Structure of a Plasmodium falciparum PfEMP1 rosetting domain reveals a role for the N-terminal segment in heparin-mediated rosette inhibition

Abstract

The human malaria parasite Plasmodium falciparum can cause infected red blood cells (iRBC) to form rosettes with uninfected RBC, a phenotype associated with severe malaria. Rosetting is mediated by a subset of the Plasmodium falciparum membrane protein 1 (PfEMP1) variant adhesins expressed on the infected host-cell surface. Heparin and other sulfated oligosaccharides, however, can disrupt rosettes, suggesting that therapeutic approaches to this form of severe malaria are feasible. We present a structural and functional study of the N-terminal domain of PfEMP1 from the VarO variant comprising the N-terminal segment (NTS) and the first DBL domain (DBL1α₁), which is directly implicated in rosetting. We demonstrate that NTS-DBL1α₁-VarO binds to RBC and that heparin inhibits this interaction in a dose-dependent manner, thus mimicking heparin-mediated rosette disruption. We have determined the crystal structure of NTS-DBL1α₁, showing that NTS, previously thought to be a structurally independent component of PfEMP1, forms an integral part of the DBL1α domain. Using mutagenesis and docking studies, we have located the heparin-binding site, which includes NTS. NTS, unique to the DBL α-class domain, is thus an intrinsic structural and functional component of the N-terminal VarO domain. The specific interaction observed with heparin opens the way for developing antirosetting therapeutic strategies.

Severe Plasmodium falciparum malaria is frequently associated with infected red blood cells (iRBC) forming rosettes with uninfected RBC (1–3). Although the relationship between rosetting and malaria pathology is not thoroughly understood, enhanced invasion of RBC (4) and microvascular obstruction caused by high rosette densities (5) are probably major contributing factors. Rosetting is mediated by a subset of the P. falciparum erythrocyte membrane protein 1 (PfEMP1) adhesins (2, 6, 7), a protein family involved in falciparum cyto-adhesion. PfEMP1 adhesins are expressed on the surface of the iRBC during the trophozoite and schizont phases and are clustered in knob-like structures, where they interact with diverse host receptors (8).

PfEMP1 has a large N-terminal extracellular region comprising an N-terminal segment (NTS), and several Duffy binding-like (DBL) domains and cysteine-rich interdomain regions (CIDR) (8–10). DBL and CIDR domains can be assigned to a small number of sequence classes. The arrangement of these domains is modular, but most variants begin with NTS followed by DBL1α. Indeed, α-class DBL domains occur only at the N-terminal position of PfEMP1 (11). Studies on laboratory parasite strains have shown that the DBL1α₁ subclass is directly implicated in rosette formation (2, 7). In spite of a diversity of rosetting phenotypes, which interact with a variety of receptors, rosettes are frequently disrupted by sulfated glycosaminoglycans (GAG), such as heparin (6, 12–16). Heparin efficiently disrupts rosettes but may enhance CD36-dependent adhesion of iRBC to microvascular endothelium cells (17). The design of sulfated glycosaminoglycans or analogs for effective antirosetting treatment for malaria thus requires an understanding of the structural basis of their inhibition (12, 18).

We have shown that heparin disrupts Palo Alto 89F5 VarO rosettes and binds to recombinant NTS-DBL1α₁-VarO with μM affinity (16). The receptor for VarO rosetting has not yet been identified. Nonetheless, heparin and the chemically related GAG, heparan sulfate, can be excluded as candidates: Heparin is present only in the granules of connective tissue mast cells and heparan sulfate, while being present on the RBC surface and reported as the receptor for one rosetting phenotype (18), disrupts VarO rosettes very weakly (16). The efficacy of VarO rosette disruption by different oligosaccharides is directly correlated with their affinity for recombinant NTS-DBL1α₁, suggesting that the ligand competes with the rosetting receptor or alters the receptor-binding site. Here we show that recombinant NTS-DBL1α₁-VarO binds to the RBC surface and that heparin inhibits this interaction, consistent with a direct role for this domain in rosetting. We present, furthermore, the crystal structure of NTS-DBL1α₁-VarO determined at 2.06-Å resolution. The domain has the characteristic fold found in structures of other DBL domains (19–23); however, it possesses novel features unique to the α-class. Extensive contacts of the NTS moiety with the DBL1α₁ core demonstrate that NTS is an integral part of the N-terminal domain. Using site-directed mutagenesis and docking simulations, we identified regions critical for heparin binding, which, importantly, include residues from NTS. Thus, in addition to showing that the regions previously identified as NTS and DBL1α form a single structural and functional domain, the crystal structure provides an important basis for developing soluble rosetting inhibitors.

