Introduction

Malaria remains one of the world’s deadliest infectious diseases, and in 2021 was responsible for 241 million clinical infections and 627,000 deaths worldwide (WHO, 2021), primarily among young children in sub-Saharan Africa. RTS,S/AS01B (RTS,S), the only approved malaria vaccine, is only partially effective, providing ~30% protection from clinical infection after four years in children aged 5–17 months1,2. Thus new tools, like next-generation vaccines and highly potent monoclonal antibodies (mAbs), the latter of which can provide prolonged, sterilizing immunity3,4,5, are needed for prevention and elimination of malaria.

PfCSP, the primary surface antigen of P. falciparum sporozoites, is a major target for vaccines and mAbs as it is both highly conserved and critical for the initiation of malaria infection. PfCSP contains an immunodominant central repeat region composed of repeating four amino-acid units, structurally defined as DPNA, NPNV, and NPNA6,7,8,9,10,11,12,13. These roughly define the junctional, minor repeat, and major repeat epitopes, respectively. Each epitope can generate potent antibodies that prevent malaria infection in animal models14,15,16, with the junctional mAb CIS43LS demonstrating high-level protection against controlled human malaria infection (CHMI) in humans3,5. Recently, we identified the minor repeat-specific mAb L9 as one of the most potent anti-PfCSP mAbs isolated to date17, which can also confer high-level sterilizing immunity against CHMI in humans4. Like many of the most potent NPNA-specific mAbs, L9 is encoded by the IGHV3-33/IGKV1-5 heavy/light chain gene combination. However, L9 is highly specific for the NPNV (minor) repeats and relies on critical contributions from the light chain for both NPNV selectivity and high potency8.

Here we used cryo-EM to understand the molecular basis of these unique functional properties. We demonstrate that L9 utilizes a distinct homotypic interface to stabilize multivalent Fab binding to the PfCSP minor repeats, and a unique paratope structure to selectively interact with NPNV repeats. In combination with MD simulations, these data indicate a key role in affinity-matured homotypic contacts in the L9 light chain for mediating high-affinity PfCSP binding and potent protection from malaria infection.

Results

L9 binds multivalently to the PfCSP minor repeats

For structure solution, a recombinant PfCSP construct was used, rsCSP, that contains the full N-terminal, junctional, minor repeat, and C-terminal regions, and about half the number of NPNA repeats as the 3D7 reference strain (Fig. 1a). To overcome both aggregation and preferred orientation of the L9 Fab-rsCSP complex in vitreous ice (see Methods), a large cryo-EM dataset was collected which resulted in a 3.36 Å resolution reconstruction (Fig. 1b; Supplementary Fig. 1; Supplementary Table 1). In the cryo-EM map, we observe three tightly packed Fabs bound to a central rsCSP, with each Fab simultaneously interacting with the peptide and the adjacent Fab via homotypic interactions10,13,18,19. In general, the complex is homogeneous and the density is well-resolved for each L9 variable region (Fv) as well as the rsCSP peptide (Fig. 1b). The structure of rsCSP, built de novo based on the EM density, consists solely of the minor repeat region (Fig. 1f). The modeled antigen sequence comprises 26 residues encompassing three complete NPNV and DPNA repeats, i.e., NA(NPNVDPNA)3; there is no additional density observed that would correspond to N-terminal, C-terminal, or major repeat regions. Moreover, we did not identify any 2D or 3D classes with more than three Fabs, indicating that any potential binding of L9 to the NPNA repeats was not stable enough to be captured by cryo-EM (Supplementary Fig. 2b); this is further supported by biolayer interferometry data showing rapid dissociation of L9 Fab to an NPNA-only peptide (NPNA8; Supplementary Fig. 2c, d). The L9 Fab and peptide cryo-EM structures correspond well with our recent X-ray structures of two chimeric precursors of L9 (L9K/F10H and F10K/L9H) in complex with a short minor repeat peptide (NANPNVDP)8 (Supplementary Fig. 3a–d). Relative to a representative Fv (Fab B) in the L9 cryo-EM structure, Cα RMSD values for both chimeric Fvs are ~0.5 Å, and ~0.1 Å when comparing only the PfCSP peptide encompassing the NPNV repeat. Within the L9-rsCSP complex, there is also a high degree of similarity between repeating components, with Cα RMSD values of ~0.5 Å between the three L9 Fvs, and 0.05–0.10 Å between the three NPNV epitopes on rsCSP (Supplementary Fig. 3e–g).

