Rosetting revisited: a critical look at the evidence for host erythrocyte receptors in Plasmodium falciparum rosetting

Malaria remains a major cause of mortality in African children, with no adjunctive treatments currently available to ameliorate the severe clinical forms of the disease. Rosetting, the adhesion of infected erythrocytes (IEs) to uninfected erythrocytes, is a parasite phenotype strongly associated with severe malaria, and hence is a potential therapeutic target. However, the molecular mechanisms of rosetting are complex and involve multiple distinct receptor–ligand interactions, with some similarities to the diverse pathways involved in P. falciparum erythrocyte invasion. This review summarizes the current understanding of the molecular interactions that lead to rosette formation, with a particular focus on host uninfected erythrocyte receptors including the A and B blood group trisaccharides, complement receptor one, heparan sulphate, glycophorin A and glycophorin C. There is strong evidence supporting blood group A trisaccharides as rosetting receptors, but evidence for other molecules is incomplete and requires further study. It is likely that additional host erythrocyte rosetting receptors remain to be discovered. A rosette-disrupting low anti-coagulant heparin derivative is being investigated as an adjunctive therapy for severe malaria, and further research into the receptor–ligand interactions underlying rosetting may reveal additional therapeutic approaches to reduce the unacceptably high mortality rate of severe malaria.


Introduction
Rosetting is a Plasmodium falciparum infected erythrocyte (IE) adhesion phenotype that is associated with severe malaria in sub-Saharan Africa (summarized in Doumbo et al., 2009). It is a form of cell adhesion in which erythrocytes infected with mature, asexual parasites bind to uninfected erythrocytes to form clusters of cells (Fig. 1). Rosetting is a phenotypically variable property, which is common in parasite isolates collected from severe malaria patients, but infrequent in parasites from uncomplicated malaria cases. For culture-adapted P. falciparum isolates, only a subset of parasite lines can be selected in vitro for the rosetting phenotype, and many of the commonly used laboratory strains such as 3D7, rosette poorly or not at all. The relative rarity of rosetting in culture-adapted parasite lines may explain why rosetting is studied infrequently, despite being a virulence-associated phenotype in clinical isolates.
Rosetting can contribute to IE sequestration and microvascular congestion, leading to obstruction to blood flow (Kaul et al., 1991), one of the major pathological events in severe falciparum malaria contributing to inflammation, tissue damage and organ failure (Miller et al., 2002;White et al., 2013). Rosetting also causes membrane changes in uninfected erythrocytes that may contribute to phagocytic removal and anaemia . In Africa, high levels of rosetting occur in parasites sampled from severe malaria patients with all clinical types of disease including cerebral malaria (Carlson et al., 1990;Treutiger et al., 1992;Ringwald et al., 1993;Rowe et al., 1995;Kun et al., 1998;Doumbo et al., 2009), severe malarial anaemia Doumbo et al., 2009) and respiratory distress (Warimwe et al., 2012). Rosette-like clusters of cells have been seen in the microvasculature in histological studies of fatal malaria cases (Dondorp et al., 2004;Barrera et al., 2018). The major Plasmodium species that infect humans are all able to form rosettes (Udomsanpetch et al., 1995;Angus et al., 1996;Chotivanich et al., 1998;Lowe et al., 1998). However, the link between severity of disease and rosetting is confined to P. falciparum, possibly due to the unique ability to bind both endothelial cells and uninfected erythrocytes simultaneously (Udomsangpetch et al., 1992;Adams et al., 2014), such that P. falciparum rosetting IEs are sequestered and are not seen in peripheral blood. Recently it has been suggested that rosetting may contribute to anaemia in Plasmodium vivax infections (Marín-Menéndez et al., 2013).