Results

RBC Binding by NTS-DBL1α₁ Is Inhibited by Heparin.

Binding of the recombinant NTS-DBL1α₁ domain to the RBC surface was demonstrated by agglutination (Fig. 1A) and immunofluoresence assays (IFA) (Fig. 1B) using specific antibodies, and confirmed by Western blot (Fig. 1C, lane 1). Binding of NTS-DBL1α₁ to RBC was inhibited by heparin in a dose-dependent manner (Fig. 1C, lanes 2–5). These experiments show that heparin binding to NTS-DBL1α₁ mimics rosette disruption and confirm data obtained with recombinant domains exposed on the surface of insect cells (7). The N-terminal domain of PfEMP1-VarO is thus the principal mediator of VarO rosetting.

Fig. 1.

Binding of NTS-DBL1α₁-VarO to RBC. (A) Nomarsky view and (B) immunofluorescence image of agglutinates formed upon binding of anti-NTS-DBL1α₁ antibodies to RBC surface-bound recombinant NTS-DBL1α₁. Anti-NTS-DBL1α₁ antibodies were revealed with Alexa Fluor-labeled antimouse antibodies. No agglutinate was formed and immunofluorescence was negative when no recombinant domain was present or in the absence of anti-NTS-DBL1α₁ antibodies. (C) Western blot analysis of RBC incubated with NTS-DBL1α₁-VarO in the presence of increasing concentrations of heparin and layered over silicone/nujol. Bound protein was detected using anti-NTS-DBL1α₁ antibodies followed by phosphatase-conjugated mouse antibodies. Lanes 1–5: 0, 0.05, 0.2, 0.5, and 1.0 mg mL^-1 heparin, respectively.

General Description of the NTS-DBL1α₁ Structure.

Crystals of NTS-DBL1α grew only upon addition of heparin in a 1∶1 mole ratio, the best being obtained with an oligosaccharide of mean molecular weight 5,000 daltons (approximately 16 sugar rings). However, no ordered electron density was observed to allow tracing of the ligand, a situation not uncommon for protein–heparin complexes (24). NTS remained associated with DBL1α₁ in spite of being cleaved after residue R69 (see Materials and Methods). The protein chain could be traced in the electron density from S12 to L52 (NTS), from D72 to F82 (DBL1α₁), and from G93 to S487 (DBL1α₁). DBL1α₁ is similar to other known DBL structures, both in secondary structure and the presence of canonical cystine bridges (19–23) (Figs. 2 and 3 and Table S1). Sequence comparison of NTS-DBL1α₁-VarO and all unique paralogous domains from the 3D7 parasite line is shown in Fig. S1.

Fig. 2.

Structure of NTS-DBL1α₁. (A and B) Two mutually orthogonal views are shown with NTS highlighted in mauve. (C and D) Detail of NTS and PEXEL regions: two mutually perpendicular views of the NTS region (mauve), showing the PEXEL sequence (RNVLE; green, with side chains), as well as the β1–β2 hairpin and helix αH7′ (both dark blue).

Fig. 3.

Sequence and secondary structure of NTS-DBL1α₁-VarO. Cylinders represent helices and arrows represent β strands. NTS is in mauve; subdomains 1, 2, and 3 are in light blue, green, and blue, respectively. Cysteines, given by their canonical nomenclature, are in yellow. K → A mutants are indicated by circles color-coded as follows: Mut1, mauve; Mut2, green; Mut3, purple; Mut4, dark blue; Mut5, brown; Mut6 yellow; Mut7, orange. In subdomains 1–3, helices up to αH8 without primes (’, ”) are common to other known DBL structures.

As with other DBL structures, DBL1α₁ can be divided into three subdomains (Fig. 3). The four-helix bundle of subdomain 2 (αH2, αH3, αH4, and αH5) and the three-helix bundle (αH6, αH7, and αH8) of subdomain 3 form the core signature observed in other DBL structures. An additional cystine bridge, canonical Cys(7b)-Cys(8a), not present in other DBL structures, is formed between C347 and C370 (Table S1). Recombinant NTS-DBL1α₁-VarO includes residues 1 to 487 and thus contains another two Cys residues, C477 and C483, which lie beyond canonical Cys(12) (C471) and form a disulfide bridge together.

