Our results identify the ABO blood group as the major VarO rosetting receptor on the host RBC and show that the presence of the CIDR1γ domain to form the Head region results in enhanced RBC binding, mimicking the ABO blood group preference of VarO rosetting. The crystal structure of the Head (DBL1α1-CIDR1γ) allowed mapping of the RBC-binding site to a structurally conserved region of rosette-forming PfEMP1 variants, involving residues from subdomains 1 and 2, with contributions from the neighbouring NTS-DBL1α1 hinge region. The RBC-binding site is distal to the heparin-binding site, indicating that although heparin prevents binding of the adhesion domain to the RBCs, it does not directly compete with the ligand-receptor interaction to prevent rosette formation.
Previous analyses of interactions at play in VarO rosetting showed that none of the common receptors of P. falciparum
cytoadherence, such as CD36, ICAM-1, CSA, nor other potential receptors (VCAM-1, HABP1, CD31/PECAM, E-selectin, Endoglin, CHO receptor “X”, and Fractalkine) were implicated in the binding of RBC to VarO-iRBCs 
. We show here that CR1/CD35, shown to be a receptor for some rosetting lines (including R29 
), is not involved in VarO rosetting. VarO rosetting shares with other rosetting lines three generic characteristics, namely an extreme sensitivity to sulphated glycosaminoglycans, the need for human serum and a marked ABO blood group preference characterised by reduced binding to group O RBCs. Our data indicate that the major determinant affecting VarO rosetting efficiency is indeed the ABO blood group. We explored VarO-iRBC binding characteristics using a monovariant culture of the Palo Alto 89F5 clone, in which >90% of the iRBCs were positively selected to express PfEMP1-VarO 
. VarO-iRBCs preferentially bind to blood group A compared to blood group B, which itself is preferred to blood group O. An identical blood group preference profile was observed with both the DBL1α1
and Head proteins.
Our dissection of blood group A preference using the recombinant domains provides, for the first time, a link between rosetting and common group A polymorphisms. The difference between the A1
subgroups is mainly quantitative, with A2
RBC displaying 4–5 times fewer blood group A determinants than A1
, although they have some qualitative differences as well 
. Our observation that DBL1α1
hardly bound Ax
RBCs is a strong indication that copy number variation of the terminal glycan is the main cause of differences in the binding behaviour between the three subgroups. Binding to A2
was similar in intensity to binding to blood group B. As the A2
RBCs expressed about 5-fold more H antigen than the B RBCs (Figure S6
), we conclude that the terminal α-1,3-linked N-acetylgalactosamine (GalNAc) of group A is preferred to the terminal galactose (Gal) of group B. This was further documented by surface plasmon resonance assays, in which we explored binding to trisaccharide-BSA conjugates immobilized on the surface of a sensor chip. Binding to trisacharide A was more efficient than to trisaccharide B (), with apparent affinities in the micromolar range. This blood group preference is fully consistent with the VarO-IRBCs blood group preference although it should be mentioned that BSA-conjugates sub-optimally mimic the RBC-displayed blood group saccharides, which present multiple branched saccharides or tandem copies of the blood group determinant.
The structure of the DBL1α1
-CIDR1γ Head reported here is the first crystal structure of a multiple domain from PfEMP1, as well as of a complete CIDR1γ domain. The structure of the DBL1α1
domain published earlier 
is confirmed and we further show that the CIDR1γ-VarO domain forms extensive contacts with the DBL1α1
region (). The interface between the DBL1α1
and CIDR1γ domains is stabilised by several charged interactions (D367-D654, N381-R518, E389-Y521, Y437-K550) and, furthermore, a significant hydrophobic surface on DBL1α1
is buried within this interface. A significant part of the CIDR1γ accessible surface (1367A2
of a total of 12477A2
) is buried within the interface, including residues from its N-terminal half, as well as from the a–b–c
helical motif (). The interface is also highly conserved within the DBL1α1
-CIDR1γ family of PfEMP1 adhesins ().
Interface between the DBL1α1 and CIDR1γ VarO domains.
