Megabodies expand the nanobody toolkit for protein structure determination by single-particle cryo-EM

Abstract

Nanobodies are popular and versatile tools for structural biology. They have a compact single immunoglobulin domain organization, bind target proteins with high affinities while reducing their conformational heterogeneity, and stabilize multi-protein complexes. Here we demonstrate that engineered nanobodies can also help overcome two major obstacles that limit the resolution of single-particle cryo-EM reconstructions: particle size and preferential orientation at the water-air interfaces. We have developed and characterised novel constructs, termed megabodies, by grafting nanobodies onto selected protein scaffolds to increase their molecular weight while retaining the full antigen binding specificity and affinity. We show that the megabody design principles are applicable to different scaffold proteins and recognition domains of compatible geometries and are amenable for efficient selection from yeast display libraries. Moreover, we demonstrate that megabodies can be used to obtain 3D reconstructions for membrane proteins that suffer from severe preferential orientation or are otherwise too small to allow accurate particle alignment.

Introduction

Single particle cryo-EM has recently become the technique of choice for the structural characterisation of membrane proteins and macromolecular complexes^1–3. Instrumentation and data analysis methods continue to improve at a spectacular pace^4,5. However, factors including small particle size, low symmetry, high flexibility and non-random distribution in ice limit the achievable resolution of 3D reconstructions. Whereas large molecules are relatively easy to identify in noisy low-dose images of frozen hydrated samples, and have sufficient features to facilitate the accurate determination of their orientation and position required for alignment and averaging⁶, structural analysis of small particles (~100 kDa or less) is considerably more difficult. Phase contrast methods such as defocusing the objective lens or the use of phase plates, especially the Volta phase plate (VPP)⁷, enable imaging of biological specimens with an acceptable (albeit still limiting) amount of radiation damage⁸. Despite the recent success in high-resolution reconstructions of small macromolecule proteins aided by a VPP including streptavidin (52 kDa tetramer, ~3.2 Å)⁹ and human haemoglobin (64 kDa tetramer, ~3.2 Å)¹⁰ or using conventional defocus-based data collection for methaemoglobin (64 kDa tetramer, ~2.8 Å and ~3.2 Å) and alcohol dehydrogenase (82 kDa dimer, ~2.7 Å)¹¹, the routine analysis of such samples remains very challenging. To circumvent this problem, several protein engineering approaches have been employed to increase the size of particles. Small proteins have been genetically fused to molecular scaffolds¹² or bound to recognition domains organised in large molecular supports of high symmetry^13,14. Attempts to align the target molecules according to the symmetry of multimeric scaffolds, with the help of local/focused classification techniques, are useful in principle but limited in practice by the flexibility of linker regions.

Regardless of particle size, particle distribution in the vitreous ice layer greatly impacts on the quality and attainable resolution of cryo-EM reconstructions¹⁵. During the EM grid preparation process, particles diffuse and interact with the water-air and/or water-support interfaces in the order of 1000 times a second^15,16. Particles are typically adsorbed to such interfaces and get denatured, or present preferential orientation due to their surface properties¹⁷. While supplementation of buffers with surfactants such as detergents is often employed to improve particle distribution in ice¹⁸, for sensitive specimens including membrane proteins this approach should be avoided. To limit the impact of preferential particle orientation, specimens can be tilted inside the microscope during data collection¹⁹. Additionally, faster plunge-freezing devices that reduce the time between sample application and vitrification²⁰, utilization of graphene-supported grids with tuneable surface properties²¹ or engineered hollow 3D-DNA structures able to trap DNA-binding proteins inside²² have been explored. Yet, to date, such approaches could not be generalised to eliminate the particle orientation problem in cryo-EM.

To help overcome these performance barriers, we designed a novel class of chimeric molecules, called megabodies (Mbs). Megabodies are built by grafting single domain antigen binding constructs such as nanobodies (~15 kDa)^23,24 or monobodies (~10 kDa)²⁵, into larger scaffold proteins to produce stable, reasonably rigid and efficiently-folding monomeric chimeras. The generic nature of this technology enables a direct selection of novel megabodies directly from nanobody immune libraries or, alternatively, a straightforward reformatting of existing nanobodies. In a first example, we show that megabodies, unlike their parental nanobodies, successfully overcome the severe preferential orientation of a human GABA_A-β3 receptor construct reconstituted in a lipid nanodisc. The excellent quality of the cryo-EM map obtained, ~2.5 Å nominal resolution, allows unambiguous identification of histamine, a small-molecule agonist composed of only eight non-hydrogen atoms. We also developed megabodies that bind to the membrane scaffold protein (MSP) commonly used to prepare nanodiscs²⁶. These generic tools help randomize the orientation of particles and will facilitate, in principle, cryo-EM studies of any membrane protein that is reconstituted in nanodiscs of different size and lipid composition. Lastly, we show that megabodies can be used as fiducial markers to obtain 3D cryo-EM reconstructions for WbaP, a small (~57 kDa) dimeric integral membrane protein reconstituted in amphipols.

Results

Design, expression and purification of megabodies

The generic megabody design consists of a small globular domain, such as a nanobody, grafted onto a scaffold protein via two short peptide linkers (Fig. 1a). We synthesized genes encoding megabodies that are built from different nanobodies linked through their first β-turn (connecting β-strands A and B) to an exposed β-hairpin of selected scaffold proteins. By visual inspection, we have identified two large secreted bacterial proteins that are amenable to circular permutation²⁷ and contain antiparallel β-strands with surface accessible β-turns: β- turn S3-S4 of the adhesin domain of H. pylori ²⁸ (HopQ, 45 kDa, PDB ID: 5LP2, Supplementary Fig. 1a-c) and β-turn A’S1-A’S2 of the E. coli K12 Glucosidase YgjK²⁹ (YgjK, 86 kDa, PDB ID: 3W7T, Supplementary Fig. 1d-f). Accordingly, we initially grafted four nanobodies onto circular permutants of these molecular scaffolds to build megabodies of about 56 and 100 kDa, respectively (Supplementary Fig. 2, Supplementary Table 1). These chimeric constructs were produced as highly soluble proteins in the periplasm of E. coli and were purified with yields in the range of 6 to 27 mg/l culture (Supplementary Fig. 2d-n). All these megabodies are monomeric and thermal shift assays indicate that they are conformationally stable, with melting temperatures in the 43 to 50 °C range (Supplementary Fig. 2o-s). Importantly, megabodies bind their cognate antigens with affinities similar to the nanobody versions (Supplementary Fig. 3) confirming that the binding properties are not affected by the nanobody insertions into the different scaffolds we selected (Fig. 1a). In our experience, these megabodies also resist at least two freeze-thaw cycles.

Figure 1. Molecular design of novel rigid antibody chimera called megabodies.

a, Megabodies are assembled from a nanobody (or a similar single-domain antigen-binding protein) and a (large) scaffold protein. The (optionally circularly permutated) scaffold protein is inserted between β-strand A and β-strand B of a nanobody. The megabody is encoded by a single gene, comprising a nanobody that is grafted into the scaffold protein via two peptide bonds. b, Crystal structure of ${\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}$ (PDB ID: 6QD6) that was built from a GFP-specific nanobody (Nb207) and a circularly permutated variant of HopQ (cHopQ). Molecule A, one out of ten molecules present in the asymmetr ic unit is represented. CDRs and β-strands of the nanobody are defined according to IMGT. c, Crystal structure of ${\text{Mb}}_{\text{Nb207}}^{\text{cYgjKNO}}$ (PDB ID: 6XUX) that was built from Nb207 and a circularly permutated variant of YgjK (cYgjK). Only one megabody molecule is present in the asymmetric unit. d, Primary structures of ${\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}$ and ${\text{Mb}}_{\text{Nb207}}^{\text{cYgjKNO}}$. The AS(GGGSG)_2/3 peptides were used to circularly permutate HopQ and YgjK. β-strands are represented by arrows. Residues of the different fragments of Nb207, cHopQ and cYgjK fragments are numbered according to the IMGT, UniProt B5Z8H1 and UniProt P42592 numbering, respectively.

To assess the structural impact of domain fusion, we crystallised a megabody assembled from the GFP-binding Nb207 and the HopQ scaffold, termed ${\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}$, and solved its structure to 2.8 Å resolution (Fig. 1b, Supplementary Fig. 4-5, Supplementary Table 2). ${\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}$ crystallised in space group P1 containing ten molecules in the asymmetric unit (PDB ID: 6QD6). The short peptide linkers that connect the nanobody to the scaffold are well-defined in the electron density map, except for molecule J, where the first twelve residues of the megabody are invisible. All ten molecules in the asymmetric unit adopt a similar elongated shape in agreement with SAXS analysis (Supplementary Fig. 6) and superimpose with root mean square deviations (RMSD) in the 0.557 to 3.972 Å range (calculated for the 532 C_α positions that could be unambiguously built, the whole megabody is 542 residues long) that are mainly caused by the flexibility of the nanobody-scaffold linker, which allows a maximal relative rotation of 22.2° when measured between molecules B and E (Supplementary Fig. 4b). Remarkably, these linkers seem to induce more structural constraints than the “elbow” regions that connect the constant and variable domains of conventional Fab antibody fragments³⁰. For example, a Fab complex that was crystallized with six molecules in the asymmetric unit (PDB ID: 4RRP) manifests a maximal relative rotation of 37° (Supplementary Fig. 7). Notably, the circular permutation and the insertion of the nanobody do not affect the overall tertiary structure of the scaffold protein (Supplementary Fig. 5). Using yeast display (Supplementary Fig. 8) followed by fluorescence-activated cell sorting (FACS, see below), we next selected several variants of ${\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}$ with different linker lengths and composition, confirmed that they possess similar biophysical characteristics (Supplementary Fig. 9) and solved the crystal structures of two of them: ${\text{Mb}}_{\text{Nb}207}^{\text{c}7\text{HopQA}12}$ (PDB ID: 6XVI, Supplementary Fig. 10-11, Supplementary Table 2) and ${\text{Mb}}_{\text{Nb}207}^{\text{c}7\text{HopQG}10}$ (PDB ID: 6XV8, Supplementary Fig. 12-13, Supplementary Table 2), both with four molecules in the asymmetric units. By comparing all structures, we found that the overall shape of these HopQ-derived megabodies is modulated by the sequence of the linkers (Supplementary Fig. 10b, Supplementary Fig. 12b).