Association of NTS with DBL1α₁-VarO.

NTS contains a 20-residue helix [αH(-2)] flanked by two short helices [αH(-3), αH(-1)] (Fig. 3). It makes extensive contacts with the DBL1α₁ core, leading to 4130 Å² of buried surface. NTS is held in a cleft where helices αH2, αH3, αH4′, αH5, αH6, αH7′, and αH8, together with the β1–β2 hairpin (F279-F289, beginning at the invariant triplet YFR) and a loop between C295 and αH6, contribute to direct contacts (Fig. 2 C and D). In particular, helix αH(-2) is held to the DBL1α₁ core by a pincer-like structure formed by the β1–β2 hairpin and αH7′, which are unique to the DBL1α fold. A total of 28 hydrogen bonds are formed between NTS and DBL1α₁.

Inclusion of NTS was essential for successful expression of DBL1α₁ in soluble form in the Escherichia coli and baculovirus/insect cell systems (16). Moreover, separation of NTS from the core domain after cleavage at R69 was only achieved under strong denaturing conditions (6M guanidine/HCl, 500 mM NaCl). We measured the stability of NTS-DBL1α₁ using circular dichroism (CD) to assess the importance of NTS for the structural integrity of the domain. The protein, cleaved at R69, was denatured with guanidine to remove NTS and then renatured (denoted DBL1α₁-Gnd). The uncleaved protein was treated in the same way as a control (denoted NTS-DBL1α₁-Gnd). We followed the change in molar ellipicity at 222 nm with temperature to compare the thermal stability of the two renatured proteins with respect to native, intact NTS-DBL1α₁ that was not treated with guanidine. The melting temperature of NTS-DBL1α₁-Gnd was indistinguishable from that of the native protein (T_m = 65.3 ± 0.3 °C and 65.7 ± 0.3 °C, respectively) (Fig. S2A). By contrast, T_m for DBL1α₁-Gnd was 47.9 ± 0.4 °C, indicating a much lower stability when NTS is absent. The far UV CD spectra measured for the three proteins before thermal denaturation indicated that all had a similar high α-helical content expected for DBL domains and were thus very similar in native conformation (Fig. S2B). Nonetheless, while the near UV CD spectra of NTS-DBL1α₁ and NTS-DBL1α₁-Gnd were indistinguishable from each other, that of DBL1α₁-Gnd was almost featureless, indicating a less well defined tertiary structure (Fig. S2C). These structural and biophysical considerations underline the importance of NTS for maintaining a stable structure and show that NTS is an integral part of the N-terminal domain of PfEMP1.

Binding of Heparin to NTS-DBL1α₁-VarO.

Because total methylation of primary amines in NTS-DBL1α₁ significantly reduced its affinity for heparin (16), we analyzed binding by surface plasmon resonance (SPR) techniques using a series of mutants where clusters of lysine residues (two to four per mutant, closely spaced on the surface) were replaced by alanine residues. Native conformation of the mutants was confirmed by comparing their near UV CD spectra with that of the native protein, and gel filtration profiles indicated that all mutants were monomeric in solution (Fig. S3 A and B). Seven mutants (Mut1 to Mut7, Table 1, Fig. 4) were prepared to give a good coverage of the protein surface. For Mut1 and Mut3, the affinity was reduced by a factor of approximately 100 and 200, respectively, compared to the wild type, while that of Mut4 was reduced by a factor of approximately 10; binding by the remaining mutants was indistinguishable from the wild type (Table 1 and Fig. S3C). The mutated residues on Mut1 and Mut3 form a long contiguous surface encompassing NTS (mutant Mut1), and helix αH7’ and the connecting loop between helices αH9 and αH10 (mutant Mut3) (Fig. 4). Mutations in Mut4 lie on helix αH4 located close by but they do not form a contiguous surface with the Mut1 and Mut3 mutations.

Table 1.

Heparin-binding constants for mutants

Mutant	Mutated residues	KD (μM)	KD/KD (nat)
Native		1.4	1
Mut1	K20, K32, K40	161	115
Mut2	K95, K166, K179	1.0	1
Mut3	K423, K424, K451, K456	305	218
Mut4	K226, K227, K230	16	11
Mut5	K404, K407, K409	1.3	1
Mut6	K117, K213	1.0	1
Mut7	K379, K393	1.1	1

Fig. 4.