Fig. 1: Cryo-EM structure of the L9 Fab-rsCSP complex.
figure 1

a Schematic of protein sequence of full-length PfCSP and rsCSP (recombinant). Each box corresponds to a single repeat. The minor repeat region is in blue and green. b Cryo-EM map of L9-rsCSP at 3.36 Å. c Ribbon diagram of the atomic model; only the Fab variable region (Fv) was built into the density. d Rotated view of c. e Zoomed-in view of c, shown in a surface representation. f Model of the minor repeat peptide, colored as in a. NPNV type-1 β-turns are highlighted with a green circle. g Buried surface area on rsCSP, color-coded to the Fab with which each rsCSP residue interacts. h Alignment of the three NPNV motifs (left), or the three DPNA motifs aligned to the central NPNV motif (right). RMSD: root mean square deviation. Source data are provided as a Source Data file.

Full size image

L9 utilizes a distinct paratope structure to confer NPNV selectivity

In the cryo-EM structure, each L9 Fab primarily engages a single NPNV repeat, while the DPNA repeats are largely unbound and serve as a linker between each NPNV (Fig. 1e, g). This binding site model is strongly supported by our previous work demonstrating the high selectivity of L9 for NPNV over both DPNA and NPNA repeats8,17, and by the structural data itself, as alternate registrations of rsCSP produced substantially worse fits to the cryo-EM map (Supplementary Fig. 4). Thus, the full epitope bound by a single L9 Fab is NPNVD (Fig. 1g). Each NPNV motif adopts a type-1 β-turn, which is frequently observed for DPNA and NPNA motifs bound to anti-PfCSP antibodies from a variety of heavy and light chain lineages6,10,20. The DPNA repeats in the L9 structure, however, are more extended and lack clear secondary structure elements (Fig. 1f, h). The L9 epitope is centered on the NPNV type-1 β-turn, which resides in a deep, central pocket on the Fab formed primarily from CDRL1, CDRL3, and CDRH3, with smaller contributions from CDRH1 and H2 (Fig. 2a; Fig. 3a, b). Interestingly, overall buried surface area (BSA) on L9 is slightly biased toward the light chain (LC; L9K) (Fig. 3a, b). Of the 550 Å2 total BSA on a single L9 Fab, L9K contributes 294 Å2 (53.5%), while the heavy chain (HC; L9H) contributes 256 Å2 (46.5%), indicating a critical role of L9K in PfCSP binding.

Fig. 2: The L9 PfCSP epitope comprises NPNVD.
figure 2

a Surface representation of L9 Fab, with central NPNVD shown in gold. Heavy and light chain CDR loops are specified as H1, L1, etc. b Structural details of PfCSP binding pocket. Key interactions are highlighted with dashed lines. c Rotated view of b, zoomed-in from d. d Rotated view of a, shown in ribbon diagram. e Alignment of L9 Fab (magenta) with a panel of NPNA-specific IGHV3-33 Fabs; sequences in Fig. S7. Note that in the main text and in figures, we use the nomenclature L9K and L9H to refer to the light chain and heavy chain, respectively, as a whole. When referring to specific amino acids within either chain, we use the more general notation of, for example, W32L and W52H.

Full size image
Fig. 3: Structural details of the L9 paratope.
figure 3

a Buried surface area (BSA) contributions of individual residues to rsCSP binding in the L9 heavy chain. Sequence alignment to the IGHV3-33 germline gene shown below. b Same as in a, for the L9 light chain. c, d Structural details of NPNV binding. e NPNA2 epitope structure in the NPNA-specific mAb 2243 (PBD 6O23), highlighting the two key CH-π interactions of germline-encoded aromatic residues (W52H and Y94L) with the repeat prolines. f Same as in e, with X-ray structures of six NPNA-specific mAbs superimposed to highlight structural conservation. These six mAbs are shown in g. g Electrostatic surface potentials from L9 cryo-EM structure +/− peptide (upper left two panels) and X-ray structures of six other NPNA-specific mAbs bound to peptide; electrostatic potentials were calculated in PyMol61. The PDB accession codes are in parentheses. Kb: Boltzmann constant; T: temperature in kelvin; ce: electron charge in coulombs. Source data are provided as a Source Data file.

Full size image

As frequently observed in anti-NPNA major repeat mAbs, many direct antigen contacts are with germline-encoded aromatic residues, which in L9 create a hydrophobic cage surrounding the NPNV motif (Fig. 2b, Supplementary Table 2). In particular, W32L in CDRL1 stacks closely against the N-terminal Asn of the NPNV motif (N1) forming a CH-π bond, while Y94L in CDRL3 engages the repeat Pro (P2) (Fig. 2c, Fig. 3c, d). L9 also utilizes the strictly conserved IGHV3-33 germline residue W52H in CDRH2, which in all structures of IGHV3-33 mAbs solved to date forms a critical CH-π interaction with P6 of the second NPNA repeat in the NPNA2 epitope9,10,13,20. However, in L9, this role is assumed by Y94L, and W52H principally acts to stabilize the Y94L:P2 interaction through a π-π stacking interaction with the Y94L side chain (Fig. 3d–f).

