Structural basis of nanobody recognition of grapevine fanleaf virus and of virus resistance loss

Significance

Grapevine fanleaf virus (GFLV), a picorna-like plant virus, severely affects vineyards worldwide. While nanobodies (Nbs) confer resistance to GFLV in plants, the underlying molecular mechanism of action is unknown. Here we present the high-resolution cryo electron microscopy structure of the GFLV–Nb complex. It uncovers the epitope on the capsid surface, which is a composite binding site into which the antigen loop is accommodated through an induced fit mechanism. Furthermore, we describe several resistance-breaking isolates of GFLV with reduced Nb binding capacity. Those escape variants that carry a C-terminal extension in the capsid protein also fail to be transmitted by nematodes. Together, these data provide structure–function insights into Nb–GFLV recognition and the molecular mechanism leading to loss of resistance.

Abstract

Grapevine fanleaf virus (GFLV) is a picorna-like plant virus transmitted by nematodes that affects vineyards worldwide. Nanobody (Nb)-mediated resistance against GFLV has been created recently, and shown to be highly effective in plants, including grapevine, but the underlying mechanism is unknown. Here we present the high-resolution cryo electron microscopy structure of the GFLV–Nb23 complex, which provides the basis for molecular recognition by the Nb. The structure reveals a composite binding site bridging over three domains of one capsid protein (CP) monomer. The structure provides a precise mapping of the Nb23 epitope on the GFLV capsid in which the antigen loop is accommodated through an induced-fit mechanism. Moreover, we uncover and characterize several resistance-breaking GFLV isolates with amino acids mapping within this epitope, including C-terminal extensions of the CP, which would sterically interfere with Nb binding. Escape variants with such extended CP fail to be transmitted by nematodes linking Nb-mediated resistance to vector transmission. Together, these data provide insights into the molecular mechanism of Nb23-mediated recognition of GFLV and of virus resistance loss.

Vineyards under production for wine grapes, table grapes, or dried grapes account for a total of 7.4 million hectares and more than 77.8 million metric tons worldwide, for a global trade value of 31 billion Euros (1). This positions grapevine (Vitis vinifera) as a highly valuable crop. However, grapevine is susceptible to a wide range of pathogens, including viruses. With more than 70 viral species belonging to 27 genera identified so far, grapevine represents the cultivated crop with the highest number of infecting viruses (2, 3). While the pathogenicity of all these viruses is not established, a number of them cause severe diseases, such as the viruses responsible for fanleaf, leafroll, and rugose wood diseases that have been reported in nearly all vine-growing areas. Grapevine fanleaf virus (GFLV) is a nematode-transmitted virus (4) and principal causal agent of grapevine fanleaf degeneration. It belongs to the genus Nepovirus of the family Secoviridae in the order Picornavirales and possesses a bipartite single-stranded positive-sense RNA genome (5). The structure of the icosahedral capsid is known from our previous crystal structure analysis (6) and follows a pseudoT = 3 triangulation. It is composed of 60 copies of the capsid protein (CP), which folds into three jelly roll β sandwiches (6).

Since their discovery (7), single-domain antigen-binding fragments of camelid-derived heavy chain-only antibodies, also known as nanobodies (Nbs) (8), have proven to be of outstanding interest as therapeutics against human diseases and pathogens (9–11), including viruses (12–14). Recent reports also revealed their effectiveness in conferring resistance against plant viruses. Thus, transient expression of Nbs against broad bean mottle virus attenuated the spreading of the cognate virus in Vicia faba (15). Recently, we showed that the constitutive expression of a single Nb (Nb23) that is specific to GFLV confers monogenic resistance to a wide range of GFLV isolates in both the model plant Nicotiana benthamiana and grapevine (16), but the molecular basis of GFLV recognition and the mechanism of resistance induced by Nb23 are unknown. Moreover, while one of the homozygous N. benthamiana lines tested was fully resistant to GFLV, another line showed infection at low frequency (3.2%), suggesting the existence of escape variants (EVs) containing resistance-breaking (RB) mutations susceptible to interfering with Nb23–GFLV interaction (16).

To address the molecular basis of Nb23–GFLV recognition, we determined the cryo electron microscopy (cryo-EM) structure of the GFLV–Nb23 complex at high resolution. The structure reveals that Nb23 bridges over 3 domains of the CP, and it provides unprecedented insights into the epitope and the residues involved in the interface, including the mechanism of molecular recognition of the antigen-binding loops. We find a perfect correlation between mutations detected in RB variants and the Nb23 epitope observed in the structure, which explains the resistance loss in EVs. In agreement with the fact that the conformational surface epitope recognized by the Nb23 partially covers a cavity involved in vector transmission (6, 16, 17), we show that EVs preexist in natural viral populations at low frequency. We also uncover that the most frequently found EVs with extended CP are deficient in transmission by nematodes.