Mutation analysis of heparin binding. Two mutually orthogonal views of the surface of NTS-DBL1α₁-VarO structure with mutated lysine residues highlighted. The two views are rotated with respect to each other by 180° about a vertical axis. The NTS is gray and the mutants are color-coded as in Fig. 3. The regions encompassing Mut1, Mut3, and Mut4 changes are circled.

Computer docking of heparin onto NTS-DBL1α was performed with models of oligosaccharides, ranging from a dimer to a dodecamer. Short oligosaccharides docked onto several regions of the protein surface, which has a high positive charge. Longer oligosaccharides, however, gave a more restricted set of solutions, suggesting a more specific interaction. Final calculations were made with a 12-mer, the longest oligomer for which accurate coordinates were available. Most of the 10 lowest energy solutions for the 12-mer docked the ligand to a region that included the mutated residues of Mut1 and Mut3, which displayed a more than 100-fold reduction in binding. The best solution docked the 12-mer onto a region exposed in a solvent channel of the crystal lattice and without steric clashes with symmetry-related domains. This channel contained diffuse electron density that was not sufficiently defined and continuous to allow modeling of heparin and had been mainly interpreted as solvent. Interestingly, a symmetry-related protein molecule faces the same channel with the surface that contains the Mut4 mutations, which also contacts the best heparin docking solution (Fig. 5). Other solvent channels do not provide sufficient volume to accommodate the ligand. Because heparin was cocrystallized with NTS-DBL1α₁ (and not soaked into the crystal), this further supports our conclusion that heparin binds primarily to a single high affinity site on the domain, because multiple strong binding sites would be incompatible with the crystal packing. Crystals grown with octa- and deca-saccharides were of poorer quality, and no crystals were obtained with tetra- and hexa-saccharides, consistent with the reduction in specificity for shorter oligosaccharides implied by docking calculations. Moreover, SPR experiments with heparins of different length showed a direct correlation between ligand size and binding strength (Fig. S4). These results parallel findings in another study on the effect of heparan sulfate on rosetting (18).

Fig. 5.

Lowest energy docking solution for the heparin 12-mer shown in the context of the crystal packing. Two protein domains, related by crystal symmetry, are in contact with a single docked heparin molecule. The domain with the larger contacting surface (cyan) includes mutations from Mut1 and Mut3; the domain with the smaller contacting surface (blue) includes Mut4 mutations. NTS and the mutated lysines are shown with the same color scheme as in Fig. 3.

The best docking solution contacts a positively charged strip on NTS-DBL1α₁ (Fig. S5A). This surface is formed by residues that are invariant or semiconserved with the NTS-DBL1α₁ domains of the two closely related IT4/R29 (2) and 3D7/PF13_0003 rosetting alleles (25) (Fig. S5B). Taken together, mutagenesis, docking calculations, and crystal packing arguments suggest that NTS-DBL1α₁ has a major binding site for heparin (Mut1 and Mut3) and probably not more than one additional site of lower affinity (Mut4).

Discussion

Structure of α-class DBL Domains.

The structure of the N-terminal domain of the rosetting variant PfEMP1-VarO shows that the DBL1α class presents the same fold found in other known DBL structures. It shows, however, that NTS is an integral component of DBL1α (we continue here to call it NTS-DBL1α₁) and is thus a unique feature distinguishing it from other DBL classes. The interface between NTS and DBL1α₁ is extensive and our biophysical studies demonstrate that NTS is crucial for preserving the stability of the domain. We note also that the var2CSA NTS-DBL1X domain aligns well with NTS-DBL1α sequences and thus should be similar in structure to the VarO domain.

The PEXEL-Like Motif in PfEMP1.

