Intrinsic curvature of HIV-1 CA hexamer underlies capsid topology and interaction with cyclophilin A

Abstract

The mature retrovirus capsid consists of a variably curved lattice of capsid protein (CA) hexamers and pentamers. High-resolution structures of the curved assembly, or in complex with host factors, have not been available. By devising new cryoEM methodologies for exceedingly flexible and pleomorphic assemblies, we have determined cryoEM structures of apo CA hexamers and in complex with cyclophilin A (CypA) at near-atomic resolutions. The CA hexamers are intrinsically curved, flexible and asymmetric, revealing the capsomere, instead of previously touted dimer or trimer interfaces, as the key contributor to capsid curvature. CypA recognizes specific geometries of the curved lattice, simultaneously interacting with three CA protomers from adjacent hexamers via two non-canonical interfaces, thus stabilizing the capsid. By determining multiple structures from various helical symmetries, we further revealed the essential plasticity of the CA molecule that allows formation of continuously curved conical capsids and the mechanism of capsid pattern-sensing by CypA.

Introduction:

The HIV-1 capsid is a conical assembly of approximately 200 hexamer and 12 pentamer capsomeres of the capsid protein (CA). CA comprises two α-helical domains, an N-terminal domain (CA_NTD) and a C-terminal domain (CA_CTD), connected by a flexible linker. CA is the most conserved among HIV-1 proteins and is essential for key functions, including virus assembly, reverse transcription, nuclear entry, and integration¹. In fact, it is the curved capsid lattice, rather than individual CA protomers, that serves as a primary binding platform for multiple host proteins that promote or inhibit HIV-1 infection^1–3.

The HIV-1 capsid has been the target of intensive study owing to its multiple roles in HIV-1 replication and its potential as a therapeutic target. Numerous efforts have focused on understanding how CA assembles into a mature fullerene cone. High-resolution crystal structures of hexameric and pentameric capsomeres have been determined alone^4–6 and in complex with various ligands (PF74^4,7, GS-CA1⁸, IP6⁹,¹⁰), but all in planar form. In addition, intermediate-resolution structures of curved assemblies have been resolved by cryogenic electron microscopy (cryoEM)¹¹ or cryo-electron tomography (cryoET) and subtomogram averaging¹², revealing a novel trimer interface critical to capsid assembly and rigid-body tilting and twisting between capsomeres to produce the curved lattice. However, a precise and reliable atomic model of the curved capsid assembly has not been tractable, which has impeded our understanding of virus-host interactions that exceed individual capsomeres. Many host cell defense proteins, including the restriction factors Trim5α¹³, TrimCyp^2,14 and MxB³, target the HIV-1 capsid and inhibit virus replication through pattern recognition of the curved assembly, but not the individual CA protomers or capsomeres.

Mature capsid assemblies are pleomorphic and highly flexible and thus represent a challenge for high-resolution structure determination. We developed a new data collection method, ArbitrEM, to automatically collect, from selected capsid assemblies, the large number of cryoEM images that are required to achieve high-resolution, thus vastly enhancing the effectiveness and throughput of data acquisition. We further devised an image processing and reconstruction workflow, including symmetry expansion, localized 3D classification and reconstruction, for exceedingly flexible and pleomorphic assemblies. Here, we present cryoEM structures of a mature capsid assembly at 3.6 Å resolution and in complex with the host protein cyclophilin A (CypA) at 4.0 Å resolution, from which we built atomic models of the hexameric capsomere and the CypA/CA lattice complex. While previous published structures of the wild-type (WT) CA hexamer⁴ or cross-linked CA hexamer and pentamer^5–6 are six (or five)-fold symmetric and in planar form due to crystalline lattice constrains, the cryoEM structure of the CA hexamer from the curved assembly is by contrast intrinsically curved and asymmetric, distinguishing it from all previous atomic structures. The curvature of CA hexamers varies depending on the helical symmetry, whereas the inter-hexamer interactions remain constant, resulting in variably curved CA surface lattices in assemblies. CypA senses the curved CA lattice, simultaneously interacting with three CA protomers from adjacent hexamers via the canonical CypA loop and additional non-canonical novel interfaces, thus stabilizing the capsid. We further determined multiple capsomere and CypA/CA structures from several helical symmetries, which provided a basis for dissecting the mechanisms of continuous curvature formation and how CypA embraces the curved CA surface lattice.

