Structural determinants of rotavirus proteolytic activation

Abstract

The infectivity of rotavirus (RV), the leading cause of childhood diarrhea, hinges on the activation of viral particles through the proteolysis of the spike protein by trypsin-like proteases in the host intestinal lumen. In order to determine the structural basis of trypsin activation, we have used cryogenic electron microscopy (cryo-EM) and advanced image processing methods to compare uncleaved and cleaved RV particles. We find that the conformation of the non-proteolyzed spike is constrained by the position of loops that surround its structure, linking the lectin domains of the spike head to its body. The proteolysis of these loops removes this structural constraint, thereby enabling the spike to undergo the necessary conformational changes required for cell membrane penetration. Thus, these loops function as regulatory elements to ensure that the spike protein is activated precisely when and where it is needed to facilitate a successful infection.

Author summary

Rotavirus is the leading cause of severe childhood diarrhea worldwide, resulting in significant morbidity and mortality each year. To cause an infection, rotavirus must first activate its spike protein, a specialized molecular structure that allows the virus to penetrate host cells in the gut. This activation occurs when trypsin, a digestive enzyme naturally found in the intestine, cleaves specific regions of the spike protein. In our study, we used advanced microscopy techniques to examine the detailed structure of the rotavirus spike before and after activation by trypsin. We discovered that the spike protein in its inactive state is locked by loops connecting different parts of its structure. Trypsin cuts these loops, releasing structural constraints and enabling the spike protein to change shape, which is critical for infection. Our findings highlight a finely tuned regulatory mechanism, ensuring that rotavirus activates only in the right location and at the appropriate time to effectively infect host cells.

Introduction

During the viral replication cycle, entry emerges as the first critical juncture facing the viral particle for successful infection. The processes involved are tightly regulated in both space and time, with success only occurring after the carefully orchestrated initial stages of infection have been successfully completed. For some viruses, these early infection steps lead to an irreversible commitment to entry, which, if unsuccessful, renders the particle non-infectious. Consequently, these viruses have evolved sensing mechanisms to ensure that entry occurs only in the correct environment. Many viral proteins responsible for crossing the host cell membrane are synthesized as inactive precursors that undergo proteolytic cleavage to become fully active [1–3]. Processing of viral surface proteins by host proteases, like furin and trypsin, is a control mechanism that has been shown to produce mature virions activated for infection in a wide range of viruses. While, typically, proteolytic activation allows membrane fusion of enveloped virus surface proteins [4–9], in many non-enveloped viruses, is essential for membrane penetration [3,10–14]. Astroviruses and rotaviruses (RV) are paradigmatic examples of non-enveloped enteric viruses whose infectivity is dependent on trypsin-like digestive enzymes present in the gastrointestinal tract [12,15–18].

RV are non-enveloped, double-stranded RNA (dsRNA) viruses of the order Reovirales that replicate in differentiated enterocytes of the intestinal tract in vertebrates, and are transmitted predominantly via the faecal-oral route [19]. The RV virion is an icosahedral triple layered particle in which the segmented viral genome and the polymerase complexes are packaged in a capsid made of the VP2 protein [20,21] (S1A Fig). This core is surrounded by an intermediate layer of the VP6 protein trimers, originating the double-layered particle (DLP) [20,22]. The external layer is formed by trimers of the VP7 glycoprotein in phase with the underlaying VP6 shell and 60 trimeric VP4 spikes, which project from the VP7 layer, that clamps them and partially covers their base [23,24]. This outer layer is responsible for the first steps of RV infection: attachment to cell surface receptors, lipid bilayer penetration, and release of the double-layered particle (DLP) into the cytosol, where transcription of the viral genome is initiated [24–26].

During RV morphogenesis, the nascent DLP exits the viroplasm just before budding into the ER lumen through interactions of the VP6 layer with the viral NSP4 protein, located in the ER membrane [27,28], resulting in a transiently enveloped DLP (eDLP). VP4 interacts with both NSP4 and VP6 in the interphase between the viroplasm and the ER, and is incorporated into the eDLP [23,27]. In the eDLP, VP4 spikes are assembled as pre-mature symmetric trimers anchored onto the VP6 hexameric cavities surrounding the five-fold axes through a ‘foot’ contributed by the C-terminal regions of the three chains (S1B and S1C Fig). Three spaced lobes, constituted by the VP4 β-barrel and lectin domains, protrude in the space between the DLP and the membrane of the eDLP [23]. It has been proposed that a rapid process, which includes the extension of VP4 to breach the membrane followed by the assembly of the VP7 layer, will lead to the loss of the transient envelope and the formation of immature TLP in the ER lumen [23].

In the spike of the non-infectious immature TLP, the three VP4 molecules, VP4A, -B, and -C, are organized into a complex asymmetric conformation held by non-covalent interactions among its chains [23,24]. While the C-terminal region of the three chains maintains a three-fold symmetry association into the ‘foot’ of the spike, the extension of the pre-mature VP4 lobes generates an upright asymmetric ‘dimeric’ conformation (S1C Fig). The ‘stalk’ region contains the β-barrel domain from the C subunit, while the equivalent β-barrels from the A and B subunits form the dimeric ‘body’ of the spike. The lectin domains from VP4A and -B form the spike ‘head,’ which covers the hydrophobic loops of the two body β-barrels, while the third VP4-C lectin domain is prone and located at the spike stalk. Newly assembled immature TLP exit the cell either through exocytosis or lysis, depending on the specific cell line [29–32]. Once in the intestinal lumen, the proteolytic processing of the virions by trypsin-like proteases result in infectious mature TLP [13,17,12]. In vitro, trypsin allows rotavirus infectivity and leads to activation of RV spikes by cleaving VP4 at three defined sites (Arg231, Arg241, and Arg247, respectively) into an N-terminal fragment, VP8*, and a C-terminal fragment, VP5* (S1C Fig). After trypsin cleavage, VP8*-A and -B lectin domains remain associated within the spike head but the stalk VP8*-C lectin domain is lost [21]. It has been suggested that an additional trypsin cleavage at Lys29, facilitates the removal of VP8*-C lectin domain [21].

