Curvature of the retroviral capsid assembly is modulated by a molecular switch

Abstract

During the maturation step, the capsid proteins (CAs) of a retrovirus assemble into polymorphic capsids, whose acute curvature is largely determined by insertion of 12 pentamers into the hexameric lattice. Despite intensive research, it remains elusive how the CA switches its conformation between the quasi-equivalent pentameric and hexameric assemblies to generate the acute curvature in the capsid. We report the high-resolution structural model of the Rous sarcoma (RSV) virus CA T=1 capsid. By comparison with our prior model of the RSV CA tubular assembly consisting entirely of hexamers, we identify the key residues for dictating the incorporation of acute curvatures in the capsid assembly. These residues undergo large torsion angle changes, which result in a 34° rotation of the C-terminal domain (CTD) relative to its N-terminal domain (NTD) around the flexible interdomain linker, without substantial changes of either the conformation of individual domains or the assembly contact interfaces.

Graphical Abstract

A mature retroviral capsid is assembled with more than 1000 copies of the capsid protein (CA) during a process in the virus life cycle called “maturation”, which finalizes assembly of the capsid structure.¹ Despite limited sequence similarity, monomeric orthoretroviral CAs share nearly identical 3D folding.² Briefly, a CA contains two independently folded domains joined by a short flexible linker,³ with the N-terminal domain (NTD) comprising a short β-hairpin followed by seven α-helices,^4-5 and the C-terminal domain (CTD) consisting of a short 3₁₀ helix followed by four additional α-helices.⁶ Intriguingly, these CAs form polymorphic capsids of distinct morphologies for different viruses, including tubular, polyhedral and conical structures.^{2, 7} It has been shown that proper capsid assembly is critical for viral propagation and infectivity.^{1, 8} Thus, CAs and their assemblies are promising antiviral drug targets.^9-10 However, due to the strong polymorphism,^{2, 11} the detailed molecular basis of retroviral capsid assembly remains elusive.

A key unresolved question is the mechanism of curvature management in the retroviral capsid assembly. Based on analyses of in vivo and in vitro assemblies, the polymorphism and particular shapes of retroviral capsids are assembled by insertion of 12 CA pentamers at varying locations in the hexameric lattice.^{7, 12} A great deal was learned about the hexameric assembly of retroviral CAs by studying various in vitro systems, while it is unclear how pentameric assembly is formed by the same CAs to incur acute assembly curvature.¹¹ Ganser et al. obtained the first high resolution structural model of the hexameric assembly formed by HIV CA R18L mutant, and identified three assembly interfaces unambiguously for the first time:¹³ the first three helices form the NTD-NTD interface that bind adjacent subunits into individual hexamers, while the CTDs are stabilized by the NTD-CTD interface between helix 4 and 8 within each subunit. Adjacent hexamers are bound together by the CTD dimer interface across helix 9 into a complete hexameric lattice. Subsequently, Pornillos et al. obtained crystal structures of isolated HIV CA hexamers^14-15 and isolated pentamers¹⁶ by crosslinking of mutant HIV CA and ccmk4¹⁷ fused CA. Both pentamer and hexamers exhibit interfaces similar to the cryogenic electron microscopy (cryoEM) model.¹³ In addition, they found that NTD-NTD and NTD-CTD interfaces are solvated with water, with little variations across all hexamer models. Instead, CTDs at the edge of isolated hexamers adopt a variety of orientations. This was interpreted as the cause of assembly plasticity. The charged residues closer to the NTD-NTD interface in the CA pentamer were suggested to act as the assembly switch. However, these crosslinked oligomers are in an isolated state with a disrupted CTD dimer interface. Around the same time, Cardone et al. and Hyun et al. obtained ~ nm resolution models of T=1 and T=3 Rous sarcoma virus (RSV) CA capsids, showing interfaces similar to those in HIV CA hexamers. Their results demonstrated that different retroviruses adopt a common capsid assembly structure and potentially similar curvature control mechanism^18-20 However, the limited resolution does not allow distinction of CA pentamers and hexamers at the molecular level. Subsequently, Zhao et al. obtained an 8 Å resolution model of the HIV CA tubular assembly by cryoEM. In this curved hexameric assembly, a fourth interface, the CTD trimer interface, was discovered, between helix 10 and 11,²¹ critical to the tubular curvature. Atomic resolution models of different shaped mature HIV capsids were constructed,²¹ using molecular dynamics (MD) simulation to insert isolated pentamers¹⁶ into their tubular model. The models indicate that variations of CTD dimers are needed for different curvatures around the capsid, which agrees with earlier work by Pornillos et al.¹⁶ In the 8.8 Å resolution cryoEM model of the authentic HIV capsid,²² pentamers seem to adopt different NTD-NTD and NTD-CTD interfaces from those in crosslinked hexamer and pentamers.^{14, 16} However, details could not be resolved at the limited resolution. Meanwhile, analysis of different RSV CA planar hexameric sheets at 4 Å resolution by Bailey et al. indicated that both NTD-NTD and CTD dimer interfaces are rigid, but the NTD-CTD orientation varies around the NTD-CTD interface in different sheets.²³ However, no curvature is present in these assemblies.