As ${\text{Mb}}_{\text{Nb}207}^{\text{cYgjKE}2}$ - our first megabody built from the same nanobody Nb207 but linked to scaffold YgjK - did not yield well-diffracting crystals, we reduced the length of its linker peptides (Supplementary Fig. 14) and solved the crystal structure of ${\text{Mb}}_{\text{Nb207}}^{\text{cYgjKNO}}$ to 1.9 Å resolution (PDB ID: 6XUX, Fig. 1c, Supplementary Fig. 15, Supplementary Table 2), confirming that nanobodies can be grafted onto different scaffolds to produce well-folded functional megabodies of distinct shapes and sizes.

To verify whether the same design principles also apply to other antigen-binding domains, we inserted the cHopQ scaffold into the first β-turn (connecting β-strands A and B) of monobody NS1 that binds to RAS³¹ (Supplementary Fig. 16). Monobodies are engineered derivatives of the fibronectin type III domain consisting of one three- and one four-stranded β-sheet folded into a seven-stranded β-sandwich²⁵ (Supplementary Fig. 16c, d). We confirmed the functionality of this megabody by flow cytometry, measuring the binding of fluorescently labelled KRAS to ${\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}$ displayed on yeast³² (Supplementary Fig. 16f).

Selection of functional megabodies from nanobody-immune repertoires by yeast display

Taking advantage of the structural conservation of the immunoglobulin fold^24,33 (Supplementary Fig. 17a), we found that one nanobody can easily be replaced by another within a particular megabody type, once a pair of peptide linkers has been validated to functionally connect a representative nanobody like Nb207 to a specific scaffold protein. Furthermore, we hypothesised that in vivo matured nanobody immune libraries can be cloned as megabody display libraries and screened by conventional phage display or yeast display for the direct selection of functional megabodies. To test this concept, we amplified the open reading frames of the nanobody repertoire of a GFP-immunised llama, ignoring the 5’-end that encodes the highly conserved β-strand A³³. All these fragments were cloned in a yeast display vector according to Fig. 2a encoding the same sequence for β-strand A and the same peptide linkers to fuse nanobody to cHopQ as in ${\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}$ described above (Fig. 2a, Supplementary Fig. 17) to generate a megabody library that was displayed on yeast in frame with the Aga2p anchor protein followed by an acyl carrier protein (ACP) for orthogonal labelling³² (Fig. 2b). Single yeast cells that display high levels of a particular megabody (high DY-647P1 fluorescence) and bind the cognate antigen (high GFP fluorescence) were identified and sorted by FACS (Fig. 2c). After three rounds of selection, we identified the original ${\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}$ and discovered eight novel GFP-specific megabodies (Supplementary Fig. 17). We confirmed that cYgjK-based megabodies can also be functionally displayed on the surface of yeast cells (Supplementary Fig. 18), indicating that many antigen-binding chimeric proteins can be selected in a straightforward manner from megabody-display libraries constructed from nanobody-immune repertoires.

Figure 2. Selection of megabodies from nanobody immune libraries by yeast display.

a, Gene fragments encoding β-strands B to G of a nanobody immune repertoire can be amplified by PCR using primers TU65 and TU64 to be cloned into yeast vector pNMB2 that encodes a display cassette containing the following elements: the appS4 leader sequence (LS) to direct secretion, a consensus sequence encoding the conserved β-strand A of the nanobody fold (A), a circularly permutated variant of HopQ (cHopQ), a multicloning site (MCS), the Aga2p anchor protein followed by an acyl carrier protein (ACP). b, The display level of a cloned megabody on the surface of a single yeast cell can be monitored through a covalent fluorophore (red star) that is attached in a single enzymatic step to the ACP tag. Antigen binding to the displayed megabody can be monitored simultaneously by following the fluorescence of the antigen (in this case GFP). c, Dot plots representing the enrichment of GFP-specific megabodies within three constitutive rounds of yeast display selection. The y- axis quantifies the megabody display level (DY-647P1 fluorescence), whereas x-axis is a measure of the amount GFP that is bound to the cell. Yeast cells displaying GFP-specific megabodies (gate Q2 represented as a blue square), were enriched from 0.2% in the first round (200 nM of GFP) to 35% in the third round of selection (10 nM of GFP).

GABA_A-specific megabodies enable high resolution cryo-EM reconstructions

GABA_A receptors (GABA_ARs) are pentameric ligand-gated ion channels (pLGICs), which mediate fast inhibitory signalling in the human brain and are targets for clinically-relevant drugs including benzodiazepines and general anaesthetics³⁴. GABA_ARs and related pLGICs are known to suffer from severe preferential orientation in free-standing ice unless detergents are added to the samples. Although this approach in combination with Fab antibody fragments has enabled structural analyses of heteromeric GABA_A receptors by cryo-EM^35–37, the transmembrane domains (TMDs) and the interfaces between TMDs and extracellular regions were damaged by detergents³⁸.

To assess the utility of megabodies for single particle cryo-EM and to enable structural studies of human GABA_ARs in lipid bilayers, we enlarged nanobody Nb25, which interacts with the extracellular domain (ECD) of the GABA_AR β3 subunit³⁹ to two megabody variants, ${\text{Mb}}_{\text{Nb25}}^{\text{c}7\text{HopQ}}$ and ${\text{Mb}}_{\text{Nb}25}^{\text{cYgjKE}2}$ (Supplementary Table 1). We reconstituted a GABA_AR-β3 homomeric receptor construct from which the intracellular M3-M4 loop has been truncated⁴⁰ into lipid nanodiscs^41,42 and subsequently vitrified these particles on electron microscopy grids alone or in complex with Nb25, ${\text{Mb}}_{\text{Nb25}}^{\text{c}7\text{HopQ}}$ or ${\text{Mb}}_{\text{Nb}25}^{\text{cYgjKE}2}$. We collected and analysed small cryo-EM datasets for these samples in order to evaluate the extent of preferential particle orientation in each condition (Fig. 3). Firstly, when GABA_AR-β3 homomer samples were frozen alone, ~84% of the particles observed had their five-fold symmetry axis perpendicular to the water-air interface (“top” views, Fig. 3a). Addition of Nb25 did not improve particle orientation (Fig. 3b). In both cases, the datasets had low specimen orientation distribution efficiencies¹⁵ (E_od= 0.18 and 0.22, respectively), leading to severely anisotropic 3D reconstructions, with large missing regions (Fig. 3a,b). Remarkably however, more than ~77% of the classified particles displayed different orientations (“tilted” views) when the GABA_AR-β3 receptor was bound to ${\text{Mb}}_{\text{Nb25}}^{\text{c}7\text{HopQ}}$ (Fig. 3c). This is reflected in a three-fold higher orientation distribution efficiency (E_od= 0.76) and a high quality 2.49 Å resolution 3D reconstruction. In contrast, ${\text{Mb}}_{\text{Nb}25}^{\text{cYgjKE}2}$ had an opposite effect and induced almost exclusively “top” GABA_AR-β3 views which led to the most anisotropic 3D reconstruction (Fig. 3d). Megabodies therefore can directly affect the orientation of particles in ice, however it is clear that multiple megabody variants, with different scaffolds, should be tested for a protein of interest.

Figure 3. Cryo-EM datasets of the homomeric GABA_A β3 receptor alone, in complex with Nb25 or bound to megabodies derived from Nb25.

a-c, Direct comparison of homomeric GABA_A β3 receptor alone (a), with addition of Nb25 (b), ${\text{Mb}}_{\text{Nb25}}^{\text{c}7\text{HopQ}}$ (c) or ${\text{Mb}}_{\text{Nb}25}^{\text{cYgjKE}2}$ (d) in single particle cryo-EM. For each sample, one representative image from 20 to 80 analysed micrographs is given. White scale bars indicate 20 nm. The 2D classes were separated into two groups: top-view (top) and tilted-view (bottom). The counts of particles with a top-view or a tilted-view orientation are indicated. The reconstructed 3D models, the corresponding particle distribution “efficiencies” (E_od) and the projections of the particles over azimuth and elevation angles are also compared amongst the different datasets.

The ${\text{Mb}}_{\text{Nb25}}^{\text{c}7\text{HopQ}}$ megabody appears to induce a new range of orientations, side views in this case, in a large number of GABA_A particles (Fig. 3c). For a 5-fold symmetric object such as GABA_AR-β3 this is beneficial, as one can fully sample the Fourier space and generate high- quality reconstructions without the need of top and bottom views. The high resolution GABA_AR-β3 cryo-EM map enabled by the ${\text{Mb}}_{\text{Nb25}}^{\text{c}7\text{HopQ}}$ megabody has allowed straightforward model building and refinement as well as unambiguous identification of the histamine density, an agonist molecule composed of only eight non-hydrogen atoms⁴³ (Fig. 4, Supplementary Fig. 19-21, Supplementary Table 4, Supplementary Video 1). The scaffold region of ${\text{Mb}}_{\text{Nb25}}^{\text{c}7\text{HopQ}}$, visible in many particles in raw images, is largely absent in the final reconstruction due to the flexibility of its linkers to the nanobody domain (Supplementary Fig. 22a).