Results:

CryoEM structure of HIV-1 CA tubular assemblies

HIV-1 CA forms tubular assemblies of diverse diameters and helical symmetries^11,15. CA tubes are flexible and fragile, reflecting the metastable nature of the capsid that protects the genome while at the same time permitting controlled disassembly (uncoating) during infection¹⁶. These features of the capsid make it a very challenging target for atomic resolution structure determination by cryoEM, which would typically require manually recording thousands of images. To facilitate this task, we developed ArbitrEM (Extended Data Fig. 1), a new data collection script and workflow built on the SerialEM platform¹⁷, to automatically record micrographs of selected targets (in this case, tubes that are suitable for high resolution analysis) (Extended Data Fig. 1, methods). The major advantages of ArbitrEM include 1) targeting is carried out on-the-fly automatically; 2) multiple targets can be positioned with minimal stage movement therefore improving throughput; and 3) additional new targets can be added offline during data collection. Once tubes of interest were identified, ArbitrEM automatically collected ~100 micrographs per hour with 30 nm positioning accuracy, a great improvement over manual data collection. Subsequently, the tube images were cropped (Extended Data Fig. 2a) and sorted based on their helical symmetries: initially by their diameters (Extended Data Fig. 2b) and then by manual inspection of their Fast Fourier Transform (FFT) patterns (Extended Data Fig. 2c). More than 12 helical symmetries were identified within a 3 nm narrow range of tube diameters (Extended Data Fig. 2b, red box), illustrating the challenge in image processing.

Assemblies carrying the A92E substitution in CA, which prevents tube aggregation¹⁵, were analyzed, initially by real-space helical reconstruction¹⁸ (Extended Data Fig. 2d–e, Table S1) and then, because CA assemblies are flexible, using a localized reconstruction method¹⁹ (Extended Data Fig. 2f, see methods). By this approach, structures of A92E assemblies from three abundant helical symmetries, (−12,11), (−13,11) and (−13,12), were determined at 4.4 Å, 4.7 Å and 4.9 Å resolution, respectively (Extended Data Fig. 3a–b, Table S1). Combining data from the three helical symmetries only marginally improved the map (4.2 Å resolution), despite nearly doubling the amount of data. For this reason, we altered our approach and used CypA to stabilize WT CA assembly^20–22, and determined the structure of the CypA-stabilized assembly at 3.6 Å resolution (Extended Data Fig. 3, Table 1). From this structure, we built an atomic model of the hexameric capsomere in a curved lattice (Fig. 1a, Table 1). Comparison of capsomere structures with the same helical symmetry (−13, 12) in the presence or absence of CypA revealed that the CypA-stabilized capsomere is essentially the same as the A92E capsomere (Extended Data Fig. 3a & Extended Data Fig. 4).

Table 1:

Cryo-EM data collection, refinement and validation statistics

	Wild-type CA hexamer (EMD-10226, PDB 6SKK)	CA/CypA complex (EMD-10739, PDB 6Y9W)
Data collection and processing
Magnification	47196	47196
Voltage (kV)	300	300
Electron exposure (e−/Å2)	40	40
Defocus range (μm)	0.8-2.0	0.8-2.0
Pixel size (Å)	1.06	1.06
Symmetry imposed	C2	C1
Initial particle images (no.)	207264	207264
Final particle images (no.)	207264	161690
Map resolution (Å)	3.6	4.1
FSC threshold	0.143	0.143
Map resolution range (Å)	3.5 - 4.3	3.8-5.4
Refinement
Initial model used	PDB 4XFX	PDB 4XFX PDB 1AK4
Model resolution (Å)
FSC threshold	0.5	0.5
Model resolution range (Å)	3.8	4.7
Map sharpening B factor (Å2)	−217	−50
Model composition
Non-hydrogen atoms	0	0
Protein residues	1320	2364
Ligands	0	0
B factors (Å2)
Protein	74.12	157.9
Ligand	0	0
R.m.s. deviations
Bond lengths (Å)	0.003	1.012
Bond angles (°)	0.596	1.065
Validation
MolProbity score	1.41	1.52
Clashscore	7.50	9.87
Poor rotamers (%)	0.18	0.45
Ramachandran plot
Favored (%)	98.32	97.99
Allowed (%)	1.68	1.97
Disallowed (%)	0	0.04

Figure 1 |. The mature CA hexamer is intrinsically curved and asymmetric.

a) Localized real-space-helical reconstruction of in vitro assembled WT CA tubes with (−13, 8) helical symmetry at 3.6 Å resolution. Three orthogonal views of the CA hexamer density, from top, side and bottom, are shown. The density map is colored according to the local resolution. CA helices H1-3, CypA-loop, CA_NTD-CA_CTD linker and CA_CTD are labeled. b) Superposition of WT CA hexamers from planar crystal (grey, PDB 4XFX) and tubular assembly (blue), which displays ~20° curvature. Close-up views of CA_NTD helices are shown on the right. c) Flexible domain connection among three asymmetric CA monomers, colored in blue, orange and cyan, overlaid with PDB 4XFX monomer (grey). d) Close-up view of the boxed region in c, displaying a novel H-bond between His62 and Tyr145 in a CA protomer along the most curved direction.

