Structure of the malaria vaccine candidate Pfs48/45 and its recognition by transmission blocking antibodies – Nature.com

A structural comparison of two antibodies targeting domain 3 of Pfs48/45

The two antibodies which have the highest transmission-blocking activity, 32F3 and 85RF45.1, both bind to the C-terminal domain of Pfs48/458,12. However, these two antibodies show substantial differences in transmission-blocking activity, with 85RF45.1 completely preventing transmission at 14g/ml, while 32F3 is only ~40% effective at 70g/ml12. To rationalise these differences, and to guide future vaccine design, we determined the structure of the C-terminal 6-cys domain of Pfs48/45 bound to 32F3 (Fig.1A, Supplementary Table1). We prepared Fab fragments from 32F3, combined these with the Pfs48/45 C-terminal domain and conducted crystallisation trials. Crystals formed and a complete dataset was collected to 1.9 resolution, allowing structure determination by molecular replacement.

A The structure of the C-terminal domain of Pfs48/45 (Pfs48/45-D3; blue) bound to antibody 32F3 (orange). The upper right inset shows a close-up of a Pfs48/45 loop which becomes ordered on 32F3 binding. The lower inset shows the equivalent view of the complex of Pfs48/45-D3 (blue) bound to 85RF45.1 (red). B An alignment of the structures of 32F3 (orange) and 85RF45.1 (red) bound to Pfs48/45-D3 (blue). C Surface plasmon resonance analysis of the binding of Pfs48/45-D3 to immobilised 32F3 and 85RF45.1. In both cases, the black lines show the responses due to a two-fold dilution series with a top concentration of 7.8nM for 85RF45.1 and 125nM for 32F3. The red dashed lines show fitting to a 1-to-1 binding model.

The structure of the C-terminal domain of Pfs48/45 was largely unchanged in conformation from that observed in complex with 85RF45.1, with a root mean square deviation of 0.66 (Fig.1A, B)12,29. The most substantial change was in the loop comprising residues 357369. This loop is disordered when bound to antibody 85RF45.1 but becomes ordered through interactions with 32F3 (Fig.1A). While antibodies 32F3 and 85RF45.1 bind to overlapping epitopes, many of the contacts are not shared and the Fabs approach Pfs48/45 at different angles.

We also determined the binding kinetics for both 85RF45.1 and 32F3 to Pfs48/45, using surface plasmon resonance measurements (Fig.1C, Supplementary Fig.1). We first captured 85RF45.1 and 32F3 on different flow paths of a chip coated with protein A/G and flowed two-fold dilution series of Pfs48/45 over these antibody-coupled surfaces. Both antibodies bound to Pfs48/45 with similar affinities in the low nanomolar range (2.5nM for 85RF45.1 and 5.2nM for 32F3). However, binding kinetics differed, with 85RF45.1 showing faster association- and dissociation-rates, while 32F3 binds more slowly but forms a more stable complex. It seems likely that the slower binding kinetics of 32F3 may be due to the need for loop 357369 to become ordered on binding, while antibody 85RF45.1 has an epitope which is unchanged in structure on antibody binding, allowing faster binding kinetics.

We next aimed to determine the structure of the complete Pfs48/45 molecule, allowing us to reveal the architecture of this three-domain protein and the locations of antibody epitopes outside the C-terminal domain. Insect cell expression systems were available to produce full-length Pfs48/45 ectodomain, or to generate a protein containing the central and C-terminal domains, Pfs48/45-D2+3 [11]. Both were expressed without the GPI-anchor modification site and were purified and mixed in different combinations with one or more Fab fragments, selected from a set of antibodies which bind to different domains of Pfs48/45 (85RF45.1, 85RF45.3, 85RF45.525, 32F38, 10D8, 9D1, 7A612). Three of these complexes generated crystals: full-length Pfs48/45 bound to 10D8; full-length Pfs48/45 bound to both 10D8 and 85RF45.1; and Pfs48/45-D2+3 bound to both 10D8 and 32F3. Datasets were collected to 4.2, 3.72 and 3.69, respectively.

We attempted to determine the structures of these complexes by molecular replacement, using the structures of the C-terminal domain bound to either 32F3 or 85RF45.1 Fab as search models. In each case, the resultant electron density maps were sufficiently detailed to allow us to build a model for the central domain of Pfs48/45. This domain contains the epitope for 10D8, also allowing us to build a model for the 10D8 Fab (Fig.2, Supplementary Tables1 and 2). However, while electron density could be observed for the N-terminal domain of Pfs48/45, this region was not sufficiently well resolved to allow a model to be built.

