4-O-Acetylation of NmA CPS is due to acetyl migration
For natural NmA-CPS contained in vaccines, total O-acetylation levels between 68–92% have been reported, of which roughly 10% is 4-O-acetylation5,11,12. To investigate, whether similar levels can be achieved by in vitro O-acetylation, we incubated CsaC with 1 mg of non-O-acetylated polymer in the presence of different donor (acetyl-CoA) to acceptor ratios, with each ManNAc moiety being considered as one acceptor site. Changing the ratio from 1:1 to 4:1 indeed increased the overall O-acetylation level from 70 to 92% (Supplementary Fig. 1a, b). Since acetyl groups can migrate between vicinal diols in a pH-dependent manner7, we next aimed at analyzing in vitro 3-O-acetylated NmA-CPS under different pH conditions to induce a potential 3,4-ester migration. Production of a larger batch of material for this experiment (20 mg) yielded 3-O-acetylated polymer, which already contained 4% of 4-O-acetylation (Fig. 2b, blue spectrum), indicating that acetyl migration occurred to a small extent already during the extended preparation time. Consequently, we aimed at reproducing these conditions in our experiment: The polymer was dissolved in phosphate buffer (pH 7.0) to mimic the enzymatic synthesis conditions, or in water (pH 5.6) to simulate the dialysis step of our processing protocol. The percentage of 3-O– and 4-O-acetylated moieties was determined by 1H NMR based on their characteristic H2 chemical shifts7. Already after one day, 4-O-acetylation was increased from 4 to 14% in phosphate buffer and remained constant upon prolonged incubation (Fig. 2b, red spectrum). In contrast, O-acetyl migration in water progressed considerably slower (Supplementary Fig. 2). Interestingly, the observed O-acetyl migration induced changes in the 1H and 31P NMR spectra (Fig. 2e, f; approximately −5.73 and −5.88 ppm) that were incongruent with previous assignments14,19. To address these discrepancies, we repeated the experiment and performed a comprehensive characterization of the O-acetylation pattern by 2D NMR (Fig. 2c–f).
Two-dimensional (2D) NMR analysis of O-acetylated NmA-CPS
A complete de novo assignment was performed starting from the considerably less complex non-O-acetylated polymer (compare Fig. 2a, c with Fig. 2b, d). To increase signals resulting from the polysaccharide termini, we used partially hydrolyzed CPS, an approach that facilitated the assignment of the corresponding spin systems (Supplementary Fig. 3). A well-resolved part of a 2D 1H-1H TOCSY (total correlation spectroscopy) spectrum was chosen (Fig. 2c, d) that shows the H2 signal of each spin system on the ω2 axis and the correlations between H2 and the other ManNAc protons on the ω1 axis. A total of nine spin systems were identified (four in the non-O-acetylated and nine in the O-acetylated sample, with the most downfield and upfield systems in Fig. 2d labeled 1 and 9, respectively (Supplementary Tables 1–3). Our data confirm the previous assignment of the major spin systems 1, 3 and 7, which belong to 3-O-acetylated, 4-O-acetylated and non-O-acetylated repeating units, respectively7.
The well-isolated, less abundant spin system 2 has been, so far, ambiguously assigned to either a 3,4-di-O-acetylated ManNAc14 or a 3OAc-ManNAc being connected to a 4OAc-ManNAc at its reducing end7. Spin system 2 shows correlations similar to those observed for 1 (3OAc-ManNAc), including the characteristic H3 chemical shift of 5.18 ppm, indicating O-acetylation at C3 (Fig. 2d and Supplementary Fig. 4). However, 1 and 2 differ by their H5 and H1 resonances (Fig. 2d, see arrows) and by the well-resolved 31P chemical shifts of the C1-linked phosphodiester (−5.28 versus −5.88 ppm, Fig. 2f). The latter finding suggests that 1 and 2 are quite similar, but connected to different neighboring units at their reducing ends. Following the vertical dotted lines of both 31P chemical shifts (−5.28 ppm for 1, −5.88 ppm for 2) and searching for correlations to the neighboring unit, reveals cross-peaks to characteristic H6a/H6b signals (Fig. 2f). The 31P resonance of the C1-linked phosphodiester of 2 correlates with H6a/H6b of 3 (4OAc-ManNAc), whereas the 31P resonance of the phosphodiester of 1 correlates to H6a/H6b of 1 (3OAc-ManNAc). Consequently, 1 belongs to a 3OAc-ManNAc that is linked at its reducing end to another 3OAc-ManNAc, whereas 2 refers to 3OAc-ManNAc linked to 4OAc-ManNAc. Owing to chemical shift degeneracies, a smaller population of 1 following nonOAc-ManNAc cannot be excluded. Importantly, the chemical shifts observed for H3/H4 (Fig. 2d) and C3/C4 (Supplementary Figs. 5 and 6) of 1 and 2 are almost identical, clearly excluding 4-O-acetylation of 2 and thus 3,4-di-O-acetylation on the same ManNAc moiety.