The three-dimensional structure of the Head allows a definitive assignment of interdomain boundaries. Thus the boundary between DBL1α1
and CIDR1γ domains lies within a rather flexible linker around residue 490, indicating that the DBL1α1
domain terminates after the disulphide Cys13–Cys14 (residues C477–C483of PfEMP1-VarO sequence). The CIDR1γ domain is more compact than the partial CIDRα structure published by 
, not only in our double-domain crystal structure but also in solution as a single domain ( and S10
). This difference could be due to a number of reasons: CIDR1γ -VarOcontains additional 98 residues at the N-terminus missing in the MC179 construct (which also lacked the preceding DBL1α domain) 
, the extensive contacts between DBL1α1
and CIDR1γ, or the dimer formation in the case of MC179. The two CIDR domains are of different sequence classes, γ for VarO and α for MC179. Moerover, the latter binds CD36, which is not the case for VarO 
. Both CIDR classes share a conserved arrangement of Cys and Trp residues in most of their sequence, with an additional disulphide (canonical Cys(8a)–Cys(8b)) in the γ class. The sequences are, however, much less well conserved in loop regions connecting both the helices and the subdomains, which could be another reason for the structural differences observed.
The similarity between DBL and CIDR domains in their general architecture, noted earlier 
, is also present in the VarO structure. Superposition of the CIDR1γ -VarO domain upon the DBL1α1
-VarO domain matches not only helices H1, H2 and H3 (CIDR1 γ) to αH6, αH7 and αH10 (DBL1α1
), but also helices a
(CIDR1γ) to αH8 and αH9 (DBL1α1
). Furthermore, strands β1 and β2 of CIDR1γ lie quite close to strands β-1 and β-2 of DBL1α1
). The similarity extends to the disulfide pattern as well, as suggested earlier 
: canonical disulfides Cys(10)–Cys(11) and Cys(7)–Cys(9) of DBL1α1
overlap with disulfides Cys(7)–Cys(9) and Cys(4)–Cys(6) of CIDR1γ, respectively, while disulfide Cys(5)–Cys(10) of CIDR1γ lies close to Cys(8)–Cys(12) of DBL1α1
. Indeed, the N-terminal half of CIDR could be classified as equivalent to subdomain 1 of DBL domains, while the second, helical domain of CIDR corresponds more closely to subdomain 3.
In the MC179 structure, the region implicated in CD36 binding 
lies near the N-terminal end of helix b
. In the CIDR1γ -VarO structure, the equivalent region corresponds to the b–c
connecting loop, which faces the H1–H2–H3 helical bundle. Within the triplet S662-I663-D664 of VarO, corresponding to the critical residues E108-I109-K110 of MC179, S662 and I663 are buried by H1 and D664 forms a salt bridge with K591 from H1. Interestingly, S662 and I663 are highly conserved among CIDRγ sequences, suggesting that this loop may have a common conformation in this domain class. If these residues were implicated in CD36 binding, they would be poorly accessible in the rosetting strains of PfEMP1, which do not bind CD36 
The structure of the Head provides critical information about the RBC-binding site. Computer docking and site-directed mutagenesis localized a blood group A binding site in a restricted area situated at the interface of subdomain1 and subdomain 2 in the vicinity of the NTS-DBL1α1
hinge region. This differs from the CSA-binding site localized on VAR2CSA DBL3X domain, which lies within subdomain 3 
, and is more in line with the location reported for P. knowlesi
Duffy Binding Protein (also a site engaging residues from subdomains 1 and 2) 
or some of the sialic acid-binding sites of P. falciparum
. The NTS-DBL1α1
hinge region, missing in our previous single domain structure, is highly exposed on the surface and proved to be crucial for the RBC-binding site. Indeed, cleavage of this sequence disrupted binding and mutations of this region reduced binding, without substantially affecting antigenicity (recognition of all mutants by ELISA was essentially unimpaired, data not shown). This reinforces the conclusion that NTS is an essential functional and structural component of the DBL1α domain 
. Importantly, the blood group A binding site and the major heparin-binding site that we mapped previously 
are distant from each other on the surface, indeed on opposite sides of the molecule. Therefore, direct competition with binding to the receptor cannot be the reason why heparin disrupts rosettes and inhibits the binding of the recombinant domains to RBC and trisaccharide-BSA conjugates, contrary to one of the previously suggested hypotheses 
. The other possible mechanism, namely that heparin could provoke the formation of oligomers that are no longer competent for receptor binding, remains an interesting possibility. The RBC-binding site could become inaccessible in heparin-aggregated adhesins, or several PfEMP1 molecules need to bind simultaneously to the ABO antigens displayed on the RBC surface for an efficient interaction to occur and this is prevented in the heparin complexes.