Parasite proteins exported across the parasitophorous vacuole membrane to different compartments of the RBC are thought to require a protein export element (PEXEL) with the consensus sequence RXLXE/Q/D (26, 27). Cleavage of the PEXEL between the third and fourth residues has been found in a number of soluble proteins, and N-terminal acylation of the product appears essential for correct trafficking of these proteins from the endoplasmic reticulum to their final destination (28). All PfEMP1 variants possess a PEXEL-like sequence in the NTS region, but it is not known if this motif plays the same role here. Unlike other exported parasite proteins, PfEMP1 has no signal sequence upstream from the PEXEL-like motif. In PfEMP1-VarO, residues 23 to 27 have the PEXEL-like sequence RNVLE. This is located at the N terminus of the NTS helix αH(-2) and is completely buried, particularly by contacts with helix αH7′ and the β1–β2 hairpin, which are not present in other DBL structures determined to date (except EBA-175 domain F2 for αH7′) (Fig. 2). With the tight association between NTS and the core DBL domain, the role of the PEXEL-like sequence in PfEMP1 cannot be similar to that of other exported parasite proteins because it is not consistent with our structural and biophysical data. If cleavage were to occur with the folded protein, the PEXEL-like motif would be inaccessible without significant structural change taking place. While cleavage could occur with the unfolded protein, loss of the N-terminal peptide would lead to an important reduction in stabilizing contacts between NTS and the core DBL1α domain.

Virulence: Cys2- and Cys4-type DBL1α Domains.

Several studies examining correlations between virulence and PfEMP1 sequences indicate that severe malaria is associated with DBL1α domains displaying particular patterns of conserved cysteines (29). These are contained within sequence motifs termed ‘positions of limited variability’ (PoLV) 1 to 4 located in subdomain 2 (30). Two major DBL1α groups are distinguished by the presence of either two or four conserved cysteines, and are referred to as Cys2 and Cys4 type domains, respectively. Cys2 types (such as DBL1α₁-VarO) correlate with severe malaria whereas Cys4 types, which are more frequent in the genome, are most often associated with mild malaria only. In Cys2-type domains, the two cysteines in the PoLV1-4 segment are canonical Cys(5a) and Cys(6). Cys4-type domains have two additional cysteines; one is five or six residues after the tyrosine in the conserved YFR motif (Y279 in DBL1α₁-VarO) while the other is two residues after canonical Cys(6); we denote these as Cys(5b) and Cys(6b) (Fig. S1). Cys(5b) and Cys(6b) align with VarO residues D284 and H297, respectively.

Examination of NTS-DBL1α sequences reveals additional patterns outside of the PoLV1-4 region. A highly conserved cysteine residue in the Cys4 type aligns with L162 in helix αH2 of NTS-DBL1α₁-VarO; we call this canonical Cys(4a) (Fig. S1). Most Cys4 domains thus have three additional cysteine residues compared to Cys2 domains: Cys(4a), Cys(5b), and Cys(6b). These contribute an odd number of sulfhydryl groups, raising the possibility of an additional disulfide bridge. Because VarO residue L162 is buried and is distant from residues equivalent to Cys(5b) and Cys(6b), it is unlikely that Cys(4a) participates in a disulfide bridge. On the other hand, VarO residue D284 [aligning with Cys(5b)] lies at the tip of the turn of the β1–β2 hairpin, while H297 [aligning with Cys(6b)] lies just beyond its base. A disulfide bridge between Cys(5b) and Cys(6b) could form but only if the conformation of this region differs from that in the crystal structure. Homology modelling of a Cys4-type DBL1α sequence (PFL1955w, 3D7 genome) suggested that the β1–β2 hairpin would bend away from the NTS helix αH(-2) if the Cys(5b)-Cys(6b) disulfide bridge were formed but would stay in the same configuration as in the VarO structure if not (Fig. S6). In the latter case, however, the free sulfhydryls would be highly accessible. No free sulfhydryls were found upon titration of recombinant NTS-DBL1α₁ from PFL1955w (25), consistent with a disulfide bridge being formed between Cys(5b) and Cys(6a), and Cys(4a) being buried. These considerations suggest that the correlation between virulence and DBL1α domain type might relate to the interaction between NTS and the DBL1α core. There are other sequence patterns distinguishing the Cys2 and Cys4 types that occur outside the PoLV1-4 regions. However, that may also play a role in malaria severity.

Heparin Binding and Rosette Disruption.

