Atomic resolution structure of human papillomavirus bound to heparin

Abstract

Human papillomavirus (HPV) is a significant health burden and leading cause of virus-induced cancers. The mechanisms of HPV receptor binding and host entry are not completely understood, although it is known that heparan sulfate proteoglycans (HSPGs) mediate entry. HPV16 quasivirus, composed of L1 and L2 capsid proteins with a packaged cottontail rabbit papillomavirus genome was incubated with heparin. The complex was vitrified, and data were collected for cryoEM single particle analysis. Subparticles were extracted and hexavalent and pentavalent capsomers were refined separately. Here we present the resulting 1.9 Å resolution structure, with heparin visualized around the capsomer at the icosahedral fivefold vertex. A model of the asymmetric unit was built unambiguously into the atomic resolution cryoEM map. Hydrogen bonds are predicted between L1 N-terminal regions, which are supported by molecular dynamics. The heparin binding site was identified, along with local L1 conformational changes and global flexibility changes. These changes induced by heparin binding likely reflect the structure of HPV during early stages of entry and provide a framework for future HPV biochemical, genetic, and biophysical studies.

Human papillomavirus (HPV) is a significant health burden, but the mechanisms of HPV receptor binding are not understood. Here, the authors solve a 1.9 Å resolution structure of HPV bound to heparin with insights into the binding interface.

Introduction

Human papillomavirus (HPV) is the most common sexually transmitted infection, with over 80% of people infected with HPV in their lifetime¹. Fifteen HPV serotypes are considered high risk, or cancer-causing, and HPV is responsible for over 90% of cervical and anal cancer cases^2,3. Oropharyngeal cancer cases have been rapidly increasing in the past three decades due to an increase in HPV related squamous cell carcinoma, and HPV16 is responsible for the majority of cases⁴. Additionally, over 60% of cervical cancer cases are attributed to HPV16 and HPV18, and because of the clinical importance, HPV16 was used in this study⁵. Most HPV infections are asymptomatic and clear spontaneously, but chronic infections occur and are more likely to cause cancer. There are currently no antivirals available to treat or cure chronic HPV infections^6,7. HPV vaccines such as Gardasil9 have been successful in reducing high-risk HPV prevalence and show evidence of herd immunity with 80% coverage⁸. HPV-associated cancer cases have also declined, but vaccine coverage in the United States remains lower than the goal of 80%, and coverage in the UK has decreased significantly in the past five years^9,10.

HPV replication relies on the differentiation of basal epithelial cells to keratinocytes, which makes the production of HPV in vitro challenging. Organotypic raft cultures are used to produce HPV, however, the yield and purity of these samples are too low for structural studies, which require a highly concentrated, pure sample¹¹. An alternative method is to produce quasiviruses, which are composed of HPV L1 and L2 proteins with a packaged papillomavirus genome¹². L1 and L2 proteins assemble and package a genome in cell culture. This system consistently produces a high yield of purified virus and has been successfully used for previous structural and vaccine studies^13,14. The quasiviruses used for this study are composed of HPV16 capsid proteins and package a cottontail rabbit papillomavirus genome¹².

Papillomaviruses are non-enveloped, double-stranded DNA viruses with an 8 kb genome. The capsid has T = 7d icosahedral symmetry and is 55–60 nm in diameter. Papillomaviruses are a special case within the T-number definition, as T = 7 symmetry would theoretically require 420 subunits, arranged as pentamers and hexamers. However, HPV is composed of 360 subunits, arranged as 72 capsomers composed of five L1 proteins. Twelve capsomers, called pentavalent capsomers, are located on the fivefold axis of symmetry and are true pentamers. The remaining sixty capsomers, called hexavalent capsomers, are composed of five proteins that have a similar beta-jellyroll motif core structure but variations in the C-terminal arm extension and N-terminus. The unique C-terminal extensions allow for coordination of the hexavalent capsomer between six other capsomers¹⁵. The interactions between the center capsomer and surrounding capsomers are not identical, and variations in the distance between capsomers have been identified¹⁶. Additionally, HPV contains between 12 and 72 copies of the minor capsid protein, L2^17,18. The location and structure of the L2 protein are unknown. L2 is not required for L1 capsid assembly, but it is essential for infectivity¹⁹.

Despite the T = 7 icosahedral symmetry, which would imply an asymmetric unit composed of seven proteins, the asymmetric unit of the virus contains six L1 proteins – one pentavalent L1 protein and five hexavalent L1 proteins. This asymmetric unit encompasses one hexavalent capsomer (with five L1 proteins) and one fifth of a pentavalent capsomer (one L1 protein). These six L1 proteins are labeled as chains A-F (Fig. 1). The pentavalent capsomers are composed of five copies of Chain A, and the hexavalent capsomers are composed of chains B-F. Each L1 protein can form a disulfide bond with an adjacent capsomer, which allows for capsid assembly and is thought to stabilize the capsid²⁰. A C-terminal loop region containing amino acids 403-446, known as the C-terminal arm extension, stretches from one capsomer to an adjacent capsomer, and a disulfide bond can be formed between Cys 428 in the C-terminal extension and Cys 175 in the adjacent capsomer. Each of the six C-terminal arms in the asymmetric unit has a different conformation¹⁴. The regions of the C-terminal extension in between capsomers (residues 403-417 and 429-446) are flexible loop regions, which allow for significant flexibility between capsomers and variability of the capsid diameter¹⁴. The first ten N-terminal amino acids are required for T = 7 capsid formation, based on work from Chen et al. 2001, and a structural role for the N-terminus of L1 was proposed by Modis et al. 2002^20,21.

Fig. 1. Major structural protein L1 arrangement in the T = 7 icosahedral capsid.

A–C Six L1 chains A-F make up the asymmetric unit, colored purple, blue, green, yellow, orange, and red, respectively. A 2.76 Å icosahedral map with density colored according to the location of the specific L1 protein in the asymmetric unit. B The asymmetric unit colored according to chain. L1 proteins in the center asymmetric unit are shown in full color and labeled while surrounding asymmetric units are in pastel colors. The L1 proteins for one surrounding asymmetric unit are labeled. One hexavalent capsomer is indicated with a hexagon and one pentavalent capsomer is indicated with a pentagon. C Graphical illustration of the arrangement of seven capsomers and capsomer arm connections between the center capsomer, not to scale. The icosahedral threefold axis is indicated with a triangle, the twofold axis is indicated with an oval, and the fivefold axis is located at the center of the pentavalent capsomer. Arm connections between the center asymmetric unit are represented by arrows, colored according to the L1 from which they arise. The chain A arm connection binds to chain B, the chain B arm connection binds to chain F’, and the chain F’ arm connection binds to chain A. Around the threefold axis, the chain C arm connection binds to chain D’, and chain D binds to chain C’. Around the twofold axis, the chain E arm connection binds to chain E’.

