Structure of Merkel Cell Polyomavirus Capsid and Interaction with Its Glycosaminoglycan Attachment Receptor

The MCPyV genome was found to be clonally integrated in 80% of cases of Merkel cell carcinoma (MCC), a rare but aggressive form of human skin cancer, strongly suggesting that this virus is tumorigenic. In the metastasizing state, the course of the disease is often fatal, especially in immunocompromised individuals, as reflected by the high mortality rate of 33 to 46% and the low 5-year survival rate (<45%). The high seroprevalence of about 60% makes MCPyV a serious health care burden and illustrates the need for targeted treatments. In this study, we present the first high-resolution structural data for this human tumor virus and demonstrate that the full capsid is required for the essential interaction with its GAG receptor(s). Together, these data can be used as a basis for future strategies in drug development.

Most PyVs utilize sialic acid-capped components of the plasma membrane as primary receptors for attachment and/or entry (12)(13)(14)(15)(16). Often, gangliosides act as PyV receptors (for instance, for simian virus 40 [SV40] [14], murine PyV [mPyV] [14], and BKPyV [13]), but glycoproteins can also fulfill this role (JCPyV) (12). On the virus side, PyV-receptor interactions are mediated entirely by the major capsid protein VP1, which forms the capsid outer layer. Sialic acid binding sites are typically composed of extended VP1 loops that form the most exposed part of the capsid (15,17), and this is also the case for MCPyV, which binds specifically to glycans containing ␣2-3-linked sialic acid (18). However, the initial attachment of MCPyV is mediated not by sialic acid but primarily by sulfated glycosaminoglycans (GAGs) such as heparan sulfate (HS) and chondroitin sulfate (CS) (16). Noncomplex gangliosides like GM3 are nevertheless also required for MCPyV transduction of cultured cells, marking MCPyV as a virus with an unusual requirement for two distinct glycan receptors (16,18,19) as well as the only PyV with a strict requirement for GAGs.
The individual sialic acid-VP1 interaction is typically weak (in the millimolar K d [dissociation constant] range [20]) but compensated for by the high multivalency of the icosahedral capsid (360 VP1 copies). VP1 assembles into 72 pentameric capsomers, giving rise to small capsids of 45 to 50 nm in diameter. Varying numbers of the minor capsid protein VP2 and its N-terminally truncated variant VP3 associate with the interior of the capsomers, with a maximum ratio of VP1 to VP2/VP3 molecules of 5:1. VP3 is not produced by MCPyV-infected cells, possibly due to an unusually weak Kozak sequence (21). On the other hand, the MCPyV VP1 sequence harbors a 37-amino-acid (aa) C-terminal extension with unknown function and no detectable homology.
In order to investigate these unique features of MCPyV, we produced highly homogeneous virus-like particles (VLPs), which enabled us to solve crystallographic as well as cryo-electron microscopy (cryo-EM) structures and to analyze its unusual receptor usage. The structural analyses reveal a unique and extensive disulfide pattern in the MCPyV capsid that indicates a close evolutionary relationship to mPyV rather than SV40, contrary to BKPyV and JCPyV. Ligand-based nuclear magnetic resonance (NMR) experiments suggest promiscuous GAG binding to the recessions of the MCPyV capsid, similar to GAG binding by human papillomavirus . Mutagenesis experiments show that the unique MCPyV VP1 C-terminal extension, which projects from the intercapsomer recessions, does not participate in receptor binding.