Results

To gain molecular insights into the mechanism of GFLV recognition by Nb23, precisely map the epitope, and decipher the interactions between Nb23 and the CP, we decided to analyze the structure of the GFLV viral particle decorated with Nb23. The structure of the GFLV–Nb23 complex was determined by high-resolution single-particle cryo-EM, using angular reconstitution (18, 19), and refined to an average resolution of 2.8 Å (Fig. 1 and SI Appendix, Figs. S1 and S2). The cryo-EM map was used to build an atomic model with proper model geometry and statistics, using the Phenix software (20), which involved manual model building and refinement notably of the antigen-binding loop region (Methods). The three-dimensional (3D) reconstruction displays the 3 core parts of the complex: the virus capsid, the Nb at the outer surface, and the genome inside (Fig. 1). The structure reveals that Nb23 binds at the surface of the GFLV capsid in the vicinity of the 5-fold axis (Fig. 1 D–F), at the level of the individual CP monomers that form the pentamers of the pseudoT = 3 icosahedral capsid. The outer isocontour surface of the GFLV–Nb23 reconstruction (Fig. 1 and SI Appendix, Fig. S1) shows that the Nb23 molecules are positioned far enough from each other to allow 60 of them to attach and reach full 1:1 stoichiometric binding without bridging neighboring CPs. Consistent with this finding and a size increase of the complex (16), we observe the same stoichiometry by native gel analysis upon titration of GFLV with increasing amounts of Nb23 (SI Appendix, Fig. S3).

Fig. 1.

Structural analysis of the GFLV–Nb23 complex by cryo-EM. (A) Typical cryo-EM micrographs of the GFLV–Nb23 complex; most particles are intact and comprise the genome inside (an empty particle is indicated at the bottom with an arrow). (Scale bar, 500 Å.) (B) Local resolution estimation, section through Nb23 and the capsid. (C) Plot showing FSC versus spatial frequency of the icosahedral averaged reconstruction. Average resolution of the reconstructions is given where the FSC curve crosses below a correlation value of 0.143. (D) Overall structure of the GFLV–Nb23 complex with Nb23 shown in cyan and the capsid in radial coloring (SI Appendix, Figs. S1 and S2), including a segmentation of the 3D reconstruction for the Nb23 and the GFLV capsid parts, and a cross section through the icosahedral reconstruction (left to right); the icosahedron is shown as an overlaid polygon. (E) Sections through the refined 3D reconstruction (from top to bottom of the 3D map). (F) Sections through the 3D reconstruction while under refinement; at lower resolution, the overall parts are clearly visible: Nb, capsid (secondary structures of the CP are visible) and genome inside the capsid (averaged out due to the icosahedral symmetry that is imposed during 3D reconstruction while the genome might have a different structural organization).

Local resolution estimation shows that while Nb23 is moderately ordered at the periphery (3.5 to 5 Å resolution toward the C terminus), it is very well defined in the region of interest, enabling a precise analysis of the interactions within the interface between Nb23 and the capsid (Fig. 2). The part of Nb23 that is oriented toward the capsid exhibits clear side-chains densities (SI Appendix, Fig. S2) and includes the important antigen-binding loops (Fig. 2 D and E). The viral capsid is well defined (including the very N- and C-terminal ends) and comprises the A, B, and C domains of each CP (Fig. 2 A and B). The interaction region between Nb23 and the capsid involves residues of the complementarity determining regions (CDRs) 2 and 3 of Nb23, as well as the two neighboring CP domains A and B (2 of the 3 jelly roll β sandwiches of the CP). These structural elements from the same CP monomer polypeptide chain jointly form a composite binding site with two core interacting regions, denoted 1 and 2 (Fig. 2 A, D, and E), adding up to a total interaction surface of ∼1,100 Ų. This large binding area is consistent with the formation of a stable Nb23–GFLV complex and with the high affinity of Nb23 for the virus, as determined by microscale thermophoresis (K_d = 9.13 ± 2.91 nM; n = 3; SI Appendix, Fig. S3).

Fig. 2.

Structural analysis of the Nb23 epitope on the GFLV capsid. (A and B) Ribbon diagram of a CP monomer and its associated Nb23, with side and top views of the CP and Nb23 (cyan), which sits onto a composite binding site comprising the 3 domains of the CP monomer shown in blue (A), red (B), and green (C); the C terminus of the CP is indicated. (C) Positions of CP and Nb23 (cyan) with respect to symmetry operators on the icosahedral reconstruction. (D and E) Close-up views of the boxed regions 1 and 2 in A, showing the detailed molecular interactions (hydrogen bonds are indicated by dotted lines; color code as in B). (F) Comparison of apo-capsid (in yellow) and Nb23-bound capsid (in blue) showing a conformational adaptation of the capsid residues Phe370 to Asn375, reminiscent of an induced fit mechanism.