The receptors and mechanisms involved in HPV cell attachment and entry are complex, and it is known that heparan sulfate proteoglycans are used for entry. Specifically, it is the heparan sulfate glycan moieties that are required for HPV infection²². Heparin has often been used as a surrogate for heparan sulfate, as it is very structurally similar, contains the same disaccharide units, and is easier to produce. However, heparin is more homogenous, has a shorter chain length, and is more highly sulfated than heparan sulfate²³. Heparin binding to the HPV capsid has been thought to induce a conformational change to the HPV capsid²². Incubation of HPV pseudovirus with heparin was shown to soften and increase capsid size²⁴. This change was only observed in heparin with at least a degree of polymerization of 45-51 (dp45-51), and not for shorter heparin constructs. The authors proposed that heparin has multiple binding sites and can mechanically lock the capsid in an extended conformation²⁴. Previously, a 4.3 Å structure of heparin bound to the HPV capsid was solved, and heparin was visualized bound in the canyon between the pentavalent capsomer and each surrounding hexavalent capsomer²⁵. However, the previous map had a modest resolution and could only predict what virus residues recognize heparin. The map resolution was insufficient to identify the HPV-heparin contacts.

Here, a high resolution cryoEM structure of the HPV-heparin complex is presented. All-atom molecular dynamics (MD) simulations of the empty HPV capsid in the absence of heparin are used to enhance interpretation of the experimental density by revealing key details of the underlying conformational ensemble. Over the past two decades, atomistic MD has been applied to study the intact capsids of numerous viruses, serving as a valuable complement to microscopy²⁶. For example, MD-derived ensembles have been used to discover rare conformations of capsid proteins that are averaged out in experimental reconstructions²⁷, establish dynamic inter-capsomer contacts that are not resolved experimentally²⁸, explain the organization of encapsidated genome apparent in micrographs²⁹, and determine whether differences between cryoEM models represent equilibrium fluctuations of the apo-capsid versus distinct structural states induced by bound host-factors^30,31.

The atomic resolution cryoEM structure of heparin bound to HPV presented here allowed for unambiguous model building of the HPV16 capsid and confident fitting of heparin. Heparin density was observed surrounding the fivefold capsomers and the L1 residues that interact with heparin were identified. Local conformational changes were identified, but no global conformational changes occurred upon heparin binding. Due to the quality of the map, water molecules were identified. N-terminal hydrogen bonds were assigned in the cryoEM map and MD supports the presence of hydrogen bonds. These findings suggest a model where both disulfide bonds and N-terminal hydrogen bonds mediate capsid stability. These details and analysis provide insights into HPV capsid structure and the first stages of viral host attachment and entry.

Results

HPV16 quasivirus incubated with heparin was applied to freshly glow discharged grids, vitrified, and cryoEM data were collected on the Pennsylvania State Titan Krios at 54,000x magnification (Table 1). Micrographs contained a high concentration of virus particles with internal density consistent with packaged genome (Supplementary Fig. 1). As with previous HPV quasivirus data collections, there was some variation in the shape of particles, and rod-shaped structures were observed^14,25. However, most particles were spherical and appeared to be the same approximate size, ~55 nm in diameter. After autopicking, particles were classified and ~75,000 were selected for further refinement. Icosahedral symmetry was imposed and higher order aberrations were corrected during 3D refinement, which resulted in a 2.98 Å HPV-heparin complex map (Supplementary Fig. 1). Density consistent with bound heparin was identified in five locations surrounding the pentavalent capsomer, in the depression between the pentavalent capsomer and each of the five surrounding hexavalent capsomers. The densities were of lower magnitude than the capsid proteins and were not visualized in between hexavalent capsomers.

Table 1.

CryoEM data collection parameters and model validation

	First data collection			Second data collection
	Full particles	Hexavalent subparticles	Pentavalent subparticles	Full particles	Hexavalent subparticles	Pentavalent subparticles
DATA COLLECTION AND PROCESSING
Microscope	Titan Krios G3	Titan Krios G3	Titan Krios G3	Titan Krios G3	Titan Krios G3	Titan Krios G3
Magnification	59,000x	59,000x	59,000x	120,000x	120,000x	120,000x
Voltage (kV)	300	300	300	300	300	300
Electron exposure (e–/Å2)	40	40	40	39.75	39.75	39.75
Defocus range (μm)	−0.5 to −2.0	−0.5 to −2.0	−0.5 to −2.0	−0.5 to −2.0	−0.5 to −2.0	−0.5 to −2.0
Pixel size (Å)	1.1	1.1	1.1	0.517	0.517	0.517
Symmetry imposed	I1	C1	C5	I1	C1	C5
Initial micrographs (no.)	3032	−	−	13,943	−	−
Final micrographs (no.)	2923	−	−	13,289	−	−
Particles (no.)	74,966	4,497,960	899,592	44,159	2,649,540	529,908
Map resolution (Å)	2.98	2.27	2.27	2.76	1.92	1.90
FSC threshold	0.143	0.143	0.143	0.143	0.143	0.143
REFINEMENT
Initial model	−	−	−	−	PDB 7KZF
Model resolution (Å)	−	−	−	−	1.3
FSC threshold	−	−	−	−	0.143
Map sharpening B factor (Å2)	−	−	−	−	63.0	63.7
Model composition
Non−hydrogen atoms	−	−	−	−	24,086
Protein residues	−	−	−	−	2864
Protein B factor (Å2)	−	−	−	−	22.45
R.M.S. deviations
Bond lengths (Å)	−	−	−	−	0.004
Bond angles (°)	−	−	−	−	0.611
Validation
MolProbity score	−	−	−	−	0.95
Clashscore	−	−	−	−	0.95
Poor rotamers (%)	−	−	−	−	0.08
Ramachandran plot
Favored (%)	−	−	−	−	98.25
Allowed (%)	−	−	−	−	1.75
Disallowed (%)	−	−	−	−	0.00

Data were processed using cryoSPARC and the model was built using ISOLDE and Phenix^47,52,53.

Data were collected on a Titan Krios G3.

The resolution improved further with subparticle extraction and refinement

As established previously, subparticle extraction and refinement improve the resolution of papillomavirus maps by reducing the effect of flexibility between capsomers^14,32. Using the established approach, 250 Å diameter subparticles centered on pentavalent and hexavalent capsomers were designated and extracted, C5 and C1 symmetry  was imposed, and 2.47 and 2.46 Å resolution maps were attained. To improve the subparticle resolution, a different strategy was used. Larger 380 Å diameter subparticles were designated, which were centered on a pentavalent capsomer and large enough to include the entirety of the five surrounding capsomers. Similar subparticles were designated and extracted for all hexavalent capsomers. We termed these large subparticles starfish. Defocus and higher-order aberrations were corrected with subsequent rounds of local refinement, global CTF refinement, and local CTF refinement, which resulted in a pentavalent starfish map of 2.62 Å resolution. We also initiated a different approach to improve the subparticle map. Each of the capsomers surrounding the center was masked and refined separately, then subtracted sequentially from the starfish refined map. Five or six capsomers were subtracted for pentavalent and hexavalent starfish subparticles, respectively, resulting in only the center capsomer. This approach resulted in 2.27 Å resolution maps of the pentavalent and hexavalent capsomers. As the data were collected at 1.1 Å pixel size, further improvement was not possible because the resolution reached the Nyquist resolution limit of 2.2 Å.