Figure 4. Megabody-enabled high-resolution structure of homopentameric β3 GABA_A receptor in lipid nanodiscs.

a-d, Side (a) and top (b) view of the sharpened cryo-EM density map of histamine-bound β3 GABA_A receptor in complex with ${\text{Mb}}_{\text{Nb25}}^{\text{c}7\text{HopQ}}$ in lipid nanodisc (EMDB-4542, PDB ID: 6QFA). Five β3 subunits are coloured light blue (subunit A) and dark blue (subunits B-E). Histamine (HSM), ${\text{Mb}}_{\text{Nb25}}^{\text{c}7\text{HopQ}}$ and glycans are coloured magenta, green and orange respectively. Sharpened density maps are contoured at 0.08. Side view (c) and top view (d) of an atomic model of the β3 GABA_AR in a complex with histamine. e-f, Histamine binding mode. Histamine occupies all five canonical agonist pockets under the sensor loop-C. The imidazole ring of histamine points towards the complementary β3^- side of the pocket where it is stacked between the side chains of Phe200 and Tyr62 and forms a hydrogen bond (black dashed lines) with Asp43 side chain (complementary face residues from E subunit marked by ‘c-’). The histamine ethylamine group faces the principal β3⁺ side of the pocket and forms hydrogen bonds with the Glu155 side chain and backbone carbonyls of Ser156 and Tyr157, and a cation-π interaction (red dashed lines) with the aromatic ring of Tyr205. g, Pore diameter of histamine (HSM) bound β3 ${\text{GABA}}_{\text{A}}\text{R}-{\text{Mb}}_{\text{Nb25}}^{\text{c}7\text{HopQ}}$ complex (blue line), benzamidine (BEN) bound β3 GABA_AR (black dotted line, PDB ID: 4COF). The HSM bound GABA_AR-β3 channel pore is in a desensitised conformation, with the 9’Leu activation gate open and the -2’ desensitization gate closed, a conformation reminiscent of the benzamidine- bound crystal structure solved in detergent⁴⁰. The upper half of the pore is relatively narrower, perhaps reflecting the impact of the lipid environment in which the GABA_AR-β3 construct described here was reconstituted.

We have extended this approach to perform cryo-EM studies of the major human heteromeric GABA_AR subtype α1β3γ2, published elsewhere^41,42. Unlike the homomeric GABA_AR-β3⁴⁰, heteromeric GABA_ARs, especially in a physiologically meaningful lipid environment, are intractable for structural analysis by X-ray crystallography. We have therefore enlarged Nb38 that binds α1 subunits between non-adjacent ECD interfaces³⁷ to generate the megabody ${\text{Mb}}_{\text{Nb}38}^{\text{cHopQ}}$. Because the ${\text{Mb}}_{\text{Nb}38}^{\text{cHopQ}}$ megabody, at the concentration used, only binds to one GABA_AR site, the range of particle orientations is more evenly distributed than we observed in the GABA_AR-β3 case. This approach has enabled high-resolution cryoEM structures of the full-length α1β3γ2 human GABA_AR in lipid nanodiscs bound to a series of small molecule modulators including the competitive antagonist bicuculline, the channel blocker picrotoxin, the agonist GABA and the classical benzodiazepines alprazolam (Xanax) and diazepam (Valium), respectively^41,42.

MSP-specific megabodies randomize the orientation of membrane protein particles

Functional and structural studies of membrane proteins are challenging and should ideally incorporate the native lipid environment^44–47. Over the years, many lipid membrane-mimicking tools have been developed, including SMALPs⁴⁸, peptidics⁴⁹, Salipros⁵⁰ and nanodiscs⁵¹. Nanodiscs are extensively utilized discoidal lipid bilayers embedded by two membrane scaffold proteins (MSPs). Several MSP variants can be combined with different lipid mixtures to control the diameter and the lipid bilayer composition of the disc⁵².

Considering the remarkable impact of ${\text{Mb}}_{\text{Nb25}}^{\text{c}7\text{HopQ}}$ on the GABA_AR-β3 particle distribution in ice, we investigated whether anti-MSP megabodies can also affect the particle orientation of membrane proteins reconstituted in nanodiscs. We have therefore generated an MSP-specific nanobody, termed NbF3 (Supplementary Table 1), and reformatted it into a megabody (${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$) that binds with high affinity to the lipid-free MSP1D1, MSP1E3D1 and MSP2N2 MSP variants (Fig. 5a-b, Supplementary Fig. 23). Furthermore, we showed that ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ forms a stable complex with nanodiscs containing lipid bilayers (Supplementary Fig. 23d).

Figure 5. ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$, an MSP-specific megabody randomizes the orientation of nanodisc- embedded β3 GABA_AR particles.

a, Recombinant MSP variants that are used to assemble nanodiscs: the His6-tag, the TEV cleavage site and the α-helical domains (H) are shown. b. Specific binding of ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ to different MSP variants analysed by ELISA. ${\text{Mb}}_{\text{Nb25}}^{\text{c}7\text{HopQ}}$, a meabody that binds GABA_AR was used as a negative control. The data are shown as mean standard error of analysed in parallel triplicates n=3. c, Representative image of 193 analysed cryo-EM micrographs of β3 GABA_AR reconstituted in MSP2N2 nanodiscs in complex with ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$. The white scale bar indicates 10 nm. d, Representative top-view and tilt-view 2D classes. e, Reconstructed 3D model of β3 GABA_AR. f, Distribution of the viewing directions over azimuth and elevation angles, calculated from this 3D model.

To assess the utility of ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ in cryo-EM, we added it to the GABA_AR-β3 receptor reconstituted in MSP2N2 nanodiscs and collected a 200 micrographs dataset (Supplementary Table 3). Remarkably, about 90 % of the particle projections represent “tilted” views and the E_od of this specimen is 0.69, allowing a 3 Å resolution reconstruction from less than 9,000 particles (Fig. 5c-f, Supplementary Fig. 24).

We applied the same approach to a serotonin 5-HT3A receptor, which belongs to the cationic branch of pentameric ligand-gated ion channels (pLGICs)⁵³. Efforts to obtain cryo-EM structures of the homopentameric mouse 5-HT3A receptor in nanodiscs were previously hampered because these particles, like GABA_AR-β3, suffer from severe preferential orientation in vitreous ice. We collected small cryo-EM datasets and analysed the particles orientation distribution of the 5-HT3A receptor reconstituted in MSP1E1D3 nanodiscs either alone or in complex with ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ (Supplementary Fig. 25). When 5-HT3AR samples were frozen alone, ~60% of the particles of the reconstituted receptor are observed as “top” views, whereas the remaining ~40% displayed a limited diversity of orientations (“tilted” views) (Supplementary Fig. 25a, upper panels). Notably, no class average presenting projections of the receptor perpendicular to the symmetry axis (“side view”) was observed. In contrast, the nanodisc reconstituted 5-HT3 receptor in complex with ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ increased the percentage of “tilted” views to ~60%, including particle orientations absent in the receptor alone sample (Supplementary Fig. 25a, middle panels). A dataset of the $5-\text{HT}3\text{R}-{\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ complex, collected on a 200 keV microscope (Supplementary Table 3), had a specimen orientation distribution efficiency of E_od= 0.8 and led to a 3.5 Å resolution reconstruction (Supplementary Fig. 25b-d). The refined model of the nanodisc reconstituted 5-HT3 receptor will be described in a forthcoming publication.

It thus appears that MSP-specific megabodies help randomize the orientation distribution of nanodisc-embedded membrane proteins. It is worth noting that the ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ density is not visible in the final reconstructions of the $5-\text{HT}3\text{R}-{\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ and ${\text{GABA}}_{\text{A}}\text{R}-\beta 3-{\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ complexes due to the axial rotation of the belt proteins around the reconstituted membrane proteins.

WbaP specific megabodies act as fiducial markers

Many membrane proteins, including most GPCRs and transporters are smaller than 100 kDa and almost completely embedded in detergent micelles, amphipols or nanodiscs for cryo-EM analysis. This prevents an accurate orientation of their projections that is required for 3D reconstructions. We hypothesized that megabodies, similarly to Fab antibody fragments (50 kDa)^54,55, can also be used as fiducial markers in cryo-EM. To test this concept, we focused on the 57 kDa inner membrane protein WbaP, a bacterial polyisoprenylphosphate hexose-1- phosphate transferase (PHPT)⁵⁶ that is involved in bacterial pathogenesis⁵⁷ and has no eukaryotic homologues. WbaP is a potential drug target however, to date, no structural information is available for this protein. We raised a WbaP-specific nanobody library and selected Nb73 (Supplementary Table 1) for cloning into the megabody format, yielding ${\text{Mb}}_{\text{Nb}73}^{\text{c}7\text{HopQ}}$. WbaP was recombinantly expressed, purified and reconstituted in amphipols for cryo-EM analysis (Supplementary Fig. 26). We collected a dataset of the $\text{WbaP}-{\text{Mb}}_{\text{Nb}73}^{\text{c}7\text{HopQ}}$ complex on a 200 keV microscope (Supplementary Table 3). 2D classification reveals a broad distribution of particle orientations as well as two clear elongated megabody densities protruding from blurred and featureless “spheres” representing the WbaP particles in amphipols (Supplementary Fig. 26b). The presence of megabodies has enabled not only 2D classification but also a 4.9 Å resolution 3D reconstruction (Supplementary Fig. 26c-e, Supplementary Table 3), which reveals a dimeric organization and clear secondary structure elements for WbaP for the first time. The resolution will likely be improved upon data collection in counting mode, and on a higher end microscope, and a detailed structural and functional analysis of WbaP will be described in a forthcoming publication.