RESULTS
Production of highly homogeneous VLPs for atomic structure determination. Established protocols for the production of PyV and papillomavirus VLPs in 293 TT cells (22,23) were successfully modified in order to obtain structural-biology-quality MCPyV VLP samples. Two species of VLPs differing in weight could be separated by isopycnic density gradient centrifugation and were characterized biophysically. Particles of higher densities were revealed to be filled with histone-DNA complexes (filled VLPs [fVLPs]), unlike the lower-density particle fraction (empty VLPs [eVLPs]) ( Fig. 1a and b). Unspecific histone-DNA complexes copurified along with MCPyV VLPs were removed by an additional monolithic ion exchange chromatography step (Fig. 1b). Here, fVLPs eluted at low ionic strength in one sharp peak and proved highly homogeneous in negativestain transmission electron microscopy (TEM) and dynamic light scattering (DLS) experiments. Empty particles, in contrast, eluted at a higher and much broader ionic strength range and appeared less homogeneous. Occasionally, we observed copurification of elongated, tubule-like structures composed of VP1 capsomers along with eVLPs (Fig. 1d). Structurally similar tubules were observed previously in mPyV-infected cells, where they were assigned a role as "viral factories" (24,25). For this reason and because of their generally higher homogeneity, we utilized exclusively filled particles for crystallographic and cryo-EM studies. Both empty and filled particles were, however, compared with respect to GAG binding in ligand-based NMR experiments. MCPyV also bound to nonhomogeneous, size-defined heparin octasaccharide (HS-dp8) and to size-defined dermatan sulfate hexasaccharide (DS-dp6) ( Fig. 2b and c). The fact that VLPs, but not pentameric VP1, bind to GAGs suggests that the interaction may be mediated by the capsid recessions rather than those exposed capsid regions that contain the sialic acid binding site. Strikingly, the HPV-16 VLP Arixtra binding epitope is indistinguishable from the MCPyV VLP Arixtra epitope ( Fig. 2d and e).
Neither in the crystallographic electron density nor in the cryo-EM map was this extension visible. Instead, the most-C-terminal residue resolved in either map is Val388␥ (from the so-called ␥-chain [see below for an explanation of the chain annotation]), meaning that a stretch of 35 amino acids is completely invisible and thus most likely flexible. Up to this amino acid, the C-terminal structure of MCPyV VP1 resembles those of SV40, mPyV, and BKPyV VP1s. However, it is clear from the electron density and electron potential maps that the missing extension must be located on the exterior of the capsid (Fig. 3a). Hence, it is potentially available for interactions with host factors. In order to probe the function of the strikingly basic C terminus (theoretical pI 9.4), we generated a VP1 mutant terminating at residue 384 (VLP 1-384 ) for STD-NMR analysis. To our surprise, however, this truncated VP1 protein does not assemble into stable VLPs and is, in fact, not expressed at all (or is expressed at very low levels) ( Fig.  3b and c). Therefore, we designed another mutant with only the second, more basic stretch missing (VLP  ). This VP1 truncation did not impact VP1 stability, i.e., capsid integrity, with VLP preparations yielding slightly smaller amounts but homogeneity similar to that of the wild-type (WT) protein ( Fig. 3b to d). STD-NMR experiments with the VLP 1-401 mutant revealed, however, that GAG binding was also intact despite the lack of the 22 most-C-terminal amino acids (Fig. 2d). We therefore conclude that the unusually long VP1 C terminus is not solely responsible for initial receptor binding but that its deletion affects the stability of VP1.
Structure of MCPyV capsid reveals extensive disulfide bridging. Although the overall structures of all PyVs solved to date are very similar in terms of the VP1 three-dimensional fold and capsomer connectivity via the VP1 termini, the SV40 and mPyV capsids differ strikingly in their disulfide patterns. Intracapsomer disulfide bridges are observed for mPyV, while intercapsomer disulfides are present only in SV40 (28). Using the fVLP preparation, we solved crystallographic and cryo-EM structures of the outer (VP1-only) MCPyV capsid. The structures have 3.5-Å and 3.4-Å resolutions, with no significant differences between them (Tables 1 and 2, Fig. 4, and Fig. 5d and e). In PyV capsids, 12 so-called "strict pentamers," located at the icosahedral vertices defined by the 5-fold symmetry axes, are surrounded by a total of 60 "local capsomers" in a pseudo-6-fold environment (28) (Fig. 5a and b). Local capsomers differ in their interactions, the fold adopted by the individual VP1 N and C termini, as well as the C and D