To improve the resolution of our HPV-heparin map, new data were collected at higher magnification using the same HPV sample described above. The additional data were collected on the Pennsylvania State Titan Krios at 120,000x magnification, which has a Nyquist limit of 1.08 Å (Table 1). After particle selection, ~44,000 particles were used for an icosahedral reconstruction that resulted in a 2.76 Å resolution map (Fig. 2) (Supplementary Fig. 2). Starfish subparticles of 760 Å diameter were extracted and refined without further treatment and resulted in 1.90 Å and 1.92 Å resolution pentavalent and hexavalent subparticle maps, respectively (Fig. 2) (Supplementary Fig. 3). After applying the sequential subtraction method described above, the resulting resolution remained 1.90 Å for the pentavalent subparticles. Thus, the subtraction approach did not improve the resolution further for these higher magnification data and was not used for the hexavalent starfish.

Fig. 2. High magnification, 120kX data collection and cryoEM icosahedral map and refinement.

A Representative micrograph, from the dataset containing 13,945 micrographs, of purified HPV16 quasivirus bound to heparin, scale bar 40 nm. B Central slab from the icosahedral 3D refinement, which included 44,159 particles. Scale bar 10 nm. C Representative 2D class average images. D Icosahedrally averaged map at 2.76 Å resolution, colored radially in Å according to the color key (below). E, F Pentavalent 1.90 Å (E) and hexavalent 1.92 Å (F) resolution subparticle maps, colored radially in Å according to the color key (below).

The pentavalent and hexavalent maps had characteristics of atomic resolution, including distinct side-chain densities and densities consistent with water molecules (Supplementary Fig. 4). One pentavalent and six hexavalent maps were recombined to build the asymmetric unit, which contained all the flexible arm connections made by L1 C-terminal extensions. The L1 build was initiated with a previous HPV structure (PDB 7KZF)¹⁴. The asymmetric unit of HPV, consisting of six copies of L1 was built unambiguously from residues 20–52, 60−129, 141−175, 183−430, 441−480. The first three gaps, between residues 52−60, 129−141, and 175−183 are regions containing surface-exposed loops. The remaining gap of 430−441 is the known flexible connecting region between capsomers. The location of these residues and all flexible regions were approximated using an unsharpened 2.4 Å hexavalent starfish subparticle map that was produced before focused refinement. Due to the high resolution and map quality, the resulting build corrected a clash between chains A and F in the previous structure. With an overall Q−score of 0.86, the model accurately reflects the cryoEM map (Supplementary Fig. 4)³³.

Extra density was observed at His 36 N1 for each of the six L1 chains in the asymmetric unit and appeared as a small protrusion on the first nitrogen on the histidine ring (Supplementary Fig. 5). This extra density did not disappear as the contour level increased, indicating that this density occurs consistently at this location throughout the capsid and is not an artifact. The appearance of the extra density is consistent with the presence of an ion, such as a possible calcium ion, or a post-translational modification, such as methylation. Previously, post-translational modifications of HPV16 L1 at Thr 129 and Thr 340, an N-acetylglucosamine and phosphorylation, respectively, have been identified with mass spectrometry, but a His 36 modification has not been previously reported³⁴. Due to the ability of histidine to coordinate ions, it is more likely that the unfilled density at His36 N1 corresponds to a cation.

The HPV capsid is inherently flexible due to capsomer movement conferred by arm connections

The flexibility of HPV bound to heparin was compared to the flexibility of HPV without heparin¹⁴. In the icosahedrally averaged map, the location of each capsomer is forced into perfect icosahedral symmetry. When the program ISECC is used to extract subparticles, the perfectly enforced symmetrical location is recorded during extraction¹⁴. Upon subparticle refinement, the actual center of each capsomer is determined. To assess global flexibility of the capsid, the icosahedrally imposed location was compared to the actual empirically determined capsomer location. Four unique interactions between neighboring capsomers were analyzed. The standard deviation of differences from perfect icosahedral symmetry for the complex was on average 2%, while the standard deviation for HPV alone was 4%. Thus, there was reduced flexibility in all arm connections of the HPV-heparin complex compared to HPV alone. The flexibility of the entire capsid with bound heparin remained the same as HPV alone, although the distribution changed (Supplementary Fig. 6). The icosahedrally averaged map had a diameter of 57.6 nm and the 95th percentile (0.986–1.014) corresponds to capsid diameters ranging from 56.8 nm to 58.4 nm, which is almost identical to diameters observed in HPV alone, which were 56.8 to 58.5 nm¹⁴ (Supplementary Fig. 7).

The above method assessed capsid flexibility by calculating the deviation of capsomers. To assess possible changes within capsomers, a local resolution map of the 2.98 Å icosahedral HPV heparin structure was generated. For the complex map, the local resolution ranged from 2.4 to 36.3 Å, with 25 and 75th percentile values of 2.9 and 4.7 Å. To compare to papillomavirus capsids without heparin, local resolution maps were calculated for HPV16 at 3.1 Å resolution¹⁴ (Supplementary Fig. 7). Compared to the HPV16 map, the HPV-heparin complex map had poorer resolution at the fivefold, suggesting more local flexibility within the pentavalent capsomers. The hexavalent capsomers had similar local resolution mapping for HPV16 and the HPV-heparin complex.

Heparin was visualized bound to the HPV capsid in between each fivefold capsomer and the surrounding hexavalent capsomers

The fivefold heparin density was resolved at low contour levels and had a lower magnitude than that of the capsid. The magnitude of the heparin density could be due to the variability in chain length (dp 40-51), flexibility of the heparin molecule, or indicate that not all binding sites were occupied. ISECC was used to extract subparticles centered on the bound heparin, and these subparticles were refined in RELION to a resolution of 3.2 Å. 3D classification resulted in three classes. Classes #1 and #2 had heparin bound (resolutions of 4.6 Å and 4.2 Å, respectively), and class #3 at a resolution of 4.7 Å had no heparin bound. The two classes with heparin bound accounted for 60.6% of subparticles, and the one class with no heparin bound contained the remaining 39.4% of subparticles. During ISECC extraction, the location of each subparticle is recorded by identifying the nearest fivefold, threefold, and twofold region. The 3D classification identified which subparticles contained bound heparin, and using the location information from ISECC, the exact location of each particle within the classes with bound heparin was determined. Thus, we were able to determine that on average, three of the five potential heparin-binding sites were occupied around each fivefold. This asymmetric binding is likely why the heparin density is not as strong as the HPV capsid density. Class #2 was chosen for further 3D classification, as the heparin density was strongest in this class. Iterative rounds of 3D classification in RELION resulted in a class with ~130,000 particles, and local refinement in cryoSPARC resulted in a 3.4 Å map with heparin density that was amorphous but continuous and distinct from L1 density.

