Integrated cryoEM structure of a spumaretrovirus reveals cross-kingdom evolutionary relationships and the molecular basis for assembly and virus entry

Foamy viruses (FVs) are an ancient lineage of retroviruses, with an evolutionary history spanning over 450 million years. Vector systems based on Prototype Foamy Virus (PFV) are promising candidates for gene and oncolytic therapies. Structural studies of PFV contribute to the understanding of the mechanisms of FV replication, cell entry and infection, and retroviral evolution. Here we combine cryoEM and cryoET to determine high-resolution in situ structures of the PFV icosahedral capsid (CA) and envelope glycoprotein (Env), including its type III transmembrane anchor and membrane-proximal external region (MPER), and show how they are organized in an integrated structure of assembled PFV particles. The atomic models reveal an ancient retroviral capsid architecture and an unexpected relationship between Env and other class 1 fusion proteins of the Mononegavirales . Our results represent the de novo structure determination of an assembled retrovirus particle.


In brief
The high-resolution in situ structural analysis of the prototype foamy virus icosahedral capsid and envelope glycoprotein provides insight into the viral infection process and evolutionary history.

INTRODUCTION
The spumaretroviruses, also known as foamy viruses (FVs), constitute the subfamily of Spumaretrovirinae within the Retroviridae family.FVs are distinguished from the Orthoretrovirinae, typified by HIV-1, by their atypical replication cycle, which has similarities to the related family of Hepadnaviridae. 1,28][9] This genetic stability corresponds to a low rate of mutation which is significantly less than in orthoretroviruses and may be in part due to FV apathogenicity and the DNA genome in the infectious particle. 10,11Vs therefore represent a source of insight into the origins and evolution of retroviruses.Simian FVs (SFVs) are capable of zoonotic transmission; prototype foamy virus (PFV) was isolated from humans but originated from a chimpanzee SFV. 1,2,12,13he broad tropism and apathogenicity of PFV, combined with its retroviral ability to integrate genes into the host genome, have attracted interest in FV-based vectors for gene therapy and oncolysis. 14,15ith a similar genome organization to orthoretroviruses, FVs possess the gag, pol, and env genes.The Gag polyprotein forms the internal structural components of the virus particle, oligomerizing to assemble a nascent virus core inside the cell before the budding and release of the enveloped virus particle.In orthoretroviruses, this immature capsid undergoes maturation through multiple proteolytic cleavages by the pol-encoded viral protease, producing MA (membrane-associated or matrix), CA (capsid), and NC (nucleocapsid) proteins, which undergo structural transitions to their mature forms. 16,17By contrast, maturation is far more limited in FVs and occurs prior to budding.Only a single cleavage occurs near the C terminus of PFV Gag, resulting in mature virions that contain a mixture of full-length (pr71) Gag and the cleavage products (p68 and p3 Gag), of which pr71 and p68 Gag are the main internal protein constituents of the virus particle. 18,19Nevertheless, the orthoretroviral nomenclature for Gag-derived proteins can be applied to the corresponding domains of FV Gag, although they are not released from the polyprotein.The N-terminal MA domain, which lacks membrane targeting, associates with membrane-anchored Env to tether the capsid to the budding envelope, the central CA domain oligomerizes to mediate capsid assembly, and the C-terminal NC domain packages the genome within the capsid. 20he FV Env polyprotein forms the trimeric transmembrane surface glycoproteins of the virus, which can organize as a hexagonal lattice on the virion surface 21 and mediate receptor recognition and membrane fusion during virus entry into the host cell. 1,22,238][29] The SU subunit contains the receptor-binding domain (RBD). 22Although FV Env binds to heparan sulfate (HS) as an attachment factor, a specific protein receptor required for cell entry remains to be identified. 30,31PFV can enter cells via fusion with plasma or endosomal membranes, indicating that HS binding, receptor binding, and/or low pH may be involved in virus entry. 32,33tructural studies of FVs will provide insight into FV-based vector systems, the mechanisms of virion assembly and fusion, and the origins and evolution of retroviruses.To date, these much-needed high-resolution data have been limited to structures of individual central CA 34 and N-terminal MA-like domains 29 of PFV Gag, which inform on the evolutionary relation-ship of FV to orthoretroviral CA 34 and the dimerization of Gag MA and its interaction with Env LP, but do not place this information in the context of assembled particles.Regarding Env, the structure of the RBD of SFV Env SU revealed a novel fold and a putative HS binding site. 35The structure and arrangement of Env on the virus surface has been studied at low resolution. 21ere we have applied cryogenic electron microscopy, in situ single-particle analysis, tomography, and subtomogram averaging to intact PFV particles, revealing the atomic structures of Env and the icosahedral Gag CA core in situ, before recombining the structures into an integrative model of the assembled virus particle.

Structure of the Env trimer in the virus envelope
The structure of an Env trimer was determined by singleparticle reconstruction from 2D images of PFV virus particles to 2.9 A ˚resolution, permitting de novo atomic model building (Figures 1A-1D; data processing described in Figures S1 and  S2; Tables S1-S3).Env (domain organization shown in Figure 1A) monomers consist of the membrane-distal receptor-binding domain (RBD, SU residues 217-570), a central domain (C, SU residues 163-216, TM residues 572-712), a lattice-forming domain (L, SU residues 127-162, TM residues 713-888), a membrane-proximal external region (MPER, TM residues 889-958), and a transmembrane domain composed of the single-span transmembrane helices of LP and TM (LP residues 59-89 and TM residues 959-982) (Figures 1D and 1E).The protomers (legend continued on next page) associate with a left-handed twist about the 3-fold axis (Figure 1C), enclosing separate cavities at the top and bottom of the ectodomain (Figures 1B and 1C).The trimer-of-heterotrimers is stabilized by extensive interactions between subunits, burying a total of $31,000 A ˚2 surface area.N-linked glycans were observed at 14 sites predicted using NetNGlyc 36 and previously experimentally confirmed, 37 excluding the position at LP residue Asn25, which is unresolved and lies inside the virus particle.Of these 14 sites, 11 had satisfactory density for modeling of the glycans (Figures S3A and S3B) as the high-mannose (LP/TM) and complex types (SU), determined previously by enzymatic digestion experiments. 37ceptor-binding domains The SU RBDs sit above the TM subunits at the trimer apex, partially enclosing the upper cavity (Figures 1B and 1C).The RBD consists of upper (residues 244-312, 375-494) and lower (residues 217-243, 313-374, 495-570) subdomains and is superimposable with the crystal structure of the gorilla SFV Env RBD. 35Notably, the conformation of the essential Asn391 glycan 37 is nearly identical to its equivalent in the gorilla SFV Env (Figure S3C).The conserved positively charged surface in the lower RBD subdomain is also observed (Figure 2A) and is likely to be the site of association to cell surface heparan sulfate (HS) which mediates cellular attachment of PFV. 30,31,35The RBD regions which are situated such as to potentially interact with neighboring RBDs have poor cryoEM map density due to disorder, resulting in them being either omitted from the model (residues 261-267, 413-431) or modeled with high temperature (B-) factors (residues 447-459) (Figure 2B).This suggests high structural variability or mobility of these regions as observed in the gorilla SFV Env RBD structure. 35

Env central and lattice-forming domains
The central domains are organized around a 6-helix bundle separating the upper and lower cavities in the trimer (Figures 1C and 1D The site of association with neighboring Env trimers is found at the lateral extremity of the lattice-forming domain.The C-terminal strand of LP (residues 101-116) packs against the outer surface of this domain.

The fusion peptide
Following cleavage, the N terminus of TM is at residue Ser572 24 and hydrophobic residues C-terminal to this have been pro-posed as a fusion peptide 23,38 (Figures 1A and 1D).The TM N terminus is only resolved after Ala578 and is solvent exposed up to residue 580 (Figure 2C), but residues 581-595, which contain the conserved hydrophobic stretch indicative of a fusion peptide, are sequestered by the RBD C terminus (SU residues 562-568) and the extended C-terminal loop of the lattice domain (TM residues 848-855).