We show here that NTS-DBL1α₁-VarO binds to the RBC surface and that binding can be inhibited with increasing amounts of heparin, mirroring the capacity of this oligosaccharide to disrupt rosettes. These observations, and the efficacy with which domain-specific antibodies disrupt rosettes (7, 16), indicate a direct role for NTS-DBL1α₁ in this phenomenon (Fig. 1). Rosette disruption by sulfated oligosaccharides is frequently observed (31), implying that these ligands compete either directly with the rosetting receptor(s) or indirectly by modifying the receptor-binding site. We had previously found a direct correlation between the critical concentration for VarO rosette disruption by different sulfated oligosaccharides and the KD of these ligands for NTS-DBL1α₁ (16). A significant loss in heparin binding occurred after alkylation of the primary amines, indicating that interactions between lysines and the sulfate groups of heparin are important. Heparin binding might be mediated by nonspecific interactions between the negatively charged oligosaccharide and positively charged lysine and arginine residues, which are abundant on DBL1α domains (and other PfEMP1 DBL types). Our mutant binding studies, however, reveal an interaction confined to a region of the NTS-DBL1α₁ surface defined by mutants Mut1, Mut3, and Mut4 (Fig. 4 and Table 1), consistent with specific binding.

NTS, hitherto considered as a structurally independent component of PfEMP1 (8–11), is shown here to be an integral part of the N-terminal VarO domain and to contribute to the binding site for rosette-disrupting heparin (Mut1). If the binding site is common to other rosetting variants, it may open the way for novel therapeutic approaches. The lowest energy heparin docking solution contacts a region of the VarO surface that is semiconserved with respect to the N-terminal domains of two other closely related rosetting alleles, R29 (2) and PF13_0003 (25) (Fig. S5B), suggesting that heparin might bind similarly to these alleles.

The overall structure of NTS-DBL1α₁ is very close to that of other DBL domains, showing that no major conformational changes occur upon binding heparin. Direct binding competition between heparin and the rosetting receptor(s) is thus a possible mechanism for disruption. Heparin, however, could also cause the formation of PfEMP1 multimers on the iRBC surface that might no longer be competent for receptor binding. We were unable to examine oligomerization of recombinant NTS-DBL1α₁ by heparin in solution because the complex precipitated, which itself is consistent with multimerization of surface PfEMP1. We could only obtain crystals of the NTS-DBL1α₁ domain in the presence of heparin, and our analysis of crystal packing suggests that PfEMP1 multimerization could also be a mechanism for VarO rosette disruption (see Fig. 5).

The capacity of heparin and similar sulfated oligosaccharides to disrupt rosettes must be viewed together with the large diversity of this P. falciparum phenotype. Rosette stability ranges from loose to tight associations mediated by a variety of reported receptors (31). Indeed, the host molecule(s) mediating VarO rosetting is (are) currently unknown. NTS, a unique feature of α-class domains, nonetheless has a direct role in heparin binding to NTS-DBL1α₁-VarO. Heparin is heterogeneous in both the degree and position of sulfation on the sugar rings. Previous studies have shown that N sulfation is important for rosette disruption while O sulfation favors binding to certain critical host molecules such as antithrombin (18). With appropriate design, rosette inhibitors might thus preferentially target the parasite without producing undesirable side effects such as inhibition of coagulation (12, 18) or enhanced adhesion of iRBC in CD36-dependent sequestration (17). The participation of NTS, which is unique to DBL1α domains, in binding of heparin suggests that the design of specific rosetting inhibitors devoid of affinity for other domains and of anticoagulant activity might be possible. The structure of NTS-DBL1α₁ provides an important basis for further understanding rosetting and for the design of novel therapeutic approaches against severe falciparum malaria.

Materials and Methods

Production of NTS-DBL1α in E. coli.

NTS-DBL1α₁-VarO was produced as previously described (16). A factor Xa site was introduced between the NTS and DBL1α₁ by replacing residues G66-YV-R69 by the IEGR sequence, using standard splicing by overlap extension PCR. K → A mutants and mutations introducing three additional methionine residues for SeMet phasing (I188 → M, L268 → M and I402 → M) were produced using PCR mutagenesis kits (Quikchange Lightning or Multi Site-Directed Mutagenesis kits, Stratagene). The SeMet protein was expressed in E. coli, Rosetta Gami strain, using a modified M9 medium supplemented with SeMet (32) and purified as the WT. Production of DBL2β-VarO and NTS-DBL1α-PFL1955w have been described (7, 25).

Binding of NTS-DBL1α to RBC.

Fresh A⁺ RBC (2 × 10⁷ cells collected by finger-prick, washed twice with PBS) were incubated at RT for 60 min with 25 pmoles NTS-DBL1α in 100 μL (final volume) RPMI/10% AB⁺ human serum with or without heparin (0.05–1 mg mL^-1). RBCs were separated from the incubation medium through 200 μL 85% silicone DC550 (Serva)/15% Nujol (Alfa Aesar) by centrifugation at 3,000 g. Bound protein was eluted in 30 μL Laemmli sample buffer and 15 μL was loaded on a 12% SDS/PAGE and detected by Western blot using mouse anti-NTS-DBL1α serum (diluted 1/500).

