Full hydrodynamic reversibility of the weak dimerization of vancomycin and elucidation of its interaction with VanS monomers at clinical concentration

The reversibility and strength of the previously established dimerization of the important glycopeptide antibiotic vancomycin in four different aqueous solvents (including a medically-used formulation) have been studied using short-column sedimentation equilibrium in the analytical ultracentrifuge and model-independent SEDFIT-MSTAR analysis across a range of loading concentrations. The change in the weight average molar mass M w with loading concentration was consistent with a monomer-dimer equilibrium. Overlap of data sets of point weight average molar masses M w(r) versus local concentration c(r) for different loading concentrations demonstrated a completely reversible equilibrium process. At the clinical infusion concentration of 5 mg.mL−1 all glycopeptide is dimerized whilst at 19 µg.mL−1 (a clinical target trough serum concentration), vancomycin was mainly monomeric (<20% dimerized). Analysis of the variation of M w with loading concentration revealed dissociation constants in the range 25-75 μM, commensurate with a relatively weak association. The effect of two-fold vancomycin (19 µg.mL−1) appears to have no effect on the monomeric enterococcal VanS kinase involved in glycopeptide resistance regulation. Therefore, the 30% increase in sedimentation coefficient of VanS on adding vancomycin observed previously is more likely to be due to a ligand-induced conformational change of VanS to a more compact form rather than a ligand-induced dimerization.

incorporation of peptidoglycan precursors leading to osmotic shock and cell lysis [4][5][6][7] . Extensive study of how vancomycin binds to Lipid II has been undertaken, yet currently there still remain important questions about the nature of the biologically relevant form of glycopeptide antibiotics. Nieto & Perkins 5 first reported the aggregation of vancomycin in aqueous solution and these workers used circular dichroism to determine a (molar) dimerization constant K 2 of 800 M −1 . Dimerisation was confirmed by crystal structural data 8 . NMR spectroscopy studies reported K 2 values for back-to-back dimers of different glycopeptides in solution of 300 -700 M −1 9 and 700 -(6.6 × 10 6 ) M −1 3, 10 . However, most studies have focused on the glycopeptide form in the presence of the ligand. For example, using model cell wall precursor peptides such as N-acetyl-D-Ala-D-Ala, binding studies in aqueous solution have established that for many glycopeptides, binding is accompanied by increased formation (1-2 orders of magnitude 11 ) of asymmetric, back-to-back homodimers 9,12-14 mediated by sugar-sugar recognition 9,15 . It has been suggested that dimerization enhances the activity of the antibiotic and this is supported by the findings that dimerization and binding of D-Ala−D-Ala in vitro are generally cooperative phenomena 3 . For example, dimerization of eremomycin was shown to enhance the affinity of the antibiotic by a factor 3 of 10. It has been proposed that the dimer binds two adjacent cell wall precursors and that binding to D-Ala−D-Ala at one site on the glycopeptide dimer anchors the dimer to the cell wall and thereby enhances the second binding event through the chelate effect resulting in a stabilised complex 3 . Although vancomycin dimerization is generally considered to be in a back-to-back configuration, crystal structural information of vancomycin in complex with N-acetyl-D-alanine revealed the existence of both back-to-back and face-to-face dimers, though the relative importance of each these configurations in antimicrobial action and affinity remains to be established 16 . Molecular dynamic simulations based on NMR data confirmed the possibility of spontaneous formation of both back-to-back and face-to-face dimers and the authors suggested a functional significance for face-to-face dimerization 3 . The crystal structural data also demonstrated higher order oligomerisation for vancomycin in the presence of ligand such as dimer-to-dimer and trimers of dimers, mediated by face-to-face and side-side interactions 17,18 . Size exclusion data provided evidence of hexameric vancomycin in aqueous solution, comprising three pairs of back-to-back dimers 17 .
For clinical use, vancomycin is prepared in 0.9% NaCl and infused at a starting concentration of 2.5-5.0 mg. mL −1 in adults so that antibiotic entry proceeds at no more than 10 mg.min −1 in order to minimise undesirable side effects of the antibiotic; the aim is to obtain a serum peak level of 25 -40 µg.mL −1 (8 x minimum inhibitory concentration (MIC) value), a trough serum concentration (TSC) maintained at 15-20 µg.mL −1 and an initial elevated TSC of 20 µg.mL −1 for severe infections 19,20 .