The model built from the 1.9 Å resolution map was fit into the 3.4 Å map with heparin density and cryoEM density attributed to L1 were removed to interpret the heparin density. The heparin density was strong, continuous, but amorphous, which was not surprising since heparin consists of a mixed population of chains of variable length. Thus, a model for heparin was fit into the density, but couldn’t be confidently built into the density. HPV contact residues were identified based on their proximity to the heparin density as well as their charge and orientation (Fig. 3). Residues from chains A, B, and F are likely to interact with heparin, and both the pentavalent and the hexavalent capsomers are involved (Fig. 3). The majority of identified contact residues are positively charged or are polar, including lysine, histidine, threonine, and asparagine, with the side chains directed towards the heparin density (Supplementary Fig. 8). The identified contacts are different than previously solved structures of HPV16 bound to heparin (Supplementary Fig. 9). The X-ray crystallography structure solved by Dasgupta et al. 2011 identified four regions where a heparin molecule composed of 8-10 subunits binds to HPV16 pentamers along the outer edge and the top of the pentamer. The L1 root mean square deviation (RMSD) between the two structures is 0.4 Å, indicating that the L1 pentamer structures are very similar (PDB 5W1O and 9NZU), and there are two overlapping heparin-binding site residues: Lys 54 and Lys 59³⁵. However, these two residues bind to separate heparin molecules in the 5W1O structure, as opposed to the same heparin molecule as presented here³⁵. The 4.3 Å cryoEM structure of T = 7 HPV16 solved by Guan et al. 2017 identified the heparin-binding site in between the pentavalent and hexavalent capsomers, which is the same site identified here²⁵. Six residues in the identified binding site are consistent between the two structures, however, the other six are not, likely due to difficulties determining the binding site at 4.3 Å resolution (Supplementary Fig. 9). The RSMD is 0.91 Å between the two structures, indicating that the overall L1 structure is not conformationally different (PDB 5KEQ and 9NZU)²⁵. At lower contour levels, the heparin density extends beyond the contacts identified and bends towards the pentavalent capsomer. The cryoEM density at this contour level is indistinct and amorphous, however the density is in close proximity to the C-terminal arm loop regions in between the capsomers, specifically residues 431–444 in chains A and F (Supplementary Fig. 8). Of interest is the proximity of the heparin density to residues 442 and 443, as these residues have been shown to be essential for infectivity and structural activation observed with AFM²⁴. As the heparin density is only seen at lower contour, these interactions are likely to be more transient than the contacts identified above.

Fig. 3. Heparin fit into the subparticle map and identification of the heparin-binding site.

A HPV-heparin map with the extent of heparin subparticle extraction indicated with magenta circle. Subparticles were centered on the heparin density present between pentavalent and hexavalent capsomers, and all symmetry-related locations were extracted with a box size of 250 Å. B Heparin subparticle refinement at 4.04 Å resolution is shown in transparent gray with a pentavalent and hexavalent capsomer indicated (purple pentagon and maroon hexagon, respectively). Four L1 proteins, Chain A (purple), Chain A’ (light purple), Chain B (blue), and Chain F (red) are fit in the map. C, D Heparin binding site shown with residues that are identified as contacts with heparin in cyan with a top (C) and side (D) view. Heparin likely interacts with four L1 chains – Chain A (purple), Chain A’ (light purple), Chain B (blue), and Chain F (red). HPV-heparin subparticle map density (transparent gray) with a zoomed view to show the L1 structural details in the contact interface. Amino acids interacting with heparin in each chain are: Chain F – Lys 54, Asn 57, Lys 59, Lys 430, Chain B – Asn 177, Chain A – Asn 57, Asn 58, Lys 59, Lys 430, His 431, Chain A’ – Thr 176, Asn 177. Zoomed views of the interface are presented in Supplementary Fig 7.

The RMSD between the previous 3.1 Å T = 7 HPV structure from Goetschius et al. 2021 (PDB 7KZF) and this structure (PDB 9NZU) is 0.454 Å, indicating that no global conformational changes occurred to L1 due to heparin binding¹⁴. However, local changes were observed among the residues predicted to be involved in heparin binding. Of the residues suspected to be involved in heparin binding, the RMSD between the model of HPV alone (PDB 7KZF) and the HPV-heparin model is 4.192 Ź⁴.

All-atom MD simulations of the HPV capsid in the absence of heparin reveal strong localization of chloride ions spanning the width of the canyon between pentavalent and hexavalent capsomers, but not the canyon between hexavalent capsomers, consistent with the location of heparin binding identified by cryoEM¹⁶ (Fig. 4a). This finding indicates a unique region of positive electrostatic potential, arising from the proximity of the positively charged and polar residues contributed by chains A, B, and F (Fig. 4b). The collective influence of these residues is sufficient to attract electronegative species, rendering the space highly complementary to the binding of sulfated proteoglycans like heparin. The strongest region of chloride occupancy corresponds to interaction with the amino group of Lys 59 from chain A, suggesting a favorable point for sulfate contact (Fig. 4b). MD simulations further show that about half of the heparin binding residues identified by cryoEM frequently ($\ge$80%) remain oriented for heparin complexation even in its absence (Fig. 4c), reflecting similarities noted between PDB 7KZF and the present heparin-bound structure. Interestingly, Lys 430 and His 431 are just beyond the range of the electropositive hotspot denoted by chloride localization.

Fig. 4. MD simulations indicate that, in the absence of heparin, chloride ions tend to localize in a region of positive electrostatic potential that corresponds with the heparin binding site.

A Representative image of the L1 starfish showing chloride ion density (cyan), which spans the width of the canyon between pentavalent and hexavalent capsomers, but not between hexavalent capsomers. B Close-up view of chloride density (contour level $\mathrm{\Delta }G=-0.4$ kcal·mol^-1) showing surrounding positively charged and polar residues. The strongest region of density (contour level $\mathrm{\Delta }G=-1.0$ kcal·mol^-1) complements the amino group of Lys 59, suggesting a favorable location for sulfate interaction. C Contact frequencies of protein residues to chloride density at contour level $\mathrm{\Delta }G=-0.4$ kcal·mol^-1. Gray highlights denote heparin binding residues identified by cryoEM (Fig. 3). Nearby histidine residues were predicted to be neutral in the absence of heparin, and their treatment as such in MD models may account for their absence amongst residues in (C). Contact frequency data are aggregated over all 60 quasi-equivalent chains resulting in an effective sampling of 42.6 μs. Source data are provided as a Source Data file.

It is known that the C-terminal arm extension of L1 is involved in capsid assembly, and disulfide bonds have been shown to form between Cys 175 and Cys 428 from L1 proteins in neighboring capsomers. However, density between Cys 175 and Cys 428 corresponding to a disulfide bond was not resolved in the 1.9 Å map in any of the L1 chains (A–F) in the asymmetric unit, even at low contour (Supplementary Fig. 10). No reducing agents were present at any point throughout the production of the HPV quasivirus particles. In the refined model built from the 1.9 Å map, the distance between sulfur atoms in Cys 175 and Cys 428 was greater than 3 Å, for all chains, significantly more distance than what is required between sulfur atoms to form disulfide bonds, ~2.05 Å. A model with all possible disulfide bonds in the asymmetric unit was refined and then compared to the model with no disulfide bonds built. There were no Ramachandran or rotamer outliers in the residues around Cys 175 and Cys 428 for either model. Q-scores for Cys 175 and Cys 428 were averaged for each model and were 0.85 and 0.86 respectively, suggesting that at each individual location, the presence or absence of a disulfide bond is plausible. During cryoEM data collection, radiation damage can reduce the cryoEM density seen for disulfide bonds and for negatively charged amino acids, and may be responsible for the lack of disulfide bonds seen in HPV structures³⁶.