beta-strand-connecting loop (CD-loop) (Fig. 5c). Overall, six VP1 conformations can be distinguished within the capsid, five from a local capsomer and one from a neighboring strict capsomer, previously annotated with the Greek letters ␣=, ␣Љ, ␤, ␤=, ␥, and ␣, respectively (Fig. 5a to c) (28)(29)(30). This asymmetric unit of six VP1 chains is sufficient to reconstruct the whole capsid by applying 60-fold icosahedral symmetry. The VP1 C termini project diametrically from the capsomers' cores and interact tightly with neighboring capsomers (the so-called "invading arm") (28,31). In MCPyV, we observed clear density for five out of potentially six disulfide bridges that connect individual VP1 chains within the strict and local pentamers, respectively, i.e., for the ␣-␣, ␣=-␣Љ, ␣Љ-␤, ␤-␤=, and ␥-␣= linkages ( Fig. 6a and b). For the ␤=-␥ pair, within the local pentamers, no disulfide bridge density was observed: the ␤=-chain density starts only with residue 26. The arrangement and proximity of the respective chains suggest that this disulfide bridge is nevertheless formed. The MCPyV cryo-EM data further support the presence of this intrapentameric disulfide bond pattern; nonetheless, only the ␣-␣ connection is clearly visible in the electrostatic potential map. For mPyV virions, intracapsomer disulfide bridges are the only disulfides observed previously (the respective mPyV residues are Cys20 and Cys115), with no reported intercapsomer disulfides. MCPyV VP1 contains two adjacent cysteines (Cys24 and Cys25) within its N-terminal region (Fig. 6e). The crystallographic electron density shows unambiguously that only Cys25 is involved in the ringlike intracapsomer disul-  Structure of MCPyV and Its Interactions with GAGs Journal of Virology fide bridges, tethering each VP1 N terminus covalently to its clockwise (cw) neighbor's CD-loop (residues 108 to 118) via Cys117, the mPyV Cys115 homolog. This observation places MCPyV in the mPyV structural clade since an equivalent disulfide pattern is observed for mPyV (28) but not for SV40 (Fig. 6e). Additionally, the MCPyV crystal structure revealed unique intercapsomer disulfide bridges, rendered possible by the presence of two additional cysteines (Cys18 and Cys24) (Fig. 6e). These two cysteines form two novel and distinct structural elements, which we term the trinity knot and disulfide handshake, respectively ( Fig. 6c and d). The trinity knot is located at the interface between two local pentamers and one strict pentamer tethering each of the ␣-, ␣=-, and ␥-chains to the respective other two. The disulfide handshake links two local pentamers whose chains run in an antiparallel fashion, with two simultaneous Cys18-  Cys24 bridges (Cys18␣Љ-Cys24␤ and Cys18␤-Cys24␣Љ). This motif is located close to the icosahedral 3-fold symmetry axis; i.e., three disulfide handshakes are close to one another. Overall, each of the 12 strict pentamers engages in 5 trinity knots (1 with each of the 5 local pentamers surrounding each strict pentamer) (Fig. 5), while each of the local pentamers forms 2 trinity knots with its respective neighboring strict pentamer and 2 with its neighboring local pentamers. Additionally, each of the 30 local pentamers that are located next to one of the 10 3-fold symmetry axes forms 2 disulfide handshakes (1 with each of the other 2 local pentamers surrounding the same 3-fold  (Continued on next page) axis) (Fig. 6d). Since all other PyVs lack a residue that is equivalent to Cys24, they are most likely unable to form such an extensive disulfide pattern. Capsid stabilization by disulfides and Ca 2؉ -chelating sites. Another conserved feature of PyV capsids is inter-VP1 Ca 2ϩ -chelating sites. A role for calcium as a promoter and stabilizer of PyV assembly was first described in in vitro assembly studies monitored with electron micrographs (29). No calcium ions were observed in the electron density maps of either SV40 or mPyV virion structures, possibly due to the absence of the element in the respective crystallization buffers. However, in the case of SV40, the location and chelation of two potential Ca 2ϩ ions could be deduced upon crystal incubation with gadolinium ions (28). For our MCPyV capsids, we provided 1 mM calcium chloride in all buffers. The crystallographic electron density suggests the presence of only one divalent ion, coordinated by the Ser236 backbone carbonyl and Glu239 side chain carboxyl groups of one monomer as well as the Glu56 side chain of the neighboring monomer and the Glu353 side chain of an invading arm of another capsomer (counterclockwise [ccw]) [i.e., involving Ser236␣=, Glu239␣=, Glu56␥, and (ccw)Glu353␥] (Fig. 7a). The location of this site corresponds to SV40 calcium site 1, formed by SV40 residues Ser213, Glu216, Glu330, and Glu48/Glu46 (28).