Conclusions

We introduce a novel, generic technology, which allows the rapid reformatting or direct library selection of high affinity binding domains such as nanobodies or monobodies into larger, conformationally and functionally stable chimeras termed megabodies. The criteria for selecting the scaffolds can be satisfied by many proteins of any size or geometry required to avoid steric clashes with the target: proteins well-expressed in the E. coli periplasm, N and C termini in proximity to allow circular permutation and an exposed “rigid” β-hairpin loop to insert globular domains. This simple but powerful approach expands the nanobody toolkit for protein structure determination by single-particle cryo-EM but also, at least theoretically, by X-ray crystallography because the relatively large and rigid scaffolds themselves offer novel opportunities for crystal contacts.

The primary benefit of megabodies is that they help mitigate the impact of preferential particle orientation, a problem that affects a broad range of samples embedded in thin layers of vitreous ice as required for cryo-EM analysis. This problem is particularly severe in membrane proteins such as pLGICs, that typically present only top views when reconstituted in amphipols or nanodiscs. We demonstrate that target specific megabodies are able to efficiently overcome this limitation and allow a 2.5 Å resolution 3D reconstruction of a human GABA_A receptor construct using a standard 300 keV microscope setup. Ongoing hardware and software developments will allow even higher resolution reconstructions, as recently exemplified by an 1.7 Å map⁵⁸, further underlining the quality of this sample and impact of megabodies. To make our technology immediately accessible to any membrane protein reconstituted in nanodiscs, for which target-specific nanobodies might not be available, we generated a megabody that binds with high affinity to the most commonly used MSPs in the field (MSP1D1, MSP1E3D1 and MSP2N2). This has enabled a straightforward reconstruction of a 5-HT3A receptor in a lipid bilayer, another sample that notoriously suffers severe orientation issues. A second benefit of megabodies is their ability to act as fiducial markers for the alignment of small particles that lack sufficient features to orient individual images. We illustrate this by reporting a ~5 Å resolution reconstruction of the integral membrane proteins WbaP (57 kDa as a monomer), reconstituted in amphipols.

It is important to note that, regardless to minor differences in the biophysical characteristics of the different megabody surfaces (Supplementary Fig. 27), we found that cHopQ-based megabodies are very efficient tools for homomeric and heteromeric GABA_A receptors whereas the same nanobodies built into a cYgjK scaffold did not help in this particular case. It thus appears that the optimal megabody design will likely vary for different target proteins, and the identification of further scaffold domains, following the principles described here, would be beneficial.

Besides the impact on the size and particle orientation we report, the megabodies (so far ~56 kDa and ~100 kDa) can also facilitate the in silico “purification”⁵⁹ of particles, sorting those that contain the stable antigen:megabody complex from those that contain the antigen only or other contaminants. When the nanobody recognition domain is specific to a particular conformation of the target protein, this particle classification step ensures that the 3D reconstruction obtained represents the antigen in the particular state trapped by the megabody. The increased particle size of antigen:megabody complexes also facilitates their identification and picking from low-contrast micrographs. Most importantly, the megabody technology has the potential to further stimulate the rapidly growing field of drug discovery using single particle cryo-EM⁶⁰ by facilitating high resolution structural analysis of difficult yet highly valuable targets such as eukaryotic membrane proteins.

Online Methods

Proteins and antigens

Lysozyme from chicken egg white (L6876) and the membrane scaffold proteins MSP1D1 (M6574), MSP1E3D1 (M7074) and MSP2N2 (MSP12) were purchased from Sigma. Activated human coagulation Factor IX (FIXa) was expressed and produced essentially as described previously³².

A codon-optimized synthetic gene encoding the human KRASG12V mutant (residues 1-169) and human SOS1 (residues 564—1049) were cloned as NdeI and XhoI fragments into pET28b (Novagen). His-tagged K-RASG12V, abbreviated KRAS in this paper, and SOS1 were expressed in E. coli strain BL21 and purified as described previously⁶¹.

GFP variant GFP+⁶² was expressed in E. coli strain DH5 alpha⁶³ as a C-terminally His6- tagged protein under the transcriptional control of the lac promoter using pUC8. Cells were grown overnight in Luria-Bertani broth supplemented with ampicillin (100 mg/l) at 37 °C, harvested by centrifugation (5,000g, 15 min), resuspended in lysis buffer (50 mM Tris pH 8, 200 mM NaCl, 15 mM EDTA) and lysed with a high-pressure homogeniser (Constant System). Lysed cells were next pelleted by centrifugation for 30 min at 10,000g and the supernatant was applied on a HisTrap FF 5 ml prepacked column (GE Healthcare). GFP was eluted with 500 mM imidazole and concentrated by centrifugation using Amicon Ultra Filters (cut-off of 3 kDa, Sigma) and polished on a Superdex 75 PG 16/90 size-exclusion column equilibrated with 20 mM Tris pH 7.3 and 140 mM NaCl.

A human β3 homopentameric GABA_AR (UniProtKB P28472) construct which contains the K279T point mutation for increased stability, an SQPARAA linker³⁴ substituting the M3-M4 loop, and a C-terminal (GGS)3GK—Rhodopsin-1D4-tag (TETSQVAPA)⁶⁴ was transiently expressed in HEK293S-GnTI^- cells as described⁴⁰. Briefly, HEK293S-GnTI^- cells were grown in protein expression medium (PEM, Thermo Fisher Scientific) supplemented with 1% foetal bovine serum (Invitrogen) at 37 °C, 8% CO₂. Cells were transfected with the DNA-PEI mix at a ~2×10⁶ cells/ml density, and 48 h post-transfection were harvested by centrifugation at 4,000g, 4 °C. Cell pellets were snap-frozen in liquid N2 and stored at -80 °C for future use.

The full-length, wild-type mouse 5-HT3A receptor was expressed and purified as previously _described ^65,66.

The E. coli homologue of WbaP (UniProtKB Q9X4C0) was expressed in E. coli from a pBAD24-derived vector, adding a TEV-protease cleavable C-terminal His10-tag. Cells were grown at 37 °C to OD 0.4-0.6, induced with 0.2% arabinose and incubated for another 3 h at 37 °C. Cells were harvested by centrifugation at 5000 g, 4 °C, and frozen at -80 °C.

Nanobody discovery

The amino acid sequences of all nanobodies are listed in Supplementary Table 1. Nanobody cAbLys3⁶⁷ that binds chicken egg white lysozyme and Nb25³⁹ that binds the extracellular domain of the GABA_AR β3 subunit were described before. All GFP, SOS1, MSP1D1, FIXa or WbaP specific nanobodies described, were selected from immune libraries and purified following standard procedures⁶⁸.

Construction of bacterial megabody-expression vectors

Plasmids pMESD2 (GenBank MT328400) and pMESD22c7 (GenBank MT338520) were constructed for subcloning nanobodies to express them as megabodies ${\text{Mb}}_{\text{Nb}}^{\text{cHopQ}}$ and ${\text{Mb}}_{\text{Nb}}^{\text{c}7\text{HopQ}}$ in the periplasm of E. coli, respectively. pMESD2 (Supplementary Fig. 2b) is a derivative of pMESy4 (GenBank KF415192).

pMESD2 contains an open reading frame encoding the DsbA leader sequence, followed by a consensus sequence for β-strand A of the nanobody fold⁶⁹ (QVQLVESGGGLVQ), followed by the C-terminal part of HopQ (residues 227-449, UniProtKB B5Z8H1), followed by peptide linker ASGGGSGGGGSG connecting the C-terminus and the N-terminus of HopQ to produce a circular permutant of the scaffold protein, followed by the N-terminal part of HopQ (residues 49-221, UniProtKB B5Z8H1), followed by a conserved Gly residue in the nanobody fold (Gly17), followed by a multi cloning site, followed by the His6 tag and the EPEA tag^70,71. This open reading frame is under the transcriptional control of the Plac promotor.

pMESD22c7 is a variant of pMESD2 in which the C-terminal end of HopQ (residues 227-446, UniProtKB B5Z8H1) is directly fused to its N-terminal end (residues 53-221, UniProtKB B5Z8H1) to encode a circular permutant of the scaffold protein (Supplementary Fig. 9a). We recommend to use this circular permutation variant to generate HopQ-based megabodies (c7HopQ) because the ASGGGSGGGGSG linker is not visible and appears to be flexible in the electron density map of ${\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}$ (PDB ID: 6QD6).