This paratope structure is distinct from most other IGHV3-33 mAbs targeting both major and minor repeats. In L9, a repositioning of the HC and LC CDR3 loops, along with a rearrangement of W52H and CDRH2, creates a compact, central PfCSP binding pocket bounded by each of the HC and LC CDRs (Fig. 2b, e). A somatically mutated residue, R96L in CDRL3, is found at the base of the pocket and creates a highly basic cavity (Fig. 3g) that is nearly fully occupied by the N3 side chain, which forms key H-bonds with R96L (Fig. 2c). V4 occupies a hydrophobic cavity at the interface of CDRL1, L3, and CDRH3 (Fig. 3c), forming hydrophobic contacts with the side chain of Y97H. Notably, Y97H accounts for the greatest amount of BSA in the L9 paratope of any HC residue (Fig. 3a), and this largely stems from the interaction with V4, suggesting that this CDRH3 residue is critical for selectivity of L9 for NPNV over NPNA. Comparison of the paratope structure of L9 to a panel of NPNA-targeting IGHV3-33/IGKV1-5 mAb structures suggests that, in addition, the unique arrangement of the L9 CDR loops is unfavorable to NPNA binding in this conformation, as superimposition of these Fab-NPNA2 cryo-EM and X-ray structures onto the L9 Fab structure revealed extensive clashing between the peptide and the L9 CDRH1, CDRH3, and CDRL3 loops (Supplementary Fig. 5b, c).

Unique homotypic interactions stabilize multivalent PfCSP binding

Another unique property of L9 is the ability to “crosslink” two NPNV motifs within the minor repeat region of PfCSP, which improves binding affinity8. Our cryo-EM structure reveals that L9 achieves this through multivalent Fab binding to sequential NPNV repeats stabilized by an extensive antibody-antibody, or homotypic, interface between adjacent Fabs (Fig. 4a). Homotypic interactions have now been identified in several anti-NPNA mAbs and appear to be a characteristic feature of the IGHV3-33 antibody family10,13,18,19,20. Importantly, we demonstrate L9 as a non-NPNA-targeting anti-PfCSP mAb to also utilize homotypic interactions, suggesting that both the major and minor PfCSP repeats can facilitate their development.

Fig. 4: L9K mediates extensive homotypic interactions.
figure 4

a Ribbon diagram of Fab B (cyan) and C (maroon); side chains of interacting residues are shown. b–d Structural details of key homotypic interactions. Dashed lines indicate specific contacts. e Buried surface area (BSA) contributions of individual residues to the homotypic interface in L9 light chain. Sequence alignment with F10K and germline IGKV1-5 gene is shown below. f Same as in e, for L9 heavy chain, with sequence alignment to F10H and germline IGHV3-33 gene. Source data are provided as a Source Data file.

Full size image

The L9 homotypic interface is distinct from that observed in NPNA-specific IGHV3-33 mAbs, which is generally conserved and derives primarily from the heavy chain13,19 (Supplementary Fig. 7). In contrast, L9K contributes numerous critical homotypic contacts, and total BSA in the interface is evenly distributed between heavy and light chains (905 Å2 and 839 Å2, respectively) (Fig. 4e, f). In the cryo-EM structure, L9K of FabC packs tightly against L9H of Fab B, and extensive polar and hydrophobic contacts are made between CDRL1 and the LC framework region 3 of FabC (LFR3) with HFR1, CDRH1, and CDRH3 of Fab B (Fig. 4b–d; Supplementary Fig. 6; Supplementary Table 3). The homotypic interface between Fab B and Fab A is nearly identical. Importantly, several residues mediating critical homotypic interactions (Fig. 4b–d) correlate with somatic hypermutation of the germline IGHV3-33 and IGKV1-5 genes (Fig. 4e, f; Supplementary Fig. 7). Four somatically mutated residues in L9K, F28L and R31L in CDRL1, and E68L and H70L in LFR3, account for the majority of BSA contributed by the LC to the homotypic interface (Fig. 4e).

E68L lies at the core of the homotypic interface in L9, where it forms a key salt bridge with the germline-encoded R94H of CDRH3B (Fig. 4b; Supplementary Fig. 6). In L9H, R94H forms a conserved interaction with Y102H to stabilize the base of CDRH3; thus E68L may also indirectly impact antigen binding through stabilization of the CDRH3 loop in the adjacent Fab. F28L coordinates a series of π-π stacking interactions in the opposing CDRH1B (Y32H) and CDRH3B (F96H and F100cH) while also packing against the E68L side chain. This pi network culminates in a cation-π bond between R31L from CDRL1C and F100cH from the opposing CDRH3B (Fig. 4c). On the other side of the homotypic interface from E68L, a mutated framework residue H70L forms a hydrogen bond with the side chain of Q1H in Fab A in addition to multiple van der Waals contacts with CDRH1B (Fig. 4d). Each of these homotypic contacts is not encoded in the germline sequence, and none directly contact rsCSP (Fig. 3a, b). These findings provide strong evidence for affinity maturation to optimize antibody-antibody binding, which may, in turn, enhance PfCSP avidity and protective efficacy, as we have shown recently for multiple NPNA-specific IGHV3-33 mAbs13.