After activation by trypsin digestion, mature TLP initiates infection by binding of VP8* to its cellular receptor, which promotes TLP internalization either by virus-directed engulfment [25] or endocytosis [33,34]. The attachment of VP8* initiates a sequence of structural changes that will ultimately facilitate membrane penetration and the delivery of DLP into the host cell cytoplasm [24,35–39]. In the initial step, VP8* head lectin domains move laterally, allowing the spike to transition from an asymmetric ‘upright’ conformation to a symmetric ‘intermediate’ conformation (S1C Fig). In this intermediate conformation, the hydrophobic loops of the three VP5* β-barrel interact with the lipid bilayer of the target membrane [35]. The interaction between the hydrophobic loops and the membrane then promotes a ‘reversed’ conformation involving the unfolding and outward extension of the three VP5* foot domains, which perforate and penetrate the lipid bilayer of the vesicle membrane (S1C Fig).

Despite the detailed structural information that has allowed the identification of the different conformational stages that the spike protein transits from its assembly into eDLP in the ER to its dissociation during membrane penetration, the structural basis underlaying the proteolytic activation from an immature to a mature spike remains unclear. Here, we use cryo-EM and computational methods to compare the structure of the uncleaved and cleaved spike of rotavirus. The resolved structures show that in the non-trypsinized spike the lateral movement of the lectin domains is restricted by the position of the loops that surround its structure and connect the lectin domains of the head with its body. Proteolysis of these loops breaks this structural constraint, allowing the necessary conformational changes to penetrate the cell membrane.

Results

Previous structural cryo-EM characterizations of immature [23,40] and mature TLP [21] have revealed that the larger difference between uncleaved and cleaved spikes lies in the presence or absence of the VP4-C head domain at the base of the spike. However, it is unclear how the release of this domain could enhance the particle infectivity. On the other hand, the structure of the loop connecting the head and body domains of VP4 remains elusive. This region contains three trypsin targets (Arg231, Arg241, and Arg247) that are processed in a defined order [41] to generate an infectious RV particle. While the icosahedral symmetry of the RV capsid has facilitated the structural study of the VP2, VP6 and VP7 layers using cryo-EM and single particle analysis (SPA), it complicates the study of the spike at high-resolution due to its flexible nature and the partial depletion of spikes around the RV particle [21,40,42–44]. By combining SPA with advanced image processing methods (as described by [43,45,46], we aimed to address the continuous flexibility and partial ocupancy of the spike.

Single particle analysis of NTR- and TR-TLP

To obtain high-resolution maps of activated and inactivated particles, TLP from cells infected with rotavirus in the presence of trypsin (TR-TLP), or in its absence in medium supplemented with the protease inhibitor leupeptin (NTR-TLP), were purified by density gradient ultracentrifugation and characterized by SDS-PAGE (S2 Fig). A single band corresponding to the 98 kDa unprocessed precursor form, VP4, was detected for the NTR-TLP spike protein. In contrast, for the TR-TLP, the proteolytic products VP8* (28 kDa) and VP5* (55 kDa) were detected as two distinct bands. Analysis of the particles by cryo-EM showed a homogeneous population of isometric TLP, some of which could be seen with VP4 (or VP8*/VP5*) spikes projecting from the TLP surface (S2 Fig, arrowheads). Three-dimensional reconstructions (3DR) of NTR- (Fig 1A–1C) and TR-TLP (Fig 1D–1F) were computed using SPA with imposition of icosahedral symmetry, achieving resolutions of 3.40 Å and 3.48 Å for NTR- and TR-TLP, respectively (S3 Fig, S1 Table). The 3D architecture of the treated and untreated spikes showed a similar average density. However, the NTR spike was slightly shorter and wider than the TR spike (Fig 1B and 1E), and had an additional density at the base (Fig 1C and 1F), consistent with the lectin domain of the VP4-C subunit previously described in these particles [23,40]. While the density of the VP2, VP6 and VP7 shells exhibited a homogeneous high local resolution, the spike showed a radial decrease in resolution as it extended away from the outer shell surface, with particularly low resolution in the body and head regions (S3 and S4 Figs). This decrease may be related to the increased motion of the spikes as we move radially away from the particle. In addition, although the VP4-C lectin domain was detected, it was observed with low density (Fig 1C), which has been associated with flexibility [40]. These limitations hinder the construction of reliable VP4 atomic coordinates in the spike density and require further density improvement for accurate interpretation.

Fig 1. Analysis of the atomic structures of SA11 NTR- and TR-TLP by cryo-EM.

(A, D) Cryo-EM 3DR of the NTR- (A) and TR-TLP (D) particles viewed along a 2-fold axis. The bar represents 100 Å. The surfaces are represented radially by colour coding: VP4 or VP8*/VP5* spikes (red), VP7 (yellow), VP6 (blue) and VP2 (green). The density is contoured to 2σ above the mean. (B, E) Cross-sections of the NTR- (B) and TR-TLP (E) maps, parallel to the central section, viewed along a 2-fold axis. (C, F) Close up view of the NTR (C) and TR (F) spikes indicated by a dashed square in B and E, respectively. The lectin domain of subunit VP4-C in the NTR spike is indicated by an arrowhead in C. (G, H) Atomic structure of the asymmetric subunit of NTR-TLP (G) and TR-TLP (H). VP2 chains are represented in dark green, VP6 in blue, and VP7 in yellow. The A, B and C chains of the VP4 or VP8*/VP5* spikes and their domains are represented as: foot (light green for the A chain, green for the B chain and dark green for C), body (red (A, B), orange (C)), head or lectin domains (purple (A, B), pink (C)). The binding loops between the different domains are represented in grey except for the α3-β13 loop, represented in cyan.