The structure allows a precise mapping of the Nb23 epitope on the GFLV capsid and identifies the key residues of Nb23 CDR2 and CDR3 and of GFLV involved in the specific molecular recognition events occurring on complex formation. Within the two core interacting regions (Fig. 2 D and E), a high level of specific interactions is provided through a series of noncovalent bonds (SI Appendix, Fig. S4). Region 1 (Fig. 2D) comprises loop region 212 to 216 (βC’’) of domain B of the CP in which Thr212 interacts with Asp100_Nb23 and Tyr32_Nb23, Lys214 forms hydrogen bonds with the carbonyl backbone of Ala101_Nb23 and Ile102_Nb23, and Tyr216 interacts with the backbone and forms hydrophobic contacts with the side chain of Leu104_Nb23. Region 2 (Fig. 2E) comprises the two strands of the capsid β-sheet region 370 to 391 (domain A of the CP) in which Asp371 forms a salt bridge and Thr373 a hydrogen bond with Arg55_Nb23. An additional anchor point of the antibody is provided through Thr58_Nb23 and Lys65_Nb23, which interact with Asn375 located in a CP loop (Fig. 2E). In the second β-strand of the 370 to 391 region, the backbone of Val379 interacts with Thr110_Nb23, and the Ser380 backbone and Met381 form contacts with Ser109_Nb23 and Trp108_Nb23, respectively. CDR3 forms a long antigen-binding loop with a short α-helical turn carrying Leu104_Nb23, which is accommodated within a hydrophobic pocket formed by Tyr216, Phe370, Phe502, Val504, and an alanine cluster formed by residues 387, 388, and 391 (Fig. 2D). Trp108_Nb23 appears to be a key residue, in that it inserts into the binding site within a cavity at the junction of the A and B domains of the CP to form π-stacking interactions with Phe370 (Fig. 2E). Importantly, the Phe502 and Val504 residues at the very C-terminal end of the CP are in direct contact with Nb23 and form hydrophobic contacts with Leu104_Nb23, Tyr107_Nb23, and Trp108_Nb23. Finally, Lys54_Nb23 interacts with the free carboxylate moiety of Val504.

Interestingly, comparison of the CP structures in the presence and absence of Nb23 (Fig. 2F) reveals that Nb23 induces a conformational change of the CP 370 to 375 loop region to accommodate the Nb23 side-chains. This significant transition indicates that the molecular recognition of the CP occurs through an induced fit mechanism in which the Phe370 and Asp371 side-chains flip over in opposite directions (Fig. 2F). Without this conformational adaptation, Asp371 would clash with Trp108_Nb23, which instead needs Phe370 to switch position to form stacking interactions with Trp108, whereas Asp371 rotates to form a stabilizing salt bridge interaction with Arg55_Nb23. This selection mechanism may contribute to Nb23 specificity toward the GFLV antigen (16).

The structure shows that the interface between Nb23 and CP is highly complementary, and also includes the C-terminal CP residue Val504 that is recognized by Lys54_Nb23. One can therefore hypothesize that disrupting the molecular interface of the epitope region would lead to resistance loss. Interestingly, Nb-mediated resistance to GFLV in N. benthamiana lines that constitutively express Nb23:EGFP displays various degrees of susceptibility to infection, indicating that GFLV could overcome Nb-mediated resistance (16). To further explore this partial resistance breakdown, we forced our two resistant lines toward the emergence of infection events by applying high inoculum pressure (3 μg vs. 300 ng of virus). Under such stringent conditions, resistance was indeed overcome by 21 d postinoculation (dpi) in 30% and 40% of plants from lines 23EG38-4 and 23EG16-9, respectively (Fig. 3 A and B). To address whether the infection events observed were due to the emergence of bona fide GFLV RB variants and considering Nb23 is specific to GFLV (16), we characterized the virus progeny by sequencing the CP from Nb23-expressing plants that showed systemic infection. Remarkably, the analysis reveals that all suspected RB events correspond to viruses with mutations in the CP coding sequence (Fig. 3C). The most frequent mutation (five independent events) is a nucleotide insertion near the 3′ end of the CP gene that leads to a frameshift responsible for nonconservative mutations of CP residues 503 and 504 and the addition of 12 extra amino acids at the C terminus of the CP (Fig. 3C). The second most represented mutation (three occurrences) was the suppression of the stop codon leading to a CP with three (CP+3) or five (CP+5) extra C-terminal amino acids, the latter being the consequence of a second mutation. The remaining two mutations are missense mutations mapping to residues 216 (Tyr216His) and 502 (Phe502Leu) of the CP, both of which are directly involved in Nb23-virus recognition (Fig. 2 D and E). Remarkably, most RB variants (8 of 10) display a mutated CP with variable C-terminal extensions of 3 (CP+3), 5, or 12 residues (Fig. 3C); such an extension would sterically clash with Nb23 and prevent recognition of the free carboxylate group Val504, as evident from the structure (Fig. 2E). Accordingly, GFLV-CP+3 and GFLV-Tyr₂₁₆His mutants are recognized only very poorly by Nb23 in double antibody sandwich–enzyme-linked immunosorbent assay (DAS-ELISA) (while they still are with conventional anti-GFLV antibodies; Fig. 3D). To confirm that C-terminal extensions of the CP are RB mutations and to exclude the contribution of non-CP residues in overcoming resistance, the CP+3 mutation was introduced into the GFLV-GHu infectious clone (21). The resulting mutant fully infects both control and Nb23-expressing plants (100% infection), in contrast to wild-type GFLV-GHu that infects only control plants (Fig. 3E), confirming that the CP+3 mutation is sufficient to overcome Nb-mediated resistance.