The four somatic mutations in L9K are atypical: F28K, E68K, and H70K are observed in <1% percent of all human IGKV1 light chain sequences, while R31K is observed in only 2% (Supplementary Fig. 7a)21. Strikingly, F28 and H70 also correspond to two of the five amino-acid differences between mature L9K and the light chain of a clonal relative and precursor of L9, F10K (S28 and D70 in F10K). Previously, we demonstrated the critical role of L9K in PfCSP binding and potency of protection by forming chimeric mAbs of L9 and F10, in which the light or heavy chain of L9 was paired with heavy or light chain of F10 (L9K/F10H and F10K/L9H). Specifically, we found key functional differences between L9 and the F10KL9H chimera: (1) reduced avidity to PfCSP minor repeats, (2) loss of the ability to bind two adjacent NPNV repeats, and (3) significantly reduced protection in vivo (p < 0.001)8. We also recently showed that mutation of residues mediating key homotypic interactions in a family of potent NPNA-specific IGHV3-33 mAbs caused similar functional effects as see in F10K chimera relative to L913. Thus, as F28 and H70 both mediate key homotypic interactions in L9, which would likely be lost in F10K, these residues may be key determinants in the minor repeat specificity and exceptional potency of L9.

Evolved homotypic contacts in L9K are critical for complex stability

To test this hypothesis, and to understand the role of homotypic contacts in L9K in general, we used molecular dynamics simulations to characterize WT L9 and a series of L9K variants. L9K residues were reverted to either the germline IGKV1-5 gene (R31S, E68G, H70E) or to the L9K precursor F10K (F28S, H70D). We first compared the free energy landscapes of the CDR loops of individual Fv domains unbound to rsCSP (Fig. 5; Supplementary Fig. 8). We find that the R31S, E68G, and H70D/E mutations in L9K result in a broader conformational space and additional highly probable minima compared to the WT L9 Fv, indicating that these residues are critical for determining the shape and the conformational flexibility of the paratope (Fig. 5b, c; Supplementary Fig. 8). These minima correspond to a substantial shift away from the binding competent conformation in combination with a higher conformational entropy, suggesting a decrease in stability and/or binding affinity (Fig. 5d). Importantly, when combined (R31S-E68G-H70D), MD simulations of the trimeric structure in complex with rsCSP predict that these mutations significantly destabilize the homotypic interface (Supplementary Table 4; p < 0.001), indicating their key role in mediating homotypic interactions. Interestingly, in the context of the trimeric complex, the H70D single mutant is predicted to stabilize homotypic interactions (Supplementary Table 4), suggesting the germline E70 or F10K D70 may have initialized the evolution of homotypic interactions during L9 maturation. Unlike other LC mutants, the F28S Fv reveals a similar conformational space and diversity in the CDR loops compared to the WT L9 Fv. However, simulations of F28S show the formation of a new intramolecular salt bridge between residues R31L and E68L, with simultaneous loss of the intermolecular salt bridge between E68L and R94H and the cation-π bond between R31L and F100cH (Fig. 5a). These results suggest that, in addition to direct homotypic interactions, F28 acts indirectly through E68L and R31L to further stabilize antibody-antibody binding. This is reflected in simulations that predict significantly decreased interaction energies of the homotypic interface in the F28S mutant relative to WT L9 (Supplementary Table 4); this is visualized in Movie S1.

Fig. 5: Molecular dynamics reveals L9K residues critical for stability of the homotypic interface and PfCSP binding.
figure 5

a Most populated structure for the F28S variant, highlighting the loss of critical homotypic interactions, which occurred in 74% of simulated structures. These contacts were maintained in 26% of F28S simulations. b, c Free energy landscapes of the L9 WT and the F28S/R31S/E68G/H70E variant projected in the same coordinate system, revealing a substantial increase in conformational space and a population shift due to the mutations. Cryo-EM structure is depicted as black diamond. k: Boltzmann constant; T: temperature. d Conformational ensemble representatives, state probabilities, and transition kinetics for the WT and the quadruple mutant, color-coded according to their dihedral entropy (blue-low flexibility, red-high variability). This mutant contains all four key homotypic contacts in the light chain mutated to the germline sequence (IGKV1-5). Note that H70 is E70 in fully germline IGKV1-5 sequence, and D70 in the L9 precursor F10K. J: joule; K: kelvin.