Enhancement of spike resolution and interpretability

Technological and computational advances in recent years have made it possible to obtain structures of macromolecular complexes at atomic or quasi-atomic resolution using SPA techniques [47,48]. This approach relies on particle averaging and, while is especially powerful for resolving highly ordered and homogeneous structures, it is challenged when conformational or compositional structural heterogeneity is present in the sample. Conventionally, SPA pipelines rely on 2D and 3D classification to obtain particle subsets as homogeneous as possible. While this approach successfully addresses the compositional heterogeneity of the RV spike (partial occupancy) it is limited in addressing its continuous conformational flexibility [23,40]. In recent years, several new methods have been developed to refine structures with flexible and/or heterogenous components [49]. To improve the interpretability of our spike densities, we developed a pipeline that integrates localized reconstruction [43], 3D classification [50], flexibility correction using optical flow [45,46] and map enhancement and sharpening with LocSpiral (Kaur et al., 2021).

Icosahedral orientations determined for individual NTR and TR particles (n = 10,815 and 22,394, respectively) were used to subtract the contribution of the VP2/VP6/VP7 layers from the experimental images. Subsequently, subparticles at the spike locations (n = 648,900 and 1,343,640, respectively) were extracted using the localized reconstruction method described by Ilca et al. [43] (S5 and S6 Figs). The subparticles were sorted through 3D classification to separate capsid locations occupied by spikes from those not occupied. This resulted in 59% and 37% spike-occupied locations for NTR- and TR-TLP subparticles, respectively. The average density corresponding to the unoccupied subparticles was used to subtract the remaining VP6 and VP7 signal from the spike-occupied subparticles. Finally, 3D subclassification of spike conformations combined with optical flow was used to merge the slightly different subparticles into a unique conformation, thereby improving local resolution and signal-to-noise ratio [46]. The amplitudes of the resulting maps were corrected by LocSpiral using a spiral phase transformation approach to enhance the interpretability of the obtained 3DR [45]. Local resolution analysis (S7 Fig) revealed a clear improvement in map quality following focused classification and signal enhancement (S4 and S7 Figs). In the original reconstructions, the foot and stalk regions reached 3.5–6.5 Å resolution, whereas the body and head domains remained at 7.5–11.5 Å. In the enhanced maps, the resolution improved to 3.5–4.5 Å in the foot and stalk, ~ 5.5 Å in the C-chain lectin domain (which was barely discernible without enhancement), 4–6 Å in the body, and 5–7.5 Å in the head region. Overall, the TR spike exhibited higher local resolution than the NTR, which we attribute to a reduced need for subclassification to recover signal in flexible regions such as the C-chain lectin domain. The resulting maps provided sufficient quality to support full atomic modelling in the foot and stalk domains, while in the body and head regions, a combination of backbone tracing and rigid-body fitting was applied (representative examples of model-to-map fitting for each region are shown in S8 Fig). These enhanced densities, together with the icosahedral 3DR of the complete NTR and TR TLPs, were used to build and refine the full asymmetric subunits, including all VP2, VP4, VP6, and VP7 chains (S1 Table, Fig 1G and 1H).

Spike rearrangement after activation

The structures of untreated and trypsin-treated TLP showed structural differences only in the spike (Figs 1G, 1H and 2), while the VP2, VP6 and VP7 layers were identical as previously described [21,23,24]. Both spikes exhibited the characteristic asymmetric arrangement, where the A and B subunits project outward from the particle as a dimeric body, while the C subunit lies prone in the stalk. As described previously [23,40], the NTR spike displayed the VP4-C lectin domain at the base of the stalk (Fig 2A, 4A pink), which was absent after proteolytic processing. The density detected at the base of the stalk in the NTR spike allowed the modelling of the lectin domain in the chain C, from the residue T73 (Fig 3A, upper panels, green sphere; Fig 4A, left panel; S1 Video) to residue N221 (Fig 3A, upper panels, blue sphere; Fig 4A, left panel). While no density was observed for the loop linking K29 (red sphere) of the α domain to the T73 (green sphere), an empty density was present in the region between the lectin domain and the β-barrel of the C subunit (Fig 3A, upper panels, arrowhead). Although the map quality in this region was insufficient for building coordinates, the position and shape of the unassigned density were consistent with the 222–245 loop previously resolved by Shah et al. [23] that directly contacts the 414–420 beta hairpin from the VP4-C stalk. Since the biochemical analysis of the NTR-TLP (S2A Fig) showed that the VP4 is not processed, the lack of density for the 29–73 and 245–253 regions (Fig 3A, upper panels, purple sphere) is likely due to a high degree of flexibility in these loops.

Fig 2. Atomic structure of the RV spike.

(A) Atomic structure of the NTR (upper panel) and TR (lower panel) spike. Each domain is named and represented following the indicated colour pattern: foot (green), stem (orange), body (red), head (violet) and VP4-C lectin domain (pink, NTR spike). (B) Each of the VP4 subunits (A, B, C) of NTR-TLP (upper panel), and TR-TLP (lower panel) are shown highlighted and represented following the same colour pattern. In both panels, the binding loops between the different domains are represented in grey except for the α3-β13 loop, represented in cyan.

Fig 4. Analysis of the structural transitions of the loops and different domains of the RV spike during cell entry.

