal-III discussed above (Fig 6). As expected the D-Galp (Dx in Fig 7) monosaccharides are highly exposed in all cases. Concerning the terminal D-Gal-II and CP regions, the antibody exposure is similar for D-Gal-I and D-Gal-III.   conformation. In addition, it may be anticipated that the process of conformational selection further facilitates the sampling of the appropriate conformation when interacting with such an antibody, analogous to those previously elucidated for HIV envelope glycans. [23] In addition, it has been shown that antibody binding can alter the conformations sampled by an oligosaccharide, which may further facilitate binding. [24,46] For an O1 OPS vaccine, despite the reduced accessibility of the β-D-Galf and α-D-Galp linear backbone monosaccharides in the RUs of D-Gal-III OPS, antibodies induced against the terminal Gal-II region should be similarly efficacious in the context of D-Dal-I or D-Gal-III given the relative insensitivity of D-Gal-II to the presence of the branches. In addition, as the Gal-II region is membrane distal, it is expected that it would be more exposed compared to the membrane proximal D-Gal-I/D-Gal-III in O1. Indeed, D-Gal-II has been found to represent the primary protective OPS epitope in O1 OPS expressing Klebsiella. [3] Furthermore, this epitope has been found to be immunodominant after immunization. Future studies will be needed to characterize the cross-reactive and antigen specific immune responses to O1 and O2 OPS molecules containing D-Gal-I or D-Gal-III.

Methods
Modeling and simulations were performed with the program CHARMM using the CHARMM36 additive force field for carbohydrates [29,30] and the CHARMM TIP3P [31] water model. Initial coordinates of the OPS were generated from the topology information present in the force field followed by minimization using the Steepest Descent (SD) and Adopted-Basis Newton-Raphson (ABNR) minimizers for 5000 steps each with end to end distance restraints on the OPS in order to maintain extended conformations. The end to end distance restraint was placed on the anhMan C1 and the terminal -Galp C1 atoms with a force constant of 15 kcal/mol/Å and with equilibrium distances of 50, 60 and 70 Å for the 3, 4 and 5 RU systems, respectively. The resulting geometries of the OPS were then immersed in a pre-equilibrated cubic water box. The size of the water box was selected based on the condition that it extend at least 10 Å beyond the non-hydrogen atoms of the fullyextended OPS. Water molecules with the oxygen within a distance of 2.8 Å of the non-hydrogen solute atoms were deleted. For all of the subsequent minimizations and MD simulations, periodic boundary conditions were employed using the CRYSTAL module implemented in the CHARMM program. [32][33][34] A list of the systems studied is shown in Table 1.
Equilibration of the solvated systems was initiated with a 500 step SD minimization followed by a 500 step ABNR minimization in which mass-weighted harmonic restraints of 1.0 kcal/mol/Å were applied on the nonhydrogen atoms of the OPS. Following minimization, each solvated system was initially heated from 100 to 298 K under constant volume and temperature (NVT) followed by 100 ps constant pressure and temperature (NPT) MD simulations at 298 K and 1 atm. In all simulations under the NVT or NPT ensembles, including the subsequent HREST-bpCMAP simulations, the temperature was maintained at 298 K using the Hoover algorithm with a thermal piston mass of 1000 kcal/mol·ps 2 . [35] A constant pressure of 1 atm was maintained using the Langevin piston algorithm with a collision frequency of 20 ps -1 and mass of 1630 amu. [36] The covalent bonds involving hydrogen atoms were constrained with the SHAKE algorithm, and a time step of 2 fs was used. [37] In the energy and force evaluations, the nonbonded Lennard-Jones interactions were computed with a cutoff of 12 Å with a switching function applied over the range from 10 to 12 Å. The electrostatic interactions were treated by the particle mesh Ewald method with a real space cutoff of 14 Å, a charge grid of 1 Å, a kappa of 0.34, and the 6-th order spline function for mesh interpolation. [38] Conformational sampling was enhanced by applying the HREST-bpCMAP method that involves concurrent solute scaling and biasing potentials. [20] All the production HREST-bpCMAP simulations were carried out in CHARMM using the replica exchange module REPDST, with BLOCK to scale the solute-solute and solute-solvent interactions [31,39,40] and with specific bpCMAPs applied as the 2D biasing potentials along selected glycosidic linkages. [22] 5RU, 4RU and 3RU (RU=Repeating Unit) OPS were simulated for both D-Gal-I and D-Gal-III, yielding a total of 6 molecules. The RU that corresponds the linear backbone region contains two monosaccharides per RU for the D-Gal-I and three monosaccharide per RU for the D-Gal-III (Scheme 1), while D-Gal-II contains three RU with two monosaccharide per each RU for all 6 systems. The bpCMAP [20] biasing potential is applied to the glycosidic linkages across monosaccharides 5 to 25 for 5RU yielding 20 bpCMAPs for D-Gal-III and 16 bpCMAPs for the D-Gal-I. The biasing potentials are applied for 4RU and 3RU in a similar fashion. The 2-dimensional grid-based bpCMAP were constructed using the corresponding disaccharide model in the gas phase as described previously [18,20,41] and was applied along the φ(O 5 -C 1 -O n -C n )/ϕ(C 1 -O n -C n -C n-1 ) dihedrals for each glycosidic linkage in the OPS. A total of 8 replicas were carried out for each system and exchanges attempted every 1000 MD steps according to the Metropolis criterion. In HREST-bpCMAP simulations, the solute scaling temperatures were assigned to 298 K, 308 K, 322 K, 336 K, 352 K, 370 K, 386 K and 405 K, with the ground-state replica temperature of 298 K selected to correspond to the experimental studies. The HREST-bpCMAP simulations were run for 150 ns on each OPS/RU combination.
The distribution of scaling factors for the bpCMAPs across the 8 perturbed replicas was determined as previously described and the acceptance ratio between different neighboring replicas was examined to guarantee that sufficient exchanges were being obtained. [18,20,23,24] Analysis of the replica walk for the D-Gal-III ground state replica ( Fig S1) showed that multiple transitions from the ground state to the highest replica occurred indicating adequate acceptance rates between the replicas and that the full replica walks from the ground state to highest replica were occurring. Analysis of convergence involved examination of the global and linear backbone endto-end distance probability distributions from the 0-50, 50-100 and 100-150 portions of the simulations (Supplement information: Fig S2). Final results are presented based on the full 150 ns of sampling obtained in the ground state, 298 K replica.
Antibody accessible surface area (AASA) was calculated using the surf tools in CHARMM by probing with a sphere radius of 10 Å, [33,[42][43][44] yielding the average AASA of the for the entire OPS as well as for the individual monosaccharides over the entire simulations. Use of AASA allows for prediction of regions of the OPS accessible to direct interactions with antibodies. The AASA is determined in a manner analogous to the solvent accessible surface area calculations and uses a probe radius of 10 Å to approximate an antibody combining region interacting with the antigen. [43,44] The OPS conformational sampling in Cartesian space was computed based on the sampled spatial volumes.
To compute the sampled spatial volumes (Cartesian space), a 3D grid with a voxel size of 1Å×1Å×1Å was