The best resolved region in the 3.6 Å cryoEM map is at the inner core of the hexamer, comprising CA helices 1-3 (Fig. 1a, arrow), whereas the CA_CTD and CypA binding loop are less well-resolved (Fig. 1a). The interdomain linker is also clearly resolved (Fig. 1a, arrow), while the N-terminal β-hairpin density is relatively weak compared to the inner core. However, at low iso-surface levels, the main chain of the β-hairpin can be unambiguously placed. The overall quality of the atomic model is indicated by the Fourier Shell Correlation between the model and the density map (Extended Data Fig. 3c) and exemplified in the overlay of the model and the density mesh, where side chain densities are clear and can be modeled unambiguously (Extended Data Fig. 3d).

The resulting atomic model revealed a highly dynamic outer surface, including the CypA-binding loop, followed by the β-hairpin and loop between H4 and H5 (Extended Data Fig. 5a–b), in addition to a dynamic region at the inner surface of the assembly, in particular the trimer interface comprising CA_CTD helix H10 of three adjacent hexamers (Extended Data Fig. 5c–e). These regions are targets for host cell protein binding, such as CypA^20,23, Trim5α^13,24, Trim-Cyp¹⁴, and MxB^3,25; therefore, structural dynamics could play important roles in such interactions.

The CA hexamer is intrinsically curved and asymmetric

A long-standing question regarding HIV-1 capsid assembly is focused on the physical properties that allow a continuously curved non-symmetrical shell to be produced from ~1200 copies of a single protomer. Our data reveal that the CA hexamer is not symmetrical and is curved by approximately 20° (Fig. 1b), distinct from all existing high-resolution structures^4,5. This asymmetry is evident in all structures from five helical symmetries. Superposition of the WT CA hexamer from a planar crystal (PDB 4XFX) and the tubular assembly reveals a large discordance (RMSD of 3.21 Å of all Cα in the hexamer) (Fig. 1b). The distances between diametrically opposed CA_CTDs within the hexamer (between N195 Cα) vary from 77.2 Å to 83.6 Å (Extended Data Fig. 6a), compared to 85.3Å in planar crystals. In fact, CA_CTD distances correlate well with the local curvature of hexamer arrays, measured from the cryoEM structures of the five helical assemblies determined in this study (Extended Data Fig. 6b), suggesting that the CA hexamer itself is naturally flexible to meet the needs of the local environment. The flexibility of the CA hexamer does not reside in the individual CA_NTD and CA_CTD domains (RMSDs of 0.55 Å and 0.84 Å respectively) but originates from variable orientations between the CA_NTD and CA_CTD through the flexible, yet well-ordered, linker (Fig. 1c). The highly conserved linker residues are critical for HIV-1 infection²⁶. Intriguingly, Tyr145 forms a novel intramolecular hydrogen bond with His62 within the most curved CA subunit (Fig. 1d), in contrast to its intermolecular contact with Arg162 at the CA_NTD and CA_CTD interface uniformly observed in planar crystal hexamers (PDB 3H4E and PDB 4XFX for example). Breaking this intramolecular hydrogen bond by substitution of either His62 (H62F, H62W)²⁷ or Tyr145 (Y145A and Y145F, Y145A)^26,28 impaired assembly in vitro and virus infection in tissue culture.

Asymmetry is also present at the hexamer centre, in particular at the β-hairpins (Extended Data Fig. 6c) and residue Arg18 (Extended Data Fig. 6d). Intriguingly, clearly resolved Arg18 side chains adopt two alternative conformations along the curved directions (Extended Data Fig. 6d, dashed box, Extended Data Fig. 6e) that differ from those observed in crystal structures with and without the pocket cofactor IP6⁹ (Extended Data Fig. 6f–g). It is tempting to speculate that Arg18 could bind and release IP6 in exchange for cytosolic nucleotides through such dual configurations. Further structure-guided molecular dynamics simulations may shed light on complex interplay between Arg18, IP6 and nucleotides.

The flexible and asymmetric nature of the hexameric capsomere provides insight into a long-puzzled observation we and others had regarding the small molecule capsid inhibitor PF74. It had been difficult to reconcile concentration-dependent effects of PF74 on CA tube assembly: while low concentrations of PF74 elicited the formation of long CA tubes, higher PF74 concentrations templated the formation of very short and rigid tubes (Extended Data Fig. 7a–f). These differing effects on CA assembly are also consistent with the two distinct antiviral mechanisms exerted by the compound at low and high concentrations (PMID: 27076642). These opposing effects can now be explained in light of the curved hexamer structure (Extended Data Fig. 7g–i). As PF74 interacts with two adjoining CA molecules at the CA_NTD and CA_CTD interface in the crystal structure, it stabilizes the hexamer and promotes capsid assembly. However, the six CA_NTD and CA_CTD interfaces in the curved hexamer are not equal. At low PF74 occupancy, PF74 promotes long tube formation by retaining the asymmetry and flexibility of the CA hexamer; at full PF74 occupancy, binding of PF74 causes induced-fit, resulting homogenization of six CA_NTD and CA_CTD interfaces, therefore losing flexibility and persistence and generating shorter, more rigid tubes. This is supported by observations that PF74 increases the stiffness of isolated cores²⁹ and accelerates capsid opening but stabilizes the remaining lattice in vitro³⁰.

