Local Stabilization of Subunit–Subunit Contacts Causes Global Destabilization of Hepatitis B Virus Capsids

Abstract

Development of antiviral molecules that bind virion is a strategy that remains in its infancy, and the details of their mechanisms are poorly understood. Here we investigate the behavior of DBT1, a dibenzothiazepine that specifically interacts with the capsid protein of hepatitis B virus (HBV). We found that DBT1 stabilizes protein–protein interaction, accelerates capsid assembly, and can induce formation of aberrant particles. Paradoxically, DBT1 can cause preformed capsids to dissociate. These activities may lead to (i) assembly of empty and defective capsids, inhibiting formation of new virus, and (ii) disruption of mature viruses, which are metastable, to inhibit new infection. Using cryo-electron microscopy, we observed that DBT1 led to asymmetric capsids where well-defined DBT1 density was bound at all intersubunit contacts. These results suggest that DBT1 can support assembly by increasing buried surface area but induce disassembly of metastable capsids by favoring asymmetry to induce structural defects.

Graphical Abstract

Hepatitis B virus (HBV) is a global health problem. In spite of an effective vaccine, more than 292 million people have chronic HBV, which can lead to cirrhosis, liver failure, and hepatocellular carcinoma and contributes to more than 880 000 deaths each year.¹ Vaccine use drastically changes the demographics of HBV² but does not act on chronic infection. Nucleotide and nucleoside analogs are effective at suppressing viral load and improve liver health but do not cure chronic infection; once started, treatment may have to continue indefinitely due to the risk of viral flares.³ Thus, there is a clear unmet medical need for new antiviral agents or treatment strategies that achieve a functional cure.⁴

HBV is among the smallest human viruses. It is an enveloped virus with an icosahedral core containing a genome of relaxed circular dsDNA (rcDNA). The viral core protein (Cp) plays many roles in the viral lifecycle and is now being explored as a target for direct-acting antiviral agents. Once HBV enters a cell, Cp mediates transport to nuclear pores by importin proteins where cores release viral DNA into the nucleus.⁵ Translation of viral transcripts yield Cp in the cytoplasm, where it self-assembles to either empty particles (about 90% of Cp⁶) or RNA-filled cores that contain a pregenomic RNA transcript (pgRNA) bound to viral reverse transcriptase.⁷ The RNA-filled core becomes a metabolic compartment for synthesis of rcDNA from the linear pgRNA template. For the resulting DNA-filled capsids, Cp can mediate transport to the nucleus, maintaining chronic infection, or alternatively Cp interacts with the viral envelope protein to initiate secretion from cells. These same fates are also available to empty capsids.

Many elements of Cp structure and assembly are well-defined.⁸ Cp has a 149-residue dimeric assembly domain, with an intradimer interface formed by a large four helix bundle. About 1200 Ų of largely hydrophobic surface is buried at this interface.⁹ Because the predominant form of HBV capsid has T = 4 icosahedral symmetry (120 dimers), the interdimer interface must adopt four different but quasi-equivalent geometries. Despite the large buried surface, the interdimer contact energy is notably weak, about −3.1 kcal/mol at physiological ionic strength and increasing with higher salt concentration.⁹ Because subunits are each tetravalent, this weak energy corresponds to a pseudo-critical concentration in the micromolar range. Since interaction energy is weak, errors during assembly can be thermodynamically edited out. Incomplete particles, overgrown particles, or aberrant assemblies can be trapped at higher ionic strengths but can relax to the 120-dimer icosahedron over time.^10–13

The biophysics of assembly show that assembly works well in a limited range of conditions.¹⁴ Pushing the reaction outside of that range using small molecules should be destructive to the virus. A Cp assembly activating molecule may have three obvious activities.⁸ (i) A molecule that stimulates Cp to adopt an assembly active state can lead to overnucleation or premature nucleation, which for HBV could mean assembly without the reverse transcriptase–pgRNA complex. (ii) A molecule that stabilizes protein–protein interactions can also stabilize mistakes and lead to defective particles. (iii) A molecule that interferes with the normal protein–protein interaction geometry can lead to aberrant structures. These activities involve inducing structural changes and therefore should be stimulated by core protein allosteric modulators (CpAMs). A CpAM that modulates assembly may also alter Cp’s ability to perform other activities.