The biological function of rosetting in vivo remains unknown. Rosettes may shield IEs from host immune attack, or close contact with uninfected erythrocytes in rosettes might enhance merozoite invasion Deans and Rowe, 2006). However, firm evidence to support either of these hypotheses is lacking. Most rosetting parasite isolates form larger, stronger rosettes with blood group A erythrocytes compared to other blood groups , and these group A rosettes may shield IEs to reduce antibody binding to parasite variant surface antigens (VSAs) (Moll et al., 2015). Whether this translates into the reduced clearance of IEs and subsequent higher parasite burdens in vivo is unclear, although some studies have noted a positive correlation between rosetting and parasitaemia (Rowe et al., 2002). Another study showed that rosetting does not prevent IgG-mediated phagocytosis of IEs (Stevenson et al., 2015a), although experiments were only performed in group O cells. Parasite invasion of erythrocytes is not increased in vitro in rosetting compared to isogenic non-rosetting parasites (Clough et al., 1998;Deans and Rowe, 2006;Ribacke et al., 2013), nor in the presence of larger rosettes (Moll et al., 2015). However, in vivo studies using splenectomized Saimiri sciureus monkeys demonstrated a 1.5 times higher parasite multiplication rate with rosetting compared to isogenic non-rosetting parasites . This suggests either increased invasion or decreased clearance of rosetting parasites in vivo, which requires further investigation.
This review will discuss the molecular mechanisms of rosetting and describe recent advances exploring the potential of rosetting as a therapeutic target in severe P. falciparum malaria. Rosetting is a complex cell adhesion phenotype involving parasite adhesion molecules on the IE surface and host receptors on uninfected erythrocytes (Fig. 2). Current evidence suggests that there are multiple distinct pathways of rosette formation, similar to the diverse pathways involved in merozoite invasion of erythrocytes (Cowman et al., 2017). Interestingly, although the parasite molecules that mediate rosetting are different from those involved in merozoite invasion, both sets of proteins have 'Duffy-Binding-Like' adhesion domains and many of the same host erythrocyte receptors are used (e.g. glycophorin A, glycophorin C and complement receptor one). The diversity in P. falciparum merozoite invasion pathways is thought to have evolved to allow parasites to successfully establish infections despite host genetic variation and/or development of host antibodies blocking single pathways. The same arguments can be applied to rosetting, and the existence of multiple rosetting pathways suggests that there has been significant selection pressure in favour of the phenotype, and that rosetting somehow improves parasite fitness.
Several recent reviews have discussed the parasite adhesion molecules involved in rosetting (Hviid and Jensen, 2015;Wang and Hviid, 2015;Yam et al., 2017), so these will not be described in detail here. Briefly, multiple studies have identified members of the VSA family P. falciparum erythrocyte membrane protein one (PfEMP1) as rosette-mediating adhesion molecules (Rowe et al., 1997;Vigan-Womas et al., 2008Albrecht et al., 2011;Ghumra et al., 2012), and recent reports suggest that other VSAs such as RIFIN (Goel et al., 2015) and STEVOR (Niang et al., 2014) may also contribute to rosette formation. Further work is needed to determine the relative contributions of the different VSAs to rosetting, especially in clinical isolates.
Host serum proteins such as IgM, α2macroglobulin, albumin and fibrinogen also contribute to rosetting, either by binding directly to parasite adhesion molecules or by non-specific erythrocyte aggregating effects (Scholander et al., 1996;Treutiger et al., 1999;Luginbuhl et al., 2007;Ghumra et al., 2008Ghumra et al., , 2012Semblat et al., 2015;Stevenson et al., 2015aStevenson et al., , 2015b. The extent to which host serum proteins influence rosetting, sequestration and microvascular obstruction in vivo is unknown, and would be a valuable area of future study.

Rosetting receptors on host erythrocytes
A number of different molecules on uninfected erythrocytes have been proposed as receptors for P. falciparum rosetting ( Fig. 2 and Table 1), and multiple receptor-ligand interactions may contribute to rosetting in any given parasite isolate. Some of the proposed rosetting receptor molecules, including blood group A and B sugars, heparan sulphate (HS)-like molecules and complement receptor one (CR1) are widely accepted as having a role in rosetting, whereas other recent candidates such glycophorin A (GYPA) and glycophorin C (GYPC) are less well-authenticated. However, a close examination of the underlying data shows that in most cases, the evidence is incomplete, as discussed in detail below.