Rearrangements of the spike proteins (VP4, VP8* and VP5*) during the transition from their inactive conformation, NTR (A), through their activated form, TR (B), and intermediate (C) to their inverted conformation. (D). The domains not observed in the atomic structures, in each case, are represented schematically (panel C: lectin domains; panel D: lectin, α-amino and foot domains). The different protein domains are coloured separately to illustrate their conformational change: α-amino domains: orange, yellow, gold; lectin domains: magenta, purple and pink; α3-β14 loops: blue; body domains: red, orange and salmon; α-coiled helix and foot domains: light green, green and dark green. Loops between the α-amino and lectin domains in panels C and D are represented by dashed lines. Arrows show different residues delimiting the different coloured domains in each panel. The panels corresponding to the body domain are shown with some transparency level and highlighting the hydrophobic loops and the β14 chain in a darker tone.

Fig 3. Analysis of the atomic structure of different domains of the NTR and TR spikes.

A close-up view from the dashed boxes in the left panels of the atomic structures that have been fitted into their density maps. Within each panel (A and B) the upper row refers to the NTR spike and the lower row to the TR spike. (A) The stalk, lectin domain (VP4-C chain) and foot domains of the spikes are represented from different angles. Different residues are indicated as spheres: K29 (red), T73 (green), N221 (blue), K258 (brown), D253 (purple), T259 (yellow). (B) Side views of the body of the spike and the trypsinized (TR-VP4) or non-trypsinized (NTR-VP4) loops. Residues R231, R241, R247 are indicated in pink spheres and adjacent residues (N232, D242, A248) in blue. The α2-β1 and α3-β13 loops are shown. The molecular swapping observed in the body domain are indicated with an asterisk (*).

For the TR spike, the map allowed for the modelling of the VP8*C chain from the first N-terminal residue to the K29 at the α2-β1 loop (Fig 3A, lower panel, red sphere; Fig 4B, left panel; S1 Video). No electron density was observed between K29 and T259 (yellow sphere) of the stalk β-barrel domain. This region encompasses the lectin domain and the loops that connect it to rest of the spike structure. Biochemical analysis of TR-TLP (S2B Fig) indicated complete proteolytic digestion of the spike. Considering the known trypsin processing targets at R231, R241, and R247 [41], the absence of density for the A248-K258 region could be attributed to either a high degree of flexibility in this region or an additional cleavage after K258 for the C subunit (Fig 3A, brown sphere). The first defined residue in the stalk β-barrel of the previously resolved structure for the RRV strain TR-TLP is Glu264, suggesting cleavage of the C-terminal to either K258 or K263 [21]. Previous analyses partially evidence a cleavage at K258 [13,51]. Our structure, as well as the structure of VP4/VP7 recoated TLP [24], support this scission. On the other hand, the structure of the inverted conformation of the spike solved by Herrmann et al. was obtained from a sample of the upright VP4/VP7 recoated TLP incubated at 37°C and shows the presence of residues 248–258 (α3-β14 loop) in the three VP5* subunits. This implies that under these conditions the chain C is not necessarily processed at K258, but rather that this segment may be present yet disordered in the upright conformation.

The 3DR of the NTR spike allowed for the modeling of the entire polypeptide chain of the A and B subunits, including the two long loops that connect the head lectin domains to the rest of the molecule: the internal α2-β1 loop, which connects the α domain with the lectin domains, and the α3-β13 external loop, which binds them to the body of the spike (Fig 2A and 2B, cyan; Fig 3B, upper panels; Fig 4A, left panel, S2 Video). Comparison with previously published structures of simian RRV [24], and human CDC9 [52], strains shows that at the resolution achieved here, the α2-β1 loop adopts a very similar conformation with no significant differences. Complete cleavage of the α3-β13 loop was evidenced as an abrupt loss of signal in the density map after R231 and before A248 (Fig 3B, lower panels; Fig 4B, left panel, S2 Video). After cleavage, the 225–231 amino region of the cleaved α3-β13 loop retracts and appears repositioned near residues 443–448 in the β-barrel domain of the opposite chain (Fig 3B, right panels; Fig 4B, left panel, S2 Video), although specific contacts cannot be resolved at the achieved resolution. The carboxy terminal region of the loop remains adjacent to the β14-sheet present in the unprocessed spike and may contribute to spike stabilization through similar backbone interactions. (Fig 3B, right panel, asterisk; Fig 4A and 4B central panels). The head lectin domains maintained a similar structure before and after cleavage, with their glycan binding sites equally accessible to their host cell targets (S9 Fig), which correlates with the binding capacity of the virions to the target cell independently of the proteolytic state of their spikes [17].

The structure of the NTR spike revealed that α2-β1 and α3-β13 loops that connect the lectin domains to the preceding α domain and the following β-barrel domain embrace the body of the spike and firmly anchor the head on top of the spike. Actually, these loops maintain the NTR spike in a conformation slightly more compact than that of to the TR spike. After trypsin activation, the suppression of the physical constrains imposed by the α3-β13 loop is reflected as an expansion of the spike domains orthogonally to the capsid surface. While the foot and stalk domains did not suffer significant movements between their α-carbon chains, with a root-mean-square deviation (rmsd) of 1.1 Å and 1.5 Å, respectively; the rmsd of the body and head domains were 3.3 Å and 5.7 Å, respectively. In this activated conformation the head domains were only anchored to the foot by the α2-β1 loop, making them competent for the lateral movement of the lectin domains that will initiate the structural transitions that finalize with the host cell membrane penetration (S1C Fig).

Discussion

Proteolytic cleavage of the spike protein is known to be essential for RV infection, but the structural details of how cleavage within VP4 transitions the RV particle from its non-infectious immature state to the infectious mature state have remained elusive. The intrinsic flexibility of the spike structure and the low occupancy of spikes on the virion have significantly hampered efforts to determine the structural determinants behind this critical molecular mechanism. Here, by employing a combination of advanced computational methods [46,53,54], we have successfully overcome these limitations to construct an atomic model of the non-processed and processed NTR spike. Together with the contributions of others, this structure completes the comprehensive model of the conformational pathway that the RV spike follows from its assembly in the ER to membrane penetration in a new host cell (Figs 4 and 5).