Fig. 3.

Molecular characterization of EV. (A) Symptom analysis of plants from lines EG11-3 (green), 23EG16-9 (red), and 23EG38-4 (orange) at 7, 14, and 21 dpi on inoculation with 3 µg of GFLV-GHu. (B) GFLV detection by DAS-ELISA at 21 dpi. Each dot corresponds to a single plant sample and represents the mean relative absorbance at 405 nm of experimental duplicates. Percentage of infections are indicated below each column. (C) Mapping of mutations found in EV. Mutations were found in the ORF encoding the CP of GFLV (red for substitutions and blue for insertions). In total, sequences from 10 independent EV were obtained and occurrence of mutations is provided for each type of mutant. CP sequence of GFLV-GHu from residues 215 to 217 and 501 to 517 is provided. CP coding sequences are boxed and corresponding protein sequences are provided here. (D) N. benthamiana infected with GFLV-GHu, GFLV-CP+3, or GFLV-Tyr₂₁₆His were tested by DAS-ELISA with either conventional antibodies or Nb23 for detection. Conventional antibodies recognize all GFLV isolates contrarily to Nb23 that fails to detect GFLV-CP+3 and GFLV-Tyr₂₁₆His, indicating reduced binding of Nb23 to GFLV EV. (E) Susceptibility of N. benthamiana transgenic lines toward GFLV GHu or GFLV-CP+3. Plants were tested by DAS-ELISA for GFLV at 21 dpi. (F) HTS data mining on 13 samples infected with GFLV (12 from grapevines, upper histograms and 1 from nematodes, lower histograms). We focused on seven positions within the CP, with residues in red involved in Nb23-CP interactions, and found in EV, in orange, those involved in Nb23-CP interactions based on the structural analysis (Fig. 2), and in black, those exposed on the outer surface of the capsid but not interacting with Nb23 nor involved in the viral transmission by nematodes. The underlined red residues were found mutated in EV. Each colored bar corresponds to the percentage of a missense mutation occurrence with ± SE (with the number of event per mutation/site being indicated above). (G) Transmission efficiency of GFLV-GHu and GFLV-CP+3 by nematodes. Dots correspond to DAS-ELISA results from individual bait plants and represent the mean relative absorbance at 405 nm of experimental replicates normalized against maximum assay values. The asterisk indicates statistically significant differences (two-tailed paired t test; P = 0.022). Number of plants tested (n) and percentage of infections (%) are provided below each column. NC, healthy plants. (H) Negative staining EM image of immunogold-labeled viral particles isolated from GFLV-CP+3 infected plants.

We next investigated the possible preexistence and frequency of such mutations in natural viral populations from grapevines and nematodes by high-throughput sequencing (HTS) data mining (Fig. 3F and Dataset S1), with a focus on the codons encoding Tyr216, Phe502, and the amber stop codon identified to be at the origin of EVs (Fig. 3F). We also looked at the codons encoding Phe370 and Asp371 due to the importance of these residues in the interaction with Nb23, as seen in the structure (Fig. 2; see list in SI Appendix, Fig. S4). As a control, we analyzed the codons encoding Met302 and Trp119, two outer-exposed residues of the CP excluded from the Nb23 epitope and likely not involved in α/β structures nor in the virus transmission by nematodes (6, 17). All point mutations discovered via Sanger sequencing were unequivocally identified in our HTS datasets. For example, the Tyr₂₁₆His was found in 2 of the 12 grapevine samples tested. In contrast, STOP₅₀₅Gln was detected in all samples at an average rate close to 0.4%, together with other missense mutations (e.g., STOP₅₀₅Tyr, STOP₅₀₅Ser, or STOP₅₀₅Glu), leading to a CP with extra C-terminal amino acid extensions. Such assortment of missense mutations was observed at all examined positions (Fig. 3F). Similarly, most RB mutations were also detected in the viral RNA isolated from viruliferous vector nematodes (Fig. 3F), underscoring that most RB variants preexist in natural viral populations at low frequencies as part of the viral quasispecies, as defined in Domingo et al. (22).

