Affinity-matured homotypic interactions induce spectrum of PfCSP-antibody structures that influence protection from malaria infection

The generation of high-quality antibody responses to PfCSP, the primary surface antigen of Plasmodium falciparum sporozoites, is paramount to the development of an effective malaria vaccine. Here we present an in-depth structural and functional analysis of a panel of potent antibodies encoded by the IGHV3-33 germline gene, which is among the most prevalent and potent antibody families induced in the anti-CSP immune response and targets the NANP repeat region. Cryo-EM reveals a remarkable spectrum of helical Fab-CSP structures stabilized by homotypic interactions between tightly packed Fabs, many of which correlate with somatic hypermutation. We demonstrate a key role of these mutated homotypic contacts for high avidity binding to CSP and in protection from P. falciparum malaria infection. These data emphasize the importance of anti-homotypic affinity maturation in the frequent selection of IGHV3-33 antibodies, advance our understanding of the mechanism(s) of antibody-mediated protection, and inform next generation CSP vaccine design.


5
. In all, the panel includes seven IGHV3-33 mAbs 116 (Table 1)  We next utilized single particle cryo-EM to elucidate the 3D organization of these distinctive 137 Fab-rsCSP structures, the potential roles of homotypic contacts, and the mechanisms governing 138 their formation. Cryo-EM datasets were collected for the seven IGHV3-33 mAbs in our panel.

139
Each complex was resolved to high resolution (Table S1)

161
The 227 Fab complex is distinct, as the NPNA repeats form two discontinuous, anti-parallel 162 disc-like structures with moderate helical pitch and left-handed curvature, with each disc bound 163 by 4 Fabs in tandem (Fig. 1). However, we note the 227 Fab structure was solved in complex 164 with NPNA8 peptide instead of rsCSP, as was done for the rest of the mAbs in the panel, due to 165 the tendency of the 227-rsCSP complex to aggregate. Thus, the two antiparallel disc structures 166 in the 227 complex likely comprise two individual NPNA8 peptides, as the four available NPNA2 7 epitopes on the peptide are fully occupied by 227 Fabs and there is no density linking the two 168 discs. Nonetheless, a NS-EM reconstruction of the 227-rsCSP complex is nearly identical to the 169 227-NPNA8 cryo-EM structure, within the 15-20Å limit of the NS data (data not shown).

170
Therefore, this antibody may induce dimerization of separate CSP molecules mediated by 171 homodimeric interactions of the Fabs themselves, which may have important implications for the 172 way this antibody engages PfCSP on sporozoites.

174 175
The IGHV3-33 NPNA2 core epitope structure is highly conserved 176 As shown previously for this family of antibodies, the epitope of each IGHV3-33 mAb comprises 177 two tandem NPNA structural units, with an N-terminal type 1 -turn followed by an Asn-

179
Interestingly, despite large differences in global helical structure, the local structure of this core 180 (NPNA)2 epitope is highly conserved and exhibits a nearly identical extended S-shaped 181 conformation in each of the seven mAbs (Fig. 2B). rsCSP binds within a deep groove running 182 along the length of each Fab that is composed entirely of the three heavy chain CDR loops and 183 CDRL3 ( Fig. 2A). Overall, the structure of the IGHV3-33 heavy chain is also highly conserved.     Table 1). The extent of 204 the full epitope footprint on rsCSP tends to correlate with light chain usage and CDRH3 and 205 CDRL3 length (Table 1). Thus, these key antibody features appear to determine the binding 206 mode, superstructure assembly and fine epitope specificity of anti-NPNA antibodies and may 207 also correlate with protective efficacy.

210
Each of the multivalent antibody-CSP structures are stabilized by homotypic interactions 211 between Fabs binding immediately adjacent NPNA2 epitopes, i.e., the primary homotypic 212 interface (Interface 1; Fig. 3). This expands the full paratope, as each Fab simultaneously binds 213 both CSP and the neighboring Fab, and substantially increases the total buried surface area 214 (BSA) on each Fab (Table S2). The architecture of the primary homotypic interface is similar 215 across the seven complexes and is composed mainly of the heavy chain CDR loops and 216 CDRL3, with polar contacts between CDRH1A-CDRH2B and CDRH3A-CDRL3B (Fig. 3B).