Recent atomic resolution X-ray diffraction models of native HIV CA hexamer crystals identified prevalent hydrogen-bonds (H-bonds) at all interfaces.²⁴ It was also shown that variation of the hydration level triggers changes of NTD-CTD orientations around the interdomain linker, which was attributed to assembly plasticity. In addition, notable differences were observed at the CTD dimer interface compared with that seen in isolated hexamers of crosslinked mutant HIV CA,^14-15 with a tighter CTD trimer interface.²¹ Solid state NMR (ssNMR) was also applied to study various HIV CA assemblies.^25-32 Residues at the ends of each domain and in the cyclophilin A binding loop were shown to be highly mobile, while the linker between the two domains undergoes millisecond motions,^26-27 which could alter the interdomain orientation and cause assembly polymorphism. However, no major distinction at the molecular level could be identified between CAs in the pentameric and hexameric assemblies.^{26, 28-29} Hence, it is still not clear how retroviral CAs switch between quasi-equivalent pentamers and hexamers, and control assembly curvature.

In our previous work, we obtained the first high-resolution model for the RSV CA tubular assembly by combining ssNMR and cryoEM constraints.³³ We showed that the curvature variations in the hexameric lattice are mediated by varying side chain configurations, while the backbone structure remains largely unchanged. In this work, we report the high-resolution structural model of the RSV CA in its T=1 capsid assembly, with 1.97 Å root-mean-square deviation (RMSD) of Carbon Cα among CA monomers. Molecular dynamics flexible fitting (MDFF) allows site-specific resolution of the molecular details of CA pentamers by combining secondary structural constraints from our ssNMR results with the previously published cryoEM model (EMDB 5773) at 8.5 Å.²⁰ Importantly, the model reveals a simple mechanism for the RSV CA to switch between hexameric and pentameric assemblies: the subunit undergoes a 34° interdomain rotation via the flexible interdomain linker, which causes large torsion angle changes to only about a dozen residues, with little adjustment of domain structure or interface contacts.

Wild-type RSV CAs tend to form either tubes or large polymorphic structures comprising mostly hexamers.³⁴ To prepare an assembly comprising pentamers, we used the I190V RSV mutant.³⁵ When solution at 6 mg/ml was mixed with an equal volume of 1 M sodium phosphate buffer at pH 8, predominantly small spherical structures were formed. To assess the morphology of the assemblies, negatively stained transmission electron microscopy (TEM) images were recorded. A representative view is shown in Figure 1A. Measurements of the assembly diameter were analyzed, and a histogram is shown in Figure 1B. The average size of the spherical assemblies is about 18.5 nm, consistent with the expected value for T=1 capsid.¹⁸ These data confirm the fact that our assembly sample consists nearly entirely of CA pentamers.

Figure 1.