Plasticity of inter-hexamer interfaces

The CA dimer interface has been extensively studied by X-ray crystallography ^4,31 and NMR^28,32, but how the interaction contributes to curvature formation is unknown. A recent cryoET study found that tilts (−1° to 29°) and twists (−12° to 12°) between neighboring hexamers could account for lattice curvature¹². We observed a similar range of tilts (0° to 31°) and twists (−8° to 8°) among five different symmetries, but only if we artificially treated the cryoEM hexamers as rigid and flat. Yet, the three dimer interfaces in the curved assembly appear similar to the planar crystal dimer (Fig. 2a–b). The deviations among three dimers and among five helical assemblies were small, with RMSDs of 0.67 Å-1.18 Å and 0.19 Å-1.39 Å, respectively. The trimer interface was also preserved (Fig. 2c–d). Therefore, the variable curvature of the capsid (0° to 31°) is contributed largely by the asymmetrically curved hexamer itself without significantly altering dimer and trimer interactions, whereas minor adjustments of hydrophobic interactions at the dimer and trimer interfaces provide some plasticity.

Figure 2 |. Plasticity of inter-hexamer interfaces in mature capsid.

a) Overlay of three dimer interfaces from the (−13, 8) helical assembly (color) superimposed with the same interface from a planar crystal (grey, PDB 4XFX). b) Enlarged view of the superimposed dimer interfaces in (a) in ribbon and stick representation. Val181 and Trp184 side chains are shown. c) Superposition of the trimer interfaces from the (−13, 8) helical assembly (color) and PDB 4XFX (grey). d) Enlarged view of the trimer interface in ribbon representation. Ala204 side chains are shown as spheres in the cryoEM model (orange) and planar crystal structure (grey). e) Close-up view of the dimer interface, displaying the interaction between Val181 and Trp184. f) Effects of Val181 and Met185 mutations on in vitro capsid assembly (top; s, supernatant; p, pellet), viral infectivity (bottom; black) and p24 in culture supernatant (bottom; grey); data are mean and s.d. of n=3, measured from independent experiments. Uncropped gel image is shown in the Source Data.

Our cryoEM structure delineates the interaction between Val181 and Trp184 at the dimer interface (Fig. 2e), in addition to the well-characterized Trp184 and Met185 interaction³¹. Although the Val181-Trp184 interaction was previously suggested by NMR experiments²⁸, its contribution to the overall capsid structure can now be appreciated. Substitutions of either Val181 or Met185 greatly impaired capsid assembly in vitro and reduced HIV-1 infectivity by ~10 to >1,000-fold (Fig. 2f). In addition, CA mutant virions exhibited a high frequency of defective particles (Extended Data Fig. 8a–b). Intriguingly, changes in V181 resulted in drastic morphological changes in vitro, from a completely flat lattice with V181C to an extremely curved assembly with V181D (17 nm diameter compared to 50 nm in WT) (Extended Data Fig. 8c–i), reinforcing the critical role of the dimer interface in capsid assembly.

CryoEM structure of CypA/CA tubular assemblies

CypA is important for HIV-1 infection of human cells³³ and disruption of the CA-CypA interaction renders HIV-1 susceptible to potent restriction by human TRIM5α³⁴. We previously proposed a 2^nd non-canonical interface between CypA and CA based on a low resolution cryoEM map and molecular dynamic simulations²⁰. To overcome the challenges of variable CypA binding modes, partial occupancy and capsid asymmetry, we devised a new data processing and 3D reconstruction strategy using symmetry expansion, localized 3D classification, alignment and reconstruction (Extended Data Fig. 2g–j), which yielded structures of CypA complexed with CA tubular assemblies from six helical symmetries to high resolution (Fig. 3, Extended Data Table S1, Extended Data Fig. 9a–b). Localized 3D classification revealed two CypA binding modes: Mode 1 with one CypA located above the CA dimer interface (Fig. 3b, orange) and Mode 2 with an additional CypA above the CA trimer interface (Fig. 3c, magenta). In both binding modes, a single CypA simultaneously interacts with three different CA molecules, one via the canonical CypA binding loop (Fig. 3b–c, green, Fig. 3e–f, CA1) and the other two from neighboring hexamers via noncanonical interactions (Fig. 3e–g, CA2, CA3 and CA2’, CA3’). In particular, the 2^nd binding interface of CypA in Mode 1 overlaps with the site we previously described²⁰ and involves residues Glu43-Lys44-Gly45-Phe46 and Pro123-Ile124-Pro125 from CA2 (Fig. 3g). This interface appears to be weaker than the canonical interface, enclosing 320.7 Ų interaction area compared to 496.5 Ų for the canonical interface. Interestingly, in binding Mode 2, the two CypA molecules always orient the same way, with the canonical binding sites on adjacent protomers within a hexameric capsomere (Fig. 3c).