Multiple CpAM chemotypes have been identified. Phenyl-propenamides (PPAs) and heteroaryldihydropyrimidines (HAPs) were discovered in naive screens and later shown to act via the Cp.^15,16 In both cases, these molecules are assembly agonists that accelerate kinetics and stabilize Cp–Cp interactions, though PPAs always lead to normal capsids while HAPs can create strikingly aberrant Cp polymers.^17,18 Sulfamoyl benzamides (SBAs) show similar properties to PPAs, forming normal capsids.¹⁹ A likely mechanism of action for these CpAMs is that they spontaneously nucleate assembly in the absence of viral pgRNA–reverse transcriptase, leading to empty and defective particles (see Feld et al.¹⁵). Each of these CpAM chemotypes binds at a similar location at the subunit interfaces but can produce very distinct outcomes in the products seen in assembly reactions. Even within a single chemotype, the products that form vary depending on substituents, stoichiometry, and solution conditions.²⁰

In a broad sense, the proper function of HBV capsid requires the core protein dimer subunits to interact with each other in a specific manner. Molecules that perturb the association energy between subunits or the spatial relationship between two adjacent dimer subunits will cause abnormal capsid function, with a proven antiviral affect. In this work, we study how a novel chemical scaffold, dibenzothiazepine (exemplified here by the molecule DBT1), can paradoxically induce HBV capsid assembly and disassembly, can lead to both normal and aberrant capsid assembly, and has capsid binding properties that are quite different from what is seen in other CpAMs.

RESULTS

DBT Locally Stabilizes Per-Contact Subunit Interfaces.

Though several hundred DBT variants have been identified,²¹ their physical chemistry and mechanisms have not been well characterized. Here we examine the behavior of DBT1 (Figure 1a, inset). First, we ask whether DBT1 modifies assembly kinetics. Assembly kinetics were monitored using 90° light scattering, where the scattering signal is proportional to the size and number of assembled products (Figure 1a). In solution conditions where capsid assembly is normally undetectable, the DBT molecule promotes assembly in a robust and dose dependent manner. Once the reaction approaches a steady state equilibrium, the fraction of assembled product can be estimated from size exclusion chromatography (Figure 1b). This comparison addresses the question of whether DBT1 affects thermodynamics as well as kinetics. Here we see that DBT promotes assembly in a dose dependent manner, shifting the reaction equilibrium away from free subunit and toward formation of larger structures. In the context of an assembly reaction model, these observations are consistent with a strengthening of the per-contact energy, ΔG_cont, of subunit–subunit association. This may be accomplished by inducing Cp to enter an assembly active transition state or by stabilizing Cp–Cp interactions or both. To quantify the effect of DBT1, we use the ratio of capsid to dimer, an equilibrium constant, to determine the change in ΔG_contact as a function of DBT1 concentration. From these data, we observe that DBT1 strengthens association energy from −3.1 kcal/mol in 150 mM NaCl to asymptotically approach −4.4 kcal/mol (Figure 1b, inset). The apparent equilibrium constant for induced assembly is 0.59 μM, indicating that DBT1 reversibly binds capsid during assembly.

Figure 1.

DBT accelerates assembly and activates misassembly. (a) Reactions of 5 μM Cp149 assembled in 150 mM NaCl (final concentrations) were monitored by 90° light scattering. The increase in scattering signal that appears with increasing DBT1 concentration (structure inset) indicates the formation of capsid and aberrant structures. In the absence of DBT1 (black line), assembly is undetectable at these conditions. (b) Size exclusion chromatography (SEC) confirms assembly and quantifies assembled products and free Cp149 dimer. Both methods indicate that DBT1 increases assembly relative to the control (absence of DBT1, black). The chromatograph is in units of absorbance. The inset in panel b fits the association energy per subunit contact (ΔG_cont) versus [DBT1] to a hyperbolic binding curve. The reference value for Cp149 without DBT1 (white circle) was previously determined.⁹ (c) Reaction products visualized by TEM correlate with light scattering. Over time, assembly products rearrange to form increasingly aberrant structures.