Full size image

To understand the molecular basis of key functional differences between L9 and F10, we next performed MD simulations of the F10 chimeras in the context of the trimeric Fab-rsCSP complex. Compared to WT L9 and L9K/F10H, simulations predict that the homotypic interface is strongly destabilized in the F10K/L9H chimera (Supplementary Table 4). This suggests that F10K/L9H would not bind multivalently to the minor repeats and would have overall reduced binding affinity, which is consistent with our previous functional data on this chimera8. Five residues differ between L9K and F10K: F28S, L33V, P40A, H70D, and E90Q (Fig. 4e). We find that the F28S mutation alone accounts for ~80% of the predicted destabilization of the homotypic interface observed with F10K/L9H compared to WT L9, while the H70D single mutant and the L33V-P40A-E90Q triple mutant Fvs are predicted to both slightly increase stability of the complex. Taken together, these data suggest that the dramatic destabilization seen in MD simulations of the F10K/L9H chimera is primarily the result of the F28S mutation. Therefore, this rare mutation in L9K (S28F), and the network of homotypic contacts it mediates, may underlie the key functional differences between L9 and F10K/L9H.

Discussion

This study reveals the structural basis for the extraordinary selectivity and binding affinity of L9 for the NPNV minor repeats and highlights the critical role of L9K for both functions. We find that rare, somatically mutated residues in L9K mediate extensive homotypic contacts between adjacent L9 Fabs and thus multivalent binding to adjacent NPNV motifs. These contacts underscore the requirement of at least two NPNV motifs for high affinity PfCSP binding by L9 (1000 nM vs 13 nM for peptides with one and two NPNV, respectively)8. Based on our recent finding that affinity-matured homotypic interactions in three potent NPNA-specific IGHV3-33 mAbs are critical for both high NPNA avidity and protective efficacy13, it is likely that L9K-mediated homotypic interactions are also critical for the potency of L9. Notably, these L9K residues (F28, R31, E68, H70) make no direct contacts with rsCSP (Fig. 3b; Supplementary Table 2), indicating that the minor repeat region facilitates antibody-antibody affinity maturation in the context of multiple adjacent NPNV motifs, as has been observed for extended NPNA repeats13,18,19.

L9 is one of the most potent anti-PfCSP mAbs identified to date and is currently undergoing clinical development as a monoclonal therapy for malaria prevention4. Thus, these structural data will be useful for rational antibody engineering to improve both the protective efficacy and pharmacokinetic properties of this mAb. The discovery of L9 and the NPNV minor repeat region as a highly protective epitope on PfCSP has led to new efforts to re-design PfCSP-based vaccines to elicit L9-like antibodies22,23. The cryo-EM structure presented here now enables a structure-based approach, which may be instrumental in developing the next-generation malaria vaccine. Future studies to identify related, NPNV-specific mAbs should enhance our understanding of this class of antibodies and their important contribution to protective immunity against malaria.

Methods

Protein production

L9 heavy and light chain sequences were synthesized and codon-optimized for mammalian expression and cloned into pHCMV3 by Genscript Inc. The full variable domain (VH/VL) and the light chain constant domain and heavy chain constant domain 1 (CL and CH1) were included in each construct (heavy chain residues 1–216; light chain residues 1–214). The recombinant Fab was expressed by transient transfection in Freestyle 239F cells (ThermoFisher, cat #R79007) grown in Freestyle 239 Medium without antibiotics (Gibco, cat #12338018); during expression, Fab was secreted into the medium due to the N-terminal signal peptide. Cells were pelleted seven days after transfection. The medium was then filtered and run over a HiTrap KappaSelect (Cytivia; cat #17545812) affinity column followed by cation exchange chromatography (Mono S, Cytivia cat #11001287). All purification steps were performed in TBS (pH 8.0). rsCSP, which contained a C-terminal 6x-His tag, was cloned into the pET28a plasmid and expressed in the Shuffle strain of E. coli (New England Biolabs; cat #C3026J). The rsCSP construct contains residues 26–159 and 240–383 of the 3D7 strain PfCSP protein sequence (UniProt Q7K740), and is identical to the 3D7 sequence over these positions. rsCSP was purified as previously described24. Briefly, E. coli SHUFFLE competent cells were transformed with the rsCSP-pET28a plasmid, and a single colony was picked for a 50 mL overnight starter culture grown in LB broth supplemented with 50 ug/mL kanamycin. Two 1 L cultures were inoculated the next day with 25 mL each of the overnight culture, and were grown at 37 °C in LB supplemented with 50 ug/mL kanamycin. When the optical density at 600 nm reached a value of 1, the cultures were induced with 1 mM isopropyl β-d-1-thiogalactopyranoside (Sigma; cat #16758) for 6 h. The cells then were harvested and lysed by microfluidization in PBS (pH 7.4). The lysate was incubated overnight with Ni cOmplete resin (Sigma; cat # 5893682001) and was eluted in PBS (pH 7.4) containing 200 mM imidazole.