Fig 5. Structural transitions of the rotavirus spike.

Schematic representation of the proposed conformational changes of the RV spike during its morphogenesis, maturation and activation. Proteins are coloured as indicated: VP6 in blue, VP7 in yellow, VP4/VP5* foot and β-barrel transparent, VP4/VP8* lectin domain A, B and C in magenta, purple and pink, respectively, and loops in grey. (A) During the late stages of ER morphogenesis, full-length VP4 monomers fold into a pre-mature three-fold symmetric structure in the eDLP. It has been proposed that the transition from this pre-mature conformation to the immature conformation is one of the driving forces for the transition of the eDLP to TLP. (B) In the immature TLP, the three unprocessed VP4 subunits are assembled as an asymmetric trimer. Two VP4 subunits. chains A and B, form the body and head of the spike joined by two internal (gray) and two external loops (light blue). The VP4-C chain folds to form the stalk of the spike to which it contributes with the β-barrel and the lectin domain. (C) Trypsin activation splits VP4 into VP5* and VP8* products in the mature spike. Trypsin proteolysis occurs at the external α3-β14 loop in three residues, R231, R241 and R247, leading to the loss of the segment 232-247 (loss of light blue loops) in the VP4-A and -B chains, and the loss of the lectin domain in the VP4-C chain (pink). (D) The activated spike is able to bind to the host cell through the interaction of VP8* lectin domains with surface glycans (light green) that result in the attachment of the virus to the cell membrane. (E) This interaction precedes the conformational change in which the lectin domains separate from the main body of the spike, exposing the hydrophobic loops of the ß-barrel domains of chains A and B. This conformational change continues with the extension of the ß-barrel domain of subunit C and ends in the intermediate conformation in which the ß-barrel domains of the three subunits adopt a symmetrical trimeric structure with the three hydrophobic loops inserted into the target membrane. F) After this interaction a further conformational transition results in the inverted conformation, in which the loops and the foot domains of the spike are insert into the membrane which is thought to provoke membrane distortions which culminate in its rupture and in the release of the DLP into the cytosol.

The lectin domain in the VP4-C of the mature NTR spike, previously described [23,40] (Figs 4 and 5), is positioned close to the capsid, making it intrinsically less accessible to cell surface receptors. The VP7 layer further restricts access to site II in the galectin-like fold structure (S7 Fig). Thus, it is highly likely that the ability of NTR-TLP to bind the cell membrane before spike activation by proteolysis [12,40,41,55] is mediated by the head lectin domains, although this binding cannot initiate a productive infection. Moreover, VP4-C lectin domain is cleaved following trypsin proteolysis in the TR spike, suggesting that its glycan-binding function may be dispensable during subsequent stages of cell entry. This domain may serve an unknown specific function in the NTR spike, or could be a vestigial component of the spike structure, which has evolved from three identical subunits into a partially trimeric (foot) and partially dimeric (body-head) structure that must transform into a fully trimeric configuration to engage with and penetrate the membrane.

Spike proteolytic activation is mediated by the cleavage of the α3-β13 loop in three target residues that have different susceptibilities to trypsin, R241 > R231 > R247 [41]. The central region of the α3-β13 loop, including the 232–242 region, is more distant from the spike body and does not appear to engage directly with the β-barrels. In contrast, the preceding and following regions, which contain residues R231 and R247, are located closer to the adjacent β-barrel. This spatial arrangement may explain the differential susceptibility to trypsin: the central region of the loop, being more exposed and less constrained, is likely more accessible to the protease, making it more susceptible to cleavage.

The determination of the structure of the α3-β14 loop in the NTR spike allows us to understand why the activation by proteolysis is necessary for infectivity. In non-activated virus the spikes have their lectin head domains covalently linked to the foot and the body domains through the internal α2-β1 and external α3-β14 loops (Fig 5). Furthermore, the external α3-β13 outer loops embrace the opposite β-barrel in the body of the spike, mediating a molecular swap that enhances the stability of this structure. In this conformation, the presence of the α3-β14 loop prevents the displacement of these lectin domains, acting as a “lock” preventing further conformational changes. The breaking of the covalent bonding and the loss of the segment 232–247 releases the structure from his locked state and allows for the lateral displacement of the VP8* lectin domains, which initiates the conformational transitions that mediate virus entry.

It is noteworthy that when the R247 trypsin site in the α3-β14 loop of the rotavirus strain KU is mutated to a furin site, endogenous furin-like proteases cleave the mutant VP4 intracellularly after TLP assembly, and results in the formation of a mutant virus that is incapable of plaque formation [56]. This unexpected outcome has been attributed to an inefficient process of virus release [56]. After its synthesis in the endoplasmic reticulum (ER), furin travels through the Golgi apparatus and reaches the trans-Golgi network (TGN), also cycling between the TGN, cell surface, and endosomes. This makes it likely that immature spikes on nascent TLPs within the ER are prematurely activated by furin, leading to their interaction with intracellular membranes and effectively trapping the nascent viruses within the cell due to the presence of cleaved spikes. Different human (Wa, DS-1) and animal (SA11, RRV, OSU) RVA species shows a clear conservation of both the canonical sites for trypsin digestion, R231, R241, and R247, as well as the K29 and K258 sites (S10 Fig). Only in the DS-1 strain is this conservation not observed at R231, although it is maintained throughout the rest of its sequence. Extending this comparison to other rotavirus groups (B-D, F-I) (S10 Fig), all of them have potential trypsin sites at or near residue 29. Except for RVA, the other groups have at least two trypsin sites in the α2-β1 loop and some of them, such as B, G, H and I, have multibasic cleavage sites, which present between 5 and 9 sites susceptible to trypsin proteolysis. Similarly, the α3-β14 loop has multibasic cleavage regions with 4–9 potential trypsin sites in all RV groups. The presence of additional monobasic residues in both, α2-β1 and α3-β14, loops in some groups of RV could serve as a mechanism to ensure the virion activation in the gastrointestinal tract, where newly assembled particles are released.