We analysed the location of the RBC-binding site on the DBL1α1
domain with respect to the position of molecular signature tags used to classify var
genes and associate them with either severe or uncomplicated malaria. Conserved tags, called positions of limited variability 1 to 4 (PolV1-4), had been identified 
. The relative combination of PoLV motifs appears characteristic of specific var
gene subsets. and show the localization of the four PolV sequence tags with respect to the identified RBC binding site of PfEMP1-VarO. All PolV tags, except for PolV1, are remote from the RBC-binding site (). Normark et al. 
identified specific PfEMP1-DBL1αamino acid motifs correlated with rosetting and severe malaria.One of the sequence signatures associated with “high rosetting”, namely H3, maps close to the binding site identified here. Palo Alto 89F5 VarO has a H3 motif (H3 K D K/A V E/Q K G) located at the beginning of αH4, which includes K216, a residue critical for RBC binding ( and ). This motif is surface-exposed and located in close proximity of the RBC-binding site.
Localisation of the PolV1-4 and High rosetting motif H3 relative to the RBC binding site on DBL1α1.
The RBC surface displays several million copies of ABO blood group determinants carried on membrane glycoproteins and glycolipids. The ABH antigens lie on terminal branches of poly-N-acetylgalactosamines, each of which may carry several ABH determinants. Although the type of branching varies, the ABH determinants displayed on the RBC surface are very dense. It is possible that PfEMP1 binding involves interaction with more than one glycan per Head region.Furthermore, although both DBL1α1
domain and the Head region are monomeric in solution, we do not know whether binding is associated with oligomerization of the adhesion domain, as reported for other RBC-binding proteins with DBL domains such as P. falciparum
and the P. vivax
Duffy Binding Protein 
-CIDR1γ Head is present in a small subset of var
genes from group A, four of which are implicated in rosetting 
. The RBC-binding site is conserved in other rosette-forming PfEMP1 variants such as R29, PF13_003 and IT-var60, indicating that data obtained here can be extrapolated to other lines and form the molecular basis of the extensively documented ABO blood group preference in rosetting 
. The presence of CIDR1γ increases binding efficiency, as indicated by the approximately 1 log unit higher MFI in flow cytometry and the increased amount of protein bound, as visualised by immunoblotting. The exact role played by CIDR1γ, however, is still unclear. It is possible that its folding back upon the DBL1α1
domain provides a structural framework for more efficient binding and increased affinity, just as the multimodular PfEMP1-VAR2CSA has been shown to require a compact fold for activity 
. The binding characteristics of the Head region resemble those of the infected red cells, except for the susbtantial residual binding in the absence of human serum and the similar enhancement by human and foetal calf serum. Serum enhancement of NTS-DBL1α1
andHead binding to RBCs may reflect a need to buffer the highly negatively charged RBC surface. As the serum component(s) implicated in VarO rosetting and Head region binding are unknown, we carried out all binding assays in the presence of human serum. VarO rosetting has an absolute requirement for human serum (Figure S1A
) that cannot be replaced by foetal calf serum. As we show that binding of NTS-DBL1α1
and/orHead does not account for this human-specific serum dependency, we suppose that interaction of serum components with downstream PfEMP1-VarO domains might contribute to modulate binding of PfEMP1-VarO, possibly by increasing affinity and optimising binding characteristics, or that other RBC surface proteins (eg. rifins or stevors) come into play in rosetting as well. Further work is needed to clarify this question.
This work provides the molecular basis underpinning the blood group preference of rosetting. The association between the ABO groups, rosetting and severe malaria 
is a strong indication that rosetting, as a contributor to severe malaria, has exerted a selective pressure that has shaped population polymorphisms at the ABO locus and has contributed to their varying geographic distribution. The data reported here expand this framework to subgroups within the susceptible blood group A. Although the genetic basis of the A1
and other rare A subgroups is well established, the physiological consequences of such phenotypes and the selective advantage they provide are unclear. The lower prevalence of A1
blood group in populations of African descent compared to populations of Asian or Caucasian origin 
is consistent with the hypothesis that P. falciparum
rosetting has contributed to subgroup selection and gene spread of blood group A variants.