Regions of the N-termini of L1 proteins in adjacent capsomers are close enough to each other for hydrogen bonds to form. Specifically, hydrogen bonds were predicted to form between chains F and C’, D and C’, and E and E’ (Fig. 5). The cryoEM density for the HPV capsomers is continuous and distinct, but the loop regions between capsomers are flexible and have different conformations throughout the capsid. Because of this heterogeneity, the majority of loop regions between capsomers cannot be resolved clearly in the 1.9 Å map. However, the N-terminus of L1, which is present as loop regions between capsomers, has distinct cryoEM densities. This finding indicates that these N-terminal regions have increased stability, consistent with hydrogen bond formation. The corresponding cryoEM density connecting the capsomers in this region is continuous and strong between C’ and F and C’ to D. Thus, the density suggests that most of the predicted H-bonds likely form at many locations throughout the capsid, although the hydrogen bonds predicted between E and E’ have less convincing density (Fig. 5). The majority of predicted H-bonds were main chain interactions and did not involve side chains. The L1 chains in the model of the asymmetric unit were superimposed and there are four regions with structural differences, which are the N-terminus, the C-terminus, the C-terminal arm extension, and amino acids 83–91. The structural differences in the C-terminal arm extension have been described previously¹⁴. Residues 83–91 are part of a loop region near the base of the capsomer, and chain D, chains A and F, and chains B, C, and E are slightly offset from each other. The C-terminus of L1 has the same conformation up to residue 481, where the chains diverge. The N-terminus of L1 has a significantly different conformation for each chain. All N-terminal residues for chains C, D, and F were modeled, while chain A was modeled up to residue 12, chain B was modeled up to residue 16, and chain E was modeled up to residue 3. Even with unmodeled regions in half of the chains, the location of the N-terminus varied between all chains (Fig. 5f). The L1 chains are in a similar conformations up to Val 21, but in the 1-20 region, the location and conformation of the N-termini are different. Varied conformations of the L1 N-termini likely enable hydrogen-bond formation between surrounding capsomers.

Fig. 5. L1 N-terminal hydrogen bonds predicted between neighboring capsomers.

A HPV asymmetric unit (full color, as in Fig. 1) with surrounding L1 proteins (pastel). Regions with N-terminal hydrogen bonds identified are bordered with rectangles. B–D N-terminal hydrogen bonds between different L1 chains are shown in blue, with shades corresponding to the hydrogen bond frequency determined with molecular dynamics (darker colors correspond to higher frequency, see color key). B The close up shows the interaction between chain F (red) and chain C’ (light green) from a neighboring capsomer. Hydrogen bonds are shown in a shade of blue corresponding to the MD H-bond frequency. C Close up of the interaction between chain E (orange) and chain E’ (light orange), with predicted hydrogen bonds in blue. D The interactions of chain C’ (light green) with chain D (yellow), and chain D’ (light yellow) are illustrated, as well as the interaction of chain D with chain D’. Hydrogen bonds are shown in a shade of blue corresponding to the MD hydrogen bond frequency. The magnitude of the cryoEM density is strongest in the regions with predicted hydrogen bonds compared to those regions without. E Chains A-F (colored purple, blue, green, yellow, orange, and red, respectively) are superimposed to show regions with different conformations, specifically the N and C terminus, the C-terminal arm extensions, and amino acids 83–91. F A zoomed in view shows the differences in the L1 N-terminus.

All-atom MD simulations of the HPV capsid in the absence of heparin support the presence of inter-capsomer hydrogen bonds predicted by cryoEM¹⁶ (Supplementary Fig. 11). Notably, the strong hydrogen bonding pattern observed between chains C’ and F is consistent with a β-sheet frequently spanning Leu 5 to Leu 13 (Supplementary Fig. 11a), and this finding is corroborated by secondary structure detection with STRIDE³⁷ (Supplementary Fig. 12a, b). Although the β-sheet is dynamic, with fluctuations in ideal strand geometry and individual backbone hydrogen bonds (Supplementary Fig. 12c–e), the two N-terminal regions remain persistently associated, including through a network of auxiliary hydrogen bonds that collectively reinforce the capsomer-capsomer interface (Supplementary Fig. 12f). Several hydrogen bonds were identified between the C- and N- termini of chains C and D, with Phe 455–Trp 4 being highly prevalent, but only three others exhibited frequency >50% (Supplementary Fig. 11b). Likewise, among hydrogen bonds identified between the C- and N- termini of chains E and E’, only two exhibited frequencies >50% (Supplementary Fig. 11c). Interestingly, a single yet prominent hydrogen bond was identified between the N-termini of chains D and D’ (Supplementary Fig. 11d). Because the reported frequencies reflect the simultaneous presence of hydrogen bonds across all L1 protein copies in the capsid, this interaction network is likely important for capsid stabilization. Further, these inter-capsomer contacts in the N-terminal region of L1 may have a role in capsid assembly.

Within the map, two cryoEM densities were observed at the inner surface of the capsid that could not be attributed to L1 density (Supplementary Fig. 13). The first unfilled density was located midway along the pore through the center of capsomers (pentavalent and hexavalent) adjacent to L1 amino acids 311-317. The second unfilled density, previously termed the “fishhook,” is located on the inner edge of capsomers, close to L1 amino acids 33–37, 457–460, and 474–476^14,32. Both unfilled densities have a distinctive protein-like appearance consistent with 2–3 amino acids at high resolution with side chain protrusions. The unknown densities are not present at a high contour level, indicating they may be asymmetric. These unknown, non-L1 densities have been visualized in ~3 Å resolution cryoEM structures of authentic papillomavirus and HPV16 quasivirus without heparin bound^14,32. Additionally, other non-L1, unfilled densities were visualized at the inner surface inside the capsid but were too noisy to be built (Supplementary Fig. 14). However, unmodeled C-terminal residues 481-503 are in the same location, which suggests the unfilled densities may be attributed to interactions between the HPV16 genome and the L1 C-terminus, which is known to interact with DNA³⁸. All-atom MD simulations of the full-length, empty capsid support this hypothesis, demonstrating that the space sampled by residues 481–503 in the absence of genome overlaps with the location of the unfilled densities along the capsid interior (Supplementary Fig. 15a–c). Results indicate that the C-termini of chains B and C are the most likely contributors, with their positions coinciding with the densities in more than 50% and 70%, respectively, of MD conformers. Density maps back-calculated from MD conformers of chains B and C show that the L1 C-terminus accounts for only 18% of the volume of the unfilled densities, suggesting that the remainder represents capsid-contacting DNA (Supplementary Fig. 15d).

Discussion

In this study, we solved a cryoEM structure of HPV bound to heparin at 1.9 Å resolution, which allowed for the identification of the heparin-binding site and investigation into the hydrogen bonds that are responsible for capsid assembly and stability. All-atom MD simulations of the capsid alone provided independent evidence for the location of heparin binding and the presence of inter-capsomer hydrogen bonds predicted by cryoEM.