TEM and ssNMR characterization of the RSV CA T=1 capsid assembly. (A). Negatively stained TEM image of the RSV CA I190V T=1 capsid assembly. The operating magnification is 15,000-fold. (B). The average diameter of observed assemblies is ~18.5 nm as shown by histogram of the distribution of assembly diameters. (C). 2D ¹⁵N-¹³C correlation spectra overlay of the T=1 capsid assembly by the RSV I190V CA (red) and the tubular assembly by the wild-type RSV CA (green). (D). Site-specific prediction of secondary structure and rigidity by TALOS-N⁴² from chemical shift assignments. Bar height represents the S² order parameter, a measurement of rigidity. Bar saturation represents prediction confidence. The thick color lines at the top highlight segments of secondary structured units, blue for β-strands and red for α-helices.

To perform NMR experiments, uniform ¹³C, ¹⁵N labeled CA (U-CA) was expressed. Its assembly produces 2D spectra with excellent resolution. A 2D ¹⁵N-¹³C correlation spectrum with cross peaks created by 50 ms dipolar assisted rotational resonance³⁶ is shown in Figure 1C. The typical linewidth is about 0.7 ppm along the ¹³C dimension and 1.2 ppm along the ¹⁵N dimension, comparable to that of our prior RSV CA tubular assembly under moderate magic angle spinning (MAS),³³ and similar to that of other highly ordered protein structures^37-40 and HIV CA assemblies.^25-32 This indicates that RSV CA adopts a highly ordered structure at the molecular level, consistent with our TEM evaluation.

To perform sequential assignments, a series of 2D and 3D correlation spectra were recorded with the U-CA, as detailed in Table S1 in the supporting information. All congested resonances in 2D spectra can be resolved in the 3D NCaCX and partially resolved in 3D NCOCX spectra. We note that only one set of resonances was identified for each residue, which suggests that the RSV CA adopts a unique and well-defined conformation in the T=1 assembly. 3D CANCX spectra were also recorded. However, given the weaker signal to noise ratio (SNR) and the similarity to assigned chemical shifts in the RSV CA tubular assembly,³³ the benefit of CANCX was very limited to help the sequential assignments. Manual sequential assignments were performed by a backbone walk to link resonances in 3D NCaCX and NCOCX. Together, 115 residues were sequentially assigned. Example stretches of the sequential walk assignments between residues 53–65, and residues 101-116 are shown in Figure S1 and S2 in the supporting information.

Additionally, 104 well resolved residues were identified in either the 3D NCaCX or NCOCX spectra. Although the backbone walk cannot be achieved for these residues, they exhibit signals nearly identical to those we observed with our prior tubular assembly.³³ These observations provide a high level of certainty regarding their identities.

Resonances in these 3D correlation spectra were acquired by cross polarization transfer, mediated by anisotropic dipolar interactions (through space). Thus, only residues in rigid segments in the assembly would contribute. To detect residues in the flexible segments, INEPT-TOBSY spectrum were acquired by polarization transfer mediated by isotropic J coupling (through bond),⁴¹ shown in Figure S3. Five residues were identified with a narrow linewidth comparable to solution NMR and chemical shifts close to random coil values. These residues undergo fast motion, as seen in the soluble state. They are assigned to P230, L231, T232, G235, and I236, like those at the tail of the CTD resolved in our prior tubular assembly work.³³ We also note that signals in the INEPT-TOBSY spectrum of our T=1 assembly sample are much weaker than those of the tubular assembly. It implies that these residues at the CTD tail are probably not as mobile as those in the tubular assembly, leading to attenuated signal intensity.

Altogether, 224 residues were assigned for the 237-residue RSV CA in its T=1 capsid assembly. The assignments were deposited in the Biological Magnetic Resonance Data Bank (BMRB), with BMRB ID 50550, and are also shown in Table S2. The following residues are missing from both through bond and through space transferred spectra: residues 47, 50, 88, 95, 117, 157, 173, 218, 219, 229, 233, 234, and 237. Except for residue 173 in the middle of helix 9, the missing residues in our T=1 capsid assembly are all in the flexible loop regions between helices. Most likely, this is because their motions are in a range not sensitive to ssNMR detection. In contrast, missing residues in the tubular assembly were residues 150, 229, and 230.³³ In addition, we note the average SNR in our T=1 capsid assembly is nearly 2.5 fold weaker than that in the tubular assembly,³³ while the spectral linewidth and acquisition conditions (experiment setup and the amount of proteins in NMR rotors) are comparable. This suggests that the molecular ordering in the hexameric and pentameric assembly is comparable, and the weaker signal intensity in the pentameric assembly is due to the shift of the overall protein dynamics towards time scale insensitive to ssNMR detection.