By electron microscopy, we observed the dose-dependent appearance of larger non-capsid structures (Figure 1c). Notably, even at substoichiometric DBT1 concentrations (1.25 μM), some particles were distinctly aberrant. This suggests that a defect in the growth of those particles occurred early during assembly, perhaps at nucleation. At high DBT1 concentrations, a larger number of progressively larger misshapen particles appeared; nonetheless, many particles with wild-type morphology appear simultaneously. Concentration-dependent misassembly is also characteristic of HAPs, where substoichiometric concentrations lead to normal capsids and higher concentrations are required to produce larger, sheet-like complexes.²² A distinctive activity of DBT1 is the slow rearrangement of reaction products with time. At the scale of several days, the fraction of icosahedral particles decreases, while the fraction of large polydisperse structures increases. This maturation effect is perhaps most striking with the appearance of cylindrical structures, which appear at high DBT1 concentrations and long times. This is significant because it suggests that the initial assembly products are not actually at thermodynamic minima, but rather are kinetic traps. At early times, more particles are icosahedral; at longer times, particles become more aberrant. This indicates that the kinetic barrier to form spherical particles is relatively low.

DBT1 Globally Destabilizes Capsids.

We hypothesize that rearrangement of smaller, spherical structures into larger, preferentially cylindrical structures occurs by transiently releasing subunits. To test for the appearance of free subunit, we added CpAMs to a homogeneous population of preassembled capsids and examined the complexes by native agarose gel electrophoresis (Figure 2). Native agarose gel electrophoresis (Figure 2a) can resolve subtle differences in the size of assembly products (>3–4 MDa) and also resolve free dimer (35 kDa). Electrophoresis also places capsids into conditions where they are metastable and could never spontaneously assemble to favor transient dissociation. By separating dimer and capsid (or other larger oligomer), we also prevent reassociation, allowing us to visualize the transiently released subunits.

Figure 2.

DBT induces capsid destabilization. (a) Native agarose gel electrophoresis can resolve both intact capsids and free subunit. We use an assembly incompetent mutant, Cp149-Y132A, as a control for free dimer. Preformed capsids treated with a 20 μM DBT1 (4:1 molar ratio of DBT to Cp) are quantitatively destabilized and comigrate with free subunit. In contrast, two other capsid-directed molecules (SBA, HAP) show very different behaviors. SBA-treated capsid migrates with untreated capsid. The band for HAP-treated capsids smear broadly to indicate the presence of products that are capsid size or larger. (b) The transition from capsid to subunit is concentration dependent, with the capsids fully dissociated only at a molar excess of DBT1 (here 5 μM subunit results in 10 μM of subunit interfaces). (c–f) The reactions shown on the gel in panel a are visualized by TEM in panels. (c) Untreated capsids provide a reference point. (d) In a nonperturbing environment, most capsids treated with DBT appear normal. Some capsids displayed what may be broken regions or holes. (e) SBA treated capsids look identical to untreated capsids. (f) HAP treated capsids become disrupted, forming larger structures, including many particles that are capsid sized but appear distorted. The presence of the larger particles suggests a process of disruption followed by reassembly.

We chose three molecules (DBT1, SBA0013, and HAP12) to represent three of the CpAM chemotypes (DBT, SBA, and HAP). DBT caused preformed capsids to comigrate with free subunit, indicating that under these conditions the molecule induces complete disassembly. In a titration of capsid by DBT1, we found that quantitative dissociation was observed at 20 μM DBT1 (with 5 μM capsid), while at 10 μM DBT1 little dissociation was observed. From the assembly based titration (Figure 1b), this suggests that we need almost every site filled (94% filled at 10 μM versus 97% filled at 20 μM, based on the simple binding model of Figure 1b, inset) to drive dissociation. Upon inspection by EM (Figure 2d), we found that under equilibrium conditions the protein remains largely in capsid form. However, reactions eluting slowly over size exclusion chromatography recapitulate the dose-dependent disassembly seen by gel electrophoresis (Figure S1b). This apparent discrepancy suggests that dimers were only transiently released under equilibrium conditions. Only under the nonequilibrium conditions of gel filtration did the broken capsids unravel into free subunit. We interpret the data showing dissociation under nonequilibrium conditions as evidence of DBT1-induced destabilization. In comparison, the SBA-treated capsids comigrate with apo-capsid and look indistinguishable from the control by electron microscopy (Figure 2e). SBAs also drive assembly and yield morphologically normal capsids; the persistence of capsids under the equilibrium-perturbing conditions of electrophoretic separation indicates that few dimers are released and suggests that drug-bound capsid is the lowest energy state. Capsids treated by HAPs also change their migration, leading to smearing of the capsid band to progressively slower migrating species. This suggests that Cp dimers are released and reassemble into large structures reminiscent of what would be expected for HAP-directed misassembly starting from free subunit (Figure 2f, S1c). Each of the three molecules has a distinct behavior: SBA does not disrupt the capsids, HAP disrupts the capsids and reassembles the subunits into larger products, and DBT causes free subunit to be ejected, poising the capsid for disassembly.