Biochemical assays confirmed the chemical identity of the Ca 2ϩ ion and the effect of bivalent cations and disulfide bridges on the capsid's chemical and thermal stability. First, we treated MCPyV capsids with reducing (dithiothreitol [DTT]) and divalent ion-chelating (EDTA) agents and monitored structural integrity at room temperature via SDS-PAGE (Fig. 7b). While intact VLPs did not enter the gel (Fig. 7b, lane 2), we observed a distinct band at around 250 kDa upon treatment with 10 mM DTT (Fig. 6b, lane 4), possibly reflecting intact capsomers (the molecular weight [MW] of monomeric VP1 is 46.6 kDa). No effect was seen at room temperature for 50 mM EDTA (Fig. 7b, lane 3), but additional heating led to complete disassembly, yielding only monomeric VP1 (Fig. 6b,  lane 1). The effect on the capsid melting temperature was significant for both reagents, as assessed by thermal shift assays (TSAs), leading to a reduction from 87.0°C for the untreated fVLPs to 84.0°C upon disulfide reduction and to 84.9°C upon incubation with EDTA (Fig. 6b). The combined effect of disulfide reduction and cation chelation further decreased the melting temperature to 82.4°C. For eVLPs, the trend was even more pronounced, with melting temperatures of 86.6°C for intact capsids, 82.2°C upon DTT and 84.4°C upon EDTA treatments, and 81.5°C for the combination of both agents.

DISCUSSION
In this study, we aimed to further the understanding of the early steps of MCPyV infection and its unusual receptor usage of two distinct types of cell surface glycoconjugates, sialic acid-containing glycans and GAG-presenting proteoglycans. Structural information on MCPyV available to date has been obtained using assemblyincompetent VP1 pentamers (18). However, since we found GAG binding to require an intact capsid, we produced MCPyV VLPs to enable structural and functional studies of the virus-GAG interaction. These VLPs are of exceptional homogeneity, as evidenced by Cys18␣Љ-Cys24␤ linkage is clearly visible in the electron map, while the spatial proximity of Cys18␤-Cys24␣Љ strongly suggests the formation of a similar "disulfide handshake." (e) Alignment of different polyomavirus sequences highlighting residues that are potentially involved in disulfide bonds. MCPyV uniquely harbors two neighboring cysteines at positions 24 and 25, thus enabling the simultaneous formation of intra-and interpentamer connections, as outlined in panels a to d. Cysteines equivalent to Cys18 in MCPyV are conserved in mPyV, SV40, and BKPyV VP1, but their VP1 proteins lack the corresponding disulfide partner in MCPyV, Cys24. Intrapentamer disulfide bridges are similar in MCPyV and mPyV, formed by cysteine residues Cys25 and Cys117 or Cys20 and Cys115, respectively. In contrast, SV40 and BKPyV pentamers are interpentamerically connected through Cys105 or Cys104, respectively, located in the CD-loop.

Structure of MCPyV and Its Interactions with GAGs
Journal of Virology their ability to form well-diffracting crystals. As high-resolution structures of recombinantly produced VLPs are exceedingly rare, the production protocol established here may help facilitate structural analyses of other VLPs, thus circumventing the need for higher biosafety levels to analyze infectious virus particles. Although both MCPyV receptors are glycoconjugates and plasma membrane constituents, they are fundamentally different biomolecules. Unlike sialylated glycoconjugates, GAGs are large and linear polysaccharides of highly but irregularly sulfated disaccharide repeats, presented by a small number of scaffold proteins (proteoglycans) (32). While binding to sialic acid by MCPyV is fairly specific (18), we observed considerable promiscuity with respect to GAG binding. One of the GAG probes employed in our STD-NMR experiments is enzymatically prepared heparin/HS-derived disaccharide (dp2 [ΔUA2S␣1-4GlcNS6S]). While dp2 represents a fully sulfated disaccharide, it differs in both pyranose ring structures in comparison to intact (high-molecular-weight) HS (through an unsaturated bond in the nonreducing ring and dynamic epimerization in the reducing-end ring). Nevertheless, an interaction between MCPyV VLPs and dp2 was unambiguously observed. MCPyV VLPs also bound to a structurally inhomogeneous heparin-derived octasaccharide preparation and DS-derived hexasaccharide as well as the synthetic, chemically pure HS-pentasaccharide Arixtra (GlcNS6S␣1-4GlcA␣1-4GlcNS3S6S␣1-4IdoA2S␣1-4GlcNS6S␣-OMet, where OMet is O-coupled methyl group). For the latter compound, a more efficient saturation transfer to the four sulfated rings than to the single nonsulfated (GlcA) ring was observed. Overall, these experiments