Finally, to address the question of whether the EVs would still be active in vector transmission, we tested GFLV-Tyr₂₁₆His, Phe₅₀₂Leu, and CP+3 mutants for transmission by nematodes. While both single-point mutants were still vectored, transmission of GFLV-CP+3 was abolished (Fig. 3G), despite its ability to form viral particles (Fig. 3H). This uncovers an important function of residues in the vicinity of the C-terminal Val504 in GFLV transmission by Xiphinema index. Remarkably, as can be seen in the structure, the Nb23 epitope (delineated in blue; Fig. 4 A and B) and the ligand-binding pocket (LBP; delineated in magenta; Fig. 4B) partially overlap on the capsid surface (Fig. 4 A and B). The LBP is a surface-exposed positively charged pocket near the 3-fold axis, whose walls are formed essentially by the βG-βH, βB-βC, and βC’βC’’ loops within the B-domain of the CP (6). Previous mutagenesis and loss-of-function studies have demonstrated the essential function of residues from the βG-βH (6) and from the βB-βC (17) loops in GFLV transmission by nematodes. However, the function of the βC’βC’’ loops and neighboring residues in this mechanism was only predicted (6). Here we show that residues, which belong to this area in the LBP (in particular Val₅₀₄), when mutated, abolish GFLV transmission. The present structure–function analysis thus clarifies why RB mutations not only hinder Nb23 binding but also, in the case of the C-terminal extensions, interfere with GFLV transmission.

Fig. 4.

Fingerprint of Nb23 on the GFLV capsid and mechanism of resistance loss. (A) Fingerprint of the Nb23 binding site on the capsid for one icosahedral asymmetric unit (black triangle, see also SI Appendix, Fig. S1) in which the polar angles θ and ϕ represent latitude and longitude, respectively. The coloring is according to the radial distance of the surface from the center of the particle; coloring of a CP monomer by domains as in Fig. 2B. (B) Fingerprint showing the projected surfaces on the GFLV capsid. The Nb23 footprints and projected surfaces are delineated in yellow and cyan, respectively, the LBP in magenta and mutated residues in GFLV EV shown in white, together with amino acid positions. The mosaic background shows the amino acids that form the viral surface. (C) Schematic representation of the molecular mechanism in which the C-terminal extension of GFLV EV would clash with Nb23, thus leading to resistance loss.

Taken together, this study provides insights into the molecular mechanism of Nb23-mediated recognition of GFLV, including the location of the Nb next to the fivefold symmetry axes on the capsid (Fig. 1) and the detailed description of the interface between the GFLV epitope and Nb23 (Fig. 2). The epitope consists of a composite binding site involving domains A, B, and C of the CP that interacts with the CDR2 and CDR3 of Nb23 (Fig. 2). Remarkably consistent with the epitope seen in the cryo-EM structure, the RB mutations all map to neighboring residues exposed on the outer surface of the GFLV capsid (Fig. 3 and SI Appendix, Fig. S4), which we show preexist at low frequency in virus quasispecies (Fig. 3). Specifically, the structure explains the mechanism of action of the C-terminal extension mutants in overcoming resistance: the extra three, five, or 12 residues at the C terminus of the CP from GFLV EVs would interfere with Nb23 binding due to steric clashes and loss of specific hydrogen bonds (in particular between the free carboxylate group of the CP last residue, Val504, and the Lys54_Nb23 region; Fig. 2E). This would prevent contacts with CP Val504 and occlude the highly specific interface necessary for Nb23-GFLV recognition (Fig. 4C). This manner of interfering with Nb23 binding and prevent GFLV neutralization is likely more potent than single-residue missense mutations such as Tyr216His or Phe502Leu, which may explain why most escape mutations consist of C-terminal extensions of the CP. Remarkably, these specific RB mutations that lead to resistance loss simultaneously result in transmission deficiency due to the overlap between the epitope and the LBP (Fig. 4). This demonstrates that the RB mechanism observed here has its fitness cost: by maintaining its capacity to infect resistant plants, GFLV has evolved in a manner leading to loss of transmission by nematodes. Although the principles are different, both Nb-mediated resistance and EVs that fail to be transmitted from plant to plant by nematodes contribute to prevent the progression of the disease. Considering that the RB mutants were isolated from a herbaceous host (N. benthamiana expressing Nb23:GFP) under laboratory conditions, it would be interesting to investigate the emergence of RB isolates, in particular the emergence of GFLV isolates with C-terminal extensions of the CP, also in grapevine upon natural transmission of the virus by nematodes. Their influence on the robustness and durability of the resistance in grapevine under natural conditions remains to be determined, especially when aiming at deploying Nb-mediated resistance (16) as an antiviral strategy in vineyards.