Using the powerful matrix-free technique of sedimentation equilibrium in the analytical ultracentrifuge, the recently available model-independent SEDFIT-MSTAR algorithm and other advanced analysis due to Roark and Yphantis and Kegeles and Rao we now study the reversibility and strength of dimerization of vancomycin in four different aqueous solvents (a) 10 mM HEPES buffer, pH = 7.9, (b) 10 mM HEPES buffer supplemented with 100 mM NaCl, pH = 7.9, (c) 10 mM HEPES buffer supplemented with 100 mM NaCl and 20% glycerol, pH = 7.9 and finally (d) the medically used formulation of 0.9% NaCl (150 mM) referred to above. To maximize sample stability, short solution columns and the Kegeles-Rao method of analysis of the strength of the interactions based on whole-cell weight average molar masses at a low temperature of 7.0 °C were successfully employed.
In addition, in a recent study 21 the monomeric state in solution of the enteroccocal A-type VanS histidine kinase (HK) was established by sedimentation equilibrium in the analytical ultracentrifuge (confirmed later by SEC-MALS 22 ). VanS is a sensor kinase and part of a signal transduction cascade leading to activation of vancomycin resistance genes 21,22 . Sedimentation velocity in the analytical ultracentrifuge (in the presence of 10 mM HEPES buffer supplemented with 100 mM NaCl and 20% glycerol to ensure stability) showed that VanS had an extended conformation in solution, of aspect ratio ~ (12 ± 2). In Phillips-Jones et al. 21 the addition of circa two-fold (in molar terms) vancomycin led to a 30% increase in sedimentation coefficient of VanS. Two possible explanations were provided: 1. Ligand-induced dimerization. The addition of vancomycin leads to a dimerization of VanS, or 2. Ligand-induced conformation change. The addition of vancomycin leads to a conformation change resulting in a decrease in asymmetry from ~12:1 to about 5:1.
We now ascertain using sedimentation equilibrium which of these explanations (1) or (2) applies, and consider the consequences of this finding, including at clinically-relevant concentrations.

Results
Vancomycin self-association. Sedimentation equilibrium in the analytical ultracentrifuge was performed at 47500 rpm (~170,000 g) at a temperature of 7.0 °C in the aqueous solvents (a)-(d) as described above. Analytical ultracentrifugation is a matrix free (no column of separation medium required) technique for the hydrodynamic characterisation of macromolecules. Although vancomycin is too small for the accurate application of sedimentation velocity (formation of a sedimentation boundary from which a distribution of sedimentation coefficients can be obtained) -speeds well in excess of the current limit of 60000 rpm would be required -it is still within the range for sedimentation equilibrium. Sedimentation equilibrium concentration profiles c(r) vs r, where c(r) is the concentration at radial position r (cm) from the centre of rotation gives a direct measure of molecular weight (molar mass g/mol) unaffected by conformation. For mixtures of more than one component, it gives principally the average molar mass (principally the weight average from concentration distributions recorded using Rayleigh interference optics) either over the whole distribution in the ultracentrifuge cell M w, or as a function of radial position M w (r). Because of the small size of the vancomycin and concentration range being studied (0.6 -10 mg. mL −1 ) no correction for thermodynamic non-ideality was deemed necessary 23 .
We employ both the M* method of Creeth and Harding 24 and the 'Hinge point' method 25 both implemented in the SEDFIT-MSTAR algorithm of Schuck, Harding and coworkers 25 to obtain M w at a series of ultracentrifuge cell loading concentrations c (g. mL −1 ). An example is given in Fig. 2 for vancomycin in 0.9% NaCl (ionic strength I = 0.15 mol. L −1 ), at a loading concentration of 1.25 mg/ml and temperature of 7.0 °C. Figure 2a gives the concentration distribution (after correction for baseline effects and the meniscus concentration) and Fig. 2b the corresponding ln c(r) vs r 2 plot. Figure 2c shows how the integral function M* function progresses along the c(r) curve and homes in on the weight average molar mass over the whole distribution M w : when the cell base is reached (r = b) the M* function M*(r → b) = M w . A value of (2.4 ± 0.1) kDa is returned, a value in excess of the monomer molar mass due to self-association. Figure 2d shows the variation in the point weight average molar masses M w (r) which increases with c(r) due to self-association. This plot also permits an estimate of the point weight average molar mass at the "hinge point", r hinge the radial position in the cell where the local concentration c(r) = the original loading concentration c, and hence since M w (r hinge ) = M w the weight average molar mass for the whole distribution. A value of (2.4 ± 0.1) kDa is obtained, in agreement with the value from the M* method.