Evidence needed to establish a role for a specific host receptor in rosetting In order to prove that a particular molecule acts as a host receptor for P. falciparum rosetting, a variety of different types of evidence have been provided. Essential data include proof that the molecule in question is found on normal human erythrocytes and that erythrocytes lacking the molecule show reduced/absent rosetting. Direct binding between IEs and/or recombinant parasite adhesion proteins and the receptor molecule should be demonstrated. Ideally, a crystal structure of the parasite adhesion moleculehost receptor complex should show the precise binding interaction site. Supportive evidence includes the ability of antibodies against the receptor or soluble receptor proteins to inhibit rosetting, and biochemical approaches to remove or alter the receptor on erythrocytes. Human genetic evidence can also provide indirect supportive evidence that particular molecules are important in life-threatening malaria. Several putative rosetting receptors have high-frequency polymorphisms in populations from malaria endemic regions that reduce rosetting and are associated with protection against severe malaria and death [reviewed in Rowe et al. (2009aRowe et al. ( , 2009b]. These various lines of evidence are summarized below for each potential host rosetting receptor.

Blood group A and B trisaccharides
The most well-validated rosetting receptors are the blood group A and B trisaccharides (Fig. 3). In vitro experiments have shown that rosetting parasites have a 'preference' for blood groups A, B or AB rather than O Udomsangpetch et al., 1993;Barragan et al., 2000b;Pipitaporn et al., 2000;Vigan-Womas et al., 2012;Moll et al., 2015). This varies by parasite genotype, with A-preference being the commonest. Clinical isolates from non-O (i.e. A, B or AB) patients show higher levels of rosetting than isolates from group O patients in studies from sub-Saharan Africa (Rowe et al., 1995 and India (Rout et al., 2012), although the same result was not seen in one Thai Inset image shows a single infected erythrocyte (centre) and three adherent uninfected erythrocytes. Images were taken using a Yenway microscope camera on a Leica DM LB2 fluorescent microscope using the ×40 and ×100 (inset) objectives.
study (Lee et al., 2014). When parasites are cultured in their 'preferred' blood group, they form larger, stronger rosettes that are more resistant to disruption by antibodies or chemical agents than in group O cells Barragan et al., 2000b;Ch'ng et al., 2016). Enzymatic removal of the terminal sugars (N-acetyl-D-galactosamine for A and D-galactose for B) results in smaller, weaker rosettes, equivalent to those seen in group O erythrocytes (Barragan et al., 2000b). Rosettes do, however, still occur with blood group O erythrocytes (that express the H antigen), and also in Bombay phenotype red cells that lack the ABO blood group core fucose residue ( Fig. 3) Rowe et al., 1997). This indicates that other red cell surface molecules in addition to the A and B antigens can act as host receptors for rosette formation.
For the blood group A-preferring parasite line, Palo Alto 89F5, direct binding between the VarO PfEMP1 adhesion molecule and the blood group A trisaccharide was shown by Surface Plasmon Resonance (Vigan-Womas et al., 2012). The VarO PfEMP1 variant also binds to the B trisaccharide, but with lower affinity (Vigan-Womas et al., 2012). A crystal structure of the PfEMP1 N-terminal region was obtained and the A-trisaccharide binding site mapped (Vigan-Womas et al., 2012). A recent study suggests that P. falciparum RIFIN molecules may also be able to interact with blood group A sugars to contribute to rosette formation (Goel et al., 2015), although direct RIFIN-A trisaccharide interaction was not shown.