Finally, our structural data based on purified particles decorated with 60 Nbs indicate that Nb23 has no virus-destabilizing activity in vitro. Hence the Nb-induced capsid destabilizing activity proposed for norovirus-specific Nbs (23) does not apply for GFLV and likely does not contribute to the Nb-mediated resistance mechanism in planta. Whether Nb23 stabilizes viral particles, interferes with cell to cell movement, and/or leads to the rapid degradation/clearance of the virus in vivo is unknown and remains to be addressed in future studies.

Methods

Transgenic N. benthamiana Lines.

Homozygous lines EG11-3 expressing EGFP and 23EG16-9 and 23EG38-4 expressing Nb23 fused to EGFP were described in Hemmer et al. (16).

Inoculations.

Plants were inoculated either mechanically by rub inoculation using carborundum powder, or naturally via nematodes. Mechanical inoculations were performed with either purified virus at 300 ng or 3 µg per inoculum or with crude sap from infected plants.

GFLV DAS-ELISA Assessment of Nb23 In Vitro Binding to EV.

Frozen leaf samples from infected plants were ground in extraction buffer (35 mM Na₂HPO₄, 15 mM KH₂PO₄ at pH 7.0) in a 1:5 wt/vol ratio. Virus detection was performed on twofold serial dilutions of clarified extracts by DAS-ELISA, using GFLV antibodies (Bioreba, catalog #120275) diluted 1,000-fold in coating buffer (15 mM Na₂CO₃, 35 mM NaHCO₃ at pH 9.6) as capture antibody and either GFLV antibodies coupled with alkaline phosphatase (Bioreba, catalog #120275) diluted 1,000-fold or Strep II-tagged Nb23 at 1 µg/mL in conjugate buffer (10 mM phosphate-buffered saline [PBS], 0.1% wt/vol bovine serum albumin [BSA], 0.05% vol/vol Tween 20 at pH 7.4) as detection antibody. In the latter case, an additional incubation with streptavidin-alkaline phosphatase (Jackson Immunoresearch, catalog #016-050-084) at 1:10,000 in conjugate buffer was performed before detection with paranitrophenyl phosphate (Interchim) at 1 mg/mL in substrate buffer (1 M diethanolamine at pH 9.8). Negative control consisted of noninoculated healthy plants. Absorbance at 405 nm (A_405nm) was recorded after 1 h of substrate hydrolysis. Results are presented as mean absorbance of experimental duplicates ±SE normalized against maximum assay value.

Characterization of RB Isolates.

Leaf samples from Nb23-expressing N. benthamiana that tested positive by ELISA for GFLV were used for immunocapture RT-PCR and gene encoding the CP of GFLV sequenced by Sanger sequencing, as described in Martin et al. (24).

Synthetic GFLV-CP+3 Construct.

GFLV-CP+3 point mutation was introduced into full-length GFLV-GHu RNA2 cDNA infectious clone (pG2) (21) as a BglII/SalI fragment following an overlap extension PCR amplification, using pG2 as template. GFLV-GHu RNA1, GFLV-GHu RNA2, and GFLV-CP+3 RNA2 capped transcripts were synthesized from corresponding SalI/XmaI or EcoRI/SalI linearized cDNAs clones by in vitro transcription with mMESSAGE mMACHINE T7 kit (Ambion) according to manufacturer’s instructions. For infections, combination of appropriate RNA1 and RNA2 transcripts was used to mechanically inoculate N. benthamiana plants at the 4- to 6-leaf stage.

Transmission Assays.

Transmission assays by nematodes were performed using a two-step transmission procedure according to Marmonier et al. (25). During a 6-wk acquisition access period, ca. 300 aviruliferous X. index per plant were left in contact with roots of systemic infected N. benthamiana grown in a 3:1:1 vol/vol ratio of sand, loess, and clay pebbles soil. Source plants were then substituted with healthy N. benthamiana for an 8- to 10-wk inoculation access period, at the end of which virus transmission was tested on roots of bait plants by DAS-ELISA. GFLV-GHu was used as positive and healthy source plants as negative controls. Values exceeding those of negative controls by a factor of 2.4 were considered positive.

His₆-tagged and Strep-Tagged II Nb23 were produced as described in Hemmer et al. (16).

Native Agarose Gel Electrophoresis.

Native gel electrophoresis of purified GFLV was performed in 1% wt/vol agarose gels in 1.0× TA buffer (20 mM TrisBase, 0.06% vol/vol acetic acid at pH 9). For nucleic acids detection, 5 µg of virus particles were diluted in loading buffer (10% vol/vol glycerol, 25 mM Hepes at pH 9) supplemented with ethidium bromide (EtBr) at 0.1 µg/mL After electrophoretic separation, the EtBr-prestained gel was first processed for nucleic acid content, using the Gel Doc system (Bio-Rad) equipped with a 302-nm excitation source and a 520- to 640-nm band-pass emission filter before Instant Blue protein staining.