We observed two weak spin systems, namely 4 and 8, that had not been described before. They show similar chemical shifts to the overlapping major spin systems 3 (4OAc-ManNAc) and 7 (nonOAc-ManNAc) (Fig. 2d), but differ by their H5 and H1 chemical shifts, reminiscent of the differences between 1 and 2. The unique H1 chemical shift of 8 (5.381 ppm) shows a clear correlation to the unique 31P resonance at −5.72 ppm, which in turn correlates with H6a/H6b of 3 (4OAc-ManNAc, Fig. 2f). Thus, 8 is nonOAc-ManNAc linked to 4OAc-ManNAc at its reducing end. In accordance, 4 seems to be 4OAc-ManNAc linked to 4OAc-ManNAc at its reducing end. Unfortunately, the H1 resonance of 4 overlaps partially with other H1 resonances, preventing the identification of an unambiguous connection to the reducing end neighboring moiety. Also, the H2-H6a/H6b correlations of 4 are not clearly visible and likely hidden under the stronger H2-H6a/H6b correlations of 3, making it difficult to deduce the connection to its non-reducing end moiety. However, since 3 and 4 belong to 4OAc-ManNAc moieties that differ only by the preceding neighbor, it is likely that the resonances of H6a/H6b are very similar. In summary, all signals of internal repeating units are present in a major (1, 3, and 7) and in a minor form (2, 4, and 8). The major forms represent units linked to 3OAc-ManNAc or nonOAc-ManNAc at their reducing ends, while the minor forms are linked to 4OAc-ManNAc. The quantification of the distinct structural motifs is presented as Supplementary Note 1.
Spin systems 5 and 9 were present in both samples (Fig. 2c, d) and were assigned to the β- and α-anomer of non-O-acetylated ManNAc at the reducing end, respectively, based on the similarity to published chemical shifts of free ManNAc20 (Supplementary Table 1). Spin system 6 originates from a terminal ManNAc with a phosphomonoester at C6 and represents the non-reducing end of the polymer as shown by a distinct 31P chemical shift of 0.63 ppm correlating to H6a/H6b of 6 in a 1H-31P HMBC (heteronuclear multiple bond correlation) spectrum (Fig. 2f and Supplementary Fig. 7).
Importantly, all termini identified in the O-acetylated sample resulted from nonOAc-ManNAc, indicating that hydrolysis of the O-acetylated polymer occurred exclusively between non-O-acetylated residues, suggesting their linkage to be less stable.
O-Acetylation increases the stability of NmA-CPS
Next, we analyzed the stability of O-acetylated and non-O-acetylated polymer under conditions that mimicked (i) hydrolysis during vaccine manufacturing (80 °C, pH 4.7, Fig. 3a) and (ii) storage of vaccines in tropical climate (45 °C, pH 7, Fig. 3b). Samples were taken at the indicated time-points and analyzed using Alcian blue/silver-stained polyacrylamide gel electrophoresis (PAGE). After both treatments, the O-acetylated polymer migrated slower than the non-O-acetylated polymer, indicating higher molecular mass and, therefore, higher stability. To confirm this finding, we quantified the change in the average degree of polymerization (DP) according to an established method19 that determines the average DP from the ratio between internal phosphodiesters and terminal phosphomonoesters obtained by 31P NMR (Fig. 2e). Again, OAc-CPS exhibited higher stability (Fig. 3c).
Overall structure of CsaC
To provide insight into the structural and mechanistic basis of CPS O-acetylation, we solved the structure of wild type CsaC (CsaC-WT, 2.0 Å resolution) in its apo state as well as complexes of CsaC-H228A with acetyl-CoA (2.0 Å resolution) and CsaC-WT with a tetrasaccharide fragment of the natural acceptor substrate (CPS-DP4, 1.95 Å resolution) (Supplementary Table 4). Consistent with the oligomeric state of CsaC in solution (Supplementary Fig. 8), the crystal packing of all CsaC structures contains a tetrameric assembly with two protein molecules in the asymmetric unit (Fig. 4a). Each protomer consists of a central eight-stranded β-sheet that is sandwiched by six α-helices and capped by a lid area formed by helices α5, α6, and α7 (Fig. 4b, c). Contact between the protomers is mainly mediated by helices α3 and α7 (Fig. 4a and Supplementary Fig. 9). A deep positively charged groove, which traverses the protein surface (Fig. 4e, f), allows substrate binding and opens into an active site that is characterized by a catalytic triad composed of S114, H228 and D198 (Fig. 5a). A DALI search21 revealed structural homology of CsaC with serine ester hydrolases of the α/β-hydrolase (ABH) fold superfamily, a large protein family that encompasses mainly hydrolytic enzymes22,23,24, as exemplified with esterase A from Streptococcus pyogenes (Supplementary Fig. 10, Z-score 18.4).