There are several factors that contribute to the different quality of the heparin density compared to that of the capsid. Heparin is inherently flexible, which would cause the heparin density to diminish due to averaging during the reconstruction process. The specific heparin used in this study had a degree of polymerization ranging from 40 to 51, presenting a heterogeneous population of molecules that could lead to additional asymmetry and loss of ordered cryoEM density. During the analysis, we found that heparin bound to an average of three out of the five possible binding sites, even though the complex was made by incubating heparin with HPV at oversaturating conditions. This asymmetric heparin binding could indicate that steric occlusion is taking place, where heparin molecules bound at a few of the possible sites around the pentavalent capsomer subsequently blocked additional heparin molecules from binding. This steric occlusion seems likely due to the length of the heparin molecules and lack of heparin saturation. The heparin bound only around the fivefold is likely due to the quasi-equivalence of the HPV icosahedral capsid, such that the canyons between hexavalent capsomers differ from canyons between pentavalent and hexavalent capsomers¹⁶.

Due to the heterogeneity, asymmetry, and flexibility of the heparin, the resulting heparin density was interpreted by fitting rather than building a model. The interactions identified are chemically appropriate since residues identified in the potential heparin binding site are either positively charged or polar, which is consistent with coordinating a highly negatively charged molecule. Side chains in this region are pointing towards the heparin density, and local conformational changes to the site where heparin binds were identified. MD simulations show strong chloride occupancy in the same region as the heparin density and suggest that Lys 59, also implicated in heparin binding by cryoEM, may be a key residue for contacting sulfates. Pronounced chloride occupancy during MD has been shown to correctly predict capsid interactions with charged molecules like nucleotides and RNA, and appears in this case to foretell the capsid’s preferred interface with HSPG receptors^39–41. There are currently no antivirals against HPV, and the identified heparin-binding site could be used to identify inhibitors to block heparin binding.

In Cerqueira et al. 2013, the authors found that antibody H16.B20 bound poorly to HPV capsids but bound more strongly to capsids incubated with heparin²². The binding site of H16.B20 includes L1 residues 396-415, part of the C-terminal arm extension between capsomers. This epitope stretches from the base of one capsomer up towards the middle of a neighboring capsomer, with the majority of the residues located between the two capsomers. The heparin-binding site is located above and slightly to the side of the L1 396-415 region, which would make antibody binding at a heparin-occupied pentavalent location difficult. However, the H16.B20 epitope is also present between hexavalent capsomers, where heparin binding was not observed. Heparin binding decreases flexibility between capsomers, which could stabilize the H16.B20 epitope and enhance antibody binding on the loop regions in between capsomers. It is also possible that the H16.B20 antibody might outcompete heparin for binding.

In Feng et al. 2024, the authors reported increased stability between capsomers, enlargement of the HPV capsid, and softening of the capsid due to heparin binding²⁴. The authors found that this structural activation only occurred when using heparin with a degree of polymerization greater than 40. Here we used heparin with a degree of polymerization of 40-51, meaning that structural activation noted by Feng et al. should have been possible. Our study corroborates the finding that heparin increases stabilization of the capsid as our analysis showed reduced flexibility between capsomers compared to the cryoEM structure of HPV alone¹⁴. However, capsid enlargement was not observed. The cryoEM structure of heparin bound to HPV had the same diameter range as the structure of HPV alone¹⁴. When analyzing cryoEM maps, the softening of the capsid, as described in AFM as the ability of the capsid to deform²⁴, cannot be measured. However, the softening reported by AFM might correspond to capsid regions with increased flexibility. The local resolution mapping of HPV bound to heparin and HPV alone did indicate some slight differences in resolution at the fivefold. The pentavalent capsomers of the complexed map may be more flexible than the hexavalent capsomers, compared to the HPV16 map alone. Although the resolution of the maps was similar, HPV heparin at 3.0 Å and HPV alone at 3.1 Å, there is a risk of overinterpreting minor differences, especially because icosahedral averaging can introduce artifacts at the fivefold. The minor differences seen at the fivefold are not compelling with cryoEM, but flexibility changes may have been more obviously detected with the biophysical approach of AFM²⁴. The Feng et al. work identified residues 442 and 443 as essential for structural activation of the capsid due to heparin binding and important for infectivity²⁴. Here, we show that the heparin density illustrated in Supplementary Fig. 8 is adjacent to these residues (442/443) and is likely to interact with heparin, agreeing with the previous findings. However, the weaker heparin cryoEM density indicates flexibility or asymmetric binding that makes certain interpretations more difficult.

The disulfide bond between HPV16 L1 Cys 175 and Cys 428 has been established as an essential bond for capsid maturation⁴². The results from this study indicate that some of the disulfide bonds are present randomly throughout the virus capsid. This finding is consistent with previous work that shows disulfide bonds are sensitive to beam damage^36,43. Our analysis of the first five frames of the data indicates that the beam damage occurs quickly, within the first frames of data collection. Both models with and without disulfide bonds had acceptable geometry. However, any further interpretation is beyond the limits of the resolution.

In Chen et al. 2000, HVP16 L1 N-terminal deletions of 10 amino acids resulted in HPV assembly defects, and a 10-residue deletion was rescued with the addition of one glycine²¹. Deletion of the 10 N-terminal amino acids disrupts the beta sheet formed between L1 chains C’ and F and disrupts four H-bonds between L1 chains C’ and D. It was previously proposed by Modis et al. 2002 that the N-terminus of L1 plugs gaps in the virus capsid and stabilizes the capsid²⁰. The hydrogen bonds that form between the N-terminal regions of certain L1 proteins are likely to stabilize the HPV capsid. Modis et al. proposed N-terminal beta-hairpins, due to beta-hairpins found in polyomavirus capsids. However, L1 N-terminal beta-hairpins were not observed. Instead, two N-terminal regions from L1 chains C and F form a beta sheet, and hydrogen bonds were observed between other L1 N-terminal regions. Similar to Modis et al., we propose that the N-terminal region of L1 and its interactions with other N-terminal L1 regions are important for capsid assembly and capsid stability, as disruption of these bonds results in HPV16 assembly defects.

Although the exact mechanisms of HPV entry are unknown, a conformational change to the capsid occurs upon attachment to the host cell. A widely accepted model is that binding to a heparan sulfate proteoglycan causes this change. We found that heparin binding induces local, fivefold conformational changes and stabilizes the arm connections between capsomers in the capsid. These alterations may prime the capsid for the next steps of entry that may include additional interactions with the host cell and furin cleavage of L2 before endocytosis. The N-terminal hydrogen bonds likely both play an important role in capsid assembly and stability.