Figure 3 |. CryoEM structure of CypA bound to mature CA lattice.

a) Surface representation of a CypA-decorated CA tube with (−13, 8) helical symmetry. The circled area includes two CA hexamers (blue) with CypA density (red) above the dimer interface that was extracted for symmetry expansion, localized reconstruction and classification. b) CryoEM structure of CypA/CA in Mode 1: one CypA interacting with two CA hexamers above the dimer interface. The canonical CypA binding loop is colored green. c) CryoEM structure of CypA/CA in Mode 2: two CypA molecules on the CA lattice surface. One CypA (orange) interacts with two CA hexamers above the dimer interface as in (b), whereas the other CypA (magenta) interacts with three CA hexamers above the trimer interface, d) Atomic model of a four-hexamer CA lattice with CypA. e) Enlarged views of the boxed region in d (dashed black) at the trimer interface. The three interacting CA molecules are labeled as CA1, CA2’ and CA3’, where CA1 harbors the canonical interface (shown in green), f) Enlarged views of boxed region in d (dashed red) at the dimer interface. The three interacting CA molecules are labeled CA1, CA2 and CA3. The CypA binding loop on CA1 is shown in green, g) Interacting residues between CA2 and CypA (boxed region in f) are shown in ball and stick representation.

CypA senses specific capsid surface geometry

We have analyzed CypA/CA tubes from different helical symmetries and observed a preferential CypA binding pattern on CA lattices (Extended Data Fig. 9c). To understand the mechanism of capsid pattern sensing by CypA, we aligned the CypA/CA_NTD complexes from 6 helical symmetries (Fig. 4a). All six symmetries with different curvatures present the same binding mode, which is distinct from the crystal structure²² (Fig. 4a, grey, Movie S1). More interestingly, the non-canonical interaction site is highly conserved among the six symmetries (Fig. 4a–b). Therefore, despite variably curved surface lattices, CypA recognizes a specific capsid surface geometry for its interaction. Since three adjacent CA molecules within a hexamer possess different geometry and curvature owing to asymmetry, it is reassuring that among three potential CypA sites above the dimer interface (Fig. 4c, dashed contours), only the site with matching geometry is occupied (Fig. 4c–e, blue). We can further map every individual CypA (which has been classified into Mode 1 or Mode 2) back onto the initial tube from where it originated (Fig. 4f–g). The result clearly illustrates how CypA decorates the CA tubular surface lattice with a specific pattern of binding preference (Fig. 4g, Movie S2). There are significant overlaps between the binding sites for CypA and putative binding sites (4 out of 6) for TRIM5α³⁵, explaining that CypA can block human TRIM5α binding to preassembled crosslinked CA tubes in vitro³⁶, and protects HIV-1 from potent restriction by human TRIM5α in cell culture³⁴. CypA is present at a reasonably high concentration (~58-109 μM) in activated CD4+ T cells (Extended Data Fig. 10), indicating that incoming virions are likely exposed to saturating levels of the host factor.

Figure 4 |. The non-canonical CypA/CA interaction is conserved and defined by CA lattice geometry.

a) Alignment of CypA molecules above the dimer interface from 6 different helical symmetries (−8, 13), (−7, 13), (−13, 7), (−13, 8), (−13, 9) and (−13, 10), shown in two orthogonal views. The crystal structure of CypA/CA_NTD complex (PDB 1AK4) is colored in grey (left panel). The non-canonical interface is marked with a dashed box. b) An enlarged view of the non-canonical interface. c) Schematic representation of three possible CypA positions (dashed contours) above the dimer interfaces in the (−13, 8) CA lattice, d) Alignment of three CA molecules (cyan, blue and pink) from the CA lattice in c, along with a bound CypA (orange), e) Enlarged view of the boxed region in d, showing the difference in distance between Cα of Lys44 (CypA) and Cα of Ilel24 of three CA protomers. f) A cryoEM image of CA tubes decorated with CypA with zoomed-in view of the boxed region, overlaid with experimental CypA positions (only Mode 1 for clarity) projected onto the tube image. g) 3D representation of the tube segment shown in f, where the actual position of each CypA was mapped onto the tube surface according to its 3D class (see Extended Data Fig. 2). Orange and blue CypAs are above the dimer interface; magenta and cyan CypAs are above the trimer interface. The orange and magenta CypAs bind in an upward pointing conformation, whereas blue and cyan CypAs bind pointing downward.