CoA binding in the acetylated H228A(-AcS114)-CoA complex
As soaking of CsaC-WT crystals with acetyl-CoA or CoA yielded only incomplete substrate density and co-crystallization trials were unsuccessful, we soaked crystals of the catalytically impaired variant H228A with acetyl-CoA. All protomers showed complete density for CoA with a free SH-group (Fig. 4e and Supplementary Fig. 11a). Concomitantly, the triad serine showed an extended electron density with a contour indicative of its O-acetylated form (AcS114) (Fig. 5b and Supplementary Fig. 11a). This suggests that CsaC adopted the double displacement mechanism of ABH esterases24 and forms a transient acetyl-enzyme intermediate that is trapped in the absence of H228 (Supplementary Fig. 12). CoA, as the first reaction product, is bound along the positively charged groove in an L-shaped conformation (Fig. 4e). The pantetheine moiety is engaged in hydrophobic contacts (I233, A39, and F40), and only the distal carbonyl oxygen is specifically recognized by S113-Oγ, a contact that directs the SH-group towards the triad serine S114 (Supplementary Fig. 13). The purine nucleobase of CoA is sandwiched between I11 and K48 and locked into position by a hydrogen bond between adenine-N7 and Y49-N, while the phosphate oxygens of the ADP-3′-phosphate moiety are encased by a hydrogen network provided by R237, K53, Y51 and I52 (Fig. 4e and Supplementary Fig. 13).
The acceptor binding site accommodates four ManNAc moieties
To investigate the structural basis of acceptor binding, crystals of CsaC-WT were soaked with a CPS fragment consisting of four ManNAc moieties (CPS-DP4), which is sufficiently long to serve as acceptor substrate (Supplementary Fig. 14). For the complete tetrasaccharide, the electron density is found in the positively charged groove (Fig. 4f and Supplementary Figs. 11b and 15). An assessment of the substrate electron densities of DP4, DP5, and DP6 soaked crystal structures revealed no significant difference in dimension and location. Since data statistics and quality of the electron density map were best for the DP4 data set, further analysis was performed on these data. A high-affinity acceptor binding site can be identified that accommodates four ManNAc moieties, designated A to D from reducing to non-reducing end. While the tetrasaccharide traverses the active site with the non-reducing end pointing towards the lid domain, the internal ManNAc moiety B faces the reactive center. The binding site of ManNAc moiety A and B overlaps with that of the pantetheine arm of CoA (Fig. 4e, f and Supplementary Fig. 15), indicating that CoA is released prior to acceptor binding. Contacts between acceptor and CsaC are mainly facilitated by residues K43, S113, S114, R148, and H228 (the latter three are illustrated in Fig. 5c and Supplementary Fig. 16).
CsaC shows an unconventional active site architecture
CsaC harbors a serine-histidine-aspartate triad composed of S114, H228, and D198, which occur in the topological arrangement of the canonical ABH-fold, with S114 being part of a nucleophilic elbow motif (GX114SXG), a typical feature of ABH-enzymes that places the nucleophile at the tip of a sharp turn22,24. In CsaC, however, the classical triad is amended by two additional residues, H201 and Q138, yielding a bifurcated network of hydrogen-bond donors and acceptors with D198 as bifurcation point and the amino acid pairs H228/S114 and H201/Q138 as the two arms (Fig. 5a). Beyond the catalytic center, the H201/Q138 arm is extended by Y79 (Supplementary Fig. 17), which allows precise positioning of Q138 by engaging its Nε2 and Oε1 via hydrogen-bonding with the flanking H201-Nε2 and Y79-OH, respectively. In this unusual active site arrangement, S114-Oγ is in hydrogen-bond distance not only to H228-Nε2 as in a classical triad, but also to Nε2 of Q138, which appears to be a unique element of CsaC. In the H228A(-AcS114)-CoA complex, Q138 positions AcS114, which is approached on the opposite side by the SH-group of CoA (Fig. 5b). In the CsaC-CPS complex, the acceptor O3 of ring B is in H-bond distance to S114-Oγ and H228-Nε2 (Fig. 5c). The latter is, thus, well-positioned to act as general base deprotonating O3. In contrast, and similar to the situation in the H228A(-AcS114)-CoA complex, Q138 is located opposite to the substrate entry site and, thus, lacks direct substrate interactions (Fig. 5c). The catalytic unit of CsaC is accomplished by an oxyanion hole provided by the backbone amides of F40 and K115 (Supplementary Fig. 18), suggesting that the CsaC-catalyzed transfer mechanism involves oxyanion-containing tetrahedral intermediates as proposed in Supplementary Fig. 12.