Cryo-EM sample preparation

To form the L9 Fab-rsCSP complex, >10 fold molar excess of L9 Fab was incubated with rsCSP in tris-buffered saline (TBS; pH 8.0) overnight at 4 °C; this would theoretically allow for full Fab occupancy of the entire minor and major repeat regions on rsCSP. The complex was purified by size exclusion chromatography (SEC) with a Superdex 200 Increase 10/300 GL column (Sigma-Aldrich; cat #GE28-9909-44) equilibrated with TBS. All complex-containing fractions were pooled and used for structural studies. For initial cryo-EM attempts, the purified complex was concentrated to ~1 mg/mL with a 30 kDa molecular weight cutoff filter (MilliporeSigma; cat #MRCF0R030), and 3 μL of this solution was applied to holey gold UltrAufoil (Quantifoil) cryo-EM grids. Grids were blotted for two to four seconds at 100% humidity, 4 °C, and plunge-frozen with a Vitrobot Mark IV into liquid ethane. Due to extensive aggregation of the complex during vitrification, and concomitant preferred orientation in vitrified ice, cryo-EM data were later collected with L9-rsCSP captured onto graphene oxide (GO) grids. For GO grid preparation, the L9-rsCSP complex was diluted to ~0.05 mg/mL in TBS, and 3 μL was applied to holey gold UltrAufoil grids containing a non-uniform layer of GO sheets on top of the grid. GO grids were made in-house; the fabrication and preparation procedure was adapted from a published protocol25. Briefly, UltrAufoil 1.2/1.3 holey gold grids (300 mesh) were washed with chloroform and allowed to dry completely. Grids were then glow-discharged, and 4 μL of 1 mg/mL PEI solution (polyethylenimine HCl, 25 mM HEPES pH 7.9) was applied to the grid and incubated for 2 min. Excess PEI was blotted with filter paper. Grids were washed with 2 drops of milli-Q water and allowed to dry completely. GO sheets (Sigma-Aldrich; cat #763705) were diluted to 0.2 mg/mL in water and centrifuged at 1500 × g. 4 μL of the supernatant was applied to grids and incubated for 2 min. Excess GO solution was blotted off, and grids were washed two times with water. Grids were allowed to dry for at least 30 min before use, and were used for sample vitrification on the same day they were prepared. This procedure resulted in ~90% coverage of the holes with GO; about half of these holes contained a monolayer of GO. For vitrification, 3 μL of L9-rsCSP complex (0.05 mg/mL) was applied to GO grids, and the sample was blotted for 2 s at 100% humidity, 4 °C, and plunge-frozen in liquid ethane.

Cryo-EM data collection

Automated data acquisition was performed with the Leginon software26 (version 3.5) on a Titan Krios (ThermoFisher) operated at 300 keV. Micrograph movies were collected in electron counting mode with a K2 Summit direct electron detector (Gatan), with an unbinned pixel size of 1.045 Å and a defocus range of −0.9 μm to −2.0 μm. The dose rate was ~6 e−/Å2/sec, with a full exposure time of 10 s; 200 ms per movie frame. This resulted in a total dose of ~60 e−/Å2 on the specimen. A total of 12,521 movies were collected over four separate data sets. Due to preferred orientation of the L9-rsCSP complex on GO grids, two of these four data sets were collected with a stage tilt of −40°, with all other imaging parameters held constant. Movies and micrographs were cataloged and stored with the aid of Appion27.

Single particle cryo-EM data processing

Movie frames were aligned and dose-weighted with MotionCor228. Subsequent processing was performed with cryoSPARCv3.329. The contrast transfer function (CTF) was calculated with the Patch CTF Estimation tool, which was critical for accurate estimation of the tilted micrographs. The Gaussian (blob) picker was used on a subset of micrographs for initial particle picking, and 2D templates were generated with multiple rounds of 2D classification. Template picking was then used on the full dataset. Multiple rounds of 2D classification resulted in a particle stack containing 842,590 particles. A starting model generated from ab initio reconstruction was used for a non-uniform refinement job to achieve a resolution of ~3.7 Å. Multiple rounds of global CTF (beam tilt) refinement and per particle defocus refinement led to a 3.35 Å map. To account for possible flexibility between each of the three L9 Fabs, 3D Variability Analysis was used specifying four principal modes30. The output was fed into a 3D Variability Display job in cluster mode, specifying 20 clusters. Close inspection of the interactive cluster plots and structural comparison of cluster maps identified rotational flexibility in the Fab (Fab A) bound at the N-terminus of the peptide relative to the other two Fabs. The most homogeneous clusters were pooled, yielding a particle stack with 451,712 particles. These were again subjected to non-uniform and CTF refinement, leading to a 3.36 Å map with significantly improved interpretability of high-resolution features, particularly for the antigen (rsCSP) density.