Continuous curvature formation and CypA binding in the conical capsid

It has been proposed that the conical mature capsid spreads its surface curvature by continuously changing dimer and trimer interfaces between CA hexamers^6,12,37. CryoEM structures of curved assemblies argue that these interfaces, on the contrary, are relatively stable. Our results indicate that the key contributor to the curvature is the intrinsically and variably curved hexameric capsomere (Fig. 5a–b), which originates from the highly conserved but flexible linker between the CA_NTD and CA_CTD To investigate whether and how various curvatures observed in helical assemblies represent the local curvature in the native conical capsid, we cross-correlated a conical capsid model (PDB 3J3Q)¹¹ and the local curvature unit of seven CA hexamers from five different symmetries. As exemplified in Fig. 5 for (−13, 8) and (−13, 12) symmetries, the best matched curvature units mapped to distinct regions of the conical capsid (Fig. 5c–d). As expected, we observed that the regions with the lowest cross-correlations correspond to locations of pentamers (Fig. 5c–d, blue). The result suggests that a continuously curved surface can be achieved through variations in local curvatures manifested in helical assemblies. Changes in local curvatures result in exposed surfaces of different sizes and/or geometries, which may alter the accessibility of different host cell factors, such as CypA²⁰, MxB^3,25 and Trim proteins, via specific capsid pattern recognition²,³,²⁰. We illustrated this by mapping CypA binding to the conical capsid surface through cross-correlation based match of underlying local curvature units from conical capsid and CypA/CA tubular assemblies (Fig. 5e, Movie S3). We suggest that the capacity for curvature adaptation is a conserved feature across the broad morphologies of different types of retroviral capsids.

Figure 5 |. Continuous curvature formation in a conical capsid assembly and its role in CypA binding.

a) Seven-hexamer local curvature unit, comprising a central curved, asymmetric hexamer (magenta) and three pairs of neighboring curved hexamers along three helical directions, with hexamers in gold nearly parallel to the tube axis (least curved) and hexamers in blue along the most curved direction, b) Clipped views with a 90° rotation from the boxed regions in (a), enclosing blue-magenta hexamers (top) and (gold-magenta hexamers (bottom). The top panel illustrates that the curvature is primarily derived from curved hexamers (blackline), instead of from tilting and twisting between flat hexamers (red-line). Bottom panel reveals the relatively flat gold-magenta-gold aspect the of local seven-hexamer lattice. c-d) Surface presentations of curvature variation in conical capsid assembly, colored according to the cross-correlation score between a seven-hexamer lattice from a conical core (PDB 3J3Q) and those from helical symmetry (−13, 8) (c) or (−13, 12) (d). The yellow circles mark the area of highest correlation value. e) Surface presentations of CypA (red and blue) binding to the conical capsid (PDB 3J3Q). The positions of CypA (Mode 1 only) on the capsid surface lattice were determined by matching the underlying 7-hexamer units of CypA/CA tubes (Fig. 4f) to the conical capsid model (PDB 3J3Q).

Discussion:

Infections by retroviruses, such as HIV-1, critically depend on their capsids. The capsid is the primary interface between the virus and host cell during viral ingress. Many host cell defense proteins, including the restriction factors Trim5α, TrimCyp and MxB, target the HIV-1 capsid during the early stages of infection and potently inhibit virus replication. These restriction factors appear to function through capsid pattern sensing that specifically recognizes the assembled capsid as opposed to individual CA molecules. While numerous atomic structures of HIV-1 CA domains, monomers, hexamers and pentamers have been determined by NMR and X-ray crystallography methods⁴–⁶,²⁸,³¹, none of these was derived from curved capsid assemblies, limiting our understanding of how the variable capsid curvature is achieved and how host factors specifically recognize the curved lattice pattern to exert their functions. The cryoEM structures of the curved CA hexamer presented here imply a different mechanism underlying capsid formation. Hexamer flexibility suggests that the CA subunits are more easily accommodated as the capsid lattice is built, rather than each CA subunit having to adopt a very precise conformation to fit into a rigid symmetrical hexamer first. Consequently, the binding of host factor CypA to the capsid surface lattice involves additional novel interfaces that are absent in the crystal structure of the binary complex. Information derived from our study will potentially allow designs of more robust therapeutic agents that exploit the intrinsic curvature of the capsid to block HIV-1 replication.