Enzymatic activity of WT and mutant forms of CsaC
CsaC-WT shows high transfer activity towards CPS and a 52-fold lower transfer activity towards water (measured as enzyme-dependent hydrolysis of acetyl-CoA in the absence of CPS) (Fig. 5d, e). As expected, single alanine replacement of the triad residues S114 and H228 resulted in a complete abrogation of detectable transfer activity, and a D198A exchange reduced transfer activity by 99.5% (+CPS) and 98.6% (−CPS). However, a severe drop in activity by 96.8% (+CPS) and 98.0% (−CPS) was also seen for Q138A, revealing a substantial contribution of this unconventional active site residue to catalysis. Exchange of H201, involved in the positioning of Q138, reduced transfer activity by over 70%. An R148A replacement extinguished transfer activity towards CPS, but not towards water, substantiating a critical role of R148 in CPS placing. The loss of activity towards CPS is only partially rescued by an R148K exchange (Fig. 5d) as only the guanidinium function allows the dual H-bonding with two acceptor sites, as seen in the CsaC-CPS complex (Fig. 5c and Supplementary Fig. 16). Replacement of Y144, which is H-bonded with R148, abolished transfer activity by 38.6% (+CPS) and 28.0% (−CPS).
To monitor the formation of the transient acetyl-enzyme intermediate, we incubated CsaC-WT in the presence of [3H]acetyl-CoA. Radiolabeled protein was detected directly after mixing and quickly disappeared thereafter due to hydrolysis of the acetyl-enzyme adduct and complete turnover of acetyl-CoA (Fig. 5f). Covalent adduct formation with S114 was confirmed by liquid chromatography-electrospray ionization-tandem mass spectrometry, while with both methods no adduct was seen for CsaC-S114A (Fig. 5f and Supplementary Figs. 19–22). CsaC-D198A, −Q138A and −H201A rapidly formed a covalent intermediate, but showed decelerated adduct hydrolysis compared to CsaC-WT (Fig. 5f). CsaC-H228A, in contrast, showed no detectable adduct hydrolysis, demonstrating an essential function of H228 in the second half-reaction, i.e., transfer of the acetyl group from AcS114 to the acceptor (CPS or water). However, how is the acetyl adduct formed by CsaC-H228A in the first half-reaction, in which the triad histidine usually enables the adduct formation by increasing the nucleophilicity of the catalytic serine? In CsaC, the only other residue that could fulfill this function is Q138, assuming the basicity of its amide function was increased, e.g., by amide-imidic acid tautomerism as proposed in Supplementary Fig. 12b. A similar imine-mediated pathway for proton elimination is discussed for ζ-fold prenyltransferases25 and the imidic form of an amide side chain as part of a proton transfer network has been directly shown for endoglucanase by neutron crystallography26. In support of our hypothesis that Q138 can substitute H228 as a general base in the first half-reaction, Q138 is optimally positioned in both CsaC-WT and the CsaC-H228A(-AcS114)-CoA complex (Supplementary Fig. 23) and impaired acetyl-adduct formation was seen upon simultaneous replacement of H228 and Q138 (Fig. 5f). In the second half-reaction, Q138 is too distant from the CPS acceptor to substitute H228. Yet, the transfer activity of CsaC-Q138A is severely impaired, suggesting a critical role of Q138 in optimal positioning of AcS114 for the incoming nucleophile and stabilization of a subsequently formed tetrahedral intermediate (Supplementary Figs. 12a and 18b, c).
Structural basis of regioselective O-acetylation of CPS
Analysis of the CsaC-CPS complex revealed that the bound tetrasaccharide adopts a step-like conformation with the central phosphodiester bridge dividing the molecule into a reducing disaccharide and a non-reducing disaccharide half (Fig. 6). Specific interactions with CsaC are limited to pyranose ring B and the phosphodiester bridge interconnecting ring C and D. The orientation of ring B within the substrate entry site allows precise positioning of O3 for nucleophilic attack (Figs. 5c and 6). Selective access of the 3-OH to the reactive center is ensured by involving the flanking functional groups in acceptor-enzyme interactions. While the carbonyl oxygen of the C2 N-acetyl group is in hydrogen bonding with the hydroxyl group of S113, the C4 hydroxyl group is engaged in a bifurcated H-bond with R148 (Figs. 5c and 6; and Supplementary Fig. 16). Anchoring of the guanidinium group of R148 by two flanking H-bonds (with Y144 and the phosphodiester between ring C and D) directs the central R148-O4 interaction away from the catalytic center, thus ensuring regioselective attack of O3.