demonstrate a broad GAG binding profile for MCPyV and suggest that sulfation is an important parameter for the interaction, possibly of larger importance than the underlying GAG pyranose scaffold. We note that GAG binding to isolated, assemblyincompetent VP1 pentamers is not observable by STD-NMR, a technique capable of detecting very weak interactions (in the millimolar K d range). Furthermore, mutations in the sialic acid binding site do not affect GAG binding to correctly assembled VLPs (18,19), also demonstrating that the GAG binding site is distant from the sialic acid binding site. Taken together, these observations suggest that the GAG binding site spans more than one VP1 pentamer and is most likely located near the capsid recession. This is in contrast to the region that engages sialic acid, which includes apical loops in an isolated VP1 pentamer. Although we think that such a scenario is unlikely, we cannot rule out the possibility that the high mutivalency of the full capsid with respect to the assembly-incompetent pentamers plays an important role in the detectability of GAG binding by MCPyV.
While we observed a direct GAG-MCPyV interaction by NMR, our crystal soaking and cryo-EM attempts to produce conclusive GAG-VLP complex structures remained unsuccessful despite the relatively high resolution obtained for the unliganded VLP structures. Recent attempts to provide HPV-16 -GAG and BKPyV-GAG cryo-EM complex structures resulted in complexes with weak electron potential in the respective capsid recessions in either case, which was attributed to GAG polymers (33,34). Instead of highly inhomogeneous polymeric GAGs, we employed the chemically homogeneous Arixtra pentasaccharide in our crystallographic and cryo-EM attempts. Therefore, and in the light of our positive NMR binding data for the same compound, we propose that the experimental difficulties that so far prevented high-resolution data on VLP-bound GAGs are not merely caused by the usage of inhomogeneous heparin preparations in the BKPyV and HPV-16 studies, as assumed by the authors of the BKPyV study (33). Instead, the fuzzy electron density could also reflect variable binding modes of chemically homogeneous and inhomogeneous GAGs within a given binding site. Strikingly, the HPV-16 VLP Arixtra binding epitope is indistinguishable from the MCPyV VLP epitope, also supporting the hypothesis that GAG binding is relatively unspecific. This scenario is in stark contrast to the highly specific interactions that restrict PyVs to a generally very small set of sialic acid glycans, as exemplified by the BKPyV/MCPyV comparison (13,18,27).
VP1 sequence comparison suggests that both BKPyV and JCPyV belong to the SV40 PyV clade (8), with high sequence identities (81.7% [33] and 76.7%, respectively). Structural evidence for an SV40-like disulfide pattern was recently confirmed for BKPyV (33). In contrast, MCPyV exhibits much lower similarity to either PyV prototype and more closely resembles mPyV than SV40, with VP1 sequence identities of 50% and 45%, respectively. The extensive disulfide pattern reported here exceeds the one thus far reported for other PyVs, which is not surprising given that MCPyV VP1 contains additional cysteine residues compared to other PyV VP1 sequences. However, the "disulfide ring" that interconnects individual protein chains within strict as well as local VP1 pentamers was reported for mPyV but not SV40 previously (28,31,35). The presence of only one Ca 2ϩ binding site in the capsid is also more reminiscent of mPyV than SV40. Thus, from a structural point of view, MCPyV more closely resembles mPyV, as was suggested by sequence analysis (5).
PyV cellular trafficking was previously shown to proceed in a retrograde manner through the endoplasmic reticulum (ER) (36,37). Presumably, disulfide bonds in the capsid are partially reduced by ER disulfide isomerases, triggering the recognition of the altered virion by the ER unfolded protein response and the translocation of the uncoating virus to the cytosol (14,36,38,58,59). Here, the low calcium concentration could promote further uncoating via destabilization of the Ca 2ϩ -mediated intra-and inter-VP1 interactions. Our finding that both conserved PyV capsid structural elements, disulfides and Ca 2ϩ -chelating sites, are important for MCPyV capsid stability suggests that this PyV-typical mechanism may also be employed during MCPyV infection. MCPyV's extensive disulfide network may reflect an ability to withstand harsh environ-mental conditions, which would presumably contribute to efficient human-to-human transmission and the presence of MCPyV in the common skin microflora.