Based on the assignments, the secondary structure composition and local dynamics were derived with TALOS-N,⁴² shown in Figure 1D. The raw secondary chemical shifts are shown in Figure S4A. Except three residues, predictions for residues without assignments were still considered as “strong” or “generous” by TALOS-N, based on assignments of adjacent residues and the match of hepta-peptide sequence.⁴² In addition, despite provided assignments, predicted torsion angles of another 27 residues are considered as “WARN” by TALOS-N, marked by light grey bars in Figure S4A. The number of “WARN” prediction is on par with that of our previous tubular assembly work (29 residues, shown in Figure S4B).³³ More, 16 of these residues overlap in two assemblies, while the rest lie adjacent to these mutual “WARN” residues. Except residue 189 that is next to the mutation site, all these “WARN” predictions are residues in structureless regions. Such “WARN” predictions are not necessarily incorrect or unreliable, due to the insensitive resolution of structureless segments in the X-ray models used by TALOS-N.⁴² Thus, the comparison of TALOS-N predictions also corroborates the quality of our assignments of the T=1 capsid sample.

Overall, the RSV CA in the T=1 capsid assembly adopts a folding nearly identical to that in the tubular assembly,³³ with the exception for a small number of residues. This notable difference can be better appreciated by overlay of the 2D ¹⁵N-¹³C correlation spectra, as shown in Figure 1C: the RSV CA demonstrates similar resonances in either T=1 or tubular assembly, except for 18 residues exhibiting large shifted resonances (≤ 2.5 ppm), which is shown specifically by the differences of secondary chemical shifts plotted in Figure S4C. This is in clear contrast to the comparison of tubular assemblies with a 2-fold change of diameters (average diameter ~ 60 nm vs. 130 nm), which showed nearly identical Cα resonances for all residues, with shifted signals associated with side chains.³³ This indicates that CAs have to reconfigure both their backbone and side chains to adapt to the pentameric assembly, while curvature changes in the hexameric lattice can be accommodated by side chain rearrangements alone.

To obtain the structural model for the RSV CA T=1 capsid, data-guided simulations were performed to combine our ssNMR constraints with the 8.5 Å cryoEM density (accession code EMDB 5773),²⁰ with protocols described in the supporting information. The simulated model described 60 CA monomers, arranged in a T=1 icosahedron, solvated with TIP3P water⁴³ and ionized to a salt concentration of 500 mM NaCl (Figure S5A and S5B). Figure S5C shows RSV CA monomers extracted from a system of the solvated T=1 capsid simulation and aligned by the backbone atoms, with 1.97 Å RMSD of Cα among CA monomers. This alignment shows the highly uniform structure of subunits in the T=1 capsid assembly, similar to those in the tubular model. Therefore, our model allows analyses to pinpoint the distinct RSV CA conformation at a site-specific resolution in different assemblies.

Next, we computed the torsion angles of the RSV CAs measured directly from the in silico tubular and T=1 assembly models. Their differences are plotted in Figure 2B. The results agree well with those predicted by TALOS-N⁴² derived directly from NMR chemical shifts (Figure 2A). Consistently, the regions in the RSV CA with the largest torsion angle differences are located in the β-hairpin around residues 1-16, the flexible loop and flexible linker regions encompassing residues 85-100 and 145-156, respectively (Figure 2C). In particular, torsion angle differences based on our NMR assignments pinpoint a dozen residues with torsion angle differences greater than 100°. Their assignments in the 3D NCaCX spectra are plotted in Figure S6. They reveal the site-specific changes RSV CA must undergo to switch from a quasi-equivalent hexameric to pentameric assembly to induce acute curvature in assemblies.