The importance of the A and B antigens in rosetting is emphasized by the fact that the non-O blood groups are associated with increased risk of severe malaria and death compared to O Fry et al., 2008;Tekeste and Petros, 2010;Rout et al., 2012;Malaria Genomic Epidemiology Network, 2014;Ndila et al., 2018;Degarege et al., 2019). Reduced rosetting in blood group O, and therefore reduced microvascular obstruction and reduced downstream pathological effects, is the proposed mechanism for the protective association with group O (Udomsangpetch et al., 1993;Rowe et al., 2007). ABO blood group does not influence parasite burden Degarege et al., 2019), and evidence for an effect of ABO on P. falciparum invasion or other host-parasite interactions is conflicting and requires further study (Chung et al., 2005;Wolofsky et al., 2012;Pathak et al., 2016;Theron et al., 2018). The ABH antigens are known to be present on endothelial cells (Ito et al., 1990) and it is likely, but has not been shown experimentally, that cytoadhesion and overall levels of sequestration of rosetting parasites are enhanced in group A/B/AB patients compared to O.
Despite the progress in identifying the A and B trisaccharides as rosetting receptors and key genetic determinants of host susceptibility to severe malaria, there have been no attempts to develop specific therapies to block P. falciparum interaction with A/B antigens. The PfEMP1-blood group A trisaccharide binding pair described above (Vigan-Womas et al., 2012) remains the most clearly defined molecular interaction between parasite ligand and host receptor in rosetting, and could be used as a starting point to develop rosette-blocking therapeutics. Vigan-Womas et al. did report that the interaction between PfEMP1 and the A and B trisaccharides is indirectly inhibited by heparin (Vigan-Womas et al., 2012), and the development of a heparinderivative as a potential adjunctive therapy for severe malaria is described below .

Complement receptor one (CR1, CD35)
CR1 is a red cell membrane glycoprotein that regulates complement activation on cell surfaces (Thielen et al., 2018) and carries the Knops Blood Group antigens (Moulds, 2010). In malaria, CR1 plays a role in both rosetting and parasite invasion of erythrocytes (Schmidt et al., 2015). CR1 was first identified as a rosetting receptor from a screen of 23 naturally occurring erythrocyte  Membrane glycoprotein responsible for regulating the complement system (Thielen et al., 2018) Polymorphisms affect CR1 copy number, molecular weight and sequence (Schmidt et al., 2015) RBC CR1 deficiency protects in medium-high (Cockburn et al., 2004;Sinha et al., 2009;Rout et al., 2011;Panda et al., 2012) but not low malaria transmission areas (Nagayasu et al., 2001;Teeranaipong et al., 2008). CR1 Knops blood group polymorphisms associated with severe malaria (Opi et al., 2018) Rosetting reduced in CR1 deficient erythrocytes (Rowe et al., 1997) Soluble CR1 and CR1 antibodies disrupt rosettes in some parasite isolates (Rowe et al., 1997;Rowe et al., 2000;Vigan-Womas et al., 2012) Essential region mapped to the C3b binding site on CR1 (Rowe et al., 2000) Further work needed to assess the relative importance of CR1 in rosetting isolates and potential as a therapeutic target Soluble recombinant CR1 has been considered for therapeutic use in humans, e.g. cardiac and renal disease (Li et al., 2006;Reddy et al., 2017) Heparan sulphate (HS) b Glycosaminoglycan Heparin is a highly sulfated form of HS that is only found in mast cells HS is a receptor for P. falciparum sporozoite invasion of hepatocytes (Frevert et al., 1993) HS is a receptor for infected erythrocyte cytoadherence to endothelial cells (Vogt et al., 2003;Adams et al., 2014) Heparin partially disrupts rosettes in some isolates Rogerson et al., 1994;Rowe et al., 1994;Barragan et al., 1999) Heparinase treatment reported to reduce rosetting in two culture-adapted parasite lines (Barragan et al., 1999) Heparin binds to rosetting IE (Barragan et al., 2000a;Heddini et al., 2001) and to rosette-mediating PfEMP1 (Barragan et al., 2000a;Vogt et al., 2003;Juillerat et al., 2010;Juillerat et al., 2011;Adams et al., 2014) Limited evidence that HS is present on mature RBCs (Vogt et al., 2004) Further work needed to determine whether HS is present on normal erythrocytes and acts as a rosetting receptor Therapeutic potential due to PfEMP1 binding and rosette disruption. Clinical trials of low anticoagulant heparin ongoing  Chondroitin sulphate (CS)