Env type III transmembrane region
The base of the Env trimer consists of the MPER and transmembrane regions (Figure 2D).The single-pass transmembrane helices of each TM (residues 959-982) and LP (residues 59-87) are associated in an antiparallel arrangement, spanning the membrane bilayer (thickness 30 A ˚) with a central TM helical bundle on the 3-fold axis associated with peripheral LP transmembrane helices.Neither of the intraviral domains of LP (residues 1-58, including the N-terminal Gag-interacting peptide) or TM (residues 983-988) were resolved.The MPER of TM (residues 889-958) adopts an arrangement where three consecutive a helices of TM sit at an oblique angle to the membrane bilayer, with the last MPER helix (residues 947-957) lying with one hydrophobic face immersed in the outer bilayer (Figure 2D).The organization of the PFV Env MPER is more reminiscent of the HIV Env MPER than the corresponding region of influenza HA 39,40 and is also similar to the MPER in the postfusion structure of the unrelated class 3 fusion protein gB of herpes simplex virus 1. 41 The PFV MPER appears to be structurally far more rigid, forming a relatively broad and stable outer-leaflet-embedded platform but with additional buttressing provided by the LP subunit.Correspondingly, no tilting of the PFV Env ectodomain with respect to the membrane normal is observed, as seen for influenza HA, SARS-CoV-2 S, and HIV Env. 40,42,43he transmembrane density also revealed bound phospholipid and cholesterol molecules in the core of the transmembrane domain (Figure 2D).These bound lipids may have roles in fusion activity, as they are in contact with residues Lys959 and Pro960 which have previously been implicated in regulation of Env fusogenicity. 44There is one phospholipid and one cholesterol per TM monomer, with an additional central lipid (not modeled) seen as a tripodal density on the 3-fold axis approximately level with the sterol moiety of the cholesterol molecules (Figure 2D).The transmembrane helices of TM make direct contact with each other only in the inner portion of the bilayer, as the lipids prevent them from making contact near the outer leaflet.LP interacts with the bound lipids and the transmembrane helices of TM at the outer bilayer leaflet and packs between adjacent MPERs of TM immediately above the membrane surface (Figure 2D).However, the splayed LP transmembrane helices sit at a $34 angle to the membrane normal and do not interact with TM in the inner leaflet of the bilayer or presumably in the intraviral domains.The conserved consecutive Cys residues 61-62 in LP are found at (D) The transmembrane and MPER domains of LP and TM.Left: Overview.The membrane density (gray) from a lowpass-filtered Env map is shown in a cutaway view.Bound lipids are shown as beige sticks.Center: Bound lipid molecules and EM density in the transmembrane domain.Lipid densities (contoured at 6.33 s, surrounding densities omitted for clarity) are colored beige (modeled) or orange (unmodelled).Right: Details of lipid binding and MPER-membrane interaction.The membrane-interacting residues 947-958 and key lipid-interacting residues Ile905, Asp906, Lys959, and Pro960 are shown as sticks and labeled, with hydrogen bonds shown as cyan dashed lines.See also Figures S5. (legend continued on next page) the inner bilayer leaflet, and the corresponding residues are palmitoylated in bovine FV, with an essential role in membrane fusion, virus assembly, and budding. 45However, we do not observe density for palmitoylation, although these residues lie in a map region with poorer density quality which may obscure the detection of these modifications.
PFV Env contains an ancient membrane fusion module found in diverse viruses Structural homology searches of the PDB using DALI 46 revealed no matches with orthoretroviral Env.However, PFV Env did show structural similarity with the F membrane fusion glycoproteins (sequence identities $10%) from the Paramyxoviridae and Pneumoviridae families (Figure 3A; Table S4).The structural similarities were confined to two domains (Figures 3B and S4): the central domain, with the best match found against the RSV F protein (PDB: 5EA6, Z score 8.3, RMSD 3.6 A ˚over 129 aligned residues), and the lattice-forming domain, with the best match found against the PIV5 F protein (PDB: 2B9B, Z score 5.6, RMSD 4. 4 A over 138 aligned residues).In both Env and F the trimers enclose a central cavity, but the structural similarity does not extend to the C-terminal region of TM, which in PFV forms the more compact a-helical MPER region when compared to the extended coiledcoil stalk seen in prefusion F1. 47 Further examination of the subunit topology suggests that the RBD of PFV SU (residues $217-558) can be viewed as a C-terminal insertion into the homologous F2 subunit (Figures 3C and S4).
In the lattice-forming domain, three disulphide bonds of PFV Env TM have strictly conserved equivalents in the paramyxo-/pneumoviruses, each corresponding to a topologically conserved subdomain (Figures 3B, 3C, and S4).The cysteine residues in these disulphide bonds are also strictly conserved among FVs, including the endogenous coelacanth and reptilian FV Env, 5,52 suggesting a conserved structural role.
An important feature for comparison in the central domain is the position of the putative fusion peptide. 23In PFV Env TM it is located on the exterior of the trimer, similar to those in F1 of the paramyxoviruses PIV5 and Nipah virus, 50,53 while in pneumoviruses the fusion peptides of RSV and hMPV are found in the interior cavity of the F trimer 54,55 (Figure 3D).Comparison of our structure with the low-resolution map of PFV Env 21 in which the SU-TM cleavage is mutated (Figure S5A) confirms that no significant structural change takes place in PFV Env after cleavage.
The recent discovery of the ancient endogenous lokiretrovirus lineage, thought to also possess an F-like Env, 56 prompted us to investigate any structural relationship with FV Env.We predicted the structure of lokiretrovirus Env using Alphafold2 57 and found that it is similar to PFV Env in the central and latticeforming domains (Table S4) but lacks an RBD, making lokiretrovirus Env more similar to paramyxo-/pneumo-virus F than to FV Env.
The Env lattice is assembled by strand exchange between neighboring trimers Dimerization of Env trimers builds the higher-order Env lattice via interactions of the lattice-forming domain.A reconstruction of Env dimers-of-trimers at 3.7 A ˚resolution (Figure 4A) allowed us to build a model of the interface between neighboring Env trimers in the lattice (Figures 4B and 4C).This revealed a strand-exchange mechanism where the N-terminal SU residues Ser127-Val134 extend into the neighboring trimer, donating a strand to the 5-stranded b-sheet formed by the symmetrically opposed SU and the TM of the neighboring trimer (Figures 4B and 4C).LP, SU, and TM all contribute to this interaction.Interestingly, four N-glycosylation sites are found surrounding this interaction, some of which stabilize the inter-trimer interaction (Figure 4C).Mutation of the Asn833 glycosylation site, which participates in the interaction, was previously found to impair particle release and infectivity, 37 although mutation of the nearby Asn141 glycosylation site did not have a significant effect.Contacts between neighboring trimers have also been described in Nipah virus and RSV, where F trimers form dimers-of-trimers, 51,53 although not by strand exchange (Figure 3E).
The inter-trimer connection is flexible, allowing the formation of pentamers-of-trimers as well as hexamers-of-trimers in the Env lattice as seen in our cryoET data (Figure 4D).The dimer-of-trimers structure was fitted into the hexamer-and pentamer-of-trimer maps, revealing the flexibility required to accommodate the varying geometry of inter-trimer interactions (Figures 4D and 4E).Multibody analysis 58 showed that the flexible linking arms forming the inter-trimer interface allow relative tilting, twisting, and rotational motions of neighboring Env trimers, enabling the Env lattice to adopt varying local arrangements (Figure 4B, 4D, and 4E).
The last resolved LP residue (Tyr116) is located proximal to both observed N termini of SU (Figure 1E), and the residues 117-126 immediately preceding the LP-SU cleavage site are not resolved, leading to an ambiguity over how the lattice is initially assembled.Two models for Env lattice assembly can be proposed, depending on whether strand exchange occurs before or after the LP-SU cleavage.In the first scenario, which we consider to be the most likely (Figure 4F), uncleaved Env forms trimers which then associate at the site of virus budding.Subsequent LP-SU cleavage would then release the SU N termini to then carry out strand exchange.In the second scenario, strand exchange occurs before cleavage; however, this would require LP and SU of the same Env molecule to be associated with different trimers before cleavage.
Structure of the capsid inside the virus particle PFV Gag (domain organization shown in Figure 5A) forms an icosahedral capsid within the virion, composed of the Gag CA domain with N-terminal (CA-NTD, residues 303-389) and C-terminal (CA-CTD, residues 390-477) subdomains (Figure 5B).The (D) Fusion peptides (pink) indicated on the ectodomain trimers of PFV Env (left), PIV5 F (middle, PDB: 4GIP), 50 and RSV F (right, PDB: 7LVW). 48For each trimer, one monomer is shown as a cartoon while the remaining two are shown as surfaces.(E) Dimers-of-trimers of PFV Env (top, SU RBDs omitted for clarity) and Nipah virus F (bottom, PDB: 8DO4). 512-fold (pseudo)symmetry axes are indicated by black ovals.See also Figures S4 and S5, Table S4.capsid structure was resolved at an overall resolution of 3.9 A (Figure 5C), having maximum and minimum diameters of 640 A ånd 560 A ˚and enclosing a volume of $91,000,000 A ˚3.No ordered density for the Gag MA, proline-rich, or NC domains was observed.The capsid has a T = 13 dextro architecture, corresponding to 780 Gag CA monomers and three types of capsomer: a 5-fold symmetric pentamer, as well as two hexamers (hexamers 1 and 2) which lack cyclic symmetry (Figures 5C-5E).The CA layer is thin ($30 A ˚) and faceted, displaying a cap around each 5-fold vertex with high intercapsomer tilt 59 (pentamer-hexamer 1 and hexamer 1-hexamer 1) and a flatter region around each 3-fold axis which is separated from the tilted cap by a boundary of high intercapsomer twist (hexamer 1-hexamer 2) (Figures 5C, 5D, S6A, and S6B).Localized reconstructions of the three capsomer types improved the resolution to $3.4 A ˚, enabling fitting and refinement of the previously determined monomeric Gag CA solution NMR structure (PDB: 5M1H). 34FV Gag capsomers are formed by extensive interactions between CA-NTDs, between CA-NTDs and CA-CTDs, and between CA-CTDs.They contain a central pore in the CA-NTD ring (Figure 5E), which in the hexamers is of sufficient diameter (7.0-7.23][64] By contrast, the pores of the pentamers are too constricted (1.7 A ˚) to permit entry even of water molecules.The pentamers are more tightly associated (955 A ˚2 buried surface area per chain) than the hexamers (768 A ˚2/704 A ˚2 buried surface area per chain for hexamers 1/2, respectively), which contain asymmetrically distributed gaps between monomers in addition to the central pores (Figure 5E).
While the structural similarity of PFV CA to that of orthoretroviral and retrotransposon CA has been noted previously, 34,65 we observe differences in both the CA monomer and the capsid architecture.In the CA monomer, the CA-NTD and CA-CTD interact extensively with each other in PFV (Figure 5B), contrasting with the case in orthoretroviruses and retrotransposons where the NTD and CTD of a given monomer typically make little to no contact with each other.In the assembled PFV capsid the CA layer is flatter, thinner, and smoother than that seen in orthoretroviruses, because the CA-NTD and CA-CTD in both pentamer and hexamer lie at almost the same radius with respect to the capsid center.By contrast, in retrotransposon-derived and orthoretroviral capsids, the centers of mass of the CA-NTD and CA-CTD layers are displaced vertically along the 6-or 5-fold axes with respect to each other, giving rise to two layers of organization (Figure 5F).Correspondingly, the domains are also more laterally displaced with respect to each other perpendicular to the capsomer 6-or 5-fold axes than in orthoretroviruses.Comparison of these displacements for orthoretroviral, endogenous retrovirus and retrotransposon CA pentamer structures 59,60,66 shows an inverse correlation (Figure 5G; Table S5).
Assembly of the capsid shell is mediated exclusively through dimeric and trimeric CTD-CTD interactions, forming a cage which links neighboring capsomers and accommodates varying tilt or twist orientations between them (Figures 5H, S6A, and  S6B).This contrasts with mature orthoretroviral and retrotransposon capsids, which are primarily assembled via dimeric CTD interactions linking neighboring capsomers, with only minimal trimeric CTD interactions also observed.Furthermore, both intercapsomer interfaces (CTD dimers and trimers) are almost contiguous in PFV, making a continuous interface around the perimeter of every capsomer (Figure 5I) differing from the more separated CTD dimer and trimer interfaces seen in orthoretroviruses and retrotransposons.