Methods:

The methods section is linked to the online version of the paper at www.nature.com/nature

Extended Data

Extended Data Fig. 1 |. Automated targeted data collection flowchart.

a) The workflow of ArbitrEM, including two steps: target selection and automated data collection. (i) Low-magnification images are firstly acquired to identify holes where tubes are present, and these holes are selected by marking the holes in the images (red crosses); (ii) The images of selected holes are acquired at hole-magnification, centered, and converted into anchor maps; (iii) The acquisition points are marked on anchor maps (boxes), and the beam-image shifts to target the individual acquisition points are calculated. (iv) After all the targets are marked on the hole magnification images, the microscope uses the anchor maps and applies the total beam-image shifts (the stage shift combined with the target-specific beam-image shift) to acquire the high magnification movies. b)A typical low-magnification cryoEM image of CypA-stabilized WT CA assemblies. Scale bar, 2 μm. c) High-magnification image of CypA-stabilized WT CA tubular assembly, illustrating tubes are variable in diameter and easily deformed. d)Gallery of targeted images collected using ArbitrEM. Scale bars, 50 nm.

Extended Data Fig. 2 |. Data processing and 3D reconstruction flowchart for mature CA hexamer and CypA/CA complex.

a) Individual tubes are cropped from motion-corrected micrographs in EMAN2. b) Tubes are sorted based on their diameters. c) The FFT of each tube is calculated and used to determine helical symmetry. In these data, 12 different helical symmetries exist within a 3 nm diameter variation (red box in b). d) Segments from the same helical symmetry tubes are classified in RELION. e) Iterative Real Space Helical Reconstruction and RELION refinement for each helical symmetry are performed. f) Symmetry expansion and localized reconstruction of individual hexamers (dashed circle) from each tube segment are performed, and C2 symmetry is applied. g) After symmetry expansion, the CA density was subtracted, leaving only the local region of CA and CypA for further classification, as highlighted in the dashed magenta box in e). h) Localized reconstruction was performed to calculate the position of each sub-particle and the extracted sub-particles were subject to 3D-classification into 10 classes without alignment, with a regularization T factor of 10 and resolution limit of 8 Å. i) Classes showing clear features of CypA and CA density were selected, further classified into Mode 1 (one CypA, blue or orange) and Mode 2 (2 CypAs, blue-cyan or orange-pink). j) The Mode 2 classes were aligned and averaged (right), whereas CypA above dimer interface from both Mode 1 and Mode 2 were aligned and averaged (left) to yield the final density maps from each helical symmetry.

Extended Data Fig. 3 |. 3D reconstructions of CA hexamer from five helical symmetries.

a) Electron density maps of CA tubes in the absence and presence of CypA and with different helical symmetries, as indicated. b) Fourier-shell-correlation (FSC) plots of CA hexamer density maps reconstructed from five helical symmetries. The highest resolution at FSC=0.143 is 3.6 Å from (−13, 8) symmetry. c) The FSC plot between the refined CA hexamer model and the 3.6 Å cryoEM density map from (−13, 8) symmetry. d) Representative density map from (−13, 8) symmetry overlapped with the refined atomic model, shown are the helices between adjacent CA monomers in different color. The map was contoured at 2σ.

Extended Data Fig. 4 |. Comparison of hexamers from CA A92E mutant and CypA-stabilized WT CA tubes with the same helical symmetry (−13, 12).

a) Overlay of rigid-body refined CA hexamers from CA A92E (cyan) and CypA-stabilized CA WT (gold) tubes. RMSD between two hexamers is 0.354 Å. CA_NTD and CA_CTD were rigid-body fitted into the density map independently. b) Overlay of a seven-hexamer lattice from a CA A92E tube with (−13, 12) symmetry (cyan) and that of a CypA-stabilized CA WT tube with the same symmetry (gold).

Extended Data Fig. 5 |. B-factors of the mature CA hexamer.

a-c) A sausage representation of B-factors mapped onto a CA monomer (a), hexamer (b), and tri-hexamer (c). The CypA-loop, H4-H5 loop, β-hairpin and H10 display high B-factors. d-e) Close-up views of a dimer interface (d) and a trimer interface (e) circled in (c). The width and coloring of the sausage are directly proportional to the B-factor, from blue (−30) to red (−100). CA_NTD, CA helices H1-3, CypA-loop, H4-H5 loop, β-hairpin and CA_CTDH10 are labeled.