The results are shown for the four different solution conditions combined in Fig. 3, and separately in Fig. 4. The results are all consistent with a monomer-dimer equilibrium extrapolating to a monomer molar mass of ~ 1500 Da (g.mol −1 ), with a suggestion of some further association beyond that above 5 mg.mL −1 in the presence of added NaCl.
Reversibility of vancomycin dimerization. The reversibility of the dimerization can be assessed from overlaying plots of the local or "point" weight average molar masses M w (r) as a function of local concentration (in fringe displacement units) c(r) at individual radial displacements r in the ultracentrifuge cell (Figs 2 and 5). For a fully reversible system the profiles must overlay 26,27 and (allowing for noise at lower concentrations) this appears to be the case for all solvent conditions (a)-(d) studied (Fig. 5). Similar behaviour is seen for example for an electron transfer flavoprotein 28 and an aminocellulose 29 . Further examples are given in reviews by Harding & Rowe 30 and Teller 31 -see also supplementary data of ref. 29 . These contrast for example with the case for a mucin glycoprotein 32 . Of clinical relevance, at 5 mg.mL −1 (a clinically important concentration for infusion), ~100% of vancomycin is dimerized (Figs 3-5). As with Figs 3 and 4, the data also indicates that for loading concentrations > 5 mg/mL the association may go further than dimerisation, though higher state complexes have so far only been reported in the presence of binding ligands 4,17,18 . Assessment of association/ dissociation constants for vancomycin dimerization. Now that full reversibility has been established we can now assess the strength of the dimerization in terms of the association (dimerization) constant k 2 (mL.g −1 ) or the corresponding molar quantity K 2 (mL mol −1 or l mol −1 ≡ M −1 ) and molar dissociation constant K d (M) = 1/K 2 , from the variation of M w with c. Following Kegeles and Rao 33 and equation III-66 of ref. 27 , for a fully reversible thermodynamically ideal dimerisation:    Fig. 6, and the corresponding values for k 2 , K 2 and K d for vancomycin dimerization are given in Table 1. For these evaluations we have used the Durchschlag-Zipper method for the evaluation of partial specific volumes from the atomic composition 34 and assumed this does not appreciably change across the range of ionic strengths studied (0.01-0.15 mol L −1 ) nor for the addition of glycerol. These assumptions are considered below. Corresponding 'affinities' or standard Gibbs free energies for dimerization ΔG° are also given. All interactions are relatively weak (dissociation constants K d ~ 25-75 μM, and free energies of association ~ 23 kJ mol −1 ). Other studies have reported dimerization constants in the absence of ligand to be in the low millimolar range: 3.9 × 10 3 M -1 (K d ~ 250 µM) 35 and 4.7 × 10 2 M −1 (K d ~ 2.1 mM) 11 and 51 M −1 (K d ~ 19.6 mM) 36 . Although the lower K d values obtained in the present study indicate a somewhat stronger self-association than previously reported, all studies to date (including our own) overall indicate only a weak dimerization, even though dimerization has been considered by some to be clinically important for antibiotic activity 3,4,9,10,14-16 .

Elucidation of the nature of the interaction of vancomycin with enterococcal VanS.
Having established the strength and fully reversible nature of the dimerization of vancomycin we can now elucidate the nature of its interaction with enterococcal A-type VanS, an interaction which may lead to triggering of glycopeptide resistance gene expression. The existence of the interaction was first shown by a combination of hydrodynamics and CD spectroscopy 21 , and like the vancomycin dimerization, was subsequently shown to be weak (K d ~ 70 μM) 22 .
In contrast to the reversible monomer-dimer equilibrium for vancomycin, VanS appears to be purely monomeric in aqueous solution 21 from sedimentation equilibrium in the analytical ultracentrifuge in the presence  Table 1. of HEPES buffer supplemented with 100 mM NaCl and 20% glycerol (i.e. solvent "c" described above), an observation confirmed by the independent method of SEC-MALS -size exclusion chromatography coupled to multi-angle light scattering 22 . Phillips-Jones et al. 21 also showed using sedimentation velocity in the analytical ultracentrifuge, and under the same solvent conditions, that VanS had an extended conformation in solution of aspect ratio (12 ± 2). They further reported that a 2-fold (on a molar basis) addition of vancomycin led to a 30% increase in sedimentation coefficient of VanS. The dissociation constant of ~70 μM (ref. 22 ) would result in no detectable change in the molar mass of VanS as a result of binding of the vancomycin, but the significant shift in the sedimentation coefficient of VanS could be due to either (1) a ligand induced dimerization or (2) a ligand induced conformation change with a decrease in asymmetry (aspect ratio from ~12:1 to ~ 5:1) as noted above and in ref. 21 .