The C-terminally truncated VP1 variants VP1 1-384 and VP1 1-402 were produced in Escherichia coli DH5␣ cells by introducing additional ochre stop codons (UAA) into the pwM plasmid using site-directed mutagenesis. All plasmids were propagated in E. coli XL10-Gold cells and purified for mammalian cell transfection using the GenElute HP plasmid maxiprep kit (Sigma-Aldrich).
Production of MCPyV VLPs. MCPyV VLPs were purified by a modification of existing protocols (22,23), optimized for obtaining a highly homogeneous sample suitable for structural studies. Siliconized Eppendorf cups were used throughout the purification procedure in order to minimize sample loss. Confluent 293 TT cells (22) were cultivated in Dulbecco's modified Eagle's medium (DMEM) (Gibco) in the presence of 10% fetal calf serum (FCS) and 250 g/ml hygromycin at 37°C with 5% carbon dioxide saturation. Transient transfection was performed at a cell count of 4. The crude lysate was adjusted to 850 mM NaCl and incubated for 30 min at 4°C. The soluble fraction was initially clarified for 10 min at 10,000 ϫ g (4°C), filtered (0.22-m-diameter cellulose membrane), and applied onto a discontinuous density gradient of 15% (wt/vol) sucrose-35% (wt/vol) CsCl in purification buffer (20 mM HEPES [pH 6.6], 150 mM NaCl, 1 mM CaCl 2 ). VLPs were isolated from the lysate at 129,840 ϫ g for 1 h (10°C) in a fixed-angle rotor (70.1 Ti rotor; Beckman).
The VLP sample quality was further improved by performing cation exchange chromatography using a monolithic CIMacSO 3 column (BiaSep). VLP fractions were separately loaded and eluted with a linear NaCl gradient (0.15 to 1 M NaCl, 20 mM HEPES [pH 6.6], 1 mM CaCl 2 ) over 20 ml at a flow rate of 1 ml/min. The purified eVLPs and fVLPs were concentrated by ultrafiltration (100,000-MW-cutoff [MWCO] Amicon filter; Merck-Millipore) to a final concentration of 1 mg/ml, flash-frozen in liquid nitrogen, and stored at Ϫ80°C.
Production of HPV-16 VLPs. L1-and L2-containing HPV-16 VLPs were produced using the MCPyV VLP procedure, except for the ion chromatography step, with plasmids pshell16 and p16L1L2 (39,40). Phosphate-buffered saline (PBS) prepared in 99% D 2 O was used as the NMR sample buffer, and VLPs were dialyzed into this buffer after the second ultracentrifugation step.
Dynamic light scattering. For the assessment of sample homogeneity and size estimation, concentrations of eVLPs and fVLPs were adjusted to 0.1 mg/ml (20 mM HEPES [pH 6.6], 150 mM NaCl, 1 mM CaCl 2 ), and dynamic light scattering (DLS) measurements were performed on a Zetasizer Nano ZS instrument (Malvern Panalytica) using a 3-mm quartz cuvette. Each sample was measured in triplicates (15 accumulations) at 20°C in noninvasive back-scatter mode at 633 nm. The average hydrodynamic radius (R H ) and the polydispersity index (PdI) were calculated based on the merged autocorrelation functions using Zetasizer software 7.11.
Negative-stain transmission electron microscopy. Soluble fractions of the clarified VLP lysates or purified VLP solutions at a concentration of 1 mg/ml were directly placed onto a glow-discharged EM grid (thin-bar, square, 300-mesh Cu grids with Formvar-carbon support film; Science Services). After adsorption, the grids were washed with double-distilled water and stained with 1% uranyl acetate. The grids were then examined using a Zeiss Libra 120 transmission electron microscope (Carl Zeiss, Oberkochen, Germany) operated at 120 kV.