Conformational variability of Gag CA accommodates the requirements for capsid assembly
The conformational plasticity required to assemble a faceted T = 13 icosahedron is distributed at multiple levels, within and between capsomers, across the intramolecular NTD-CTD, intracapsomer NTD-NTD, NTD-CTD, and CTD-CTD, and intercapsomer dimeric and trimeric CTD interfaces (Figure 6).The plasticity of the interfaces results in residues that participate differently in interactions at different locations in the capsid.This can be expressed by a per-residue equivalence score,  (legend continued on next page) similar to a previously described per-interface measure of quasi-equivalence, 67 denoting the frequency with which a given residue participates in a particular interaction across the capsid (see STAR Methods).This score identifies key invariant residues that are used in different interfaces (Table S6; Figure S7) and those that make interactions only in more specific contexts (Figure S7).According to the measure described above, high plasticity is particularly apparent in residues proximal to the NTD-CTD linker which participate in the NTD-CTD intramolecular, intracapsomer NTD-CTD, and intercapsomer dimeric CTD-CTD interactions (Figure S7).Averaging the equivalence scores of interface residues allows quantitative comparison of the plasticity of the different interface types (Figure S7).This shows the CTD trimer interface to be the least plastic, followed by the intracapsomer interface, the CTD dimer interface, and finally the most plastic, the intramolecular NTD-CTD interface (Figure S7).Superposition of the 13 Gag CA monomers in the asymmetric unit (ASU) revealed three predominant conformations (Figures 6A and S7A).The conformations are distinguished by the relative pose of the CA-NTD and CA-CTD, mediated by the high plasticity of the interdomain interface, as well as the conformations of the inter-domain linker (residues 380-393).Interestingly, the ''loop-turn'' conformation observed in the pentamer is also found in both hexamers (Figure S6C), suggesting that pentamer formation does not require a specific conformational switch in the CA domain.Plasticity in intracapsomer NTD-NTD interactions is mediated by the loop consisting of residues Gly356-Ser362 (Figures 6B and S7B), but a specific conformation of this region unique to the pentamer is also not observed.The intercapsomer dimeric CTD-CTD interfaces are rather small and highly plastic, with the NTD-CTD linker playing a significant role (Figures 6C and S7C).However, the trimeric CTD interfaces are larger and less structurally plastic than the dimeric interfaces (Figures 6C and  S7C), containing previously identified capsid assembly motifs (Figure S7C). 34,68tegrated structure of a PFV virion In order to obtain an integrated structure for whole virions, we applied cryoET and subtomogram averaging to Env and Gag CA assemblies within PFV particles (Figures 4D, 7A-7C, and  S8; Tables S1 and S7) and additionally applied rotational averaging to the capsid map to obtain a radial density function for entire PFV particles (Figure 7D).CryoET datasets were acquired from PFV samples containing either WT Gag (pr71/p68/p3 Gag) or p68 Gag (p68 Gag only).No significant difference in capsid ar-chitecture or virus morphology was observed between the datasets (Figure S8).
Subtomogram averaging of Env in the WT Gag cryoET dataset produced maps of a hexamer and a pentamer of Env trimers at 10 and 12 A ˚resolution, respectively (Figure 4D), which accommodate the high-resolution Env trimer structure well.Backplotting of the maps into the tomograms revealed the organization of the Env lattice on virus particles, giving a hexamer-hexamer lattice spacing of approximately 160 A ˚and with an average of 27 but up to 240 prefusion Env trimers on each virus particle (Figure 7B).The arrangement of Env trimers in the lattice orients the HS-binding surfaces of the SU RBDs of neighboring trimers to face each other (Figure 4B), at a minimum distance of $33 A ˚.
This may permit cooperative binding of two or more trimers to the same HS or receptor molecule, increasing the avidity of the interaction and attaching the virus more strongly to the cell surface.The observed curvature and spacing of the Env lattice is geometrically compatible with a T = 13 icosahedrally symmetric lattice at the envelope radius of 500 A ˚, which would contain 260 Env trimers in a 1:1 Gag:Env stoichiometry, greater than the average number of Env trimers observed in the cryoET data, and could completely cover the virion surface (Figure 7C).This idealized arrangement would also require a symmetric distribution of Env hexamers and pentamers.Given our experimental data, which feature asymmetrically distributed pentamers and hexamers, significant empty areas, and lattice defects (Figures 4D and 7B), we favor a model of the PFV envelope in which locally ordered patches of Env lattice, with asymmetrically distributed pentamers and hexamers of Env trimers, decorate the virus membrane without completely covering it and are sub-stoichiometric with respect to Gag.Analysis of the radial density profile of the rotationally averaged capsid map shows distinct peaks corresponding to the averaged radial position of each Gag domain, the viral envelope, and the packaged genome.The profile has three shells of high density at radii of $300 A ˚, $400 A ˚, and $500 A ˚(Figures 7D  and 7E).The 300 A ˚and 500 A ˚layers are explained by the capsid and by the virion membrane, respectively.The layer at 400 A likely corresponds to the Gag MA layer.A distinct peak of density at 260 A ˚, with an interior shoulder, likely represents the Gag NC domain and packaged nucleic acid.Placement of 780 monomers from the Gag MA domain crystal structure 29 with their head domains at the radius of the $400 A ˚peak requires that they pack under a level of spatial constraint, consistent with the layer of high protein density observed (Figure 7D).Moreover, the observation that Gag MA is the determinant of FV restriction by TRIM5a, which recognizes capsid lattices, 69 is also (F) Comparison of relative radial displacements between CA-NTD and CA-CTD in PFV, retrotransposon-derived, and orthoretroviral capsid pentamers.Left, PFV; middle, dArc1 (PDB: 6TAR) 60 ; right, HIV-1 (PDB: 5MCY). 59Pentamers are viewed perpendicular to the 5-fold axis.Colored circles represent centers of mass (CoMs) for the CA-NTD and CA-CTD rings.Dz is the relative vertical displacement and Dxy is the relative lateral displacement (see STAR Methods).See also Table S5.consistent with a degree of organization within the MA layer. 29,70nspection of the cryoET data reveals only rare connections between the MA layer and the envelope (Figure 7A), suggesting that a sub-stoichiometric interaction between the Gag MA head domains and Env LP tethers the viral core to the envelope, at least after virion assembly.