Plasmids pMESP23E2 (GenBank MT338521) and pMESP23NO (GenBank MT338522) were constructed for subcloning nanobodies to express them as megabodies ${\text{Mb}}_{\text{Nb}}^{\text{cYgjKE}2}$ and ${\text{Mb}}_{\text{Nb}}^{\text{cYgjKNO}}$ in the periplasm of E. coli, respectively. pMESP23E2 (Supplementary Fig. 2c) contains an open reading frame encoding the PelB leader sequence, followed by a consensus sequence for β-strand A of the nanobody fold⁶⁹ (QVQLVESGGGLV), followed by a Tyr, followed by the C-terminal part of YgjK (residues 487-783, UniProtKB P42592), followed by peptide linker ASGGGSGGGGSGGGGSG connecting the C-terminus and the N-terminus of YgjK to produce a circular permutant of the scaffold protein, followed by the N-terminal part of YgjK (residues 24-484, UniProtKB P42592), followed by Asp, followed by a multi cloning site (MCS), followed by the His6 tag and the EPEA tag^70,71. This open reading frame is under the transcriptional control of the Plac promotor. pMESP23NO is a variant of pMESP23E2 in which the Tyr and Asp residues were omitted from the peptide linkers connecting the nanobody to cYgjK (Supplementary Fig. 14a), to crystallize and solve the crystal structure of ${\text{Mb}}_{\text{Nb207}}^{\text{cYgjKNO}}$ (PDB ID: 6XUX).

Nanobody reformatting and bacterial expression of megabodies

Nanobody reformatting and subsequent production of related megabodies was performed as described⁷². Briefly, to clone and express nanobodies in a megabody format of choice, gene fragments encoding β-strands B to G of the parental nanobodies (residues 18-128, Supplementary Fig. 2a) were amplified by PCR with primers TU89 and EP230 (Supplementary Table 5) and cloned as a SapI fragment⁷³ in the desired expression vector described above. To express and secrete megabodies to the periplasm of E. coli, WK6 cells⁷⁴ bearing a megabody-expression plasmid were grown in Terrific Broth medium supplemented with ampicillin (100 mg/l) at 120 rpm and 37 °C to OD₆₀₀ = 4, induced overnight with 1 mM IPTG at 28 °C and harvested by centrifugation (5,000g, 15 min). Recombinant megabodies (Supplementary Table 1) were released from the periplasm by resuspending the pellets in 20% w/v sucrose supplemented with 0.5 mg/ml lysozyme (L6876 Sigma), 50 mM Tris pH 8, 1 mM EDTA and 150 mM NaCl for 30 min at 4 °C or by applying a multistep osmotic shock⁶⁸. Soluble megabodies were next separated from the protoplasts by centrifugation, supplemented with 500 mM NaCl and 5 mM MgCl2 (final concentration) and recovered from the clarified supernatant on a HisTrap FF 5 ml prepacked column. Proteins were next eluted from the Ni-NTA resin by applying 500 mM imidazole and concentrated by centrifugation using Amicon Ultra Filters (Sigma, cut-off). All megabodies were ran on Superdex 200 PG size exclusion columns (GE Healthcare), equilibrated with 10 mM Tris pH 7.3 and 140 mM NaCl, as single peaks (Supplementary Fig. 2). All purification steps were done at 4 °C or on ice, and typical protein yields were 6-27 mg/l of culture.

Thermal stability measurements

Thermal stabilities of nanobodies and megabodies were compared by measuring the increase in fluorescent intensity of the partition hydrophobic-binding dye SYPRO^® Orange that binds to unfolding proteins upon thermal melting. The thermal shift assays were performed in triplicate (n=3) in a 96-well qPCR microplate (BioRad) in a final volume of 20 μL containing 5x SYPRO^® Orange (Thermo Fisher Scientific) in 10 mM Tris pH 7.3, 140 mM NaCl and 0.2 mg/ml of the nanobody or 2 mg/ml of the megabodies, respectively. Thermally-induced protein melting was performed in a CFX qPCR instrument (BioRad) using a temperature gradient from 25 to 100 °C at a heating rate of 1 °C per minute (Supplementary Fig. 2o-s). Experimental data were fitted with GraphPad Prism 7 using Boltzmann’s equation Y=Bottom+(Top-Bottom)/(1+exp((V50-X)/Slope)).

Antigen binding kinetics

We used bio-layer interferometry (BLI) on an OctetRED96 (ForteBio) to measure the binding kinetics of the nanobodies and the corresponding megabodies onto immobilized antigens. For preparing the biosensors, GFP, Lysozme, FIXa, SOS1 and MSP1D1 were biotinylated with a five-fold molar excess of EZ-link NHS-Biotin (Thermo Fisher Scientific) following the manufacturer’s instructions and separated from unreacted NHS-biotin on a NAP10 column (GE Healthcare). The biotin/antigen ratios were determined in the range of 2 - 2.5 using the Pierce Biotin Quantitation kit (Thermo Fisher Scientific). For BLI, biotinylated antigens were diluted to 0.75 μg/ml in 10 mM Tris pH 7.3, 140 mM NaCl, 1% BSA and 0.05% Tween20 for GFP, lysozyme, SOS1 and MSP1D1, and in 10 mM HEPES pH 8.0, 300mM NaCl, 2.5 mM CaCl2, 1% BSA and 0.04% Tween 20 for FIXa, and directly immobilized on streptavidin biosensors (ForteBio) at about 1 nm response. After two equilibration steps of 100-300 s, the binding isotherms were monitored by exposing separate sensors simultaneously to different concentrations of the cognate nanobodies and megabodies, respectively. Association kinetics were followed for 300 s at 30 °C under constant stirring at 1000 rpm, tailed by dissociation experiments for 2800 s for GFP or 700 s for GFP, lysozyme, SOS1, FIXa (Supplementary Fig. 3) and MSP1D1 (Supplementary Fig. 23a). Association and dissociation rates were estimated by fitting the sensograms using the 1:1 binding model included in the Octet Data Analysis software 9.1 (ForteBio).

Enzyme-linked immunosorbent assay (ELISA)

For ELISAs, GFP, MSP1D1, MSP1E3D1 or MSP2N2 were diluted to 1 μg/mL in 100 μl of 10 mM Tris pH 7.3 and 140 mM NaCl (buffer A) and immobilized overnight at 4 °C on flat-bottom maxisorp microtiter plates (Thermo Fisher Scientific). Wells were then blocked with buffer A supplemented with 2% (w/v) milk powder. Megabodies were diluted to 100 nM in buffer A containing 0.2% (w/v) milk powder and incubated for 30 min at RT with the immobilized antigen. After three washes with buffer A, bound megabodies were labeled with a conjugate of the CaptureSelect antibody (Life Technologies) and Streptavidin Alkaline Phosphatase (Sigma), diluted 1:4000 and 1:1000 in buffer A, respectively. After 30 min at RT and three washing steps with buffer A, ELISAs were developed by adding Disodium 4-nitrophenyl phosphate hexahydrate at 2 mg/ml in 100 mM Tris HCl pH 9.5, 100 mM MgCl₂, 100 mM NaCl and the absorption of the colorimetric product was measured at 405 nm with a SpectroStarNano plate reader (BMG LABTECH).

Structure determination of ${\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}$, ${\text{Mb}}_{\text{Nb}207}^{\text{c}7\text{HopQA}12}$, ${\text{Mb}}_{\text{Nb}207}^{\text{c}7\text{HopQG}10}$ and ${\text{Mb}}_{\text{Nb207}}^{\text{cYgjKNO}}$ by X-ray crystallography

Megabodies ${\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}$, ${\text{Mb}}_{\text{Nb}207}^{\text{c}7\text{HopQA}12}$, ${\text{Mb}}_{\text{Nb}207}^{\text{c}7\text{HopQG}10}$ and ${\text{Mb}}_{\text{Nb207}}^{\text{cYgjKNO}}$ were purified and subjected to a number of commercial sparse-matrix crystallization screens (JSCG, Proplex, PEGion, Wizard12, Morpheus) in 0.1 μL sitting drops supplemented with 0.1 μL of the mother liquor.

Small crystals of ${\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}$ were obtained in the A2 condition of the JSCG screen (0.1 M sodium citrate, pH 5.5, 20/ w/v PEG3000). Well-diffracting crystals were grown by seeding ${\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}$ at 48 mg/ml in 0.2 M ammonium citrate, 17% PEG3350, 10% glycerol with the small crystals. Data were collected at 100K on the I03 source at the Diamond Light Source synchrotron (Oxfordshire, UK) and the structure was refined to 2.84 Å resolution. The megabody crystallized in P1 with ten molecules in the asymmetric unit (PDB ID: 6QD6).

Small crystals of ${\text{Mb}}_{\text{Nb}207}^{\text{c}7\text{HopQG}10}$ were obtained in the B5 condition of the Proplex screen (0.1 M magnesium chloride, 0.1 HEPES pH 7.5, 10/ w/v PEG4000). Well-diffracting crystals were grown by seeding ${\text{Mb}}_{\text{Nb}207}^{\text{c}7\text{HopQG}10}$ at 44 mg/ml in 0.1 M magnesium chloride, 0.1 HEPES pH 7.5, 19% w/v PEG4000 with the small crystals. Data were collected at 100K on the I03 source at the Diamond Light Source synchrotron (Oxfordshire, UK) and the structure was refined to 3.15 Å resolution. This megabody crystallized in P1 21 1 with four molecules in the asymmetric unit (PDB ID: 6XV8).

Small crystals ${\text{Mb}}_{\text{Nb}207}^{\text{c}7\text{HopQA}12}$ were obtained in the H11 condition of the JCSG screen (0.2 M magnesium chloride hexahydrate, 0.1 M BIS-Tris pH 5.5, 25% w/v PEG3350). Welldiffracting crystals were grown in 0.2 M Magnesium chloride hexahydrate, 0.1 M BIS-Tris pH 5.0, 21% w/v PEG3350 and ${\text{Mb}}_{\text{Nb}207}^{\text{c}7\text{HopQA}12}$ at 42 mg/ml. Data were collected at 100K on the I24 source at the Diamond Light Source synchrotron (Oxfordshire, UK) and the structure was refined to 2.6 Å resolution. The megabody crystallized in P1 21 1 with four molecules in the asymmetric unit (PDB ID: 6XVI).