Thermal shift assay. The effects of disulfide bond formation and calcium complexation on capsid stability were estimated in thermal shift assays (TSAs) performed on a QuantStudio 5 real-time PCR cycler (Applied Biosystems, Thermo Fisher Scientific). VLPs were diluted in purification buffer to a concentration of 0.1 mg/ml and incubated with 10 mM dithiothreitol (DTT) and/or 10 mM EDTA for 1 h at room temperature prior to the addition of 1ϫ Protein Thermal Shift dye (Applied Biosystems, Thermo Fisher Scientific). The samples were equilibrated at 4°C and gradually heated to 95°C over 30 min while monitoring dye fluorescence at 586 nm. The averaged protein melting temperature for each reaction was calculated from the melting curve inflection points from three replicates using Protein Thermal Shift software v1.3 (Applied Biosystems, Thermo Fisher Scientific). All reagents used were preadjusted to pH 6.6 to avoid pH-dependent effects on the thermal stability of the VLPs.
Nonreducing SDS-PAGE. Nonreducing SDS-PAGE was performed using 4 to 20% Tris-glycine precast gradient gels (Rosetta Stone Biotech) or 8% Tris-glycine gels. All samples were incubated in purification or lysis buffer with 10 mM DTT and/or 50 mM EDTA in the presence of 2% (wt/vol) SDS for 1 h at ambient temperature. Per well, 0.2 g of capsid or a 1:10 dilution of the normalized crude lysate was applied, and a heat-treated sample (5 min at 95°C) with 10 mM DTT and 50 mM EDTA was used as the VP1 monomer reference sample. The molecular weight was estimated using a Spectra multicolor broad-range protein ladder (Thermo Fisher Scientific), and electrophoresis was performed using Tris-glycine buffer (Laemmli buffer [41]) at a constant current of 35 mA/gel until the marker reached the end of the gel.
Crystallization and data collection. MCPyV fVLPs were crystallized in a hanging-drop vapor diffusion setup at protein concentrations of 1 to 1.5 mg/ml in a broad range of sodium citrate buffers containing 2-methyl-2,4-pentanediol (MPD) and NaCl. Diffraction was strictly crystal size dependent. The best diffraction data were obtained from a crystal grown for approximately 1 year in a solution containing 0.1 M sodium citrate (pH 4.6), 30% (vol/vol) MPD, and 0.2 M NaCl at 20°C. Crystals were flash-frozen in liquid nitrogen without an additional cryoprotectant, and X-ray diffraction data were collected at the I03 beamline of the Diamond Light Source (Didcot, UK). Diffraction images from three data sets collected on one crystal were integrated, reduced, and merged to a 3.5-Å resolution using the XDS/XSCALE program package (42).
Phasing, model building, and refinement. The initial phases for the MCPyV capsid structure were obtained by molecular replacement with PHASER (v2.5.0) (43) using a CHAINSAW-modified (44) search model generated from the SV40 capsid structure (Protein Data Bank [PDB] accession number 1SVA) (28). For model building, 15-fold-constraint noncrystallographic symmetry (NCS) parameterization was applied to the six conformationally unique VP1 monomers ␣, ␣=, ␣Љ, ␤, ␤=, and ␥ in order to reconstruct the crystal asymmetric unit of 90 VP1 chains. The model was completed by several cycles of manual real-space correction in COOT (45) (v0.8.9.1) and reciprocal space refinement using PHENIX (46) (v1.13). Each of the six monomers was separately built and copied with respect to the NCS symmetry using a Python-based PyMOL script. One rigid-body group was defined for each monomer of the asymmetric unit and used during the first macrocycle of refinement, followed by NCS constraint coordinate and group-B-factor optimization. Due to a mild deviation from a perfectly spherical shape, a different density appeared systematically in the unit cell during refinement. Therefore, the pentamer-connecting so-called invading arms of the individual VP1 chains were excluded from the NCS, and a separate NCS group was defined for the ␣-monomers in order to eliminate the otherwise restrained angle between the strict and local capsomers. Additionally, B-factor sharpening of Ϫ45.6 Å 2 was applied using PHENIX.autosharp (47), which further increased the map interpretability. Model and map accuracies were evaluated by NCSbased B-factor analysis and calculation of real-space correlation coefficients (RSCCs). Figures were generated using PyMOL (v1.8.4; Schrödinger Inc.).