Cryo-EM Structure.

To gain insight into the mechanism of the drug-induced stabilization and destabilization, we carried out cryo-EM reconstructions of preformed capsids treated with DBT. Initially we imposed icosahedral symmetry during image reconstruction (Figure 3a) as usually this is an accurate assumption and enhances signal-to-noise ratio by 60-fold on average. Quasi-equivalence explains how chemically identical subunits fit into structurally distinct but similar locations in an icosahedrally symmetric structure. For HBV, with T = 4 quasi-equivalence, the 240 interfaces have four unique conformations, labeled A, B, C, or D. In our icosahedral reconstruction, we saw density that could be attributed to DBT1 with the strongest density in the A and D sites. However, the density for the drug was ill-defined (Figure S2a). We speculated that this blurring was due to imposing icosahedral averaging on a conformational ensemble.

Figure 3.

Cryo-EM reconstruction reveals density for the DBT molecule. (a) A T = 4 HBV capsid shows the arrangement of AB dimers (green and orange, respectively) and CD dimers (purple and blue) and results in four “quasi-equivalent” protein interfaces (B–C, C–D, D–B, A–A). Each interface is comprised of a “base” subunit, which is “capped” by helix-5 of the adjacent subunit. We refer to each interface by the letter of the “base” subunit (i.e., B–C, a molecule binding the B–C interface binds to the B site). A reconstruction of the capsid–DBT complex that employed icosahedral averaging had insufficient detail to evaluate DBT binding at every interface (Figure S2). (b) Asymmetric reconstructions that focus on localized hexameric and pentameric regions of the capsid surface provide the clarity needed to identify the DBT molecule in all four quasi-equivalent sites (A, B, C, D) without any ambiguity. Protein density (gray) and DBT1 density (colored as in panel a) for each class of dimer is shown. (c, d) The DBT1 molecule binds each site in a similar pose for all sites in either the pentamer focused reconstruction (c) or the hexamer reconstruction (d). The DBT1 wraps around helix 5 of the adjacent capping subunit with the sulfonamide group solvent-exposed toward the capsid lumen.

For this reason, we revisited the reconstruction process without assuming symmetry; performing asymmetric reconstruction focused on either a 5-fold or a quasi-6-fold region of the capsid surface. DBT1 structural detail was enhanced and its binding pose became unambiguous (Figure S2b). The CpAM density appeared as a bent plane that was consistent with the DBT1 structure (Figures 1a, inset, and 3b,c,d). The DBT1 density fits into the CpAM pocket, wrapping around helix 5 from the neighboring “capping” subunit. We find a DBT1 molecule in every quasi-equivalent protein–protein interface. This presents a very different distribution than seen with previous CpAM structures, where only B and C of the four quasi-equivalent sites were filled.^23–25 We note that the local environment of each of the four quasi-equivalent sites varies substantially (Figure S3). In capsids without a CpAM, the available volume at the A and D interfaces is much more restricted than that in the B and C sites. However, DBT1 has the flexibility to fit into the spatially restricted sites, and consequently it has a direct effect on every dimer–dimer interaction in the capsid.

Bound CpAM affects capsid quaternary structure, typically expanding the capsid diameter by about 5%. For structural comparison, we use capsid bound to HAP-TAMRA (6BVF²⁵), in part because the binding stoichiometry was verified via the TAMRA fluorophore; also the 6BVF structure was determined by cryo-EM and is not constrained by crystal packing contacts. In 6BVF, HAP-TAMRA fills the B and C sites resulting in flattening of the quasi-6-fold vertex with attendant flexing of the 5-fold (where HAPs do not bind). These concerted motions make the capsid appear more faceted than a typical apo-capsid (3J2V). In comparison, the icosahedrally averaged DBT1 capsid appears to have local distortion that is intermediate between the apo-capsid and the HAP-TAMRA capsid. The presence of DBT1 at the 5-fold limits the flexing of the vertex, and similarly, the loose fits of the DBT1 in the two B and two C sites do not require the quasi-6-fold to be flat. However, the fact that focused reconstruction was required to visualize details of the DBT1 structure indicates that to maintain local structure in most vertices, capsids were asymmetric and may include a subset of capsomers that were locally deformed to absorb defects over the rest of the capsid.