To help us delineate between explanation (1) or (2) we now use sedimentation equilibrium to explore the oligomeric state of VanS in the presence of 2-fold vancomycin (19 μg.mL −1 ). This vancomycin concentration is equivalent to the target initial trough serum concentration desired in the clinic of 20 µg.mL −1 and the target maintenance level of 15 -20 µg.mL −1 desired thereafter 19,20,37 and therefore chosen here to investigate bacterial VanS-vancomycin interactions. Figure 7 (VanS with ligand) gives the comparable SEDFIT-MSTAR set of plots to those of Fig. 5 (VanS without the ligand) of Phillips-Jones et al. 21 . Figure 7a shows the concentration distribution c(r) in fringe displacement units as a function of radial position r, Fig. 7b the corresponding plot of lnc(r) vs r 2 : the ~ linear nature of this plot is our first indicator of a highly monodisperse solution. The M* extrapolation (Fig. 7c) and hinge point estimation methods (Fig. 7d)

Discussion
Reversible weak dimerization of vancomycin: Proportion of monomers and dimers at administered concentrations. Vancomycin has to be administered at concentrations high enough so as to be effective but also low enough to minimize the risks of damage to human tissue such as thrombophlebitis (deep vein thrombosis) and nephrotoxicity 19,20,[38][39][40] . In infusion formulations administration doses are typically at 5 mg.mL −1 or less. This concentration of 5 mg.mL −1 would correspond to ~ 100% dimers (Eq. 1 ; Figs 3 and 4). However, this is substantially diluted in the serum to 15 -20 μg.mL −1 , minimizing the risk of toxicity 38,39 , however, the present study has established that the monomeric form of vancomycin predominates at this concentration: the proportion of dimers will be < 20%.

Monomeric nature of VanS.
A previous hydrodynamic study demonstrated that the purified intact VanS histidine kinase occurs as a monomer under the aqueous solution conditions used in the present study 21 . The monomeric nature of the protein was also found to be the case using the alternative approach of SEC-MALS and in the presence of 0.025% n-dodecyl-β-D-maltoside detergent 22 . Classic prototypical HKs are considered to be membrane-bound homodimers that require two cytoplasmic catalytic ATP-binding (CA) domains to achieve trans-autophosphorylation (reviewed in ref. 41 ). However, a full structural understanding of the autophosphorylation reaction has yet to be elucidated and there are examples of histidine kinases, which deviate from this scheme, including those possessing highly dynamic CA domains. For example, autophosphorylation may occur in cis, rather than in trans, as documented for the HK853 kinase 41 . So cis-autophosphorylation occurs by an intrasubunit mechanism; that is, the ATP bound in the CA domain of one subunit phosphorylates the histidine residue in the same subunit monomer 41,42 . As VanS has been shown in the present study to occur as a monomer in our experimental system, and it has also previously been shown that VanS protein stored in these same conditions possesses strong autophosphorylation activity even in the absence of vancomycin ligand 21 , it is possible that under the conditions of our experiments, VanS performs cis-autophosphorylation reactions.
The versatility and accessibility of monomeric VanS in aqueous solvent makes it a highly suitable model system for investigating ligand-induced conformational changes. Under these conditions, monomeric VanS demonstrates weak interactions with vancomycin 21 with a K d value of 70 µM 22 . The finding here of a conformational compaction of the elongated VanS protein in response to vancomycin addition; that is, a decrease in asymmetry from ~12:1 to about 5:1, is consistent with one of the conformational responses expected for a HK during cis-autophosphorylation (or indeed trans-autophosphorylation), namely, an approach of the C-terminal CA domain towards the phospho-accepting histidine in the central DHp (dimerization histidine phosphotransfer) domain. Further studies are required to confirm whether or not this is the case following VanS-vancomycin interactions and indeed if the situation we observe in aqueous solution occurs similarly within the membrane. It is intriguing that under the conditions used in the present study and previously 21 , that the VanS membrane protein remained soluble in the absence of added detergent and that no detectable detergent remained bound after dialysis prior to AUC analysis and indeed post-purification (as revealed by mass spectrometry) 21 . This may be attributable to the large soluble non-membrane portion (predicted to be ~90%) of this sensor kinase. This feature has been useful (and indeed highly suitable) for our initial AUC-based interrogations (here and in ref. 21 ) to characterise VanS conformational changes in simple aqueous solvent. Further studies are now underway to determine whether the VanS-vancomycin interactions observed in aqueous solution here also occurs similarly within the membrane environment.