Small crystals ${\text{Mb}}_{\text{Nb207}}^{\text{cYgjKNO}}$ were obtained in the E11 condition of the Proplex screen (0.1 M sodium citrate pH 5.0, 20% w/v PEG8000). Well-diffracting crystals were grown in 0.1 M sodium citrate pH 7.5, 20% w/v PEG8000, 10% glycerol and ${\text{Mb}}_{\text{Nb207}}^{\text{cYgjKNO}}$ at 20 mg/ml. Data were collected at 100K on the I03 source at the Diamond Light Source synchrotron (Oxfordshire, UK) and the structure was refined to 1.9 Å resolution. The megabody crystallised in P21 21 21 with one molecule in the asymmetric unit (PDB ID: 6XUX).

All diffraction data were integrated and scaled with XDS⁷⁵. Models were built by iterative cycles of refinement with Phenix and Buster-TNT⁷⁶ and manual building in Coot⁷⁷. MolProbity was used for structure validation⁷⁸. Data collection and refinement statistics are summarized in Supplementary Table 2. Root mean square deviations (RMSD), rotation angles, kdHydrophobicity⁷⁹ and Poisson-Boltzmann electrostatic⁸⁰ potentials were calculated using UCSF Chimera⁸¹. RMSD calculation settings and obtained values for megabody molecules present in the asymmetric units are listed in Supplementary Table 6. RMSD of circularly permutated HopQ and YgjK scaffold proteins were calculated using the parental H. pylori adhesin domain (PDB ID: 5LP2) and E. coli K12 Glucosidase YgjK (PDB ID: 3W7S) as references, respectively, using 30 Å distance cut-off.

Small-angle X-ray scattering of ${\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}$

Small-angle X-ray scattering (SAXS) data on ${\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}$ were collected in a mail-in session at the B21 beamline of Diamond Light Source synchrotron (Oxfordshire, UK). The scattering intensities were recorded in a Size-Exclusion Chromatography coupled SAXS (SEC-SAXS) experiment after injection of the protein on a Superdex 200 Increase 3.2/300 size exclusion column (GE Healthcare) equilibrated with 10 mM Tris-HCl pH 7.3, 140 mM NaCl. The averaging and buffer subtraction of the resulting data frames were performed using DATASW⁸² and processed using ATSAS⁸³. The scattering curve was generated with PRIMUS and was subjected to indirect Fourier transform using GNOM to yield the pair-distance distribution function P(r), from which the radius of gyration (Rg) and the maximum particle dimension (Dmax) were estimated. Rg was obtained also from the slope of the Guinier plot in PRIMUS. Further interpretation of the SAXS data involved using the dimensionless Kratky plot. The theoretical scattering profile of the X-ray structure of ${\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}$ was calculated using CRYSOL⁸⁴.

Selection of GFP-specific megabodies from Nanobody-immune libraries by yeast display and FACS

Plasmid pNMB2 (GenBank MT338523) is a derivative of yeast surface display vector pNACP³². A new megabody display cassette encoding the following elements was synthesized and cloned as an EcoRI-BglII fragment to replace the original displayed fusion protein: appS4 leader sequence (LS) for secretion⁸⁵, followed by a consensus sequence for β- strand A of the nanobody fold⁶⁹ (QVQLVESGGGLVQ), the C-terminal part of HopQ²⁸ (residues 227-449, UniProtKB B5Z8H1), a peptide linker ASGGGSGGGGSG connecting the C-terminus and the N-terminus of HopQ to produce a circular permutant of the scaffold protein, the N-terminal part of HopQ (residues 49-221, UniProtKB B5Z8H1), a multi cloning site (MCS), the Aga2p anchor protein followed by the ACP and the Myc tag (Fig. 2a). The MCS of the cloning vector pNMB2 contains an in frame stop codon to avoid background display and orthogonal staining of ACP from plasmids that do not contain an insert.

A llama was immunized with GFP+⁶² and DNA fragments encoding the entire nanobody repertoire of the immunized animal were isolated as described⁶⁸. Briefly, total RNA was isolated from the peripheral blood lymphocytes (PBLs) of the immunized animal to prepare cDNA and the open reading frames encoding all immunoglobulin heavy-chains were amplified by RT-PCR with primers call001 and call002. The ~700 bp fragment representing the heavy chain-only antibody repertoire is purified from gel. Fragments encoding the nanobodies from β-strands B to G were amplified thereof through a nested PCR using primers TU64 and TU65 (Fig. 2a, Supplementary Fig. 17, Supplementary Table 5). 10 μg of this PCR product was mixed with 10 μg of BamHI/HindIII linearised pNMB2 and transformed into electrocompetent EBY100 Saccharomyces cerevisiae cells for GAP repair homologous recombination⁸⁶ to produce a library of yeast cells that display diverse cHopQ megabodies derived from the nanobody repertoire of the GFP-immunized llama.

This yeast library displaying diverse megabodies was inoculated, induced and orthogonally stained with CoA-647 to monitor the display level of the Mb-Aga2p-ACP fusion on each yeast cell as described³². For each round of selection, 4 × 10⁷ of CoA-647-stained yeast cells were incubated for 60 min with GFP at 4 °C in 500 μl of cold PBS supplemented with 0.2% (w/v) BSA at pH 7.4 (PBS-BSA). Yeast cells were next washed three times with and resuspended in 2 ml of PBS-BSA and sorted on a FACS Aria (BD Biosciences). Selected yeast cells were recovered into SDCAA medium, grown and induced, then stained again for subsequent rounds of selection. Individual yeast clones expressing a GFP binding megabody were grown separately in 96-well plates for sequencing and further characterization as described³². Routinely, ~10,000 yeast cells derived from a particular clone were incubated with 100 nM GFP, washed 3 times and analyzed for green fluorescence by flow cytometry using FACS Fortessa (BD Biosciences). For each clone, the GFP mean fluorescence intensity (MFI) was calculated using FlowJo software (FlowJo, LLC) and compared to the fluorescence of a yeast clone displaying an irrelevant megabody as the negative control.

Selection of functional variants of ${\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}$ by yeast display and FACS

Plasmid pNMB1m_C_Nb207 (GenBank MT543226) is a derivative of pNACP³². A new display cassette encoding the following elements was synthesized and cloned as an EcoRI-BglII fragment to replace the original cassette: the appS4 leader sequence (LS) for secretion⁸⁵, followed by a consensus sequence of β-strand A of the nanobody fold⁶⁹ (QVQLVESGGGLV), a multi cloning site (MCS), the C-terminal part of Nb207 (residues 18128), the Aga2p anchor protein followed by the ACP and the Myc tag (Supplementary Fig. 8).

Site-directed randomization of the peptide linkers connecting Nb207 to scaffold cHopQ was achieved by PCR amplification of the cHopQ gene with mixtures of primers containing randomized codons. Accordingly, four parallel PCR reactions were performed using pNMB2 as a template with four pairs of primers TU131/TU133, TU131/TU134, TU132/TU133 and TU132/TU134 (Supplementary Table 5) to generate 1-1, 1-2, 2-1 and 2-2 random amino acid linkers, respectively. These four PCR products were gel-purified, pulled together (8 μg of each), mixed with 12 μg of Sapl-linearized pNMB1m_C_Nb207 to transform electrocompetent EBY100 as described above. This yeast library displaying diverse MbNb2û7 variants was orthogonally stained with CoA-647, incubated with 100 nM GFP and subjected to one round of selection by FACS as described above.

Construction and display of ${\text{Mb}}_{\text{NS}1}^{\text{cHopQ}}$ on yeast

pNS1MB (GenBank MT543227, Supplementary Fig. 16) is a derivative of pNACP³² encoding the following elements: the appS4 leader sequence (LS)⁸⁵, β_strand A of monobody NS1³¹ (residues 4-16), followed by Phe, the C-terminal part of HopQ (residues 227-449, UniProtKB B5Z8H1), peptide linker ASGGGSGGGGSG connecting the C-terminus and the N-terminus of HopQ to produce a circular permutant of the scaffold protein, an N-terminal part of HopQ (residues 49-221, UniProtKB B5Z8H1), a Gly residue, the C-terminal part of monobody NS1 (residues 19-97), the Aga2p anchor protein, the ACP and the Myc tag (Supplementary Fig. 16).

For applications in FACS, purified KRAS was labelled with a five-fold molar excess of the DyLight650-NHS ester (Thermo Fisher Scientific) following the manufacturer’s instructions. After a 30 min incubation at room temperature, unreacted label was quenched with 50 mM Tris pH 8.0 and the labelled protein was separated from free label by size-exclusion chromatography on a Superdex 200 PG 16/90 column (GE Healthcare).

Yeast cells containing the pNS1MB vector were inoculated, induced and orthogonally stained with CoA-488 to monitor the display level of the ${\text{Mb}}_{\text{NS}1}^{\text{cHopQ}}-\text{Aga}2\text{p}-\text{ACP}$ fusion on each yeast cell as described³². Cells transformed with $\text{pNMB}2\_{\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}$ that display ${\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}-\text{Aga}2\text{p}-\text{ACP}$ were analyzed in parallel as a control. 10⁵ cells stained with CoA-488 were incubated for 60 min at 4 °C with KRAS-Dylight650 in 100 μl of cold PBS—BSA, washed three times with PBS—BSA, resuspended in 100 μl, applied on FACS Fortessa (BD Biosciences) and analyzed using FlowJo software (FlowJo, LLC).