Methods

Sample preparation

HPV16 quasivirus containing L1 and L2 proteins with an encapsidated cottontail rabbit papillomavirus (CRPV) genome with the SV40 origin of replication was prepared as described previously⁴⁴. In brief, HPV16 sheLL plasmid (kindly provided by John Schiller, NIH, Addgene plasmid # 37320; http://n2t.net/addgene:37320; RRID:Addgene_37320) was transfected together with CRPV/SV40ori DNA into 293TT cells (ATCC CRL-3467)^12,45,46. After 48 h, the 293TT cells were harvested, lysed at high cell density with 0.22% Brij58, and HPV16 was allowed to mature for 24 h at 37 °C. The lysate was chilled, NaCl was added to achieve a final concentration of 0.85 M, and incubated on ice for 10 min. The lysate was clarified with centrifugation at 2000 x g, and the supernatant was collected. The pellet was resuspended in 1 M NaCl and 0.2 M Tris (pH 7.4), centrifuged, and the supernatant collected. A stepwise CsCl gradient was prepared for HPV16 purification, and the sample was centrifuged for 3 h at 110,862.3 x g with a Beckman Colter SW 40Ti swinging bucket rotor. Two bands were observed, and the lower band was collected. The sample was pelleted with a sucrose cushion by centrifuging for 1 h at 87,501.3 x g and resuspended in 1x PBS (pH 7.4). The concentrated HPV16 quasivirus particles were applied to 300-mesh carbon-coated copper grids and stained with 2% phosphotungstic acid. The sample was analyzed for integrity and concentration on an FEI Tecnai G2 Spirit BioTwin transmission electron microscope. The sample was further concentrated using the previous sucrose cushion technique. The sample was incubated with heparin in excess, assuming 60 binding sites per capsid, for one hour at 37 °C. Samples were vitrified using a Vitrobot on 2/1 300-mesh copper Quantifoil grids using 3.5 uL of complex.

Data collection

Micrographs were collected on the Titan Krios at the Huck Institutes at Penn State University. An atlas image was taken at 165 × magnification, and suitable areas were selected for imaging on the FEI Falcon 4 direct electron detector. The microscope was operated at 300 kV with a 70 µm condenser aperture and a 100 µm objective aperture. For the first data collection, magnification was set at 59,000x, yielding a calibrated pixel size of 1.1 Å. One image was acquired at the center of each selected 2-µm-diameter hole of the grid for an overall collection of 3032 movies. The total dose per exposure was set to 40 e⁻/Ų. For the second data collection, a magnification of 120,000x, with a calibrated pixel size of 0.54 Å was used to collect 13,945 movies. The total exposure dose was 40 e⁻/Ų.

Refinement approach

Data was processed using cryoSPARC, RELION, pyem, and ISECC^14,47,48. Movies underwent patch motion correction and patch CTF estimation in cryoSPARC. Particles were manually picked from approximately 100 micrographs, then used to generate templates for autopicking. Particles were picked, extracted, and underwent 2D classification in cryoSPARC. Particles from poor-quality classes were rejected. The first dataset had a final particle number of 74,966 particles, and the second dataset had a final particle number of 44,159 particles. Final particle images were obtained via local motion correction at box sizes of 800 and 1600, respectively. Particles underwent icosahedral refinement with Local and Global CTF refinement and Ewald Sphere correction, for a final resolution of 2.98 Å and 2.76 Å, respectively.

Subparticle extraction

Following established protocol, subparticles were extracted using a program called Icosahedral Subvolume Extraction and Correlated Classification (ISECC)¹⁴. Briefly, icosahedrally refined particle data were converted to RELION star format using pyem. For the first dataset, pentavalent and hexavalent subparticles were extracted using ISECC_star_subparticle_extract with a box size of 400 and vectors of 0 142.14 230 (pentavalent) and 44.103 110.038 249.419 (hexavalent). For the second dataset, pentavalent and hexavalent subparticles were extracted using ISECC_star_subparticle_extract with a box size of 760 and vectors of 0 119.892 194 (pentavalent) and 36.202 36.129 221.861 (hexavalent).

Subparticle refinement

Subparticles were reimported to cryoSPARC for subparticle refinement. For the pentavalent subparticle, a mask was created corresponding to the pentavalent capsomer and five surrounding hexavalent capsomers, using USCF Chimera and cryoSPARC^49,50. In cryoSPARC, this collection of capsomers underwent iterative local refinement, global CTF refinement, and local CTF refinement until the resolution stopped improving. Then, each of the five hexavalent capsomers underwent masked local refinement, followed by image subtraction. Once all five hexavalent capsomers were subtracted, the remaining pentavalent capsomer underwent iterative masked local refinement, global CTF refinement, and local CTF refinement until the resolution stopped improving. This process was repeated for the hexavalent capsomer, using an initial subparticle centered on a hexavalent capsomer with the six surrounding capsomers. The final resolution of the first dataset of pentavalent and hexavalent capsomers was 2.2 Å and the resolution of the second dataset of pentavalent and hexavalent capsomers were 1.90 Å and 1.92 Å respectively. In Chimera, the individual pentavalent and hexavalent capsomer maps were fit into a 2.42 Å hexavalent subparticle starfish map. A high-resolution composite map of seven subparticles was generated using the command ‘vop maximum‘. The heparin subparticles from the first data collection were extracted as described above, with a vector of 44.1 110.04 249.4 and a box size of 250. The subparticles were refined in RELION. Iterative 3D classification and 3D refinement jobs were run, resulting in a 3.42 Å map with distinct heparin density. A single heparin molecule isolated from PDB 5W1O was fit into heparin density³⁵.

Model building and refinement

The pixel size of the map was adjusted by fitting an HPV pentamer structure solved with X-ray diffraction, PDB 2R5H, into the pentavalent subparticle map⁵¹. The pixel size of the map was adjusted repeatedly, and the model was fit into the map each time. The number of atoms present outside the map was recorded. The fewest number of atoms outside the map occurred at a map pixel size of 0.517, and this is the pixel size used throughout model building and analysis. The model of the asymmetric unit was initiated using PDB 7KZF, with manual and automated refinement in ISOLDE and Phenix^14,52–54. The model was built into the high-resolution composite map, and the flexible regions were built into an unsharpened 2.4 Å  hexavalent subparticle starfish map. Water molecules and magnesium ions were built into the map with Segger Segmentation-guided Water and Ion Modeling (SWIM)^55,56. Assigned water molecules and magnesium ions were visually inspected and approved or rejected.

Flexibility analysis and local resolution

Flexibility analysis of the first data collection was conducted using ISECC as previously described in Goetschius et al. 2021 and Hartmann et al. 2021^14,32. Because this analysis requires RELION star file input, the pentavalent and hexavalent capsomers were locally refined to 3.1 Å in RELION. Briefly, the ISECC_local_motions script used the locally refined star files to identify the subparticle origins and compare them to the idealized icosahedral origin. Then the deviation of each capsomer from the idealized icosahedral location was determined. Qualified particles were identified as those with a hexavalent center within 25 Å x,y distance of the particle center, and neighboring capsomers were identified. The ratio of observed to expected distances was determined by dividing the locally refined distance by the icosahedrally defined distance, resulting in a ratio with no units. For the capsid diameter deviation analysis, the distance between pentavalent capsomers on opposite ends of the capsid within +/- 5% of the central plane was recorded and compared to the ideal icosahedral distance. The ratio of observed to expected distances was plotted. Local resolution jobs were run in cryoSPARC, with an FSC threshold of 0.5. The icosahedral HPV-heparin map was compared to the icosahedral maps of HPV alone and MmuPV1^14,32. The HPV data from Goetschius et al. 2021 was reprocessed to achieve a higher icosahedral resolution¹⁴. The icosahedral refinement approach was the same as described for the HPV-heparin icosahedral refinement and resulted in a resolution of 3.06 Å. To compare maps of similar resolution, the 2.98 Å resolution HPV-heparin map was used. The local resolution jobs were run in cryoSPARC, and icosahedral half maps were inputted and an FSC threshold of 0.5 was used to determine local resolution for each map.