Structures of Pfs48/45 bound to different combinations of Fab fragments or scFvs. Pfs48/45 is shown with the three domains in different shades of blue. The N-terminal domain is light blue, the central domain is mid-blue and the C-terminal domain is dark blue. 10D8 is yellow, 85RF45.1 is red and 32F3 and its scFv are orange. Each figure also shows the electron density in pale blue, shown at a contour threshold of 1.0.

The hinge between the variable and constant domains of Fabs is flexible. In case this flexibility reduced the degree of order within the crystals, we also generated scFv constructs for 85RF45.1, 32F3 and 10D8. These were purified, mixed with full-length Pfs48/45 and complexes were subjected to crystallisation trials. Crystals formed for the complex containing 32F3 scFv and a dataset was collected to 2.13 resolution, allowing structure determination by molecular replacement. This higher resolution data allowed us to build an improved model for the central domain of Pfs48/45 (Fig.2). However, the density for the N-terminal domain was still too poorly resolved to allow building of a complete molecular model of this domain. Analysis of packing within the different crystals suggested that, in each case, the N-terminal domain lies within a substantial cavity within the crystal lattice, and the lack of packing against other regions of Pfs48/45 contributed to disorder of this domain, even within a well-ordered lattice.

Recent improvements in protein structure prediction from AlphaFold230 provided a solution, allowing us to interpret the electron density maps for the N-terminal domain. AlphaFold2 correctly predicted the architecture of the central and C-terminal domains (with root mean square deviation of 1.26 for the central domain and 0.43 for the core 753/1044 atoms of the C-terminal domain), albeit with differences in loop structure (accession code Q8I6T1 at alphafold.ebi.ac.uk) (Supplementary Fig.2). We therefore compared the AlphaFold2 model for the N-terminal domain with the electron density for this domain within the three different maps. The most complete electron density for this region of Pfs48/45 was obtained from crystals of full-length Pfs48/45 bound to 10D8. The AlphaFold2 model was therefore docked as single rigid body into this electron density and was rebuilt to fit the density. This generated a model for 122 residues of the 150 residue-long N-terminal domain, with a root mean square deviation of 1.63 from the AlphaFold2 model. The N-terminus and loops 6268 and 163168 were unresolved. This model was then used to guide building of the observed portions of the N-terminal domain in full-length Pfs48/45 bound to 10D8 and 85RF45.1 (95 residues were built) and Pfs48/45-D2+3 bound to both 10D8 and 32F3 (20 residues were built). Through this approach, we generated the first molecular models of full-length Pfs48/45 derived from experimental crystallographic data, with an AlphaFold2 model used to guide building of the N-terminal domain (Fig.2, Supplementary Tables13).

We next used our structures, together with molecular dynamics simulations, to understand the organisation of Pfs48/45. Our three structures of Pfs48/45 all reveal a flattened, disc-like architecture (Fig.3A), with a 20 residue long linker before the GPI anchor. This membrane-attachment site emerges from a flat surface of the molecule, suggesting that the disc may lie, on average, parallel to the membrane, with all three domains equally exposed at the gametocyte surface.

A The structure of Pfs48/45 taken from that of full-length Pfs48/45 bound to 10D8, viewed from two different directions. The N-terminal (D1), central (D2) and C-terminal domains (D3) are light, medium and dark blue. The right-hand panel also indicates the membrane and the 20 residue long linker not included in our constructs. B Comparison of the three full-length Pfs48/45 structures. Three structures were constructed by taking the models of Pfs48/45 from the three different crystal structures of Pfs48/45 bound to different antibody combinations and aligning the structure of the N-terminal domain onto the fragments of the domain built into the electron density. These composite structures have been aligned based on the central and C-terminal domains, showing the motion of the N-terminal domain, highlighted by the red arrow. C Histogram showing the observed interdomain angles in full-length Pfs48/45 during atomistic molecular dynamics simulations. To define these angles two lines were drawn, linking the centres of mass of the N-terminal and central domains and those of the central and C-terminal domains. The angle shown is that between these lines. The three coloured histograms refer to outcomes of three independent simulations which started with the models in panel (B) with histograms being constructed by pooling five replicates for each starting structure. The black histogram is the pooled distribution of all replicates in all three starting models. D The average angle observed in the simulations from (C). is shown top-centre with the most closed observed model bottom-left and the most open model bottom-right. Red lines are as described in C. E shows the fitting of models from the molecular dynamic simulations to data from small angle X-ray scattering. Each point represents a different model and the 2 gives the quality of fit to the scattering curve. Horizontal lines show the 2 for the fit corresponding to interdomain angles found in the three crystal structures, as seen in B.