The precise timing and location of the spike conformational transitions are highly regulated to ensure the success of the viral infectious cycle. The fact that the non-enveloped rotavirus virion assembles within the ER is particularly remarkable and likely relates to the intricate conformational dynamics of the virion spike, an extreme example of polymorphism for the VP4 polypeptide. Previous studies [23] have demonstrated that VP4 initially assembles as a membrane-interacting, 3-fold symmetric pre-mature structure, with the lectin domains positioned on the sides of the β-barrel. The transition from this form to an immature state involves a simultaneous extension of VP4 into an upright, elongated ‘dimeric’ asymmetric conformation, accompanied by the addition of the VP7 layer and the loss of the transient envelope. Stabilization of the immature spike occurs through increased lateral interactions between the elongated A and B subunits, while the C-subunit head domain remains positioned at the base of the uncleaved VP4 spike. This conformation is maintained within the ER in nascent TLPs and continues to exist in the extracellular immature TLP [23,40]. To ensure successful infection, the spike structure is ‘locked’ by the α2-β1 and α3-β13 loops, which anchor the lectin domain that conceal the hydrophobic loops until the moment of infection. To avoid premature activation, RV has evolved the structure and sequence of these loops to ensure activation occurs only when they encounter trypsin-like proteases in the intestine, preventing early conformational changes and exposure of the hydrophobic loops.

Materials and methods

Virus production, purification and titration

In this study, the C4111 clone of the SA11 simian rotavirus A strain (RVA/Simian-tc/ESP/SA11-C411/2009/G3P[2]) [40] and the African green monkey kidney epithelial cell line MA104 (ECACC 85102918) were used to obtain rotavirus particles. Cells were grown in MEM medium supplemented with 2 mM glutamine, 50 µg/ml gentamicin and 10% foetal serum, maintained at 37ºC with 5% C02 and 95% humidity and used between passages 7 and 24. The purification of TLP produced in the presence or in the absence of trypsin was performed as previously described [40]. Viral infectivity was determined by plaque assays as previously described [40].

Electron microscopy and image processing

For cryo-EM analysis, copper-rhodium or gold Quantifoil R2/2 holey grids were ionized by ion discharge (Q150T, Quorum). Samples were applied, blotted, and plunged into liquid ethane using a Leica EM GP2 automatic cryo-fixation unit. Grids were analyzed under low-dose conditions in different Titan Krios microscopes (Thermofisher Scientific) operated at 300 kV using different direct detectors in linear mode. In all cases, data collection was performed with the automatic acquisition program EPU (Thermofisher Scientific) for single particle analysis. The acquisition conditions for each of the samples are indicated in S1 Table.

Image processing operations were performed using Xmipp [57] and RELION [50,58] integrated into the Scipion [57] platform and graphic representations were produced by UCSF Chimera [59]. MotionCor2 [60] routine was used to align, dose correct and average frames of the acquired movies. The contrast transfer function (CTF) and defocus were later estimate with CTFFIND4 [61] and Xmipp automatic picking routine was used to select particles (indicated in S1 Table). Images were 2D classified using the corresponding Relion routine and the selected particles were analyzed using the 3D classification routine. Given the size of the particles, and to obtain greater computational efficiency, during the 2D and 3D classification steps, the particles were subsampled three times with respect to the original sampling frequency. We imposed icosahedral symmetry for alignment and reconstruction. The corresponding Relion auto-refinement routine was used with the original, non-downsampled, particles to obtain a 3DR and resolution was assessed by gold standard Fourier Shell Correlation (FSC) [62] between two independently processed half datasets [63]. The resolution of each reconstruction (S1 Table) was determined applying a correlation limit of 0.143. Additionally, the local resolution of each map was determined with the MonoRes routine [64,65]. Next, a correction of the amplitudes at high frequencies was performed using the RELION Postprocessing routine, which automatically determines a B-factor to apply to the cryo-EM map weighted by the curve of the FSC. The values obtained from B-factor for each map are found in S1 Table. Additionally, amplitudes at high frequencies were corrected using the Xmipp LocalDeblur routine [66] which perform a correction based on the local resolution.

The asymmetric subunits of rotavirus particles were extracted using the UCSF Chimera program [59] or the Xmipp program (ExtractUnitCell) [65,67].

Determination of the spike 3D maps

The localized reconstruction method [43] was used for both 3D maps, NTR- and TR-TLP. This method allows the generation of 3DR of VP4 spike regions that present higher flexibility, different symmetry, and degrees of occupancy. These regions, hereinafter called subparticles, are studied as isolated particles, thus facilitating their classification and subsequent refinement and 3DR steps (S5 Fig). The signal contribution of the VP2, VP6 and VP7 layers was subtracted from the original TLP 3DR using the density map of the corresponding TLP and a mask that encompasses these three layers. Subsequently, considering the position of the spikes in the final 3DR and its icosahedral symmetry, the 2D projections of each spike were extracted from the difference images calculated previously. Subsequent 3D classification of these VP4 subparticles of the TLP spikes generated in the absence (NTR spikes) and presence of trypsin (TR spikes) revealed in both cases a class with a density corresponding to VP4 and another class in which the density of VP4 was absent. The surrounding VP7 and VP6 density was subtracted using the non-spike density class from the map with the density of VP4. In the case of NTR spikes, a second 3D classification round was performed to identify subparticles with a density associated with the VP4-C lectin domain. Next, 3D maps of both NTR and TR spikes were reconstructed. Additionally, the EnRICH (Resolution Improvement in Conformational Heterogeneity maps) method was implemented for NTR and TR spike reconstructions as previously described [46]. This method transforms the particles from their initial conformation to an established reference conformation. For this, the following successive steps were carried out: 1) 3D classification into 6 classes to obtain different conformations of the map and particles associated with them; 2) 3D refinement of each class to obtain the precise angular assignments of each particle; 3) translocation of the particles with different conformations to a reference one through an optical flow process and 4) alignment and 3D reconstruction of the set of all the particles with the new conformation with better resolution, contrast and signal-to-noise ratio As a final step, the LocSpiral algorithm [45] was applied to improve the interpretability of the 3DR.