Microscale Thermophoresis.

Microscale thermophoresis (MST) measurements were performed using a Monolith NT115 instrument (NanoTemper Technologies GmbH), using Nb23 and purified GFLV-derived virus-like particles in which enhanced GFP is encaged instead of mRFP, as described in Belval et al. (26). The concentration of the fluorescent virus-like particles was kept constant at a concentration of 100 nM, while the concentration of Nb23 varied from 0.07 to 1.0 μM. Samples were diluted in MST optimized buffer (20 mM Tris⋅HCl buffer at pH 9.5). Each experiment consisted of 15-point titration series in which the Nb concentrations were generated as a 1:1 dilution of virus-like particles. For the measurements, 10 μL of the different dilutions of samples were transferred into hydrophilic-treated capillaries (premium coated capillaries, NanoTemper Technologies) and measured after a 10-min equilibration at room temperature. The measurements were performed (n = 3) at 40% light-emitting diode (LED), and 20%, 40%, and 80% MST power. Laser-on time was 30 s, and laser-off time was 5 s. MST data were analyzed using the program PALMIST (biophysics.swmed.edu/MBR/software.html), and evaluated using the appropriate model (1:1), according to Scheuermann et al. (27).

Immunosorbent Electron Microscopy.

GFLV-CP+3-infected leaves were ground in phosphate buffer (35 mM Na₂HPO₄, 15 mM KH₂PO₄) and centrifuged at 3,000 × g for 5 min, and clarified samples were incubated for 2 h on carbon-coated Formvar nickel grids (300 mesh; Electron Microscopy Science, PA) coated with anti-GFLV polyclonal antibodies (Bioreba AG, catalog #120475) at 1:200 dilution. After blocking with 2% wt/vol BSA, 10% vol/vol normal goat serum, 0.05% Triton-X100 in 22.5 mM Hepes at pH 8 for 30 min, grids were incubated with monoclonal anti-GFLV (Bioreba, catalog #120475) at 1:150 dilution for 30 min and immunogold labeling performed using anti-mouse antibodies conjugated to 10 nm colloidal gold particles at 1:50 dilution for 30 min (British Biocell International, catalog #EM.GAMA10). The experiment was performed at room temperature, and washes were done with phosphate buffer between all steps. Observations were realized after negative staining with 1% m/v ammonium molybdate, using a Philips EM208 transmission electron microscope.

Sample Preparation for Cryo-EM.

GFLV was purified from infected N. benthamiana plants according to Schellenberger et al. (6). GFLV-specific single-domain antibody Nbs (Nb23) were generated as described in Hemmer et al. (16). Dynamic light-scattering analyses were used both to find the buffer compositions and pH values at which GFLV and GFLV bound to Nb23 are soluble and to determine their diameters and monodispersity. Measurements were performed using a Malvern Zetasizer NanoZS instrument and quartz cuvettes. For every 20-µL sample, 10 successive measurements were performed at 20 °C with constant amounts of GFLV (0.1 mg/mL). Data were processed with manufacturer’s DTS software (version 6.01). GFLV-F13 remained soluble in 20 mM Tris⋅HCl at pH 8.3 to 8.5 upon addition of Nb23. The latter composition was chosen for the preparation of cryo-EM grids in which GFLV at 0.5 mg/mL led to exploitable images.

Structure Determination.

Nb23 was mixed with purified GFLV-F13 at a 70:1 or 100:1 molecular ratio, so as to saturate the theoretical maximum number of 60 possible binding Nbs per capsid. Samples of the freshly prepared GFLV–Nb23 complex (2.5 µL of a 0.5 mg/mL solution) were applied to 300 mesh holey carbon Quantifoil R1.2/1.3 grids (Quantifoil Micro Tools, Jena, Germany), blotted with filter paper from both sides in the temperature- and humidity-controlled Vitrobot apparatus (FEI, Eindhoven, Netherlands), T = 20 °C, humidity 99%, blot force 4, blot time 0.5 s, and vitrified in liquid ethane precooled by liquid nitrogen. Images were collected on the in-house spherical aberration-corrected Titan Krios S-FEG instrument (FEI) operating at 300 kV acceleration voltage and at an actual underfocus of −0.6 to −6.0 μm, using the second-generation back-thinned direct electron detector CMOS (Falcon II) 4,096 × 4,096 camera and automated data collection with EPU software (FEI), using seven frames over a 1-s exposure (dose rate of ∼40 e⁻ Å⁻²⋅s⁻¹). The calibrated magnification (based on the fit of the GFLV crystal structure into the cryo-EM map) is 127,272×, resulting in 0.527 Å pixel size at the specimen level (which is close to the nominal pixel size of 0.533 Å, 1.1% difference). Before semiautomatic particle picking using IMAGIC (19), stack alignment was performed using the whole image motion correction method (28), and correction of the contrast transfer function was done by phase flipping using IMAGIC. After centering by alignment against the total sum reference image, 2D classification in IMAGIC was used to remove bad or empty particles, leaving 6,139 of 9,502 particles (selected from 1,055 images). For the first steps of image processing, the images were coarsened by a factor of four (binned 4×), and further refinement was achieved with two times coarsened data. The structure was determined via common lines angle assignment and refined with the anchor set angular reconstitution procedure, as implemented in IMAGIC, using final search angles of 0.05°.