To further improve the quality of the reconstruction of the rsCSP epitope and L9 Fab paratope structures, we masked the region surrounding the antigen for Local Refinement in CryoSPARC v4.0 using the following parameters: (1) a dilation radius of 5 and (2) a soft padding width of 10 (highlighted in orange in Supplemental Fig. 1g). All 842,509 particles that were used for the consensus refinement (Supplemental Fig. 1c), were used here. The Local refinement job generated a 3.34 Å map that enabled us to improve the overall density of the antigen and model interpretability, highlighted by the black circle in the second row.

Atomic model building

The X-ray structure of 239 Fab bound to NPNA2 (6W00), which contains matching germline heavy chain (IGHV3-33) and light chain (IGKV1-5) genes, was used to generate a homology model of L9 Fab. This model was then used as the template for re-building of the structure with RosettaCM31. At first, only the central Fab was modeled. On the resulting lowest energy model, the CDR loops were removed and built manually in Coot32. This structure was docked into the density of the two neighboring Fabs, and the trimeric Fab complex was refined with PHENIX real-space refine33 (v1.20.1). Based on the known preferred epitope of L9, and inspection of the cryo-EM density, the structure of the PfCSP minor repeat region was built manually in Coot. The L9 Fab-rsCSP complex was again refined with PHENIX and errors were iteratively corrected with Coot. Rosetta Relax was used for a final all-atom refinement34.

Structural analysis

BSA and root mean square deviation (RMSD) calculations were performed in UCSF Chimera35. For general structural interpretation, UCSF Chimera and Coot were used. Calculation of electrostatic potential surfaces was performed with PyMol (The PyMOL Molecular Graphics System, Version 2.0 Schrödinger, LLC). The Epitope Analyzer webtool was used to assess direct contacts within the homotypic interface and between L9 Fab and rsCSP36. Structure figures were made with UCSF Chimera, UCSF ChimeraX37, and PyMol.

Molecular dynamics simulations

Based on the cryo-EM structure of the WT L9 (this study), containing three Fvs bound to rsCSP, we performed five replicas each of 1 µs of classical molecular dynamics simulations of the complex to identify critical residues that stabilize/favor the homotypic interface. For the other investigated variants (Supplementary Table 4), we derived the starting structures for our simulations from the WT L9 structure by replacing the respective amino acids, followed by a local energy minimization in MOE (Molecular Operating Environment, Chemical Computing Group, version 2020.09). The starting structures for simulations were prepared in MOE using the Protonate3D tool38. To neutralize the charges, we used the uniform background charge, which is required to compute long-range electrostatic interactions39. Using the tleap tool of the AmberTools2040 package, the structures were soaked in cubic water boxes of TIP3P water molecules with a minimum wall distance of 12 Å to the protein41,42. For all simulations, parameters of the AMBER force field 14SB were used43. Molecular dynamics simulations were performed in an NpT ensemble using pmemd.cuda44. Bonds involving hydrogen atoms were restrained by applying the SHAKE algorithm45, allowing a time step of 2 fs. Atmospheric pressure of the system was preserved by weak coupling to an external bath using the Berendsen algorithm46. The Langevin thermostat was used to maintain the temperature during simulations at 300 K. The interaction energies were calculated with cpptraj by using the linear interaction energy (LIE) tool40. We calculated the electrostatic and van der Waals interaction energies for all frames of each simulation (10000 frames/simulation) and provided the simulation-averages of these interaction energies in Supplementary Table 4.

A previously published method characterizing the CDR loop ensembles in solution47 was used to investigate the conformational diversity of the six CDR loops of the free (apo) L9 Fv and the respective variants. To enhance the sampling of the conformational space, well-tempered bias-exchange metadynamics48,49 simulations were performed in GROMACS50,51 with the PLUMED 2 implementation52. We chose metadynamics as it enhances sampling on predefined collective variables (CV). The sampling is accelerated by a history-dependent bias potential, which is constructed in the space of the CVs53. As collective variables, we used a well-established protocol, boosting a linear combination of sine and cosine of the ψ torsion angles of all six CDR loops calculated with functions MATHEVAL and COMBINE implemented in PLUMED 247. As discussed previously, the ψ torsion angle captures conformational transitions comprehensively54. The underlying method presented in this paper has been validated in various studies against a large number of experimental results47,55. The simulations were performed at 300 K in an NpT ensemble using the GPU implementation of the pmemd module44 to be as close to the experimental conditions as possible and to obtain the correct density distributions of both protein and water. We used a Gaussian height of 10.0 kJ/mol and a width of 0.3 rad. Gaussian deposition occurred every 1000 steps and a biasfactor of 10 was used. 500 ns of bias-exchange metadynamics simulations were performed for the prepared Fv structures. The resulting trajectories were aligned to the whole Fv and clustered with cpptraj40 using the average linkage hierarchical clustering algorithm with a RMSD cutoff criterion of 1.2 Å resulting in a large number of clusters. The cluster representatives for the antibody fragments were equilibrated and simulated for 100 ns using the AMBER 20 simulation package. The accumulated simulation times for the investigated L9 variants are summarized in Table S5.