The availability of three structures of Pfs48/45 allowed us to assess its degree of flexibility within crystals. We docked the most complete model of the N-terminal domain, from the structure of full-length Pfs48/45 bound to 10D8, onto the fragments of the domain observed in the two other crystal forms and the three resultant models for the full Pfs48/45 ectodomain were aligned (Fig.3B). While the relative positions of the central and C-terminal domains were largely unchanged across these molecules, the position of the N-terminal domain changed, due to flexibility in the linker joining the N-terminal and central domains. Through these movements, the separation between the N- and C-terminal domains varied.

To further assess the degree of motion within Pfs48/45, we used molecular dynamics simulations. Three separate simulations were run, with our three independent structures of Pfs48/45,each modified to include the aligned N-terminal domain, used as three distinct starting points. In each case, we simulated the system for 500ns with 5 repeats. Analysis of these simulations revealed motion of the N-terminal domain in excess of that seen in the three crystal structures, leading to further separation of the N- and C-terminal domains. The two peaks seen in some of the individual simulations were not observed in a pooled simulation, suggesting that each individual simulation had not converged, with pooled simulations reaching a more representative endpoint. We plotted two lines though the structure; one linking the centres of mass of the N-terminal and central domains and one linking the centres of mass of the central and C-terminal domains. The angle between these lines was used as a measure of the degree of opening of the complex. This varied from 56.4 to 114.9, with an average of 72.8 and a standard deviation of 7.3 (Fig.3C, D, Supplementary Fig.3). The major peak varied from 56.4 to 92.1, although we also observed a single instance of extreme opening to 112.7 (Supplementary Fig.3A). This compares with angles of 60.3, 63.1 and 67.5 in the three crystal structures.

To compare the fit of our molecular dynamic simulations to experimental data, we subjected Pfs48/45 to small angle X-ray scattering (Supplementary Fig.4). We then sampled the molecular dynamic simulation trajectories to generate a PDB file for every 50th frame and fitted each of these to the scattering data using FoxS31 thereby generating a 2 which allows us to quantify the fit. For comparison, we also fit the input models used for simulations, with domain positions matching those found in the three crystals. We then plotted the 2 against the interdomain angle (Fig.3E). This analysis shows that no single angle fits most closely to the SAXS data, with a broad range of models, with interdomain angle 56 from 93, all fitting with a lower 2 than any of the crystal structures. Indeed 66% of the models derived from simulation fitted the SAXS scattering data more closely than the three crystal structures. This confirms that Pfs48/45 in solution is a dynamic molecule, sampling a wide range of interdomain angles, through movement of the N-terminal domain relative to the central and C-terminal domains. This flexibility may allow Pfs48/45 to adopt different conformations on binding to partners, such as Pfs230 and will also leave all three domains exposed to antibody recognition.

We next combined crystallography, electron microscopy and molecular dynamics simulations to assess how different monoclonal antibodies bind to Pfs48/45. In addition to visualising antibodies 32F3 and 85RF45.1 (Fig.1), our crystal structures revealed the epitope for 10D8 (Fig.2), an antibody with weak transmission-reducing potential which binds to the central domain of Pfs48/4512. We combined the structure of Pfs48/45 bound to 10D8 with those of the C-terminal domain bound to 32F3 and 85RF45.1 to generate a composite model, showing the structure of Pfs48/45 bound to these three antibodies (Fig.4A).

A The structure of Pfs48/45 bound to 10D8 is used as a template, and combined with structures of the D3 domain of Pfs48/45 bound to either 85RF45.1 or 32F3 to generate a model, showing the relative locations of the three antibody epitopes. B A measure of the flexibility of the residues which form the epitopes for 32F3, 85RF45.1 and 10D8, as derived from molecular dynamics simulations. Each epitope was analysed in all 15 replicates and significance levels are based on a two-sided MannWhitney U tests where *** indicates p<0.0005 and **** indicates p<0.00005. The box bounds are interquartile ranges, and the lines within boxes are the median values. Whiskers extend to 1.5-fold the interquartile range. For 32F3 vs 85RF45.1 p=0.00008; for 32F3 vs 10D8 p=0.00031; for 10D8 vs 85RF45.1 p=0.58974. C A composite model, derived from negative-stain electron microscopy structures. In each case, the complexes contained Pfs48/45 bound to 85RF45.1, together with either 9A6, 3H6, 10D8, 1F10 and 6A10. These were imaged by negative stain electron microscopy, and models were fitting into the resultant envelopes. These models were used to align the envelopes, allowing us to derive a composite model, showing the location for the five epitopes.