Molecular dynamics

The empty L1 capsid is known experimentally to represent a stable assembly, suitable as a model for studying emergent particle dynamics⁵⁷. MD simulation of the empty HPV type 16 capsid was carried out on the Frontera supercomputer. The complete model of the intact capsid, composed of 360 L1 proteins, was developed based on the previously reported cryoEM structure (PDB 7KZF, Goetschius et al. 2021), as described in Antolínez et al. 2025^14,16. All 360 possible Cys175-Cys428 disulfide bonds were linked. The capsid model was immersed in a truncated octahedron of explicit solvent containing 150 mM NaCl for a total system size of ~16.8 million atoms. System composition is summarized in Supplementary Table 1. The simulation was performed with NAMD3.0b5⁵⁸ using the AMBERff-in-NAMD⁵⁹ implementation of the ff14SB forcefield⁶⁰ with TIP3P⁶¹, as described in Antolínez et al. 2025¹⁶. A total of 710 ns of conformational sampling for the intact capsid was considered here for analysis.

Hydrogen bond, secondary structure, and density contact frequencies were determined using VMD 1.9.4⁶². Contact frequency data was aggregated over all 60 quasi-equivalent chains to yield cumulative sampling of 60 $\times$ 710 $=$ 42.6 μs. The Jensen-Shannon distance (JSD) was used as a metric to assess convergence of these MD-derived data. The JSD serves as a measure of similarity between two distributions, where a distance of zero signifies identical distributions and unity signals maximally different distributions. The JSD between two distributions $P$ and $Q$ is defined in Eq. (1)

$$\mathrm{JSD}\left(P\mid \mid Q\right)=\sqrt{\frac{\left(D\left(P\mid \mid M\right)+D\left(Q\mid \mid M\right)\right)}{2}}$$ 1

where M represents the mixture distribution, $M=\left(P+Q\right)/2$, and D corresponds to the Kullback-Leibler divergence, defined in Eq. (2):

$$D\left(P\mid \mid Q\right)={\int }_{-\infty }^{\infty }p\left(x\right)\log \left(\frac{p\left(x\right)}{q\left(x\right)}\right)dx$$ 2

Pairwise comparison of JSD values based on varying amounts of sampling demonstrated a high degree of similarity between all distributions (all values < 0.14, Supplementary Fig. 16), justifying the use of the complete dataset.

Chloride ion occupancies around pentavalent starfish were averaged over all 12 subunit copies in the capsid to yield cumulative sampling of 12 $\times$ 710 $=$ 8.52 µs. Occupancy maps were calculated using the volmap plugin in VMD with a grid spacing of 1.0 Å. Potentials of mean force, relating ion occupancy to the Gibbs free energy of binding (∆G), were calculated using Eq. (3)⁶³, where ${\rho }_0=0.0015$ was used as the bulk occupancy of chloride, determined from the bulk solvent region in the center of the capsid interior⁶³:

$$W\left(\mathbf{r}\right)=-k_{B}T\ln \frac{\rho \left(\mathbf{r}\right)}{{\rho }_0}$$ 3

Back-calculated C-terminal densities were averaged over all 20 copies of the threefold pores of the capsid to yield cumulative sampling of 20 $\times$ 710 $=$ 14.2 μs. Simulated density maps were calculated using the volmap plugin in VMD with a grid spacing of 1.1 Å. This grid spacing was chosen for consistency with the experimental density map derived by cryoEM. Overlap between the simulated and unfilled experimental densities was quantified in ChimeraX 1.9 following processing to facilitate comparison: both maps were resampled onto a common grid and normalized by scaling by $p^{\prime }=\left(p-\mu \right)/\sigma$.

First five frames reconstruction

To save computational resources, the first data collection at 59,000x magnification was selected for reprocessing. Patch motion correction was run on the entire dataset of 3032 movies, but only the first five frames of each movie were processed. The refinement approach described above was used to process the dataset, resulting in an icosahedral resolution of 2.98 Å. Subparticles were extracted and refined as described above, resulting in pentavalent and hexavalent subparticles at 2.97 Å and 2.96 Å, respectively.

Hydrogen bonds and unknown densities

Potential hydrogen bonds were identified from the model built from the 1.9 Å map. Six models of the asymmetric unit were fit into the icosahedral map to investigate interactions between capsomers. In ChimeraX, the command “hbonds” was used to identify potential hydrogen bonds based on the atom type and the distance between atoms^50,64. The 1.9 Å map and the model of the asymmetric unit were opened in ChimeraX at the same time. Using the ColorZone tool, the cryoEM map density within 2 Å of the model was split from the rest of the cryoEM density. The remainder of the cryoEM density could not be attributed to L1 and was considered unknown density.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

C.H.L., N.D.C., and S.L.H. conceptualized the cryoEM experiment. S.A.B. and N.D.C. provided reagents. C.H.L. prepared the samples. C.H.L. vitrified the samples. C.M.B. collected the data. D.J.G. planned and optimized the refinement approach. D.J.G. and C.H.L. processed the data. C.H.L. built the model. C.H.L. interpreted the maps and model. C.H.L., E.P.O., and S.L.H. analyzed cryoEM results. J.A.H.-P. conceptualized the MD simulation. S.A. performed the MD simulation and trajectory processing. S.A. and J.A.H.-P. analyzed MD simulation results. C.H.L., N.D.C., S.A., J.A.H.-P., and S.L.H. wrote the paper.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This research was supported in part by the J. Gittlen Memorial Golf Tournament and the Pennsylvania Department of Health CURE funds. Research reported in this publication was supported by the Office of the Director, National Institutes of Health, under Award Number 1 R01 Al134910-01A1 (SH) and IMVTP, T32 AI83196 (CHL). MD simulation work was supported by NIH administrative supplement 3P20-GM-104316-09S1 to JAH-P, the Delaware Advanced Research Workforce and Innovation Network (DARWIN, NSF award OAC-1919839), and the BioStore resource (NIH awards P20GM103446 and S10OD028725). Supercomputer time on Frontera at the Texas Advanced Computing Center at the University of Texas at Austin was provided by allocation MCB-23072. Supercomputer time on Anvil at the Rosen Center for Advanced Computing at Purdue University was provided by allocation BIO240029 from the Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support program (ACCESS, NSF awards 2138259, 2138286, 2138307, 2137603, and 2138296).

Data availability

The HPV structures have been deposited in the EM database under accession code EMD-49960 (1.9 Å recombined map), EMD-70163 (pentavalent capsomer), EMD-70164 (hexavalent capsomer), EMD-70167 (hexavalent starfish consensus map), and EMD-70186 (icosahedral map). Coordinates for the model of the HPV16 L1 asymmetric unit have been deposited in the protein database under accession code PDB 9NZU. MD simulation files for the HPV16 L1 capsid are provided on Zenodo [https://doi.org/10.5281/zenodo.18842850]. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77019-9.