Figure 2.

Quasi-equivalent conformation of the RSV CA in its T=1 capsid and tubular assembly. (A) and (B) are the torsion angle differences derived by TALOS-N⁴² from NMR assignments, and computed in silico from the T=1 capsid and prior tubular model.³³ The thick color lines at the top highlight segments of secondary structured units, blue for β-strands and red for α-helices. The grey bars in (A) mark those lack assignments, but their torsion angles were still predicted by TALOS -N based on assignments of adjacent residues and the match of hepta-sequence. (C). Residues with large torsion angle differences between two assemblies colored in green on a subunit from the T=1 capsid model.

Our T=1 assembly model also resolves the four intermolecular CA interfaces, critical for capsid stability and morphology, shown in Figure 3. Interestingly, torsion angle analysis shows that differences in backbone dihedral angles of these interfacial residues are limited compared to other regions of CA, such as residues of the flexible linker and other flexible regions. In the NTD, residues Leu61 and Gly62 situated near the H1-H2-H3 bundle show notable torsion angle differences although they are positioned away from the helical bundle interior (Figure 3C). Residues comprising the NTD-CTD interface (Figure 3E) show consistent torsion angles, while Ser178 – downstream of H8 and the residues which interface with H4 and H11 – shows a dramatic change of the psi dihedral angle. Ser178 precedes the dimeric interface in CA which, like other interfaces, does not demonstrate large torsion angle differences among assembly models. This observation follows the trend of torsion angle differences concentrated among residues preceding and succeeding structured interfaces. Continuing down the CA sequence and preceding the trimer interface is Lys196, which additionally substantiates our observation as to where torsion angles differ in distinct CA assemblies.

Figure 3.

The four intermolecular assembly interfaces in the T=1 capsid model in (A) at the residue specific level. (B). The dimeric interface is between helix 9 of adjacent subunits. (C). The NTD-NTD interface comprises helix 1 to 3 of each subunit in the pentamer. (D). The CTD trimeric interface involves helix 11 and 10. (E). The NTD-CTD interface is between helix 8 and 11.

As shown in Figure 3A to D, close inspection confirms that the same set of residues are employed to stabilize each interface in both pentamers in the RSV CA T=1 capsid assembly, and hexamers in the tubular assembly.³³ Thus, the molecular differences in torsion angles do not cause large variations of assembly interfaces in the CA pentameric assembly compared to its hexameric assembly. It shows the remarkable efficiency of the viral CA to achieve assembly versatility with minimum modulation.

Thus, our above analyses show only subtle differences exist in the RSV CA folding at the molecular level, in hexameric vs. pentameric assemblies. Then, the question arises: How do such subtle differences in folding translate into one less subunit in the quasi-equivalent assemblies?

To answer this, we first aligned the monomers extracted from the RSV tubular and T=1 assembly by their backbone atoms in the NTD, as shown in Figure 4A. Then, Principal Component Analysis (PCA) was applied to identify the dynamical mode or essential component that set apart the molecular folding in the hexameric vs. pentameric assembly, with their contribution ranked by the overall variance of the dataset.

Figure 4.

The differences of RSV CA in quasi-equivalent assemblies revealed by PCA. (A). 360° view of CA monomers from the T=1 capsid (orange) aligned by its NTD against that from the tubular (blue) assembly. (B). Principal components reveal a rotation plus translation motion around the flexible interdomain linker to transform a CA in its pentameric assembly to that in the hexameric assembly. The blue vectors denote the direction and magnitude of the motion. (C). Ranking of the contributions of principal components to the differences of the RSV CA conformation in the pentameric vs hexameric assemblies. (D). The residue specific motions of the largest three principal components are shown by colored vectors.