Evolution of retroviruses
The integrated cryoEM structure of PFV provides insight into the FV life cycle and its evolutionary relationships to other viruses.The capsid's native icosahedral structure is a fullerene shell assembled via interactions of the Gag CA domain.By contrast with the orthoretroviruses, the PFV capsid is remarkably flat and smooth and shows a different arrangement; though similar to a mature orthoretroviral CA arrangement, the intermolecular CA interfaces are more extensive than in orthoretroviruses.
The two-domain architecture of CA is thought to have arisen from an ancient domain duplication, 65 with subsequent divergence giving rise to the CA-NTD and CA-CTD.The CA subdomains are more similar to each other in PFV and retrotransposons than in orthoretroviruses.In orthoretroviruses, the CA-NTD is larger and has additional structural elements, such as the N-terminal b-hairpin and additional loops, which interact with cellular factors such as TRIM5a, CypA, and CPSF6. 69,71,72Moreover, the orthoretroviral CA-NTD forms an additional layer above the CA-CTD layer, suggesting that different modes of capsid assembly and maturation in FVs and orthoretroviruses may produce different structural constraints on capsid architecture.
The structure of PFV Env reveals that an ancient class 1 membrane fusion module has been acquired by diverse groups of enveloped viruses.The structural similarity of PFV Env to paramyxo-/pneumo-virus F is surprising, given the lack of any other observed relationship between the Spumaretrovirinae and the Mononegavirales and the fact that they are classified into different viral kingdoms. 73The core fold shared by PFV Env and F is also present in the S membrane fusion proteins of the coronaviruses. 74These class 1 fusion proteins from disparate viral taxa share a common viral or cellular ancestor and are likely to be the result of several horizontal gene transfer events, such as is thought to have occurred for orthoretroviral Env and filoviral GP. 75 FV Env and orthoretroviral Env share the same overall class 1 organization but differ in detailed structure.The recent discovery of a class 2 membrane fusion protein in an endogenous retroelement from the belpaovirus family, 76 as well as the large structural differences between spuma-and ortho-retroviral Env, suggests that the retroviruses and retrotransposons have acquired envelope genes from different sources on multiple occasions over evolutionary history, resulting in the currently observed GP-type (class 1, similar to filovirus GP; gamma-/beta-type Env), F-type (class 1, similar to pneumo-/paramyxo-virus  S6. F; spuma-/loki-retrovirus Env), and G-type (class 2, similar to Phenuiviridae G C ; belpaovirus Env) Env glycoproteins in the Ortervirales order.However, the conserved structure seen in Env of ancient lokiretroviruses and endogenous foamy viruses demonstrates that the PFV Env structure presented here is representa-tive of Env of the earliest FVs.Spuma-/loki-retroviral Env may have been acquired prior to or concurrent with the divergence of lokiretroviruses from retroviruses. 56,77Orthoretroviral Env may have existed prior to this divergence, or been acquired afterward, replacing spuma-/loki-retroviral Env.

Env-mediated virus entry
The structural similarity of PFV Env to paramyxo-/pneumo-virus F allows us to propose a similar mechanism of membrane fusion (Figure S5B).The domains with structural similarity to PFV Env are found in both prefusion and postfusion forms of F. 49,54 They do not undergo dramatic refolding within the domains or relative to each other (Figure 3A) and can be thought of as a relatively invariant fulcrum during the large-scale conformational changes that mediate membrane fusion.Pneumo-/paramyxoviral F2 remains associated with F1 throughout membrane fusion, 49,78,79 suggesting that PFV Env SU also remains associated to TM via its N-terminal region.It is possible that Env LP interactions with the MPER and transmembrane regions help to stabilize the prefusion Env conformation, and LP may dissociate during membrane fusion to permit the C-terminal region of TM to refold. 21The observation of lipids bound in the core of the transmembrane domain, associated with conserved residues of functional importance, 44 suggests that lipids may have pivotal roles in the virus assembly and/or membrane fusion functions of Env, potentially by modulating the stability or dynamics of the transmembrane domains as has been suggested for HIV Env and influenza HA. 80,81 In paramyxo-/pneumo-viruses, the RBD function is carried out by a separate receptor-binding protein.While these receptorbinding proteins do not have structural similarity to the PFV Env RBD, [82][83][84][85] they are reported to pack against the apex of the F trimers on the virus surface, in a similar position to the PFV RBD, 86 which can be thought of as a C-terminal insertion into the homologous F2 subunit (Figure 3C).In prefusion F structures, the heptad repeat 1, immediately C-terminal to the F1 fusion peptide, packs against the trimer apex instead (Figures 3A and 3D). 54,87onversely, in coronaviruses there are additional insertions of the S1 NTD, domain 0, and an RBD unrelated to that of FV SU into the region homologous to the N-terminal region of F2. 88 By analogy to paramyxo-/pneumo-/corona-viruses, FV RBD binding to its receptor may trigger structural transitions associated with membrane fusion.[91]

Virion assembly
The virion structure contains features important for aspects of PFV virus assembly that differ from orthoretroviruses.Firstly, PFV Gag does not appear to undergo the same process of structural maturation as that seen for orthoretroviral Gag.The structures of the WT Gag and p68 Gag (truncated at the cleavage site) capsids are identical, and previous studies have shown the identical appearance of intracellular and extracellular capsids, 92 suggesting that mature capsids pre-assemble in the cytoplasm before budding.
Secondly, Env has a significant role in assembly through recruitment of the capsid via the LP cytoplasmic region, in contrast to orthoretroviruses where direct Gag-membrane interaction is required for budding.Env lattice formation on the virion surface is also an important driver of budding; subviral particles can still be released from cells by Env expression alone. 93,94PFV Env covers the virion surface more completely and with a greater number of Env trimers than observed in HIV-1. 95By comparison, paramyxo-/pneumo-virus glycoprotein arrangements may be coordinated by the lattice of matrix protein underlying the membrane, a feature not observed in PFV particles. 96,97inally, the combination of our cryoET observations and our high-resolution structures suggests that the partially ordered Gag MA layer is not membrane associated and is linked to the CA layer by an unstructured Pro-rich region.The flexible attachments of MA to CA and of the LP cytoplasmic domain to the transmembrane domain could promote capsid recruitment in the absence of direct membrane targeting of MA.Local concentration of flexibly linked LP at the inner face of the membrane may increase the efficiency of recruiting pre-assembled capsids.Interactions via LP couple the external Env and internal Gag assemblies.
Structural comparisons between FVs and other viruses provide insight into the evolution of (spuma)retroviruses and how structural and functional differences are associated with specific modes of assembly and entry.Further structural investigations of distant homologies between viruses from different taxonomic kingdoms may offer new insights into virus evolution.In the future, elucidation of the structural basis for the broad tropism of FVs awaits the identification of the viral receptor and the mechanism by which membrane fusion is triggered, all of which will be important to application of FV-based vectors in therapeutic settings.

Limitations of the study
Our cryoEM/cryoET study provides information on structural components of the virus (including the prefusion form of Env and the icosahedral capsid) and their organization in the particle but determines detailed structural features or molecular identity only where we have been able to apply image averaging.Given that we do not observe order within the core of the particles and only a diffuse MA domain layer, we are unable to provide detailed features important for understanding their roles in genome and viral enzyme packaging, egress, and interaction with host cell factors.Our study is based on purified particles, whereas additional structural interactions important to assembly, such as ordered Gag-Env interactions, may be observable only during viral budding.These additional features of the foamy virus life cycle may be addressed by high-resolution imaging of virion assembly directly in infected cells.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

METHOD DETAILS
Virus preparation PFV samples were produced as described previously. 98,99In brief, a 4-plasmid vector system was used to transfect HEK293T cells with the viral env, gag and pol genes and a genome transfer vector.Virus particles were then purified from the FCS-and cell-free tissue culture supernatant into phosphate buffered saline (PBS) by filtration through a 0.45mm filter, sucrose density gradient centrifugation (SW-32Ti rotor, 5 mL 20% sucrose cushion under 35 mL filtered supernatant, 90 min at 25,500rpm/82,600g), resuspension of the viral pellet in PBS, and a final concentration using centrifugal concentrators (Amicon, 0.5 mL volume, 100kDa cutoff).Samples were frozen in liquid nitrogen, then shipped and stored at À80 C until use.The p68 and WT Gag samples for tomography were produced with the 4-plasmid vector system using polyethylenimine (PEI) transfection and wild type 293T packaging cells, while the WT Gag sample for single particle analysis was produced with the 4-plasmid system using calcium phosphate transfection and glycosaminoglycan-deficient 293T-25A packaging cells. 98Particle production was verified by Western blotting and cell infection assays.

CryoEM sample preparation
For cryoET samples, Quantifoil grids (R2/2, 200 mesh Cu) were glow discharged in an amylamine atmosphere (25mA, 30 s) using a EmiTech K100X glow discharger (Quorum Technologies).After thawing on ice, PFV samples were mixed with 10 nm gold/Protein A fiducial beads (BBI Solutions) at a ratio of 1 part fiducials to 5 parts virus sample. 2 mL of sample was applied to the grid in the chamber of a Vitrobot mark 3 (FEI) maintained at 100% relative humidity and 4 C.The grids were then blotted for 1 s using a double layer of filter paper, and another 2 mL of sample applied.After a further 4 s blot, the grids were then plunge frozen in liquid ethane, then transferred to a grid box and stored in liquid nitrogen until use.
For the WT Gag PFV single particle samples, holey carbon Quantifoil grids (R2/2, 200 mesh Cu) were glow discharged in an amylamine atmosphere (25mA for 30 s) using a EmiTech K100X glow discharger (Quorum Technologies).After thawing on ice, 2.5 mL of PFV sample was applied to the grid in the chamber of a Vitrobot mark 3 (FEI) kept at 100% relative humidity and 4 C.The grid was then blotted for 4 s using a double layer of filter paper and plunge frozen in liquid ethane, before being transferred to a grid box and stored in liquid nitrogen until use.