To investigate the difference in quasi-equivalent specificity between DBT1 and HAPs, we docked the HAP molecule NVR10–001E2 from a relatively high resolution structure²⁶ into the DBT1 structure (Figure 4c). We find there are major clashes that would not permit binding without large scale movements of the capping dimer. Residue V124 on helix-5 of the adjacent capping subunit is the primary clash that we observe. In contrast, DBT1 avoids such a clash because of its flexible chemical structure, twisting around the capping helix to avoid V124. These clashes highlight the importance of molecular scaffold as a determinant of site specificity. The V124 clash is with the core of the HAP molecule, not with a substituent. Thus, it seems unlikely that any HAP derivative could be synthesized that binds in the same manner as DBT. Even though both molecules bind the same protein interface, the specifics of their molecular structure matter and lead to different structural properties and assembly phenotypes.

Figure 4.

DBT maximizes occupancy of sites on the capsid surface. (a) Other CpAMs will only bind a subset of available interfaces. HAPs²⁴ (and PPAs²³) bind to the B and C sites. DBT1 also binds A and D sites. (b) Crystal structure of a HAP molecule at the intersubunit interface.⁵⁰ In capsid structures where HAP binds, the B and C pockets allosterically enlarge to accommodate the HAP molecule, while the A and D pockets do not display bound HAP.⁵¹ To compare HAP and DBT1 at the A and D sites, we docked a HAP molecule into the A and D site of the new DBT1 structure and observed steric clashes (c). In contrast, DBT1 fits into both sites without collisions. The primary clash is with residue V124 on helix-5 of the adjacent “capping” subunit. To alleviate this clash and allow HAP to bind the A or D sites would require large changes in the quaternary relationship between adjacent subunits and may explain why HAPs are not observed at these sites. All distances are measured in angstroms. In the lower right panel of part c, the distance values are for the same relationships measured in the upper right panel.

To compare how multiple CpAM chemotypes bind the same pocket, in Figure 5, we summarize the binding poses of unique chemotypes of known structure. The degeneracy of the interaction between core protein and CpAMs is striking; a single CpAM will bind to multiple quasi-equivalent interfaces, and each interface will bind multiple CpAM chemotypes. With DBT1, we introduce a molecule that maximizes degeneracy by binding every quasi-equivalent site on the capsid surface.

Figure 5.

DBT1 adopts a distinct binding pose. For clarity, only one-half of the interface is shown, with the CpAM binding pocket visible as a cavity in the subunit surface. (a) The interactions at subunit interfaces are dominated by hydrophobic contacts (tan is more hydrophobic). There is a remarkable diversity of molecular scaffold for molecules that bind the interdimer interface. Shown are dibenzothiazepine (DBT, b), heteroaryldihydropyrimidine (HAP, c), ciclopirox (d), sulfamoylbenzamide (SBA, e), and phenylpropenamide (PPA, f).

DISCUSSION

The virus capsid as a target for therapeutics has some important advantages over other targets. Most viral structural proteins have little similarity to host proteins. Polymeric complexes, like capsids, may suppress the effect of resistant mutants, because the presence of any wild-type protein will exert a dominant sensitive effect.²⁷ HBV is not the only icosahedral virus that has been targeted. A series of anti-picornavirus molecules stabilized the capsids to inhibit genome release.²⁸ A series of HIV maturation inhibitors prevent proteolysis of the Gag polyprotein presumably by stabilizing Gag–Gag interactions.²⁹ These capsid-directed molecules either are or were in clinical trials. All of them are capsid stabilizing. This makes sense from a physical chemical perspective, because assembly is a reaction with a steeply downhill energetic path.^14,30 These molecules take advantage of mechanisms that nature has already established.