Reconstitution of human β3 homopentameric GABA_AR in nanodiscs

All purification and reconstitution steps were performed at 4 °C or on ice. Each of three cell pellets from 0.8 l culture were resuspended by vortexing in the dilution buffer: 50 mM HEPES pH 7.6, 300 mM NaCl, 1 mM histamine, 1% (w/v) mammalian protease inhibitor cocktail (Sigma-Aldrich). Solubilisation was performed for 1 h by adding 1% (w/v) lauryl maltose neopentyl glycol (LMNG, Anatrace) and cholesterol hemisuccinate (CHS, Anatrace) at a 10:1 (w/w) ratio. Solubilised GABA_AR was separated from insoluble material by centrifugation (10,000g, 15 min) and captured on a 1D4 affinity resin (250 μl, University of British Columbia) by slow rotation for 2 h. The resin was harvested (300g, 5 min) and washed three times with 50 ml of washing buffer: 50 mM HEPES pH 7.6, 300 mM NaCl, 1 mM histamine (Sigma-Aldrich), 1% (w/v) LMNG and 0.1% CHS. The washed resin was equilibrated with 1 ml of dilution buffer and supplemented with 240 μl of a mixture containing 80% (w/v) phosphatidylcholine (POPC, Avanti) and 20% of a bovine brain lipid (BBL) extract (Sigma-Aldrich). After 30 min incubation, the resin was equally divided in five Eppendorf tubes and collected by centrifugation. For nanodisc reconstitution, Bio-Beads (10 mg/ml final concentration, BioRad) with an excess of MSP2N2⁵¹ (0.6 mg/ml final concentration) were added to each sample. 100 μl of Nb25, ${\text{Mb}}_{\text{Nb25}}^{\text{c}7\text{HopQ}}$ or ${\text{Mb}}_{\text{Nb}25}^{\text{cYgjKE}2}$ (~120 μM) were added to corresponding sample tubes and incubated for 1 h slowly rotating. Resin samples were harvested (300g, 5 min), washed six times with dilution buffer, resuspended in 50 μl of elution buffer: 12.5 mM HEPES pH 7.6, 75 mM NaCl, 0.25 mM histamine, 1.5 mM 1D4 peptide (Cube Biotech) and incubated overnight. Beads were pelleted by centrifugation (300g, 5 min) to collect the supernatants. These supernatants were supplemented once more with 0.4 μl of ~ 120 μM of Nb25, ${\text{Mb}}_{\text{Nb25}}^{\text{c}7\text{HopQ}}$, and ${\text{Mb}}_{\text{Nb}25}^{\text{cYgjKE}2}$, respectively and incubated for 30 minutes on ice prior to cryo-EM grid preparation.

Cryo-EM sample preparation, image collection and processing of β3 homopentameric GABA_AR alone, in complex with Nb25, ${\text{Mb}}_{\text{Nb25}}^{\text{c}7\text{HopQ}}$, and ${\text{Mb}}_{\text{Nb}25}^{\text{cYgjKE}2}$

We used the same batch of β3_K279T GABA_AR reconstituted in lipid nanodiscs to prepare cryo-EM grids of the receptor alone, in complex with Nb25, ${\text{Mb}}_{\text{Nb25}}^{\text{c}7\text{HopQ}}$ or ${\text{Mb}}_{\text{Nb}25}^{\text{cYgjKE}2}$, respectively. 3.5 μl of each sample was applied onto glow-discharged gold R1.2/1.3 300 mesh UltraAuFoil grids (Quantifoil) for 30 s and blotted for 5.5 s before vitrification in liquid ethane. A Vitrobot Mark IV (Thermo Fisher Scientific) was used for plunge-freezing at ~100% humidity and 14.5 °C.

Cryo-EM data of all samples were collected on a 300 kV Titan Krios microscope (Thermo Fisher Scientific) using a Falcon 3EC (Thermo Fisher Scientific) direct electron detector in counting mode and a Volta Phase Plate (VPP, Thermo Fisher Scientific). Data collection parameters were identical as for the high resolution structure of β3 GABA_AR bound to ${\text{Mb}}_{\text{Nb25}}^{\text{c}7\text{HopQ}}$ (see below, Supplementary Table 4).

In order to investigate the proportion of preferential particle views of β3 GABA_AR particles in samples where β3 homomer was alone or complexed with Nb25, ${\text{Mb}}_{\text{Nb25}}^{\text{c}7\text{HopQ}}$ or ${\text{Mb}}_{\text{Nb}25}^{\text{cYgjKE}2}$, small cryo-EM datasets were analysed by using the same basic data processing procedure. First, MotionCor2⁸⁷ was used to motion-correct the movies and Warp⁸⁸ was applied to estimate the contrast transfer function (CTF), phase shift parameters and to pick, and extract particles. The reference-free 2D classification was performed using RELION 3.0⁸⁹. One round of 2D classification was performed and well-aligned 2D classes showing clear GABA_AR particle projections were used to determine the proportion of preferred particle orientations in each sample (around 6,000 particles for each of the conditions). Next, the particles from the 2D classification were subjected to reference-free 3D model generation and 3D refinement using cryoSPARC⁹⁰. The efficiency of the particle orientation distribution (E_od values) for each 3D model was calculated using cryoEF¹⁵.

For the high-resolution reconstruction of β3 GABA_AR bound to ${\text{Mb}}_{\text{Nb}25}^{\text{cYgjKE}2}$, a larger dataset was processed using RELION 3.0 as described below. MotionCor2 and Gctf⁹¹ wrappers inside RELION 3.0 package were used to motion-correct movies and to estimate the contrast transfer function and phase shift parameters, respectively. Manual inspection was used to discard poor quality movies. Next, particles were auto-picked using a Gaussian blob autopicker function in RELION 3.0. The resulting particles were 2D classified and the best classes were selected for further processing. Stochastic gradient descent (SGD) methodology⁹⁰ (RELION 3.0) was used to generate an initial reference-free 3D model. A ‘gold standard’ 3D-refinement was performed and subjected to Bayesian particle polishing⁹². Next, particles were sorted using 3D classification jobs without particle alignment and the highest-resolution classes were used for a 3D-refinement using a soft mask and solvent-flattened Fourier shell correlations (FSCs). Further beam tilt correction and per-particle contrast transfer function refinement implementations in RELION 3.0 were applied, yielding the final cryoEM map of 2.49 Å resolution (FSC criteria of 0.143, Supplementary Fig. 19). Local map resolution was estimated with ResMap⁹³.

For atomic model generation, the coordinates of β3 GABA_AR (PDB ID: 4COF) and α,5β3 chimera (PDB ID: 5O8F) bound to Nb25 were used as templates. First, the atomic coordinates of β3 homomer ECD, TMD and Nb25 were fitted to the highest resolution map (2.49 Å) as rigid bodies using UCSF Chimera. Then, the coordinates were manually adjusted using COOT⁷⁷ followed by several rounds of global refinement and minimisation in real space with phenix_real_space_refine⁹. The geometry constraint files for the histamine molecule were generated using Grade Web Server (Global Phasing). The model geometry quality assessment was performed using the MolProbity⁷⁸ web server. To validate the refinement protocol, the coordinates of the final model were displaced by 0.5 Å using CCP-EM software⁹⁵. The resulting model was refined with phenix_real_space_refine against one of the half-maps produced by RELION 3.0. FSC curves were calculated between this model and the half-map used for refinement (‘work’) and the half-map, which was not used for refinement, (‘free’) usingphenix.mtriage⁹⁶. In addition, the FSC curve was calculated for the refined model vs the final sharpened map (‘full’). The separation between FSC_work and FSCf_ree curves was not significant, indicating that the model was not over-refined. Pore diameters were calculated using the HOLE⁹⁷ plug-in available in Coot. RMSD of GABA_AR β3 subunits bound to different molecules (Supplementary Fig. 20) were calculated for equivalent C_α positions 8447 and 8-217 for full β3 subunits and ECD, respectively, using UCSF Chimera⁸¹ and 30 Å distance cut-off.

Cryo-EM sample preparation, image collection and processing of β3 homopentameric GABA_AR in complex with anti-nanodisc ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$

Purification, MSP2N2 lipid nanodisc reconstruction, cryo-EM sample preparation for β3 GABA_AR in complex with ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ was performed as described above, except for the step that the β3 GABA_AR was only once supplemented with ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ at a molar ratio 1:3, 30 min prior vitrification step. Cryo-EM data were collected on a 300 kV Titan Krios microscope (Thermo Fisher Scientific) using a Gatan K2 (Gatan) direct electron detector in counting mode. Data collection parameters are provided in Supplementary Table 3. High-resolution reconstruction of β3 GABA_AR bound to ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ was performed as described above, yielding the final cryo-EM map of 3.0 Å resolution (FSC criteria of 0.143, Supplementary Fig. 24).