CryoEM and cryoET data collection
All cryoEM and cryoET datasets were collected on a FEI Titan Krios microscope at 300 kV, using a Gatan K2 Summit camera in counting mode and Gatan GIF Quantum energy filter in zero-loss mode with a slit width of 20 eV.
For the WT Gag single particle dataset, the nominal magnification was 130,0003, corresponding to a calibrated pixel size of 1.08 A åt the specimen level.Automated data collection was carried out using EPU software (FEI), with aberration-free image shift and a total applied electron dose of 48 e À /A ˚2, fractionated over 32 movie frames (dose of 1.5 e À /A ˚2/frame) in a 10 s exposure.Applied defocus ranged from À1 mm to À4 mm, incremented in 0.5 mm steps throughout the data collection.
For the WT Gag cryoET dataset, the nominal magnification was 105,0003, corresponding to a calibrated pixel size of 1. 38 A ˚at the specimen level.The data collection was carried out using FEI Tomography 5 software.A dose-symmetric tilt scheme 125 was used, starting at 0 tilt and going to ±60 tilt in 3 increments, with grouping of 2. The total applied dose per tilt series was 107 e À /A ˚2, corresponding to a dose per tilt image of 2.62 e À /A ˚2, fractionated over 4 movie frames.The exposure time at each tilt was 1s, at a dose rate of 5 e À /pix/sec.Nominal applied defocus values ranged from À2.0 mm to À4.5 mm, incremented in 0.5 mm steps throughout the data collection.
For the p68 Gag cryoET dataset, the nominal magnification was 64,0003, corresponding to a calibrated pixel size of 2. 22 A ˚at the specimen level.The data collection was carried out using FEI Tomography 4 software.Two tilt schemes were used; both tilt schemes were bidirectional, starting at 0 tilt and going to either À42 or À54 , then returning to 3 and continuing onwards to 54 , in 3 increments.The total dose for the whole tilt series was either 61 e À /A ˚2 (À54 to +54 scheme) or 54 e À /A ˚2 (À42 to +54 scheme), with a dose per tilt image of 1.65 e À /A ˚2, fractionated over 4 movie frames.The exposure time at each tilt was 1 s, at a dose rate (incident on the camera) of 8 e À /pix/sec.24 tilt series were collected with the À54 to +54 tilt scheme, followed by an additional 134 tilt series with the À42 to +54 tilt scheme.158 tilt series were collected, at nominal applied defocus values between À2 mm and À4.5 mm, incremented in 0.5mm steps throughout the data collection.

CryoEM single particle data processing
Processing workflows are summarised in Figures S1 and S2.Single particle data processing was carried out using Relion 3.1. 100ovies were motion corrected using Relion's implementation of the MotionCor2 algorithm, 126 and initial CTF parameters were estimated using Gctf. 102Env and core particles were picked with crYOLO, 103 using models for each particle type which were trained manually on a subset of micrographs.For the first stages of processing, only the first third of the movies in the dataset ($11,000 movies) were used.Core particles were extracted into 1120-pixel (1210 A ˚) boxes and Env particles were initially extracted into 360 pixel (389 A ˚) boxes (sizes without downsampling).Particles were first extracted at 4-fold (cores, 4.32 A ˚/pixel) or 2-fold downsampling (Env, 2. 16 A ˚/pixel) to speed up the first rounds of 2D and 3D classification.After initial rounds of 2D classification to select good particles, 3D classifications were carried out, using the subtomogram averaging maps as references.Particles contributing to good classes were then selected and refined against the best 3D classes to create references for the full dataset.C3 symmetry was imposed for Env trimer particles (C2 for dimers-of-trimers) and I3 icosahedral symmetry for the core particles.
The full set of movies was then processed using the reference maps obtained from the first third of the movies.After an initial 2D classification to remove poor quality particles, particles were subjected to 3D classification against the references from the first third of the data, identifying populations of particles which produced high resolution maps.These particle sets were taken forward for 3D refinement.After the 3D classification, overlapping Env particles within a distance cutoff of 50 A ˚were removed, and the remaining particles were re-extracted in larger 480-pixel (518 A ˚) boxes, so that all high-resolution signal delocalised due to the effect of the CTF could be included.
For picking of Env trimers crYOLO was trained to identify particles at the periphery of virions, in side-view.For this reason, the second Euler angle (rlnAngleTilt), describing the out-of-plane tilt, for Env particles was set to 90 before carrying out initial 3D classifications, and only local searches of the second Euler angle, with ±30 search range, were permitted during classification.The other two Euler angles were left unrestrained.This was only done for the initial 3D classifications of Env particles from the full dataset, after 2D class averages and 3D orientation distributions from the first third of the data showed that nearly all particles were assigned tilt angles clustering around 90 , consistent with picking of almost exclusively side views.Imposing this prior angular restraint improved 3D classification results and increased the number of retained particles but had no effect on the resolutions achieved during refinement at later stages.After convergence of 3D refinements, rounds of CTF refinement and Bayesian polishing were used to refine perparticle motion and CTF parameters before final 3D refinements.The final Env maps were obtained from standard Relion postprocessing.To aid model building, denoised maps were produced from the half maps using LAFTER. 104For multi-body analysis of Env dimers-of-trimers, multi-body refinement was carried out in Relion, 58 using one body and corresponding mask for each Env trimer.
The final icosahedral capsid map was obtained by reconstructing the two independent half-maps with Ewald sphere and CTF correction using relion_reconstruct, 127 before cropping both the half-maps and the mask to an 800-pixel (864 A ˚) box size from the starting box size of 1120 pixels (1210 A ˚). Postprocessing was then carried out using relion_postprocess.This resulted in a slight improvement of the map resolution from 3.97 A ˚to 3.89 A ˚and the fitted sharpening B factor from À98.4 A ˚2 to À84.0 A ˚2 when compared to the map without Ewald sphere correction.
Maps of the pentamers and 2 hexamers were produced by symmetry expansion of the full capsid particle set using relion_parti-cle_symmetry_expand, followed by signal subtraction of the remainder of the capsid shell and re-windowing into a 320-pixel box centered on each capsomer.All masks used for signal subtraction were spheres of 160 A ˚diameter with an additional 8-pixel (8.61A ˚) soft edge, centered on the capsomer.For the pentamers situated on the icosahedral 5-fold axis, redundant symmetry copies from the I3 symmetry expansion were removed by selecting only particles with a first Euler angle (rlnAngleRot) within the range ±36 i.e., the C5 asymmetric unit.This procedure produced signal-subtracted particle images for every individual capsomer in each full capsid particle, which were then independently classified without alignment and with a T value of 10, before further refinement, CTF refinement and a final re-refinement.This produced maps at 3.4-3.5A ˚resolution, with C5 symmetry for the pentamer and no symmetry for the hexamers.All refinements of subtracted capsomer particles used local angular searches only (±15 ).

CryoET/subtomogram averaging data processing
For the WT Gag data, movie frames were aligned using the alignframes program in IMOD 4.10. 105,128Tilt series showing poor tracking, no virions in the field of view, or with otherwise poor quality were discarded, leaving 250 tilt series for further processing.Bad tilt micrographs with poor tracking or significant ice contamination were then removed from each tilt series before further processing.The tilt series were aligned using their fiducial markers with eTomo in IMOD 4.10.CTF estimation was carried out using CTFFIND 4.1.5, 106before tomograms were reconstructed using novaCTF 107 with 3D CTF correction by phase-flipping and a defocus step size of 150 A ˚.This produced a full-size unbinned tomogram reconstructed by weighted back projection (WBP), and a 4-fold binned tomogram with a SIRT-like filter to boost contrast for particle picking.The 4-fold binned tomograms were also denoised using 20 iterations of nonlinear anisotropic diffusion using Bsoft 108 with default parameters, to further boost contrast and interpretability for particle picking and later visualisation.
For particle picking, oversampled surface models of the Env layer were generated using Dynamo. 109Whole cores were picked manually using the e2spt_boxer.pytool in EMAN2.3. 110For the Env particles, initial Euler angles were assigned using the Dynamo models and converted to Relion format using a custom script, and initial angular searches during refinement and classification were restricted to a range of ±45 around the assigned orientations.
Env particles were reconstructed from the tilt series using Relion 4 101 at 4-fold downsampling (5.52A ˚/pixel) in a 440 A box.
Reference-free alignment and classification were used to create initial low-resolution models.After initial alignment at low resolution without application of symmetry, a distance cutoff of 50 A ˚was used to remove duplicate subtomograms which aligned to the same position in the tomogram.Asymmetric classification against the initial models produced classes which centered either on a single Env trimer or a hexamer-of-trimers, which were then separated and refined with C3 and C6 symmetry respectively.The class of hexamers-of-trimers was then reclassified, using the symmetry relaxation capability in Relion to relax C30 symmetry (lowest common multiple of C5 and C6), which revealed a population of pentamers-of-trimers.The classification procedures resulted in 3 classes: Env trimers, pentamers-of-trimers and hexamers-of-trimers, which were then refined with C3, C5 and C6 symmetry respectively.Pentamers and hexamers of Env trimers were only retained for processing if the corresponding class average maps exhibited full occupancy at all trimer positions.Particles were then re-extracted at 2-fold downsampling (440 A ˚box, 2.76 A ˚/pixel) after convergence of refinements at 4-fold downsampling.Refinement of the hexamers and pentamers at 2-fold downsampling produced the final maps at 10 and 12 A ˚resolution, respectively.The single-trimer particles further refined at full sampling (276 A ˚box, 1.38 A ˚/pixel) in a smaller box for computational efficiency.Refinement of the tilt series frame alignment and CTF parameters in Relion 4 resulted in a final map of the Env trimer at 4.9 A ˚resolution.
Capsid particles (WT Gag) were reconstructed from the tilt series using Relion 4 at 4-fold downsampling (5.52A ˚/pixel, 180 pixel box).A 3D reference was constructed by the ab initio reference-free initial model generation routine before refinement with I3 icosahedral symmetry imposed.Refinement initially used only a spherical mask around the capsid density, then was carried out with a mask over the icosahedrally symmetric CA domain assembly.After refinement converged at 4-fold binning, the map was reconstructed from the tilt series images at full sampling (1.38 A ˚/pixel, 600 pixel box).
For the p68 Gag data, movie frames were aligned using the alignframes program in IMOD 4.10, producing tilt series stacks with and without dose-weighting.The dose-weighted tilt series were aligned by their fiducial markers using eTomo in IMOD 4.9.CTF estimation was carried out on the non-dose-weighted stacks using CTFFIND 4.1.5,before tomograms were reconstructed using novaCTF with 3D CTF correction by phase-flipping and a defocus step size of 150 A ˚.This produced tomograms reconstructed by weighted back projection for particle extraction and tomograms with a SIRT-like filter to enhance contrast for particle picking.Whole cores were picked manually using the e2spt_boxer.pytool in EMAN2.3.
Subtomogram averaging was performed using Relion 4.0.Capsid particles were reconstructed from the tilt series images at 4-fold binning (8.88A ˚/pix) in a 100-pixel box.A 3D reference was constructed by the ab initio reference-free initial model generation routine before refinement with I3 icosahedral symmetry imposed.Refinement initially used only a spherical mask around the capsid density, then was carried out with a mask over the icosahedrally symmetric CA domain assembly.After refinement converged at 4-fold binning, the map was reconstructed from the tilt series images at full sampling (2.22A ˚/pix, 400 pixel box).
For backplotting of refined maps into tomograms, a custom script was used to place copies of the refined map at subtomogram positions and orientations using Bsoft.