In this paper, we show that DBT1 is an assembly agonist with a range of functions. DBT1 stabilizes each protein–protein interaction by approximately 1.3 kcal/mol; this effect is amplified because each dimer is tetravalent, and the resulting capsid has 240 protein contacts. Because stronger association energy stabilizes intermediates, this thermodynamic effect contributes to faster assembly kinetics and kinetic traps.^31–35 At a structural level, DBT1 has a unique pose in the CpAM pocket, requiring less volume and interacting more strongly with helix 5 of the capping subunit. DBT1 also fits into all four quasi-equivalent sites of a T = 4 particle, a promiscuity never previously seen. Depending on conditions, DBT1 can lead to noncapsid polymers indicating a competition between normative and aberrant assembly paths.³⁶ Accelerated assembly and misassembly are thus a first basis of antiviral activity. Paradoxically, we also show that DBT1 can destabilize capsids leading to their dissociation, an effect that may be of particular importance for antiviral activity on metastable, mature rcDNA-filled capsids.^37,38

These observations present a puzzle: how can a molecule that increases the association energy between subunits, presumably a stabilizing effect, simultaneously cause capsid rupture? To reconcile these seemingly conflicting ideas, we outline the relevant reactions and propose corresponding energy diagrams in Figure 6. Starting with free subunit in the absence of any CpAM, assembly will produce normal icosahedral capsids (Figure 6, reaction 1a). Because each incoming subunit during assembly makes progressively more contacts, the assembly energy surface has a progressively steeper downhill slope.³⁹ When a CpAM is added to preformed capsids, it will bind the capsid (Figure 6, reaction 1b). Each binding event increases the association between adjacent subunits and is locally favorable. However, because CpAM binding modifies the intersubunit interaction, as the capsid fills up it becomes “strained”, as each subunit interface is no longer interacting with the ideal geometry for a well-formed icosahedron. We introduce the concept of a drug-bound “strained” capsid to explain the puzzle of a drug that increases association between subunits but also leads to capsid rupture. The “strained” capsid has drug bound and must be higher in energy than a ruptured capsid; this is demonstrated by the observation that the capsids do in fact rupture.

Figure 6.

Free energy schematic for CpAM mechanisms of action. On the left side, we describe a series of reactions; on the right side, we describe free energy diagrams for these reactions. (Reaction 1) In step 1a, free subunit assembles into a normal capsid. In step 1b, CpAM binds to the preassembled capsid. We know that this is a lower free energy state, because the CpAM binds. We assume all bound capsids are globally strained, because CpAMs bind to subunit interfaces, and we observed that PPAs, which do not cause aberrant assembly, cause capsids to expand by 5%.²³ In step 1c, if the strain is sufficient the capsid ruptures releasing free subunit. (Reaction 2) Capsid and CpAM coassemble to form capsids. These capsids assemble into icosahedral or nearly icosahedral particles, but because they have bound CpAM, they are strained in that they are conformationally and energetically distinct from the apo capsid. (Reaction 3) Co-assembly of free subunit and CpAM leads to aberrant misassembled products. We depict this reaction with saturating CpAM, with one drug at every subunit interface. For this reaction, a CpAM at every subunit interface is required for misassembly (as seen with BAY 41–4109²²). (Reaction 4) Assembly produces similar morphology to reaction 3 but at substoichiometric CpAM. This implies kinetic regulation, perhaps at the level of nucleation, leading to aberrant assembly. Free subunit from ruptured capsid (product of reaction 1c) may enter reactions 3 or 4 to form aberrant structures.

In free energy terms, the capsid takes on a global penalty, a positive free energy (ΔG_strain), which is paid for in part by CpAMs having a high affinity for the local subunit interface. If the strain penalty is sufficiently large, the capsid may be able to reach a lower free energy through rupture (reaction 1c). The rupture event breaks some local contacts, but it relieves the global strain penalty. Formally, the free energy of a strained capsid is the sum of its local contact terms minus a global strain penalty:

$$\Delta G_{\text{strained}}=N_{\text{contacts}}\left(\Delta G_{\text{cont}}+\Delta G_{\text{CpAM}}\right)-\Delta G_{\text{strain}}$$ (1)

where the subunits in a capsid are connected by N_contacts, each with an association energy of ΔG_cont. Previously we have demonstrated that the contribution of a CpAM to this association energy (ΔG_CpAM) can be quantified.²⁰ For Cp149, ΔG_cont is −3.1 kcal·mol⁻¹·contact⁻¹, and DBT1 adds an extra −1.3 kcal·mol⁻¹·contact⁻¹ (Figure 1b, inset).