Nanodisc reconstitution of mouse 5-HT3A receptor

All steps were performed at 4 °C. The peak fractions after size-exclusion chromatography in the C12E9 detergent (Anatrace) containing 5-HT3A receptor were pooled and concentrated to 1 mg/ml and mixed with asolectin lipids (Sigma-Aldrich) solubilized at 5 mg/ml in 5% DDM (Anatrace). After 30 minutes incubation, MSP1E3D1(-) (a gift from Stephen Sligar, Addgene plasmid #20066, expressed and purified as previously described⁹⁸) was added to the mixture, which was incubated for 30 additional minutes, before the addition of Bio-Beads (Sigma-Aldrich) at 10 mg/ml final concentration. The molar ratio of the receptor over the MSP and the lipids was 1:7:200. The mixture was incubated under gentle rotation overnight for detergent removal and nanodiscs reconstitution. Bio-Beads were removed by centrifugation (250g, 10 min) and the supernatant was subjected to size-exclusion chromatography in a Superose 6 Increase column (GE healthcare) equilibrated in SEC buffer (50 mM Tris-HCl, 125 mM NaCl, pH 7.5). The fractions containing the reconstituted receptor in nanodiscs were pooled, concentrated to 0.5 mg/ml, aliquoted, snap frozen in liquid nitrogen and stored at -80°C.

Cryo-EM sample preparation, image collection and processing of 5-HT3A receptor in complex with anti-nanodisc ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$

We used aliquots of the same batch of the 5-HT3A receptor reconstituted in nanodiscs to prepare cryo-EM grids of either the receptor alone or in complex with ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$. For the complex formation, the 5-HT3A receptor and ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ were mixed at a molar ratio 1:3, so that their final concentration was 0.35 and 0.25 mg/ml, respectively. The mixture was incubated for 30 minutes on ice prior to its application onto a grid. For each sample, 3.5 μl were deposited on a glow-discharged (30 mA, 50 s) gold-carbon R 1.2/1.3 grid (Quantifoil) for 5 s, blotted for 6 s with force 0, at 8 °C and 100% humidity using a Mark IV Vitrobot (FEI, Thermo Fisher Scientific) and plunge-frozen in liquid ethane, for sample vitrification.

For screening purposes two small datasets (250 movies for the reconstituted receptor alone and 499 movies for its complex with ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$) were recorded on a Glacios (FEI, Thermo Fisher Scientific) electron microscope at the IBS, Grenoble, where movies of 29 frames were acquired on a Falcon II direct electron detector, in counting mode. The raw movies were aligned and summed with MotionCor2⁸⁷ and CTF estimation for non dose-weighted sums was calculated using Gctf⁹¹. Particles were auto-picked with crYOLO 1.5.6⁹⁹, using the general model for low-pass filtered images, and extracted in RELION 3.0⁸⁹ with 256x256 pixel box size. Particles stacks were imported to cryoSPARC⁹⁰ for 2D classification and class averages that resembled a pentameric ion channel were selected and used to determine the proportion of top over tilted or side views in each sample (Supplementary Fig. 25).

For the sample incubated with ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$, a larger dataset was recorded with a Gatan K2 Summit direct electron detector, in counting mode. Data collection parameters are provided in Supplementary Table 3. Raw movies were aligned and summed with MotionCor2 and imported to cryoSPARC for CTF estimation, particle blob auto-picking and extraction (256x256 pixel box size) and all subsequent steps. After the initial 2D classification cleaning, the selected particles were subjected to a second round of 2D classification and the particles from well-aligned class averages were further sorted in 3 classes by reference-free 3D model generation and 3D classification. The class that presented typical 5-HT3 receptor features was selected and subjected to Non-Uniform refinement, with C5 symmetry, yielding a masked map at 3.5 Å resolution (FSC criteria of 0.143, Supplementary Fig. 25). The efficiency of the particle orientation distribution (E_od value) for the final 3D reconstruction was calculated using cryoEF¹⁵. This reconstruction and the associated model will be described in a forthcoming publication.

Cryo-EM sample preparation, image collection and processing of WbaP in complex with ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$

Cell pellets were lysed by cell disruption, cleared at 10,000g for 30 min and membranes pelleted at 186,000g. WbaP-containing membranes were solubilised in PBS containing 0.1% (w/v) lauryl maltose neopentyl glycol (LMNG, Anatrace) overnight at 4 °C and cleared by centrifugation at 186 000g. Solubilised WbaP was bound to 5 ml nickel resin (ABT), washed with 50 ml wash buffer (50 mM sodium phosphate pH 8, 250 mM NaCl, 45 mM imidazole, 2 mM β-mercaptoethanol and 0.003/LMNG) and eluted in 10 ml elution buffer (wash buffer with 300 mM imidazole). After buffer exchange into 20mM Tris-Cl pH8.0, 2mM β-mercaptoethanol, 10% glycerol and 0.003/ LMNG, the His-tag was removed by incubation with TEV protease and WbaP was separated by reverse nickel chromatography. ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ was added to WbaP at a 1.2:1 molar ratio and incubated for 3 h, at 4 °C on a rotating wheel. Amphipol A8-35 was added at 4x mass excess to WbaP, incubated overnight at 4 °C and detergent was removed by the addition of Bio-Beads. The complex was applied to a Superdex 200 (GE Healthcare) and run in 20 mM Tris-HCl pH8.0, 150mM NaCl and 2 mM β-mercaptoethanol. A single fraction was concentrated to 3 mg/ml and applied onto glow- discharged gold R1.2/1.3 200 mesh grids (Quantifoil) and blotted for 3 sec before vitrification in liquid ethane. A Vitrobot Mark IV (Thermo Fisher Scientific) was used for plunge-freezing at ~100% humidity and 4 °C.

A dataset of 2072 movies was recorded on a Glacios (FEI, Thermo Fisher Scientific) electron microscope at OPIC, Oxford, using a Falcon3 direct electron detector in integration mode. Raw movies were motion-corrected using RELION’s implementation of MotionCorr2⁸⁷. CTF estimation was done with CTFFIND-4.1¹⁰⁰ using the sums of power spectra from combined frames corresponding to an accumulated dose of 3.846 e-/Å2 and micrographs with estimated resolution lower than 5 Å were discarded, leaving 2016 micrographs for downstream processing. Autopicking was done using the Laplacian-of-Gaussian picker in RELION and 2,063,047 particles were extracted at 5.12 Å/pix in a box of 245 Å and subjected to 2D classification. A total of 419,117 good particles were selected and re-extracted at 1.92 Å/pix then subjected to a round of 3D classification. After selecting the best class, 209,298 particles were selected for refinement using a 3D reference generated in cryoSPARC (v2.14.1- live_privatebeta)⁹⁰. To generate the 3D reference, all particles picked in RELION (2,063,047) were extracted at full pixel size (0.96 Å/pix), imported into cryoSPARC and subjected to three rounds of 2D classification, ab-initio model generation (2 classes, window inner radius 0.65, class similarity 0.5, C1 symmetry). Best class was selected and subjected to two rounds of NU-refinement¹⁰¹ (window inner radius 0.45 in first and 0.65 in second round; C2 symmetry imposed in both) and a single round of homogeneous refinement (window inner radius 0.65; C2 symmetry). This map was then downsampled to 1.92 A/pix, imported into RELION and used a 3D reference map in the standard 3D auto-refinement of the selected 209,298 particles. This yielded an initial reconstruction with an estimated resolution of 5.3 Å. Particles were then subjected to Bayesian polishing⁹² to optimise per-particle beam-induced motion tracks, followed by another round of auto-refinement that resulted in the final map with estimated resolution of 4.9 Å (FSC threshold of 0.143, Supplementary Fig. 26).

Analysis of empty MSP1D1 nanodiscs in complex with ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ by size exclusion chromatography

The reconstitution of empty MSP1D1 nanodiscs containing phosphatidylcholine (Avanti) only was performed as described previously⁵¹. Next, a fraction of these nanodiscs was incubated for 30 min at 4 °C with ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ at a molar ratio 1:3 and analyzed on a Bio SEC-3 HPLC column (Agilent). The chromatogram of the nanodiscs incubated with ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ was compared to the chromatograms of the empty nanodiscs and of the megabody ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ only (Supplementary Fig. 23).

Data Availability

DNA sequences of pMESD2, pMESD22c7, pMESP23E2, pMESP23NO, pNMB2, pNMB1m_C_Nb207 and pNS1MB plasmids have been deposited in GenBank with accession codes MT328400, MT338520, MT338521, MT338522, MT338523, MT543226 and MT543227, respectively. All megabody expression plasmids are available from the Steyaert Lab upon request.

X-ray structure coordinates and structure factors for ${\text{Mb}}_{\text{Nb}207}^{\text{cHopQ}}$, ${\text{Mb}}_{\text{Nb}207}^{\text{c}7\text{HopQA}12}$, ${\text{Mb}}_{\text{Nb}207}^{\text{c}7\text{HopQG}10}$ and ${\text{Mb}}_{\text{Nb207}}^{\text{cYgjKNO}}$ structures have been deposited in the Protein Data Bank under accession codes 6QD6, 6XVI, 6XV8 and 6XUX, respectively. Atomic coordinates and cryo-EM density maps of β3 GABA_AR - ${\text{Mb}}_{\text{Nb25}}^{\text{c}7\text{HopQ}}$ have been deposited in the Protein Data Bank and the Electron Microscopy Data Bank under accession code 6QFA and EMD-4542, respectively. Cryo-EM density map of β3 GABA_AR - ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ complex has been deposited in the Electron Microscopy Data Bank under accession code EMD-11610. The refined atomic coordinates and cryo-EM maps of WbaP - ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ and 5-HT3A - ${\text{Mb}}_{\text{NbF}3}^{\text{c}7\text{HopQ}}$ complexes will be published elsewhere.

Other data that support the findings of this study are available from the corresponding authors on request.