With the obtained trajectories, we performed a time-lagged independent component analysis (tICA) using the python library PyEMMA 2 employing a lag time of 10 ns. tICA was applied to identify the slowest movements of the investigated Fv fragments and consequently to obtain a kinetic discretization of the sampled conformational space56. tICA is a dimensionality reduction technique that detects the slowest-relaxing degrees of freedom and facilitates kinetic clustering, which is a crucial pre-requisite for building a Markov-state model. It linearly transforms a set of high-dimensional input coordinates to a set of output coordinates, by finding a subspace of “good reaction coordinates”. Thereby, tICA finds coordinates of maximal autocorrelation at a given lag time. The lag time sets a lower limit to the timescales considered in the tICA and the Markov-state model. Accordingly, tIC1 and tIC2 represent the two slowest degrees of freedom of the systems.

Based on the tICA conformational spaces, thermodynamics and kinetics were calculated with a Markov-state model (MSM)57 by using PyEMMA 2, which uses the k-means clustering algorithm to define microstates and the PCCA+ clustering algorithm58 to coarse-grain the microstates to macrostates. Markov-state models are network models which provide valuable insights for conformational states and transition probabilities between them, as it allows identification of the boundaries between two states57. Basically, MSMs coarse-grain the system’s dynamics, which reflect the free energy surface and ultimately determine the system’s structure and dynamics. Thus, MSMs provide important insights and enhance the understanding of states and transition probabilities and facilitates a quantitative connection with experimental data59.

The sampling efficiency and the reliability of the Markov-state model (e.g., defining optimal feature mappings) has been evaluated with the Chapman-Kolmogorov test by using the variational approach for Markov processes and monitoring the fraction of states used, since the network states must be fully connected to calculate probabilities of transitions and the relative equilibrium probabilities. To build the Markov-state model, we used the backbone torsions of the respective CDR loops, defined 100 microstates using the k-means clustering algorithm and applied a lag time of 10 ns.

Additionally, we calculated the residue-wise dihedral entropies with the recently published X-entropy python package, which calculates the entropy of a given dihedral angle distribution60. This approach uses a Gaussian kernel density estimation (KDE) with a plug-in bandwidth selection, which is fully implemented in C++ and parallelized with OpenMP. The obtained residue-wise dihedral entropies were projected onto the respective structures (Fig. 5d).

Biolayer interferometry (BLI)

To evaluate the binding of L9 to the major and minor PfCSP repeats, BLI experiments were performed using the Octet Red96 system (ForteBio). A basic kinetics experiment was used to measure interaction of L9 and 311 Fabs to NPNA8 (major repeat only) and rsCSP (major + minor repeats). 311 was used as a positive control for NPNA8 binding. Kinetics buffer (PBS + 0.01% BSA, 0.002% Tween-20, pH 7.4) was used for all dilutions, baseline measurements, and reference subtractions. Biotinylated NPNA8 or Twin-Strep tagged rsCSP was diluted to 5 μg/mL in kinetics buffer (KB) and immobilized onto Streptavidin BLI biosensors (Sartorius). Binding kinetics for each antibody were measured across a dilution series comprising the following concentrations of Fab (in nM): 6.25, 12.5, 25, 50, 100, 200. The steps of the kinetics experiment were as follows: baseline, 60 s (KB only), antigen loading, 600 s (KB + antigen), baseline 2, 60 s (KB only), association, 600 s (KB + antibody), dissociation, 1200 s (KB only). BLI data were processed with the ForteBio Data Analysis 9.0 software to evaluate kinetic parameters. In each case, global (full) fitting was performed with a 2:1 binding model, as there were at least two binding sites per peptide that are likely non-independent (4 sites for NPNA8, 11 sites for rsCSP), and a 1:1 kinetic model yielded substantially lower R2 values. Using a 2:1 kinetic model, two KD values are reported; for comparison across mAbs and peptides, an overall affinity to each peptide was calculated as an average of these two values, which were in turn averaged across at least 4 concentrations of Fab with an R2 of ≥0.95.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.