Cryo-EM sample preparation, data acquisition, and processing. Three microliters of the MCPyV sample at a concentration of ca. 10 mg ml Ϫ1 was applied onto freshly glow-discharged CF 2/2 grids (Microchips) and plunge-frozen in liquid ethane using Vitro bot Mark IV (Thermo Fisher). Micrographs were recorded automatically (EMU software; Thermo Fisher), using a Titan Krios microscope operated at 300 K, with a Falcon 3EC direct electron detector (Thermo Fisher) at a nominal magnification of ϫ75,000, corresponding to a pixel size of 1.053 Å. Dose-fractionated movies were acquired at an electron flux of 0.45 e Ϫ /pixel/s over 76 s (67 frames), corresponding to a total electron dose of ϳ30 e Ϫ /Å 2 . Images were recorded in a defocus range from Ϫ0.5 to Ϫ1.5 m. Frame-based motion correction was performed using MotionCor2 (48) with a dose filter of 0.45 e Ϫ /Å 2 /frame. The contrast transfer function (CTF) was estimated from the non-dose-weighted images using CTFFIND4 (49). A total of 23,349 particles were picked using Dog-picker (50) as implemented in Appion (51) and refined in cisTEM (52), applying icosahedral symmetry. After obtaining a consensus map, the particles were subclassified into 3 classes and locally refined. This procedure combined with CTF refinement yielded a map at a 3.4-Å resolution, containing 95% of all picked particles. The reported resolution was estimated according to the goldstandard criterion of a Fourier shell correlation (FSC) value of 0.143 (53). Cryo-EM model building and refinement. The map was initially sharpened by applying a B-factor of Ϫ90 Å 2 using a hollow-sphere mask with radii of 265 Å and Ϫ150 Å. The crystal structure of the MCPyV capsid was positioned within the cryo-EM map using rigid-body fitting in CHIMERA (54) (v.1.11.2) based on the map-to-model correlation coefficient (CC ϭ 0.7). For model building, the map was segmented, covering the asymmetric unit, followed by a simulated annealing protocol in PHENIX (v.1.14). Several iterations of manual model correction were performed in COOT (v0.8.9.1), and the icosahedral asymmetric unit was refined using PHENIX.real_space_refine (55), applying NCS constraints. The model was evaluated by using MOLPROBITY (56), and calculations of RSCC, half-map-based FSC, model-to-map FSC (data not shown), as well as CC (Table 2) were performed using PHENIX.validation_cryoem (57). Figures were generated using PyMOL (v. 1.8.4).
NMR spectroscopy. All STD-NMR spectra were recorded on a Bruker AVIII 600-MHz spectrometer equipped with a room-temperature probe head at temperatures of 283 K using 3-mm-internal-diameter (ID) Match tubes. The MCPyV NMR buffer contained 150 mM NaCl and 1 mM CaCl 2 in 99% D 2 O (Cortecnet). The HPV-16 and MCPyV VP1 NMR buffer contained 150 mM NaCl and 20 mM KH 2 PO 4 (pH 7.4) in 99% D 2 O. Samples contained about 10 M VLPs or 20 M VP1 and 1 mM the respective oligosaccharide(s). HS-dp2, HS-dp8, and DS-dp6 oligosaccharides were purchased from Dextra Laboratories and resuspended in 99% D 2 O to yield 40 mM stock solutions. Arixtra (Aspen Laboratories) was dialyzed in NMR buffer using 500-MWCO Slide-A-Lyzer conical cups with cellulose membranes (Thermo Fisher Scientific). Off-and on-resonance frequencies in STD-NMR experiments were set to Ϫ30 ppm and Ϫ0.5 ppm, respectively. The irradiation power and length of the selective pulse train were 57 Hz and 2 s, respectively. A strength of 3.2 kHz was employed to suppress residual protein resonances in a continuous-wave spin-lock pulse. A total of 5,000 scans were recorded, and the relaxation delay was 3 s. Prior to Fourier transformation, NMR data were multiplied with a Gaussian window function. Data were processed with TOPSPIN 3.0 (Bruker).
Data availability. The MCPyV structures described here were deposited in the Protein Data Bank under accession numbers 6ZLZ and 6ZML. Cryo-EM data are available at the EM Data Resource (www.emdataresource.org) under accession number EMD-11293.