Structure modelling, refinement and validation

The asymmetric unit of the NTR- and TR-TLP SA11 rotavirus strain under study consist of the trimetric VP4 spike protein and 4 + 1/3 trimers of each VP6 and VP7. We initially generated a homology atomic model for each polypeptide chain (VP2, VP4, VP6 and VP7) that forms the asymmetric subunit of the capsid using the published atomic structures of the rotavirus RRV TLP (PDB: 4V7Q [21] and 6WXE [24]. The reference PDB was adjusted in the density map and the chains were generated by homology modeling with the Modeller program [68] within UCSF Chimera [59] by pairwise alignment of sequences. The chains were adjusted as a rigid body in the density map with the UCSF Chimera program [59]. Coot program [69] was used to manually trace or adjust the polypeptide chains in the map. In each case, the map with the amplitude correction that best allowed us to carry out successive atomic modeling processes was selected. TR spike densities, having higher local resolution, were modeled first and used as a reference for fitting NTR counterparts.

Finally, iterative rounds of refinement and validation of the model were carried out using PHENIX programs [70] (S1 Table). In each case, the quality of the model and the correct geometry of the atomic structure was validated through the statistics determined by Molprobity [71]. We used standard stereochemical and B-factor restraints, as well as Ramachandran, rotamer, and secondary structure restraints. Model-to-map correlation coefficients were calculated yielding global CC values of 0.68 for the NTR spike and 0.78 for the TR spike. Residue-by-residue model-to-map correlation coefficients oscillated around the global CC values—0.67 for NTR and 0.78 for TR—throughout the polypeptide chains, supporting the consistency and overall reliability of the model fitting. Local CC values for individual residues in the α3–β13 loop were > 0.7 for all but Pro235–Ser237, where density was weaker and CC ~ 0.5. Residues included in the models and model statistics are summarized in S1 Table. Molecular graphics and analyses were performed with UCSF Chimera (1.18), developed by the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco, with support from NIH P41-GM103311 [59].

Multiple sequence alignment analysis

Multiple sequence alignments of VP4 protein sequences of representative strains of the RVA specie (Wa, DS-1, OSU, RRV, SA11) and the reference strains for the different RV species were performed using the Clustal Omega server [72,73] and analyzed using the ESPript server (ESPript - https://espript.ibcp.fr; [74]). We retrieved rotavirus sequences from GenBank [75]. The accession numbers of the VP4 genomic segments of the different strains used are indicated in S2 Table.

Accession codes

Structures described in this manuscript have been deposited in Protein Data Bank under accession code 8OLB and 8OLC for the NTR- and TR-TLP structures, respectively. The cryo-EM data has been deposited and issued the code EMD-16954 and EMD-16955, for the NTR- and TR-TLP structures, respectively.

The NTR- and TR-spikes structures have been deposited in Protein Data Bank under accession code 8OLE and 8QTZ, respectively. The cryo-EM data has been deposited and issued the code EMD-16956 and EMD-18655, in each case (S3 Table).

Supporting information

S1 Fig. Rotavirus structure and structural transitions of the RV spike during the viral cycle.

(A). Representation of the RVA viral particle. The different protein layers ensembled and their structural proteins are indicated with the colour code. (B). Schematic representation of the structure and domain organization of VP4. The atomic structure (PDB: 4V7Q) is represented showing each of the monomers that forms the RV spike (VPA-A, VP4-B, VP4-C) and the names of its domains. The panel represent the primary structure of the spike indicating the different domains: α (yellow), lectin (magenta), β barrel (red), and C-terminal (green) domains. The VP4 proteolytic products (VP5* and VP8*) and domains are labelled. Residues delimiting domains and trypsin cleavage sites are indicated. (C). Structural transition of the rotavirus spike during the infectious cycle. Proteins are coloured as indicated: VP6 in blue, VP7 in yellow, VP4/VP5* foot in green, VP4/VP5* β-barrel in red, VP4/VP8* lectin domain in magenta, and loops in grey. During the last stages of the morphogenesis in the endosome, the full-length VP4 monomers in the pre-mature TLP form a flexible 3-fold symmetry structure which carry out a conformational change into the upright structure found in the immature TLP. In this immature spike, two VP4 subunits assemble forming the body and head of the spike (A and B chains) joined by loops (gray). The third VP4 subunit folds to for the stalk with a β-barrel and a lectin domain (C chain). Trypsin proteolysis cleaves the VP4 chains into VP5* and VP8* subproducts. The trypsinization of the α3-β14 loop in three residues, R231, R241 and R247, leads to the loss of the segment 232–247 in the VP4A-B chains and the loss of the lectin domain in VP4C in the mature spike. The activated spike a to the host cell through the interaction of VP8* lectin domains (attachment) with surface glycans (light green) of the cell membrane. This interaction precedes the conformational change in which the lectin domains separate and expose the hydrophobic loops resulting in an intermediate conformation in which these loops interact and insert themselves into the membrane, causing its distortion, rupture and DLP releasing in the cytosol (reversed conformation).