Glycosaminoglycan
Receptor for infected erythrocyte placental sequestration in pregnancy malaria (Fried and Duffy, 1996) Soluble CS did not disrupt rosettes (Rogerson et al., 1994;Rowe et al., 1994) Chondroitinase treatment reduced rosetting in one parasite line only (Barragan et al., 1999) No evidence that CS is present on mature RBC Minimal evidence for a role in rosetting CD36 Widely distributed membrane protein and scavenger receptor (Silverstein and Febbraio, 2009) Deficiency is common in Africa but not associated with severe malaria (Fry et al., 2009) Antibodies disrupt rosettes in single culture-adapted line only  PfEMP1 variants that mediate rosetting are group A types that do not bind CD36 (Robinson et al., 2003) Minimal evidence for a widespread role in rosetting Glycophorin C (GYPC) Red cell membrane protein responsible for Gerbich blood group (Jaskiewicz et al., 2018) Receptor for merozoite invasion of erythrocytes (Maier et al., 2003) 'Gerbich-negative' blood group common in Melanesians (Patel et al., 2001), but no evidence yet for association with protection against severe malaria Reduced rosetting with GYPC antibodies and GYPC knockdown RBCs (Niang et al., 2014) (single culture-adapted parasite line tested) Gerbich-negative erythrocytes formed rosettes normally with five P. falciparum lines (Rowe et al., 1997) Possible role in P. vivax rosetting (Lee et al., 2014) Further work needed to assess the relative importance of GYPC in P. falciparum rosetting isolates and potential as a therapeutic target  (Rowe et al., 1997). The only variant to show substantially reduced rosetting with five P. falciparum parasite lines was Knops null cells, which are deficient in CR1. Normally, erythrocytes have between 100 and 1000 molecules of CR1 per cell (Wilson et al., 1986), whereas Knops null cells have fewer than 100 molecules per cell (Moulds et al., 1992). Erythrocytes with fewer than 50 CR1 molecules per cell form rosettes poorly (Rowe et al., 1997), with normal rosetting occurring above a threshold of around 100 molecules per cell (JA Rowe, unpublished data). Soluble CR1 and CR1 antibodies were shown to inhibit rosetting in some but, not all P. falciparum rosetting laboratory strains and clinical isolates, with only monoclonal antibodies (mAbs) that map to the C3b binding site on CR1 being effective inhibitors (Rowe et al., 1997(Rowe et al., , 2000Vigan-Womas et al., 2012). A recent paper suggested that the commercially available CR1 mAb E11 that recognizes epitopes outside the C3b binding site (Nickells et al., 1998) may inhibit P. falciparum rosetting (Lee et al., 2014), but this was not seen in our hands (Rowe et al., 2000). Further evidence of a role for CR1 in rosetting came from the expression of recombinant PfEMP1 domains in COS-7 cells, which bound to normal erythrocytes but not to CR1-deficient cells (Rowe et al., 1997).
Despite these supportive data, direct binding of IEs to CR1 protein has not been demonstrated, and recombinant rosettemediating PfEMP1 proteins produced in E. coli (Ghumra et al., 2012) do not bind to CR1 in Surface Plasmon Resonance experiments (Tetteh-Quarcoo et al., 2012). This could reflect a genuine lack of interaction between the two molecules, or could be due to technical reasons (e.g. the recombinant CR1 used in experiments was produced in mouse rather than human cells, whereas CR1 glycosylation, which may affect function, is cell-type specific) (Lublin et al., 1986). It is also possible that a serum protein mediates the interaction between PfEMP1 on IEs and CR1 on uninfected erythrocytes, as the original experiments were all performed in the presence of serum (Rowe et al., 1997).