Methods
Source of vancomycin. Vancomycin was obtained from Duchefa Biochemie, Haarlem, The Netherlands.
A partial specific volume ῡ of 0.67 mL g −1 was calculated from the composition (Fig. 1) and equation 4 of Durchschlag & Zipper 34 which allows for corrections for covolume, ring formation and electrostriction from charged groups. This value is in good agreement with the value of 0.69 ml g −1 of Linsdell et al. 35 based on amino acid and carbohydrate composition. The validity of the Durchschlag-Zipper approach for aqueous systems had been confirmed by a systematic comparison of calculated and experimental partial specific volumes of different classes of organic and biochemical compounds, including small molecules and polymers of ionic nature 34 . We make the assumption that there is no ionic strength (I) dependence of ῡ in the range studied (I = 0.01-0.15 mM) through electrostriction or related effects (see, for example ref. 43 ) and no effect on ῡ through the addition of glycerol. Jaenicke and Lauffer 44 have shown that the 'apparent' partial specific volumes for tobacco mosaic virus (TMV) and TMV protein increase by ~ 0.1% and 0.8% respectively (see for example ref. 44 ) in 25% glycerol.
Solution concentrations of vancomycin c (g. mL −1 ) were determined densimetrically from the relation 45 where ρ and ρ o are the solution and solvent densities respectively, measured on an Anton-Paar (Graz, Austria) digital density meter.

Source of VanS.
Purified VanS protein was prepared as described previously 21  Sedimentation equilibrium in the analytical ultracentrifuge. Sedimentation equilibrium experiments were performed using a Beckman (Palo Alto, CA, USA) Optima XL-I analytical ultracentrifuge equipped with Rayleigh interference optics and an automatic on-line data capture system, to record equilibrium SCientifiC REPoRTS | 7: 12697 | DOI:10.1038/s41598-017-12620-z concentration distribution profiles. For the characterisation of vancomycin dimerization, standard 12 mm path length cells with aluminium epoxy centrepieces were used and loaded with matched 100 μL of solution (solution channel) and reference solvent (reference channel). A rotor speed of 47500 rpm was used at a temperature of 7.0 °C. The rotor speed was found to be optimum for giving an appropriate concentration (fringe) increment from meniscus to base (see Fig. 2). Short solution columns -resulting in relatively rapid equilibrium (<48 h) and low temperature ensured the stability of the samples. With the short columns solvent redistribution effects were negligible. Rinde 46 calculations for the redistribution of glycerol (M = 92 Da) show a ratio of concentration at the base to meniscus at equilibrium of ~1.04. Long columns or multiple-speed averaging were inappropriate. Concentration distributions at sedimentation equilibrium were analysed using the model-independent SEDFIT-MSTAR procedure of Schuck, Harding & coworkers 25 based on the M* function of Creeth and Harding (1982) 24 . Because loading concentrations up to only 10 mg.mL −1 are used, and particularly because of the small size of the vancomycin, non-ideality effects (which tend to lead to underestimates of the molar mass) will also be relatively small and we make the approximation that apparent weight average molar masses M w,app are equal to the true weight average molar masses M w (ref. 25 ).
For the characterisation of VanS in the presence of vancomycin we followed the procedure of Phillips-Jones et al. 21 using the modified long (20mm) optical path length double-sector titanium cells with sapphire windows. These were loaded with 140 µL of solution (containing a final protein concentration of 5.6 µM (260 μg.mL −1 ) in the presence of 12.8 µM (19 μg.mL −1 ) (final concentration) vancomycin as described above), and a matching amount of reference solvent dialysate in the appropriate channels. An equilibrium speed of 25000 rpm was employed at a temperature of 7.0 °C to ensure stability over the long time course (96 hours) of the experiment. The use of long path length cells meant that low loading concentrations could be used to give a sufficient signal (~0.3 mg.mL −1 ) for records to be interpreted. Although VanS is much larger than vancomycin, at such low concentration, non-ideality effects will also be relatively small and we can also make the approximation 23,25 that the apparent weight average molar mass M w,app is equal to the true weight average molar mass M w .