(TIF)

S2 Fig. Purified RV SA11 NTR- and TR-TLP analyzed by SDS-PAGE and cryo-electron microscopy.

(A, B) Coomassie blue-stained SDS-PAGE of purified TLP, cultured in the absence (A) or presence (B) of trypsin. The positions of RV structural proteins (VPs) are indicated. (C, D) Cryo-electron micrographs of NTR- (C) and TR-TLP (D). The position of some spikes projected from the surface of the particles is indicated with white arrowheads. The bar represents 100 nm.

(TIF)

S3 Fig. Resolution of the atomic models of the NTR-TLP and TR-TLP.

The resolution values of the NTR-TLP, 3.40 Å, and TR-TLP, 3.48 Å, are based on the FSC criterion at 0.143.

(TIF)

S4 Fig. Analysis of the local resolution of the 3D maps obtained from the NTR- and TR-TLP.

(A-B) Representation of the 3D maps of the NTR- (A) and TR-TLP (B) particle viewed along the icosahedral axis of symmetry 2. The densities observed in panels C and D are indicated with a dashed square. (C-D) Close view of NTR- (C) and TR-TLP (D) cross sections of each 3D map. The sections are parallel but offset 14.7Å from the central section of the maps. The surfaces are coloured according to the local resolution calculated for each 3DR. The colour code is shown with the corresponding resolutions in Å. Densities are contoured at 2σ above the mean. Scale bar represents 100 Å.

(TIF)

S5 Fig. Digital processing performed to obtain the 3DR of the spikes.

Firstly, we used the localized reconstruction method [43]: the VP2, VP6 and VP7 layers were subtracted using the corresponding TLP maps and a mask that encompasses these three layers. Subsequently, the spikes from all positions were extracted from the calculated difference images and treated as individual particles for their 3D classification, refinement and reconstruction. A 3D classification separated the positions occupied and not occupied by spikes, with an occupancy level of 59 and 37% for NTR- and TR-TLP, in each case. The unoccupied 3DR were used to subtract the VP6 and VP7 signal from the spike-occupied subparticles. Finally, the EnRICH method was applied [46] to obtain aligned subparticles whose 3D reconstructions showed a significant increase in their local resolutions, contrast and signal-to-noise ratio in the stalk, body, and head regions for both spikes.

(TIF)

S6 Fig. Analysis of the local resolution of the NTR- and TR-spikes at the different stages of the subparticle refinement.

1.34 Å thick cross sections of the maps obtained at the different stages of refinement of the VP4 NTR and TR subparticles. The panels show sections of each 3D map parallel to the central section of the maps and offset by 14.7Å (top panels) and 6.7Å (bottom panels).

(TIF)

S7 Fig. Local resolution estimates of NTR and TR spike reconstructions before and after enhancement.

Left panel: Local resolution maps of the original reconstructions of the NTR (left) and TR (right) spikes, prior to 3D classification and signal enhancement. Right panel: Local resolution maps of the final enhanced reconstructions of the NTR (left) and TR (right) spikes following focused classification and EnRICH-based signal optimization. Resolutions are color-coded from blue (higher resolution) to red (lower resolution), as indicated by the scale bar.

(TIF)

S8 Fig. Representative model-to-map fitting in distinct regions of the spike.

Examples of atomic model fitting into the cryo-EM density are shown for key structural regions of the TR spike: the foot, stalk, body, and head domains. The α3–β13 loop is shown from the NTR spike. Each panel displays the model overlaid on the corresponding enhanced map, illustrating the quality of the fit and interpretability in each region.

(TIF)

S9 Fig. Accessibility of the glycan binding sites of the lectin domains.

Representation of the VP8* domain of SA11 (PDB 1KQR) coloured according to its secondary structure. The glycan binding sites located in the cleft between the β-sheets (I) and adjacent to it (II) are indicated. (B-D) Accessibility of sites I (green line) and II (blue line) in the head (B, C) and stem (D) lectin domains of the NTR (B, D) and TR (C) spike.

(TIF)

S10 Fig. Multiple alignment of the amino acid sequences of protein VP4 from different RV strains in the regions surrounding the trypsin digestion sites.

(A) Sequences corresponding to the α2- β1 loop (between the α and lectin domains, residues 25–35). (B) Sequences corresponding to the α3 − β14 loop (between the lectin and β-barrel domains of the spike body, residues 225–261). The sequences shown correspond to representative strains of the RVA species (top panel) and to the reference strain of the different RV species (bottom panel). Residues susceptible to being cut by trypsin, lysine (K) and arginine (R), are marked in blue. All aa are presented with the single letter code.

(TIF)

S1 Video. Structural rearrangement in the stalk region.

(MOV)

S2 Video. Conformational changes in the spike body associated with α3–β13 loop cleavage.

(MOV)

S1 Table. Cryo-EM data collection and model statistics.

(DOCX)

S2 Table. Accession numbers of used VP4 segments for sequence alignment.

(DOCX)

S3 Table. Atomic coordinates and EM maps accession codes.

(DOCX)

Data Availability

Structures described in this manuscript have been deposited in Protein Data Bank under accession code 8OLB and 8OLC for the NTR- and TR-TLP structures, respectively. The cryo EM data has been deposited and issued the code EMD-16954 and EMD-16955, for the NTR- and TR-TLP structures, respectively. The NTR- and TR-spikes structures have been deposited in Protein Data Bank under accession code 8OLE and 8QTZ, respectively. The cryo-EM data has been deposited and issued the code EMD-16956 and EMD-18655, in each case.

Funding Statement

This work was supported by grants of the Instituto de Salud Carlos III (PI20CIII-00014) to DL and of the Spanish Ministerio de Ciencia e Innovación (PID2022-137548OB-I00 funded by MCIN/AEI/10.13039/501100011033/) and by ERDF A way of making Europe to JV. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.