The free energy of a ruptured capsid has fewer local contacts but lacks a significant global strain penalty:

$$\Delta G_{\text{ruptured}}=\left(N-N_{\text{broken}}\right)\left(\Delta G_{\text{cont}}+\Delta G_{\text{CpAM}}\right)$$ (2)

This perspective provides an energetic framework for thinking about how a CpAM affects the geometry of a capsid. The key inequality is ΔG_ruptured < ΔG_strained, which is the condition where a capsid will rupture. It represents the tipping point where the strain penalty is enough that the capsid can enter a lower energy state by losing subunit contacts. For a capsid without drug and without strain, losing contacts would be an unfavorable transition. Our observation that this rupturing transition does occur provides evidence for the existence of the strain penalty.

The extent of global strain that can be tolerated will be dependent on two competing factors: (i) the propensity of the CpAM to modify intersubunit geometry and induce global capsid strain and (ii) the stabilizing effect at each local interface where the association energy is increased by the presence of the CpAM. For a capsid that is full of DBT1, we can make some reasonable estimates:

$$\Delta G_{\text{strained}}={240}_{\text{contacts}}(-3.1\text{ kcal}/\text{mol}+-1.3\text{ kcal}/\text{mol}/)-\Delta G_{\text{strain}}$$

$$\Delta G_{\text{strained}}=-1056\text{ kcal}/\text{mol}-\Delta G_{\text{strain}}$$

For a capsid that has ruptured along a seam that breaks eight contacts and loses eight DBT1 molecules,

$$\Delta G_{\text{ruptured}}=(240-8)(-3.1\text{ kcal}/\text{mol}+-1.3\text{kcal}/\text{mol})$$

$$\Delta G_{\text{ruptured}}=-1020.8\text{ kcal}/\text{mol}$$

If ΔG_ruptured < ΔG_strained is to remain valid in this example, the value for the global strain penalty (ΔG_strain) must be ≥35.2 kcal/mol. Of course, this value is meant to be illustrative and relies on several assumptions. It assumes that a capsid completely fills DBT sites before rupturing, but this is consistent with our own observations (Figure 2b), where the capsid appears to fall apart only after >97% of the sites have been filled. It assumes there is no residual strain. Also shifting the equilibrium is the possibility that drug and subunits that have been released are free to interact with other free subunits and initiate a new assembly reaction. When capsids are treated with a potent HAP, such as HAP12, we often see the larger misassembled complexes (Figure 2a,f) that are the ultimate products of freed subunits. DBT1 allowed us to observe free subunit released upon rupture (Figure 2a,b). In comparison, the SBA that we tested showed no evidence of capsid rupture, indicating that some CpAMs induce only a modest strain penalty (ΔG_ruptured ≫ ΔG_strained).

Co-assembly with CpAMs can lead to approximately normal or clearly abnormal assembly products, depending on the CpAM and reaction conditions (Figure 6, reactions 2–4). Assembly with an SBA or PPA leads to “on-path” capsids (Figure 6, reaction 2), but we expect they are still strained to a modest degree; for example, PPAs that do not cause capsid rupture still cause capsids to expand by 5%.²³ Assembly with substoichiometric concentrations of HAP (e.g., BAY41–4109) also leads to morphologically normal capsids.^22,40 Conversely, assembly with superstoichiometric concentrations of HAP leads to aberrant assembly (Figure 6, reaction 3). We reason that with saturating HAP, the formation of these larger, flatter products involves less strain penalty than assembly into icosahedral capsids. While coassembly with DBT1 at substoichiometric concentrations preferentially forms morphologically normal capsids, it will also produce some aberrant structures (Figure 2c). This suggests that a CpAM can induce aberrant geometry at an early point in assembly, possibly nucleation, and that defect can persist (Figure 6, reaction 4).

In a biological environment, where mature capsids are at vanishingly low concentration and are destabilized due to the encapsidation of a dsDNA coiled spring,^37,38 the addition of a drug that favors aberrant geometry may be enough to cause capsid rupture to disrupt the normal path of infection. In a recent paper, it was observed that the CpAM JNJ6379 rendered dsDNA-filled capsids but not ssDNA-filled capsids to be DNase sensitive and inhibited formation of cccDNA during infection.⁴¹ CpAM activities may also include interference with reverse transcription⁴² or RNA transport.⁴³ We show that DBT1 has distinct behavior in capsid assembly and destabilization relative to other CpAMs. The experiments presented here demonstrate that DBT1 paradoxically induces assembly and misassembly while simultaneously globally destabilizing capsid structure. There can also be significant variations within a CpAM chemotype. We expect that development of the most effective CpAM variants will favor molecules that make core protein dysfunctional during multiple processes of the viral lifecycle, capsid assembly, packaging pgRNA, mediating reverse transcription, delivering DNA to the nucleus, and maintaining the critical pool of cccDNA, toward the ultimate goal of achieving a functional cure for patients living with chronic HBV.