Extended Data Fig. 6 |. Asymmetric and flexible configuration of CA in helical assemblies.

a) Distances between opposing CA_CTD domains in CA hexamer (measured between C⍺of N195) from helical symmetry (−13, 8). b) Correlation plot of distances between CA_CTDdomains versus hexamer array angles from five different helical assemblies. c-d) Close-up views of asymmetric β-hairpins (c, orange) and Arg18 residues adopting two conformations at the hexamer center (d). e) Overlay of electron density map with atomic model, showing two Arg18 side chain positions. f) Overlay of three Arg18 side chains in CA tubular structure (gold) with those in crystal structures in the absence of IP6 (blue, PDB 4WYM, PDB 4XFX, PDB 4XFY, PDB 4XFZ, PDB 5HGL). g) Overlay of Arg18 in crystal structures in complex with IP6 (PDB 6bht), same view as in f. 12 CA molecules from two hexamers (colored in pink and cyan, respectively) in one asymmetric unit were aligned on CA_NTD.

Extended Data Fig. 7 |. Binding of PF74 to CA hexamer and its effects on CA tubular assembly.

a) Light scattering of CA assemblies in the presence of PF74 at the indicated concentrations. b) Average number of tubes in each EM micrograph taken at the same magnification, with representatives shown in d-f. c) Distribution of the length of tubular assemblies in the presence of 0 μM (black), 10 μM (green) and 50 μM (pink) PF74. d-f) Representative images of negatively stained CA assemblies in the presence of 0 μM (d), 10 μM (e) and 50 μM (f) PF74. Scale bars, 0.5 μm. g) Atomic model of CA hexamer, clipped to show CA_NTD. Each CA monomer is labeled. The vertical dashed line indicates the tube axis. h-j) Overlay of density map with atomic model, showing the position of Arg173, which is critical for the binding of PF74 to hexamer⁵⁴. Three CA_NTD-CA_CTD interfaces within the CA hexamer are shown between chains A-F (h), chains B-A (i) and chains C-B (j). Intermolecular hydrogen bonds between CA_CTD and CA_NTD are marked by red dashed lines with corresponding distances indicated. k) Overlay of WT and PF74-bound CA crystal structures, colored in grey and cyan, respectively. l) Overlay of three different PF74 pockets from the asymmetric hexamer in (−13,8) helical symmetry with crystal structures(PDB 4XFX and PDB 4XFZ). Arg173 side chains are shown in gold, red and blue, corresponding to panels h-j. The chain C-B interface is more similar to the crystal structure than are the chain A-F and chain B-A interfaces.

Extended Data Fig. 8 |. Analysis of CA mutant on virion morphologies and CA assemblies by TEM.

a) Representative images of mature, immature, empty, and eccentric particle morphologies; particles with 1-3 eccentric nucleoids alongside an otherwise electron-lucent core structure were grouped together. Magnification is 30,000X (scale bar, 100 nm). b) Quantitation of core morphology frequencies (average ± SD for n = 2 experiments) for WT and indicated mutant viruses. More than 200 particles were counted for each. Two-tailed Student’s t-test revealed significant differences for indicated categories versus the WT (****, P < 0.0001; ***, P < 0.001; **, P < 0.01; *, P < 0.05). For V181T, only 14 particles (predominantly immature) were observed in total. V181T was accordingly omitted from the graph and statistical analyses. c-g) Negative-stained images of WT and mutant CA assemblies. V181C forms two-dimensional crystalline sheets, whereas V181D forms tubular assemblies with much smaller diameters compared to WT tubes (17 nm vs 50 nm). Other mutants failed to form ordered assemblies. h-i) Close-up view of WT (h) and V181D (i) assemblies on the same scale. Scale bars, 200 nm in c-g, 50 nm in h-i.

Extended Data Fig. 9 |. 3D reconstruction of CypA/CA complexes. Local resolution estimation of CypA/CA complex from different helical symmetries.

a) Local resolution estimation of CypA/CA complex from four different helical symmetries, colored by local resolution from blue to red. b) Fourier-shell-correlation (FSC) plots of CypA/CA density maps reconstructed from four helical symmetries. The highest resolution at FSC=0.143 is 4.0Å from (−13, 8) symmetry. c) Gallery of CypA/CA complexes. Eight helical reconstructions are shown in cross-sectional view (left) and face-on view (middle). Density maps are colored by the radius, with regions corresponding to CA_CTD, CA_NTDand CypA colored in yellow, green and blue, respectively. The outer diameters of the tubes are indicated. Enlarged views from regions marked by red dashed ovals are shown on the right.

Extended Data Fig. 10 |. Semi-quantitative analysis of CypAlevels in primary CD4+ T cells.

a) CD4+ T cell CypA levels from two different blood donors (labeled D1 and D2) were compared by immunoblotting versus a dilution series of purified recombinant CypA protein. Where indicated, cells were cultured with 5 μg/ml of phytohaemagglutinin(+PHA) for 48 h prior to lysis. Lamin B1 was monitored as a loading control. Approximate ng amount of Cyclophilin A in each of the last four lanes is marked in red color. b)Signal intensity of each band in the dilution series in A was plotted against protein amount to generate a standard curve.