To understand the degree of flexibility of these three epitopes we analysed the molecular dynamics simulations presented in Fig.3, specifically assessing the motion of residues directly in contact with each of the three antibodies. This revealed that the epitope for 32F3 is significantly more flexible than those for 85RF45.1 and 10D8 (Fig.4B), predominantly due to motion of the 357369 loop, which becomes ordered on 32F3 binding due to direct interactions with the antibody (Supplementary Fig.3). This indicates that the slower association-rate for 32F3, compared to the association rates for either 85RF45.1 or 10D8, is due to this flexibility and to the need for the epitope to adopt the correct structure on antibody binding. In contrast, off-rate is likely to be determined by the degree of complementarity of the epitope and paratope when forming an interaction. Indeed, 32F3 has the slowest off-rate (Supplementary Fig.1), which correlates with a larger buried surface area for the epitope (11072 for 32F3, 8752 for 85RF45.1 and 7152 for 10D8) and a greater number of direct interactions (Supplementary Fig.3)12.

Finally, we used negative stain electron microscopy to visualise the approximate location on Pfs48/45 of the epitopes of four more antibodies. 1F10 and 6A10 are in the same competition group on the central domain as 10D8 while 9A6 is in a different competition group, binding to the same domain12. In contrast, 3H6 binds to the N-terminal domain. While 1F10 and 6A10 show some transmission-reducing activity, 9A6 and 3H6 do not12. In each case, we assembled complexes containing Pfs48/45, 85RF45.1 Fab and one other antibody Fab, with 85RF45.1 included to provide a clear marker which could be used to align and position each additional antibody. We then used negative stain electron microscopy to image the complexes and single particle analysis to determine a low-resolution structure. The Pfs48/45:85RF45.1 complex and an additional Fab model were then docked into these structures, taking into account to which domain the additional Fab bound to determine the organisation of the complex and the approximate location of the Fab. These five models were then aligned on Pfs48/45 and assembled together to show the locations of the five antibodies (Fig.4C, Supplementary Fig.5, Supplementary Table4). The locations of 10D8, 1F10 and 6A10, which are part of the same competition group, were superimposable. In contrast, 9A6 and 3H6 adopted different locations on Pfs48/45, with both protruding in approximately the same plane as that shared by the three domains of Pfs48/45.

These findings suggest that both accessibility and also unknown functional properties of Pfs48/45 contribute to the transmission-blocking efficacy of antibodies which target Pfs48/45. When attached to the membrane, through a C-terminal GPI anchor, we would expect the Pfs48/45 disc to be on average arranged horizontal to the membrane plane (Fig.3A). In this orientation, the epitope for 85RF45.1 will be most exposed on the membrane surface with the 32F3 epitope exposed to a lesser degree and the 10D8 epitope least accessible. This correlates with the order of transmission-blocking activity of these three antibodies, with 85RF45.1 as most potent and 10D8 as least potent. However, the epitopes for 9A6 and 3H6 emerge in the plane of Pfs48/45 and we would expect them to be more regularly exposed on the membrane surface than the epitope for 10D8, and yet they are not transmission-blocking12. It is also notable that neither 3H6 or 9A6 stain gametocytes, suggesting these epitopes to be buried in that context12. A possible explanation for this is that the region of Pfs48/45 bound by these antibodies may be occluded by its binding partners, such as Pfs230. Further studies of Pfs48/45 function are required to determine whether this is the case.

The structure of Pfs48/45 indicates that each of its three domains will be exposed on the gametocyte surface (Fig.3) and antibodies binding to each domain have been shown to stain gametocytes12. However, the most effective known transmission-blocking monoclonal antibodies bind to the C-terminal domain8,25. We therefore aimed to determine the degree to which antibodies against the N-terminal and central domains contribute to transmission-blocking activity, using a combination of immunisation and antibody depletion experiments.