EXPERIMENTAL METHODS

Preparation of Material.

The hepatitis B core protein assembly domain was expressed in E. coli and purified as previously described.⁴⁴ To prepare purified empty capsids, core protein homodimers were induced to assemble using 300 mM NaCl, 20 mM Tris, pH = 7.5, overnight and purified using a Superose 6 10/300 GL size exclusion column.

Quantification of Assembly Reactions.

For the assembly reactions involving Cp149 and DBT1, fresh dimer was added at a 1:1 ratio by volume to a premixed reactant mixture of NaCl, buffer, and DBT1 so that the postmixing concentrations were 5 μM dimer and each DBT1 concentration specified. Assembly kinetics were monitored by 90° light scattering using a Photon Technology fluorometer with excitation and emission monochromators at 320 nm. The mass fractions of assembled and unassembled subunit were resolved by size exclusion chromatography using a Superose 6 10/300 GL column and quantified on absorbance at 280 nm.

Detection of CpAM-Induced Capsid Disruption.

Serial dilution of DBT-1 with DMSO was performed to produce drug stocks over the desired range of concentrations, which were mixed at a 1:1 ratio with 10 μM Cp149 purified capsids (20 mM Tris, pH = 7.5), for a final concentration of 5 μM Cp149 dimer in capsid form (in 1% DMSO). Samples were given 12 h to incubate before analysis. Following incubation, samples were loaded in a 1% native agarose gel in TBE running buffer for electrophoresis at 50 V for 4 h, then blotted for 8 h on a polyvinylidene membrane. The membrane was incubated with rabbit anti-Cp primary and horseradish peroxidase-labeled secondary antibody (1:5000). Luminol and hydrogen peroxide at 1:1 ratio were added to the membrane for imaging. To verify the results of the gel electrophoresis for DBT1, the reactions were analyzed by size exclusion chromatography on a Superose 6 10/300 GL column, using the same buffer as that used for gel electrophoresis (TBE) and at a low flow rate (0.125 mL/min) for comparable running times between the electrophoresis and chromatography.

Negative Stain Electron Microscopy.

Samples were adsorbed to the glow discharged carbon film coated 300 mesh copper grids, washed with water, stained, and air-dried. Samples were stained with 2% uranyl acetate (Figure 1) or 6% ammonium molybdate (6%) and trehalose (0.5%) (Figure 2). Trehalose is added to minimize the collapse of particles that occurs on the grid upon dehydration. Images were collected with a JEOL TEM 1010 (Figure 1) or a JEOL 1400plus microscope (Figure 2).

Cryo-electron Microscopy and Image Processing.

For single particle Cryo-EM, the Cp150 construct was used to prevent capsids from dissociating before imaging. The Cp150 capsids are prepared in the same way as Cp149 capsids, but the construct spontaneously forms an interdimer disulfide cross-link. Purified Cp150 capsids were then incubated with a molar excess of DBT1 overnight before concentration to 10 mg mL⁻¹ using an Amicon centrifugal concentrator. Concentrated samples were applied to glow-discharged Quantifoil holey-carbon grids (R2/2). The grids were blotted with filter paper for 4 s before automated plunging into liquid ethane using an FEI Vitrobot. Imaging was carried out on a FEI Titan Krios operated at 300 kV at a nominal magnification of 22 500. Images were recorded on a Gatan K2 Summit detector operating in super-resolution mode, resulting in a pixel size of 0.65 Å with a dose of ~33 e⁻/Ų. Each exposure was 8 s long and was collected as 35 individual frames. Cryo-EM classification and reconstruction were implemented by adapting standard protocols of the EMAN2 and Relion software.⁴⁵ The use of focused subparticle reconstruction was inspired by previous descriptions from Ilca et al.⁴⁶

Model Determination and Structural Analysis.

The crystal structures of both apo capsid (1qgt) and HAP-TAMRA bound capsid (6bvf)²⁵ were used as starting points for flexible model refinement using the PHENIX and COOT software programs.^47,48 The model validation statistics that we report were obtained from the final output of the Phenix real space refinement tool. Structural comparisons, figure creation, and molecular docking tasks were carried out in UCSF Chimera.⁴⁹