217
Importantly, this asymmetric, edge-to-edge interaction, in which FabA and FabB contribute 218 different residues to the interface, is distinct from the asymmetric head-to-head configuration 9 observed in the crystal structure of mAb 1210-NANP5, another potent IGHV3-33/IGKV1-5 220 antibody (Imkeller et al., 2018). This mode of binding also differs from our previous crystal 221 structure of 399-NPNA6 (IGHV3-49/IGKV2-29), which forms a symmetric head-to-head 222 homotypic interface between adjacent Fabs (Pholcharee et al., 2021). As these latter two mAbs 223 are not known to form stable structures on extended repeats, the edge-to-edge binding mode 224 seen here is likely necessary for optimal geometry and packing of Fabs to promote long-range 225 helical order.   symmetry plane for the complex (Fig. S5B). This interface is therefore symmetric and consists 251 solely of apposing heavy chain framework residues. In general, the secondary homotypic 252 interface contributes about half of the total BSA relative to the primary interface (Table S2).

253
Strikingly, however, the reverse is true with 334, where the total BSA of the secondary interface 254 is roughly twice that of the primary, suggesting a critical role for framework region residues in 255 the stability and/or formation of this complex.

257
Key contacts within the secondary interface are also linked to somatic hypermutation of the 258 germline heavy and light chain genes. In HFR1B of 334, a mutated residue T19 appears critical 259 for the interface and mediates a key hydrogen bond with S65 of LFR3E ( Fig S5D). In the 260 symmetric secondary interface of 227, H82A of HFR3B mediates a cation-pi bond with R75 of 261 HFR3G (Fig. S8C). Both were mutated from highly conserved residues in the germline IGHV3-

281
We first determined the impact of these mutations on binding to various CSP peptides with 282 biolayer interferometry (BLI). We tested the hypothesis that homotypic interactions underlie the 283 large increase in apparent affinity to peptides with increasing NANP content that is observed for 284 this family of antibodies. Thus, we compared binding of WT and mutant Fabs to NPNA4, 285 NPNA8, and rsCSP. In terms of NPNA4, the apparent affinity of 311R was essentially 286 unchanged relative to 311 (p=0.17), while 356R and 239R were ~two-fold higher (improved) 287 (p=,0.005) and ~two-fold lower (p=0.01) than 356 and 239, respectively (Fig. 4I,J; Table S10).

288
These BLI data suggest that binding to minimal NPNA repeats is largely unperturbed by the 289 germline mutations. As expected, for each WT Fab we observed a large increase in apparent 290 affinity to both NPNA8 and rsCSP relative to NPNA4, largely driven by substantial reductions in 291 the dissociation rate (k off ). However, for the reverted mutants, both affinity and k off remained 292 roughly constant across each peptide and rsCSP. Thus, homotypic interactions are critical for 293 high avidity binding to extended NANP repeats.

295
We next used NS-EM to assess the impact of the germline mutations on the structure of the 296 Fab-rsCSP complex. As shown previously for 311R, 2D class averages of both 239R and 356R 12 were highly variable, both in structure and stoichiometry of the Fabs (Fig. 4H). Interestingly, we 298 observed some helical propensity in the 239R-rsCSP complex, similar to 311R, suggesting 299 helical structure formation is at least partially germline-encoded or that CSP has a preferential 300 bias toward a helical conformation. Nonetheless, we were unable to obtain stable 3D 301 reconstructions for each mutant, indicative of a high degree of structural disorder. In contrast, 302 the WT versions formed stable helical structures on rsCSP (Fig. 1, Fig. S1). Thus, somatically 303 mutated homotypic interactions are crucial for both high avidity and for the formation and 304 stability of long-range, helical order on rsCSP, both of which may impact protective efficacy.

306
To ensure these effects were due to the loss of homotypic interactions rather than unanticipated 307 changes in the structure of the antibody paratope, which could impact the structure of the bound 308 NPNA2 epitope, we solved a 1.9Å co-crystal structure of Fab311R in complex with NPNA3 and 309 compared this to our previous X-ray structure of Fab311 bound to NPNA3 (

311
with an overall RSMD of 0.28Å. Due to their similarity with 311R, we expect this to also be true 312 for 239R and 356R, although we did not obtain crystals of these mAbs. Therefore, the effects of