Thus, the structure determination, and also refinement, were done entirely using common line methods for Euler angle assignments and refinement, which are less biased and require no starting reference. The resolution was estimated by Fourier shell correlation (FSC) calculations, indicating an average resolution of 2.8 Å according to the 0.143 and half-bit resolution criteria (29, 30), using IMAGIC (19); map-model correlation calculations [using Phenix (20)] suggest an average resolution of ∼3.5 Å (d_model in Phenix mtriage, d₉₉ is 3.2 Å resolution for the unmasked reconstruction), probably because there is averaging in the resolution estimation between the well-defined CP and CP–Nb23 interface and the less ordered solvent-exposed parts of Nb23 (3.5 to 5 Å resolution toward the C terminus but still well visible in medium-resolution maps [SI Appendix, Fig. S1 A and B]; we kept the atomic model of these Nb23 regions in the overall structure as the fold is known from other structures and places well into the nonfiltered map). Because the exterior part of the Nb23 structure is less well ordered, and therefore less resolved (Fig. 1B), this might also have an influence on the shape of the FSC curve. However, there is no doubt that the features resolved in the map allow visualizing side-chains in the capsid and in the interface region with Nb23, which is required for the description of the residues at the CP–Nb23 interface, consistent also with the fact that the atomic model refined well, as illustrated from the good fit in the well-resolved regions (SI Appendix, Fig. S2) and the overall good geometry (SI Appendix, Fig. S1E). Local resolution values were calculated with ResMap (31). Map interpretation was done using Chimera (32) and COOT (33), starting from the available atomic model of the GFLV crystal structure (PDB IDs: 4V5T and 2Y7T) (6) and of a homology model of Nb23 that we derived from the cAb-Lys3 VHH antibody domain, using the Swiss-Model Server (https://swissmodel.expasy.org/, starting model: PDB ID: 1XFP; ref. 34, 68% sequence identity with Nb23). Initial atomic model building was done by a rigid body fitting, followed by extensive manual model building in COOT and real space refinement of the atomic model against the experimental map, using Phenix, including restrained temperature factor refinement (20). The final atomic model comprises 295,800 atoms (60 × 4,930 atoms excluding hydrogens across the 504 and 137 amino acids for each monomer of the CP and Nb23, respectively). Protein residues of the final atomic model show well-refined geometrical parameters (most favored regions, 96.7%; additionally allowed regions, 3.3%; and 0.0% of outliers in Ramachandran plots, r.m.s. bond deviations of 0.009 Å and angle deviations of 0.8°; SI Appendix, Fig. S1 for data collection and structure refinement statistics). Solvent-accessible surface was calculated with the program GetArea (probe radius 1.4 Å, ref. 35). Figures were prepared using the software Chimera (32) and Pymol (36).

HTS Data Mining.

To determine whether RB mutations preexisted within natural viral populations or were acquired in Nb23-expressing plants, we analyzed HTS datasets (2 × 150 bp RNAseq performed on an Illumina Hiseq 3000, Illumina, San Diego, CA) from a previous study by Hily et al. (37). We specifically focused on 12 repetitions of grafted and nongrafted grapevines naturally infected by a vineyard population of GFLV. In addition, total RNA was extracted from a single sample containing 300 viruliferous nematodes previously fed on a GFLV-F13-infected plant according to Demangeat et al. (38), and a cDNA library was made as described in Hily et al. (37). Analyses of datasets were performed using CLC Genomics Workbench 8.5.1 software (Qiagen), with mapping parameters set at length fraction of 0.5 and similarity of 0.7. Postmapping, SNPs at each position were recovered (only SNPs above the Illumina error rate of 0.1% per base were recorded), and missense mutations (consisting of one SNP per codon) were reported with their percentage rate. Data are available in Dataset S1.

Statistical Analyses.

For the GFLV-CP+3 transmission assay, statistical analysis was performed by two-tailed paired t test (P = 0.022).

Data Availability.

Materials are available upon reasonable request. Raw data of the HTS data mining are available in the Dataset S1 file. Cryo-EM map and coordinates of the atomic model can be found in the Protein Data Bank and the Electron Microscopy Data Bank under the following accession numbers: PDB, 5FOJ; EMD, 3246.