Model building
Using the postprocessed maps from subtomogram averaging and in situ single particle analysis, 13 copies of the NMR structure of monomeric PFV Gag CA domain (PDB 5M1H) 34 were docked as rigid bodies into one asymmetric unit of the capsid using UCSF Chi-meraX. 111This showed that the CA-NTD and CA-CTD adopt different orientations relative to each other depending on the Gag monomer's position in the capsid asymmetric unit.Therefore, rigid-body fitting against the density was performed on each CA-NTD and CA-CTD separately in Coot. 112This procedure was then repeated for each a helix in the model, producing a model in which each domain, then each a helix, had been fit to the density as a rigid body.The NTD-CTD linkers and inter-helical loops were then manually rebuilt in Coot to make an initial ASU model, then symmetrised using UCSF ChimeraX to make an initial full capsid model.
From the initial full capsid model, chains comprising a single pentamer, hexamer 1 or hexamer 2 were extracted to fit into the higher resolution capsomer maps derived from signal subtraction.Further cycles of iterative manual rebuilding and refinement using Coot and Servalcat/Refmac5 113 in the CCPEM software package 114 produced the final capsomer models.For the pentamer, 5-fold NCS was imposed during refinement, but no NCS was specified for either hexamer 1 or hexamer 2. All model refinements in which coordinate shifts were permitted were carried out with hydrogen atoms added in their riding positions during refinement, and applying secondary structure restraints generated using ProSMART. 115fter refinement of the capsomer models was complete capsomers were placed back into the full capsid map as rigid bodies and then symmetrised to make an improved full-capsid model.In order to correctly model the interfaces between neighboring capsomers and ASUs, chains corresponding to one ASU plus all adjacent contacting chains from neighboring ASUs were then extracted from the improved full capsid model and refined against the full capsid map using Servalcat/Refmac5, imposing NCS between each chain in the ASU and any of its contacting symmetry mates from neighboring ASUs, but not imposing NCS between chains within the ASU.Two consecutive refinements were carried out at this stage; the first treating all chains as rigid bodies, and the second refining atomic B factors only, with no coordinate shifts allowed.This refined model was then used for analysis of the interfaces between capsomers.To create the final full capsid model, the adjacent contacting chains were removed, and the ASU was symmetrised using UCSF ChimeraX.
The model for the Env trimer was manually built de novo in Coot, using both the Relion-postprocessed and LAFTER-denoised maps, and refined against the Relion half maps using Servalcat/Refmac5 in the CCPEM software package, imposing 3-fold symmetry constraints between each protomer and with secondary structure/hydrogen bonding restraints generated using ProSMART, and torsion angle restraints on all glycan residues.All model refinements in which coordinate shifts were permitted were carried out with hydrogen atoms added in their riding positions during refinement.The final model resulted from iterative rounds of building and refinement.Modeled glycans were built using Coot's inbuilt glycan building functionalities and their geometry was validated using Privateer. 116or the Env dimer-of-trimers model, two copies of the trimer model were docked into the lower-resolution dimer map, and the dimeric interface between the trimers was rebuilt manually in Coot.One further round of refinement, using symmetry constraints, and H-bond and self-restraints generated using ProSMART, produced the final dimer structure.
The Env dimer-of-trimer model was then fitted into the lower-resolution hexamer and pentamer maps from subtomogram averaging.For the hexamer, this was done by fitting as a rigid body in UCSF ChimeraX.For the pentamer, initial rigid body fitting showed a conformational difference compared to the hexamer.Consequently, the model was fitted to the pentamer map using heavily restrained molecular dynamics flexible fitting in ISOLDE, 117,129 with restraints on all interatomic distances %20 A ˚to minimise the deviation from the initial model at the sidechain/backbone level while still allowing for larger-scale domain movements.
All models were validated using the CCPEM model validation task, with Molprobity, 118 Refmac5 and TEMPy. 119Map-model FSC curves were calculated using Servalcat/Refmac5.

Structure analysis
The approximate internal volume of the capsid was calculated by simulating a density map from the atomic model at 12 A ˚resolution using the molmap command in ChimeraX, and adjusting the threshold such that there were no holes in the map surface.The enclosed volume was estimated as the interior volume enclosed by the isosurface.
For analysis of the relative tilts and twists of the 7 unique capsomer-capsomer interfaces, each capsomer's orientation was described as the normal to a plane, which was fitted to the atom coordinates of each capsomer using the define plane function in UCSF ChimeraX.Two points along each normal (the capsomer centroid and a point displaced from the centroid by 20 A ˚along the normal, making 4 points for each capsomer pair) were then used to measure the tilt and twist angles in PyMol. 120The tilt is defined as the angle between capsomer normals parallel to the inter-capsomer vector, and the twist is defined as the torsion angle between the capsomer normals about the inter-capsomer vector, as described previously. 59or comparison of retroviral CA pentamer structures, coordinates were extracted from the PDB (dArc1/dArc2 -PDB 6TAR/ 6TAT 60 ; Ty3 -PDB 6R24 130 ; HML2 -PDB 6SSJ 66 ; HIV-1 -PDB 5MCY 59 ; MLV -PDB 6HWY 131 ; RSV -PDB 7NO0 64 ) and manually aligned to a common 5-fold axis in UCSF ChimeraX.Centers of mass (CoMs) of all Ca atoms in both the CA-NTD and CA-CTD were then calculated for each structure using the measure center command, and used to calculate the relative lateral displacements in the XY plane, as the difference between the average radii of the CA-NTD and CA-CTD rings with respect to the 5-fold axis, and the relative vertical displacements, as the distance along the z axis.Analysis of the hexamer and pentamer pores was carried out using HOLE2. 121o analyze the quasi-equivalence of capsid interfaces, lists of contacting residues were compiled for each interaction type using the interfaces command in UCSF ChimeraX, with inclusion criteria of >100 A ˚2 total interface area and >10 A ˚2 per-residue interface area.The frequency of the occurrence of each residue in each given interface type was then calculated using Microsoft Excel.These frequencies were then normalised by dividing by the number of unique instances of each interface type in the ASU, then further dividing by the number of chains contributing to each instance.The normalisation produces a score which we term the equivalence score, ranging from 0 to 1 in most cases; scores over 1 reflect that the residue in question can participate in multiple interaction instances simultaneously.For example, residue Gln308 takes part in the intracapsomer NTD-NTD interaction at the center of the NTD ring, often contacting the adjacent monomers on both sides simultaneously.It is found to make an intermolecular NTD-NTD contact 17 times across the ASU.Dividing by 13 (the number of CA monomers in the ASU) gives its equivalence score of 1.308.Residue Arg477 sometimes takes part in the CTD trimer interaction, making contact 11 times across the ASU.Dividing by 5 (the number of unique CTD trimers in the ASU) and then again by 3 (the number of chains in each CTD trimer) gives its equivalence score of 0.733.The average of all scores in each residue taking part in a given interface type gives the overall score for that interface type.The per-residue equivalence score described here can be considered to be conceptually similar to the per-interface score proposed by Damodaran and colleagues. 67equence alignment of selected simian FV Env proteins was carried out using Clustal Omega. 122The sequences used were: PFV (GenBank KX087159), eastern chimpanzee SFV (SFVptr, GenBank U04327), macaque SFV (SFVmac, GenBank X54482), marmoset SFV (SFVcja, GenBank NC_039030), orangutan SFV (SFVppy, GenBank NC_039085) and spider monkey SFV (SFVspm, GenBank NC_039027).
The electrostatic surface of Env was calculated in UCSF ChimeraX using the coulombic command.Structural homology searches using the Env model were carried out against the full PDB using the DALI server. 46Protein topology diagrams were generated using Pro-origami 123 and manually rearranged in Adobe Illustrator.Lokiretrovirus Env structures were predicted using Alphafold2 57 using all reported consensus lokiretrovirus Env sequences, 56 all of which produced similar structures.The Loki-Ame (Astynax mexicanus lokiretrovirus) Env prediction was then chosen for DALI comparison to PFV Env.