356
As each of these mAbs target the same epitope(s) on CSP, affinity to the NPNA repeat may 357 underlie differences in protection. To test this notion, we measured apparent affinities of each 358 Fab to NPNA4, NPNA8, and rsCSP with BLI (Table S10). Except for the germline mutants, 359 apparent affinity increased substantially between NPNA4 and NPNA8, and again between 360 NPNA8 and rsCSP; this is likely due to the high avidity afforded by homotypic interactions, as 361 increases in avidity were largely driven by reductions in the dissociation rate (koff). In general, 362 rsCSP apparent affinity was very high (10 -9 M or higher) for the WT Fabs, and lower for the three 363 mutants (10 -7 to 10 -8 M). We then correlated these data with percent inhibition from the liver 364 burden experiment. Interestingly, we observe no correlation between protection and NPNA4 365 affinity, while there is a moderate correlation with NPNA8 and rsCSP apparent affinity (R 2 =0.61 366 and 0.68, respectively), as well as rsCSP dissociation rate (R 2 =0.65). These data suggest 367 avidity to extended NPNA repeats, which is facilitated by homotypic interactions, is a key 368 determinant of protective efficacy among IGHV3-33 mAbs. However, apparent affinity to NPNA8 369 or rsCSP poorly discriminates protective efficacy among the WT mAbs in the panel, which all 370 have rsCSP apparent affinities of 10 -9 M or lower. Therefore, high avidity to extended repeats is 371 likely necessary but on its own insufficient to confer high level protection in IGHV3-33 mAbs.

372
Thus other parameters, likely concerning specifics of the interaction of antibodies with PfCSP on 373 live sporozoites, also appear to be involved in determining protective efficacy.

391
We also demonstrate somatically mutated homotypic interactions, and possibly the CSP 392 structures that they stabilize, play a key role in the mechanism of protection from P. falciparum 393 infection. Specifically, we show reversion of these somatically mutated residues to their 394 germline identities, in both heavy and light chains, abolishes well-ordered, extended helical CSP 395 structures and eliminates the high avidity to extended NANP repeats characteristic of this 396 antibody family, without significantly impacting affinity to the core epitope or the ability to 397 assemble multiple Fabs onto CSP. Importantly, these effects are accompanied by a significant

558
Geometric mean and SD are indicated as black and colored lines, respectively. A Mann-559 Whitney U-test was used to compare efficacy relative to naïve (no mAb) and between WT and 560 mutant mAbs. Percent inhibition listed is relative to naïve. Significance: * p<0.05; *** p<0.001.

656
In all cases, Leginon was used for automated data collection. CtfFind4. The Laplacian-of-Gaussian picker was used for initial autopicking on a subset of 663 micrographs, and initial 2D templates were generated with multiple rounds of 2D classification.

27
High quality templates were selected as input for the automated template picker in RELION for 665 use on the whole dataset. Multiple rounds of 2D classification were used to eliminate low-666 quality particles, after which a total of 605,000 particles were re-extracted for 3D classification.

667
Our previous cryo-EM reconstruction of 311-rsCSP was used as the initial reference, low-pass 668 filtered to 60Å. A global angular search was used in the initial round of 3D classification, 669 followed by multiple rounds of 3D classification without alignment. This process resulted in a 670 final stack of ~400,000 particles that were re-extracted to generate a consensus refinement at 671 3.38Å, which is the same resolution of our previous 311-rsCSP cryo-EM map generated from 672 these same data (EMD-9065). Further processing in RELION3.0 was used to improve the 673 resolution of this complex. Per particle defocus values were refined in RELION, followed by 674 another round of 3D refinement and then Bayesian polishing, which refines per-particle beam-

687
The remaining datasets were all processed according to a similar protocol in cryoSPARC.

688
Frames were motion-corrected with MotionCor2, and the aligned and dose-weighted 689 micrographs were imported into cryoSPARCv3.3. CTF estimation was performed with CtfFind4.

28
Autopicking was performed initially with the blob picker in cryoSPARC, and multiple rounds of 691 2D classification were used to select high quality 2D templates for subsequent template picking.

692
Multiple rounds of 2D classification were used followed by a single round of Ab-initio 693 reconstruction with two classes. The high-quality class was selected for further processing and 694 was also used as the initial model. Multiple rounds of homogenous refinement, global and local 695 CTF refinement, followed by non-uniform refinement were performed which led to the final 696 reconstructions for each data set.    Table   721 S3-S9 were derived from the Epitope Analyzer software, part of the ViperDB webserver      (Table S10). Significance 784 was calculated with a student's T-test. For antibody pharmacokinetics studies in mice (N=5), 785 non-linear regression was used to analyze the ELISA data using Prism 9 software, and 786 circulating human IgG concentrations were interpolated based on a standard curve; data were 787 then reported as the geometric mean +/-the SD, in mg/mL ( Figure S7).