Agglutination and IFA were performed on RBC incubated with NTS-DBL1α₁ as above. RBC were washed with 2% fetal calf serum in PBS, centrifuged (2 min /1,000 g), incubated with anti-NTS-DBL1α₁ serum (1/500), then washed and incubated with Alexa Fluor 488-conjugated goat antimouse antibodies (Molecular Probes) at 1/1,000 dilution for 30 mn. After washing and centrifugation, cells were observed with a Leica DM 5000 B microscope.

Circular Dichroism.

CD spectra were measured as previously described (16) to verify that all mutants were expressed with the native protein fold (Fig. S3A). Thermal denaturation was recorded at 222 nm from 10 °C to 99 °C using a temperature increase of 1 °C for each step.

Crystallization, Diffraction Data and Structure Determination.

Protein (10 mg mL^-1) was mixed with heparin 5000 (0.7 mg mL^-1, Sigma-Aldrich) in 250 mM NaCl, 20 mM Tris pH 8. Crystals were grown by the hanging-drop method using a 1∶1 dilution with the reservoir solution (10% PEG 3350, 0.2 M L-proline, 100 mM HEPES pH7.5). Crystals of selenomethionine-labeled protein grew in the same conditions.

Native data were collected on Proxima1 (SOLEIL, Ile de France). SeMet MAD data were collected on ID23-1 (ESRF, Grenoble) (Table S2). Data were treated with XDS (33) and the CCP4 program suite (34). The structure was determined by Se MAD phasing to 2.6-Å resolution using AutoSHARP (35) and refined with autoBUSTER (Global Phasing Ltd.) (Table 2). There are no outliers in the Ramachandran plot of the refined structure (97.3% in most favored region, 2.7% in allowed regions). Structural figures were prepared with CCP4mg (36).

Table 2.

Refinement statistics

Resolution (Å)	28.7-2.06
No. reflections	35007
Rwork/Rfree (%)	18.18/21.61
No. atoms
Protein	3690
Ligand/ion	33
Water	269
B-factors (Å2)
Protein	49.6
Ligand/ion	60.4
Water	52.1
R.m.s deviations
Bond lengths (Å)	0.010
Bond angles (°)	1.06

Heparin Binding Studies.

Interaction between NTS-DBL1α₁ and heparin (gift from H. Lortat-Jacob) was studied by SPR as previously described (16). Experiments were performed in 20 mM Tris at pH 7.4, 150 mM NaCl, 0.005% (w/v) surfactant P20. A total of 50–70 RU of biotinylated heparin was captured on the sensor chip. Association between soluble NTS-DBL1α₁ (concentration 0.78–100 μM) and the immobilized glycan was measured at 25 °C and a flow rate of 50 μL min^-1, corrected by control curves from a blank channel. Affinity constants were calculated from the variation of the steady-state binding amplitudes with protein concentration as the 1∶1 kinetic binding model did not apply (BIAevaluation 3.1 software, Biacore AB).

Heparin Docking and Homology Modelling.

Computer docking of heparin onto the NTS-DBL1α₁-VarO structure was performed using AutoDock version 4.2 (37). Coordinates for heparin were taken from heparin complexes in the Protein Data Bank (ID codes 1bfb, 1e0o, and 1hpn). The search was performed over the entire protein surface. Default values were used for all docking parameters, except the number of search runs (used 100). The 10 best solutions were retained. Further docking runs used the dodecamer structure, with rotation about glycosidic bonds between dimers and all N-sulfate and C-sulfate bonds (32 torsions; the limit in AutoDock) and with a 0.74-Å grid. Two equal grid boxes were chosen so as to cover half the volume needed for each run and 100 search runs were performed for each grid position. The 10 best solutions were retained.

NTS-DBL1α of PFL1955w was modeled by homology using Modeller version 9.7 (38). The first run used the automatic protocol, with a manually adjusted alignment, over the entire sequence. The second run was a loop refinement starting from the lowest-energy result, using different loop lengths ranging from 18 to 32 residues and a disulfide bridge between cysteines Cys(5b) and Cys(6b) either imposed or not.