Integrated virion structure
The integrated model of a PFV virion was constructed by first defining the envelope relative to the icosahedral capsid model as a sphere of 500 A ˚radius, with the same center as the capsid.Copies of the Env trimer model were then manually placed on the virion surface such that their transmembrane regions were embedded in the membrane and that the lattice organisation followed the T = 13 icosahedral symmetry of the capsid, with the linking arms of neighboring Env trimers positioned to face each other.For modeling of the MA layer, one protomer from the crystal structure of the Gag MA domain (PDB: 4JMR) 29 was extracted and placed such that its C-terminal helix pointed toward the N-terminus of the CA domain and its N-terminal head domain was oriented toward the envelope, while positioning the head domain at the radius of the shell of higher protein density in the capsid map.
The radial density functions in Figure 7 were produced by 4-fold downsampling of the raw capsid map from 3D refinement, before applying rotational averaging using the math.rotationalaveragefunction of e2proc3d.pyfrom EMAN2.3.The radial line profile of the resulting rotationally averaged map was then traced in ImageJ 124 to produce the radial density function.For the simulated density, the molmap function of UCSF ChimeraX was used to simulate density to 3.9 A ˚resolution on the same map grid as the experimental map.This map was then downsampled 4-fold and the radial density function derived as described for the experimental map.

Figure 1 .Figure 2 .
Figure 1.In situ structure of the full-length PFV Env trimer (A) Env subunit organization (LP, blue; SU, green; TM, red).Proteolytic cleavage sites indicated by black arrowheads.Black diagonal cross-hatched regions denote transmembrane helices in LP and TM.Horizontal cross-hatched region indicates the TM fusion peptide.The green/gold cross-hatched region indicates the N terminus of SU from both the central trimer (green) and the neighboring trimer in the Env lattice (gold).RBD, receptor-binding domain; C, central domain; L, lattice domain; MPER, membrane-proximal external region.Beneath the bar, positions of free cysteines are shown by lines, disulphide bonds by connecting lines, and putative disulphide bonds which lacked density for modeling by dashed lines.(B) CryoEM map of the Env trimer, contoured at 11.85 s. (C) Atomic model of the Env trimer.Left: Side view as in (B), with the location of the membrane (determined from the cryoEM density as in Figure 2D) indicated by the gray rectangle.Right: Top view down the 3-fold axis.(D) Domain organization of a single Env protomer.The location of the membrane is shown by the gray rectangle.(E) Same view as in (D) but indicating the positions of the observed N and C termini of LP, SU, and TM.See also Figures S1-S3.
), in which 3 central helices of TM (residues 632-652) are surrounded by 3 SU helices (residues 186-206).A 4-stranded b-sheet (SU residues 163-174, TM residues 609-619, 684-712) packs against the outer face of the 6-helix bundle in each protomer, with the RBD in turn packing against the exterior face of the b-sheet.The lattice-forming domains enclose the lower cavity of the trimer and are of predominantly b-sheet secondary structure, with the N-terminal strand of SU incorporated into 3 TM b-sheets.

Figure 4 .
Figure 4.The Env lattice on the virus surface is assembled through a strand-exchange mode of interaction Subunits colored as in Figure 1.(A) CryoEM map of a dimer of Env trimers, contoured at 9.28 s. (B) Atomic model of a dimer of Env trimers, shown from side (left) and top (right) views.The two SU subunits forming the strand-exchange interaction between Env trimers are colored, with other protomers transparent.2-fold and 3-fold symmetry axes are indicated by black oval and triangles, respectively.The two symmetrically opposed HS-binding surfaces are indicated by dashed ellipses.Modes of relative tilting, twisting, and rotating motions between Env trimers are indicated by dashed arrows.(C) Details of the inter-trimer interaction.Key residues and glycans close to the exchanging strands shown as sticks.2-fold axis indicated by the black oval.Hydrogen bonds and salt bridges shown as cyan dashed lines.Side view at left, top view at right.(D) Subtomogram averaging of the Env lattice.Top: Subtomogram average map of an Env hexamer-of-trimers (cyan), from top view (left) and side view (right), contoured at 2.33 s.Middle: Subtomogram average map of a pentamer of Env trimers (pink), from top view (left) and side view (right), contoured at 2.26 s.Symmetry axes are indicated by white polygons and dashed lines indicate planes for slab cutaway in (E).Bottom: Gallery of central slices through subtomogram class averages, showing oligomers with (pentagons/hexagons) and without (occluded pentagons/hexagons) full occupancy at all Env trimer positions.Scale bar indicates 100 A ˚. (E) Fitting of the Env model into maps (transparent gray surfaces, shown as slab cutaways of the linking arm region) of the hexamer (top, contoured at 4.03 s) and pentamer of trimers (middle, contoured at 4.19 s) shows the transformation from the hexamer to the pentamer.Bottom: Rotation about each trimeric axis and bending about the 2-fold axis is indicated.Pro135 is labeled as the halfway point.(F) Proposed lattice assembly pathway.LP-SU cleavage takes place in immature Env trimers (left) before strand exchange between adjacent trimers, yielding the mature lattice observed in our data (right).Mature LP originates from the same trimer as the cleaved SU.LP helix shown as blue circle, N-terminal SU strand shown as green, and TM subunits shown as red cartoons.See also Figures S1 and S2.

Figure 5 .
Figure 5. PFV Gag forms a T = 13 icosahedral capsid in the core of virus particles (A) Domain organization of Gag, with MA (yellow), proline-rich region (gray), CA-NTD (red), CA-CTD (blue), NC (pale green), and p3 (pink) domains.The p68-p3 proteolytic cleavage site is indicated by the black arrowhead.(B) PFV Gag CA monomer.(C) CryoEM map of the T = 13 icosahedral PFV capsid viewed down the 5-fold axis, contoured at 4.39 s.Density for the CA-NTD and CA-CTD domains is colored according to (B) but with pentamers shown in paler tones.Locations of hexamers 1 and 2 relative to a central pentamer are indicated.The 5-, 3-, and 2-fold symmetry axes are represented by the yellow pentagon, triangle, and oval, respectively.(D) The icosahedral ASU, viewed normal to (left) and in the plane of (right) the capsid surface.One copy of each unique capsomer, colored as in (C), and the surrounding contacting chains (semi-transparent) is shown.The ASU, comprising a single CA monomer from a pentamer, six from hexamer 1 and six from hexamer 2, is delineated by the dashed line.Symmetry axes are indicated by the yellow symbols as in (C).(E) The individual pentamer, hexamer 1, and hexamer 2 capsomers viewed from the top (left shows molecular surface & cartoon) and side (right shows cartoon).
(G) Plot of the relative vertical (Dz) versus lateral (Dxy) displacements of the CoMs of CA-NTDs and CA-CTDs in PFV, transposon-derived, and orthoretroviral capsid pentamers.(H) The underlying network of CTD-CTD interactions connecting the capsomers of the capsid.Pentamers and hexamers are colored as in (E), with rear portion of the assembly removed for clarity.(I) Mapping of intermolecular interactions.A view of each unique capsomer and neighboring chains is shown as in (D).Atoms of residues participating in intermolecular interactions are shown as spheres, with coloring by interface type according to legend.See also Figures S1 and S2.

Figure 6 .
Figure 6.Conformational variability of the Gag CA domain Domains colored as in Figure 5A.(A) The ensemble of Gag CA conformations (see Figures 5, S6C, and S7A).Left: Alignment of all 13 CA chains in the ASU on the CA-CTD.The arrow indicates flexibility in the relative pose of the CA-NTD and CA-CTD.Middle: Superposition places CA monomers into three clusters based on the interdomain linker conformation.Clusters are shown with their respective fractions of the 13 ASU chains and Ca RMSDs.Right: Detail of each linker conformation.(B) Plasticity at intracapsomer interfaces (see Figures 5D, 5E, 5I, and S7B).Viewed from outside the capsid.Variable loop Gly356-Ser362 is indicated.Each chain in the ASU (solid color) is aligned on the CA-NTD to show the interactions with its neighbors (transparent).(C) Plasticity at CA-CTD dimer (left) and trimer (right) interfaces (see Figures 5H, 5I, and S7C).All CA-CTD dimers/trimers are aligned on one monomer to show the variability of the interfaces.Ca RMSD indicated.See also Figures S6 and S7, TableS6.