Interestingly, our PCA reveals a dominant mode that shifts an RSV CA monomer in the hexameric tubular assembly to that in the T=1 capsid assembly, contributing to more than half of all the variance in the dataset. As shown in Figures 4B and 4C, this mode corresponds to a 34° rotation and 5.94 Å center-of-mass translation of the CTD mediated by residues in the flexible linker, encompassing residues 147 to 149, as well as residues 151 to 156 in the 3₁₀ helix. In fact, nearly the entire structural rearrangements of CTDs can be captured by the top three PCs, as shown in Figure 4D. An animation of the top three PCs is shown in the movie in the supporting information. The animation also shows that the additional structural changes in the flexible loop and β-hairpin region are concomitant with the structural rearrangements of the CTD.

Previously, ssNMR were applied to study HIV CA assemblies of tubular, spherical and conical assemblies. No major differences of chemical shifts were found to correlate with assembly morphologies.^{29, 44} This contrasts with our ssNMR results that unambiguously correlate the residues in different conformations in T=1 capsid and tubular assemblies. The different findings are probably due to the low concentration of pentamers in those large, polymorphic HIV conical and spherical assemblies, which masked their structural signatures in the overwhelmingly hexameric subunits. In contrast, our sample comprises nearly exclusively and uniform pentameric assemblies (Figure 1B). Hence, the observed chemical shift differences in this work by ssNMR testify to its unique strength to detect even the subtle differences at a site-specific resolution between quasi-equivalent pentameric and hexameric assemblies, which is challenging to obtain otherwise. Combined with MDFF, the ssNMR restraints are critical to refine the 8.5 Å resolution cryoEM density map to the high-resolution model reported in this work. Our PCA results corroborate our ssNMR observation and further revealed how these key residues in the RSV CA orchestrate the structural transformation to form a pentamer or hexamer, without major modulations of the assembly interfaces or the structures of NTD and CTD. Moreover, as shown in Figure S7, this structural transformation leads to a NTD-CTD orientation in the pentamer that matches with what was identified by cryoEM in the authentic HIV virion, where the CPSF6 binding pocket in the NTD of the neighboring subunit is exposed.²² It suggests that a common structural transition of subunits is employed to switch between the pentameric and hexameric assembly for both HIV and RSV CAs. Hence, our results provide critical new insights to unraveling the mystery of how retroviruses control the curvature in their capsid assemblies.

The function of the flexible interdomain linker revealed by our work, was hypothesized years ago, by both Pornillos et al.¹⁶ and Worthylake et al.⁴⁵ Brun et al.⁴⁶ and Wacharapornin et al.⁴⁷ first demonstrated the detrimental effect of the S149A mutation on HIV viral infectivity. Subsequently, Jiang et al. performed a systematic site-directed mutagenesis study of the HIV CA flexible linker (residues 145-151). Their results showed that residues in the non-structured linker regions are critical for the correct assembly and stability of HIV capsid and that specific mutations attenuate or completely abolish infectivity.⁸ The flexible linker in the HIV CA tubular assemblies adopts four distinct conformations,³² undergoing millisecond scale dynamics by ssNMR,^{44, 48} which would allow such a modulation to take place. Our results support these reports and elucidate the detailed molecular basis to correlate the motions of the flexible linker with the curvature control.

Nevertheless, it is not yet clear how this motion of the flexible linker is triggered and regulated in actual viral particle assembly. Solution NMR studies by Lalit et al. defined the conformation space of interdomain orientation sampled by HIV CA dimers in low salt solution and suggested that the selection of the pentameric assembly over hexameric assembly is tightly regulated by such motion.³ Accounting for the conformation space defined by solution NMR, coarse-grained simulations by Xin et al. also showed that the HIV CA pentamers can be readily incorporated into a hexameric lattice at concentrations comparable to those used in experimental assembly; this requires a highly specific NTD to CTD orientation, not sensitive to particular structural variations within individual domains.⁴⁹ Subsequent coarse-gained simulations by Grime et al. suggest a higher concentration favors the assembly of an HIV CA pentamer over a hexamer.⁵⁰ Thus, a possible mechanism for different retroviruses to control their preferred assembly morphology, would be to position the locations of such factors that promote the specific torsion angles of the interdomain linker suitable for pentameric CA assembly. Slight randomization of their locations would displace the locations of resulted pentamers and give rise to assembly polymorphism.