We first produced protein consisting of the N-terminal and central domains of Pfs48/45 (Pfs48/45-D1+2) and used this to immunise mice. IgG was purified from these mice and the titres of antibodies targeting Pfs48/45-D1+2 were determined by end-point ELISA (Fig.5A). The transmission-blocking activity of these antibodies was then assessed using a standard membrane feeding assay. Antibodies were added to blood containing Plasmodium falciparum gametocytes, fed to mosquitos, and the number of ookinetes formed in mosquito midguts was counted. Total IgG, purified from mice immunised with either 0.1g or 1g of Pfs48/45-D1+2 was tested at a concentration of 750g/ml, causing almost complete inhibition (96 and 100% reductions resulting from 0.1g and 1g immunisations, respectively) in oocyte numbers (Fig.5B). When tested at 375 and 188g/ml, the total IgG from the 1g group also showed significant inhibition of 92 and 68%, respectively (Fig.5C). Therefore, protein immunogens containing just the N-terminal and central domains of Pfs48/45 are able to induce highly potent transmission-blocking antibodies.

AC Female CD1 mice (n=6 mice) were immunised twice with either 0.1g or 1g of Pfs48/45-D1+2, or 1g of Pfs25. Sera was collected three weeks after the final dose for analysis. A IgG titres as measured by endpoint ELISA using Pfs48/45-D1+2 as the coating antigen. Each symbol is for the serum sample from an individual mouse; lines represent the median of each group. MannWhitney two-tailed test was performed to compare the two Pfs48/45-D1+2 groups (p=0.0043). B, C Transmission-blocking efficacy of IgG induced by immunisation with Pfs48/45-D1+2. Total IgG was purified from the pooled serum of each group (3 weeks post-boost) and mixed with P. falciparum NF54 cultured gametocytes at 750g/mL (B) and in a separate experiment at 375g/mL and 83g/mL (C) and fed to A. stephensi mosquitoes (n=20 mosquitoes per test group of pooled IgG in a single feed experiment). IgG from naive mice was used as a negative control (normal mouse Ab); the transmission-blocking anti-Pfs25 mAb 4B7 was used as a positive control at a concentration of 94g/mL. Data points represent the number of oocysts in individual mosquitoes 8 days post-feeding; horizontal lines show the arithmetic mean. D, E Female CD1 mice were immunised three times with 5g of Pfs48/45-FL. Three weeks after the final dose sera were collected and total IgG purified. Total IgG was depleted of Pfs48/45-D3-targeting IgG using a column coupled with Pfs48/45-D3, resulting in Pfs48/45-D3. D Pfs48/45-D3 specific ELISA showing lack of recognition of Pfs48/45-D3 by Pfs48/45-D3. Pfs25-specific IgG (Pfs25) was included as a negative control. Data presented as mean+/standard deviation. E Transmission-blocking efficacy of total IgG from Pfs48/45-FL or Pfs48/45-D3 fed to A. stephensi mosquitoes at concentrations of 750g/mL, 250g/mL and 83g/mL (n=20 mosquitoes per test group of pooled IgG in a single feed experiment). IgG from naive mice was used as a negative control (normal mouse Ab); transmission-blocking anti-Pfs25 mAb 4B7 was used as a positive control. Data points represent the number of oocysts in individual mosquitoes; lines show the arithmetic mean (TRA, transmission reducing activity [% inhibition in mean oocyst count per mosquito]).

We next used a depletion approach to assess the contribution that antibodies which target the N-terminal and central domains of Pfs48/45 make to the transmission-blocking activity of IgG from mice immunised with full-length Pfs48/45. Total IgG from mice immunised with 5g of full-length Pfs48/45 was tested at a concentration of 750g/ml and showed 100% transmission-blocking activity (Fig.5D, E). This IgG was depleted of antibodies which bind to the C-terminal domain of Pfs48/45 using an affinity column and was tested for transmission-blocking activity (Fig.5D, E). Depletion of C-terminal domain (D3) specific antibodies from purified IgG was confirmed by ELISA, using Pfs48/45-D3 as the coating antigen, with no signal observed in the depleted IgG compared to the original un-depleted IgG (Fig.5D).

At all three total IgG concentrations tested, the sera depleted of C-terminal domain-targeting antibodies showed no significant difference to the sera prior to depletion (Fig.5E). This shows that the majority of the antibodies with transmission-blocking activity, induced through immunisation with full-length Pfs48/45, bind to either the N-terminal or central domains.

Continue reading here:

Structure of the malaria vaccine candidate Pfs48/45 and its recognition by transmission blocking antibodies - Nature.com