Figure 7 .
Figure 7. Integrative model of the PFV virion Subunits/domains colored as in Figures 1 (Env) and 5 (Gag).(A) Slice through a tomogram of PFV particles.White arrowhead indicates density connecting the capsid to the envelope.Scale bar indicates 1,000 A ˚. (B) Backplots of maps from subtomogram averaging into the tomogram region shown in (A), featuring Env hexamers-of-trimers (cyan), Env pentamers-of-trimers (pink), and the capsid in the lower virus particle.(C) A theoretical PFV virion model featuring a T = 13 icosahedrally symmetric Env lattice on the surface, viewed along the 5-fold axis.Proteins are shown as molecular surfaces.Top: Exterior view.Middle: Cutaway view, showing the capsid aligned with the symmetry of the exterior Env lattice.Virus membrane is colored yellow.Bottom: Domain and subunit organizations of Gag and Env, with unmodelled regions colored gray.(D) Central slices through full-virion maps, viewed along the capsid 5-fold axis (indicated).Top left: Simulated density at 3.9 A ˚resolution, calculated from the fullvirion model (see STAR Methods).Bottom left: Rotational average of the simulated map.Top right: Unmasked experimental capsid map.Bottom right: Rotational average of the experimental capsid map.Inset bottom right: Radial density profile of the rotationally averaged experimental map, with peaks corresponding to virus structural features indicated.(E) Radial density profiles of rotational averages of the unmasked experimental capsid map (black) and the theoretical virion model (blue).Rotationally averaged density value is plotted against map radius.Peaks corresponding to Gag domains, the membrane, and Env are labeled.Vertical dotted lines show minimum and maximum radii of the CA shell.Approximate radial extent of Env and Gag shown schematically below the x axis.See also Figures S2 and S8.

Figure S1 .
Figure S1.Data processing workflow, related to STAR Methods, Figures 1, 4, and 5 Overview of the image processing and refinement procedures used to produce maps and determine structures.Particle counts and classes retained along with the procedures and software packages used at each stage are indicated.

Figure
Figure S2.FSC curves, related to STAR Methods, Figures 1, 4, 5, and 7 Half-map FSC curves (green: unmasked, blue: masked, red: masked phase-randomised, black: final corrected (including Ewald sphere correction for the full capsid SPA map only, where orange is the curve without Ewald sphere correction)) and model-map FSC curves (magenta) for each EM map and model.Resolutions in A ˚are indicated for half-map FSCs at the cutoff of 0.143 (black text) and for model-map FSCs at the cutoff of 0.5 (magenta text).

(
Figure S3.Glycosylation of the Env trimer and conservation of essential glycan structure, related to Figure1

( A )
Figure S4.-Topology diagrams of PFV Env and RSV F, related to Figure 3 Topology diagrams of the central domains (top) and lattice domains (bottom) are shown for side-by-side comparison of the folds of PFV Env (left) and RSV (right).Disulphide bonds are indicated by yellow lines, and conserved disulphide bonds are numbered as in Figure 3C.Structurally similar regions are indicated by gray areas.

( A )
Docking the atomic model of the wild-type PFV Env trimer into a map of a PFV SU-TM cleavage site mutant (EMDB-4013).21Map is shown as gray transparent surface and the Env backbone is shown as cartoon colored as in Figure1, with the fusion peptide in pink and the trajectory of the uncleaved fusion peptide sequence indicated by the dashed line.Fitting of the processed structure into the map of the unprocessed Env shows that the cleavage which releases the fusion peptide does not cause substantial structural changes.(B) Schematic of the proposed pathway of PFV Env-mediated membrane fusion, illustrating 4 key stages.Color scheme as in Figure1.Env TM fusion peptide is colored pink.From left to right: prefusion Env, extended TM N-terminal coiled coil, foldback of TM C-terminus, postfusion structure.SU and LP subunits may or may not dissociate from the TM trimer at unknown intermediate stages of this process.

Figure S6 .
Figure S6.Gag capsomers assemble a faceted icosahedral assembly, related to 6 (Left) Tilt/twist analysis of capsomers.Models rendered as colored as 5.A plane is fitted through the atoms of each capsomer in the ASU and surrounding capsomers that are labeled P (pentamer), 1 (hexamer 1) and 2 (hexamer 2).The centroids and plane normal for each capsomer are used to compute angles of relative tilt (angle between normals) and twist (torsion angle with respect to the vector between capsomer centroids), illustrated.(Right) Plot of tilt against twist for each unique pair of capsomers.(B) Mapping of tilt and twist to positions in the assembled capsid.Models are colored by tilt and twist according to color bars below.Capsomer boundaries indicated using a cage model in yellow.Mapping of tilt and twist shows a tilted ''cap'' at each vertex (left), delineated by a ''twisted'' boundary (right).(C) Mapping of the loop-turn, tight-loop and loose-loop conformations of Gag CA to their positions in the assembled capsid.CA-monomers are colored according to conformation as indicated in legend.Atoms are shown as spheres and the capsomer boundaries are delineated using a cage model in yellow.

Figure S7 .
Figure S7.Quasi-equivalence of the Gag CA domain, related to Figure 5 and 6 Quasi-equivalence plots for Gag CA allow identification of invariant interacting residues.Positions of a-helices in the sequence are illustrated beneath the x axis.(A) Quasi-equivalence plots for the intramolecular NTD-CTD interaction, comparing the 3 conformations of Gag CA. (Left) Plots of equivalence score (above), and of the difference in equivalence score with respect to the ASU average (below).Average equivalence score: 0.50.(Right) Details of the NTD-CTD interaction in the pentamer.Interacting residues shown as sticks.Residues that make invariant NTD-CTD interactions in all three conformations are labeled.(B) (Left) Quasi-equivalence plots for the intracapsomer interactions, comparing pentamers to hexamers 1 and 2 via equivalence score (above) and difference in equivalence score with respect to the ASU average (below).Average equivalence score: 0.63.(Right) Details of intracapsomer interactions in the pentamer.Interacting residues in the pentamer are shown as sticks.Residues that make invariant interactions in all capsomers are labeled.Dashed line indicates the boundary between adjacent monomers.(C) (Left) Quasi-equivalence plot of equivalence score for intercapsomer CtD dimer and trimer interactions.Average equivalence scores; CA-CTD dimer -0.54 and CA-CTD trimer -0.69) (Right) Details of the pentamer-hexamer 1 CA-CTD dimer and trimer interactions.Interacting residues are shown as sticks.Residues that make invariant residues in all CA-CTD dimers/trimers are labeled.Residues which are part of the conserved YxxLGL and PGQA motifs are color coded as indicated.(D) Quasi-equivalence analysis of molecular interfaces mediating the conformational variability of Gag CA in the assembled capsid.The bar plot shows the equivalence score (y axis) of each Gag CA residue (x axis) (see STAR Methods).The horizontal dotted line indicates the threshold where a residue participates in every interface in the asymmetric unit (1.0) and is identified as an invariant contact residue.Vertical dotted lines indicate the boundaries of the NTD, CTD and interdomain linker.Bars are colored according to the class of interaction shown in the legend.(E) (Left) Mapping of invariant residues onto the icosahedral asymmetric unit.The ASU and surrounding chains are shown in cartoon as depicted in Figure 5I.Invariant residues at molecular interfaces are shown as spheres with coloring according to interface type as in E. (Right) same view but with cartoon of ASU removed to show the distribution of invariant residues at each interface.

Figure S8 .
Figure S8.The WT Gag and p68 Gag capsids are identical, related to Figure 7

( A )
Slice through a tomogram of WT Gag PFV particles.Scale bar = 100 nm (B) Slice through a tomogram of p68 Gag PFV particles.Scale bar = 100 nm (C) (Top) Subtomogram averaging map of WT Gag capsid at 10 A ˚resolution, viewed along the 5-fold symmetry axis.(Bottom) Central slice through subtomogram averaging map, viewed along the 2-fold symmetry axis.(D) (Top) Subtomogram averaging map of p68 Gag capsids at 13.5 A ˚resolution, viewed along the 5-fold symmetry axis.(Bottom) Central slice through subtomogram averaging map, viewed along the 2-fold symmetry axis.

TABLE
Atomic models have been deposited in the PDB and cryoEM maps deposited in the EMDB, with accession codes: Env trimer (PDB 8OZH, EMD-17309, EMD-17316), Env dimer of trimers (PDB 8OZJ, EMD-17311), Env pentamer of trimers (PDB 8OZP, Data are publicly available as of the date of publication.Accession codes for deposited data and a list of software used in this study can be found in the key resources table.d This paper does not report original code.d Any additional information required to reanalyse the data reported in this work paper is available from the lead contact upon request.
d RESOURCE AVAILABILITY B Lead contact B Materials availability B Data and code availability d METHOD DETAILS B Virus preparation B CryoEM sample preparation B CryoEM and cryoET data collection B CryoEM single particle data processing B CryoET/subtomogram averaging data processing Data and code availability d