Human genetic studies provide additional support for the importance of CR1 in malaria host-parasite interactions.
Erythrocyte CR1 deficiency is common in some malaria-endemic countries such as Papua New Guinea (Cockburn et al., 2004) and India (Sinha et al., 2009), and is associated with protection against severe malaria in medium to high transmission areas (Cockburn et al., 2004;Sinha et al., 2009;Rout et al., 2011;Panda et al., 2012). However, erythrocyte CR1 deficiency may be detrimental in areas such as Thailand, where malaria transmission is low (Nagayasu et al., 2001;Teeranaipong et al., 2008). There is also evidence that the CR1 Swain Langley 2 (Sl2) Knops blood group polymorphism that is common in African populations (Moulds, 2010) is associated with protection against severe malaria (Thathy et al., 2005;Opi et al., 2018). Red cells carrying the Sl2 antigen on CR1 show reduced rosetting (Rowe et al., 1997;Opi et al., 2018), and Sl2 may have additional effects on complement activation and regulation (Opi et al., 2018).
Overall, the ability of CR1 mAbs and soluble protein to reverse rosettes suggests that CR1 plays a role in rosetting for some P. falciparum isolates. However, further work is needed to fully investigate the molecular interactions between parasite adhesion molecules and CR1, and to explore the potential for CR1 reagents (Li et al., 2006;Reddy et al., 2017) as therapeutic disruptors of rosetting.

Heparan sulphate and chondroitin sulphate
The glycosaminoglycans HS and chondroitin sulphate (CS) are found on cell surfaces and in the extracellular matrix of many tissues, and have a role in multiple aspects of the P. falciparum life cycle including hepatocyte invasion (Frevert et al., 1993), endothelial cell cytoadherence (Vogt et al., 2003;Adams et al., 2014) and, for CS, placental sequestration (Fried and Duffy, 1996). A number of papers have showed that heparin (which is a highly-sulphated form of HS found only in mast cells) can partially disrupt rosettes in about one-third to one-half of P. falciparum clinical isolates in vitro Rogerson et al., 1994;Rowe et al., 1994;Barragan et al., 1999). It was shown that treating erythrocytes with heparinase III, which selectively cleaves HS chains, reduces rosetting in two P. falciparum lines (Barragan et al., 1999), and therefore suggested that 'HS-like' molecules on red cells are receptors for rosetting (Chen et al., 2000). However, there has been only one paper reporting the existence of HS on normal human erythrocytes (Vogt et al., 2004) and we have been unable to confirm this, and unable to detect any rosette-reducing effect of heparinase III in a range of parasite lines (McQuaid and Rowe, unpublished data).
Fluorescently-labelled heparin does bind to the surface of erythrocytes infected with rosetting parasites more than nonrosetting lines (Barragan et al., 2000a;Heddini et al., 2001), and some rosette-mediating PfEMP1 variants bind directly to heparin (Barragan et al., 2000a;Vogt et al., 2003;Juillerat et al., 2010Juillerat et al., , 2011Adams et al., 2014). The heparin binding site in the N-terminal region of the varO PfEMP1 variant was mapped onto a crystal structure , and shown to be on the opposite side of the molecule from the erythrocyte binding site (Vigan-Womas et al., 2012). Hence, the rosette-disrupting effect of heparin is not due to direct blocking of receptor binding, but may result from aggregating PfEMP1 monomers and preventing their interaction with erythrocyte receptors (Vigan-Womas et al., 2012). Similarly, for another rosette-mediating PfEMP1 variant IT4var60, site-directed mutagenesis studies of recombinant proteins showed that mutations that disrupt heparin binding are distinct from mutations that disrupt erythrocyte binding, indicating that heparin-like molecules are not the main host rosetting receptor in this case (Angeletti et al., 2015).
Overall, whether HS is present on normal erythrocytes and is a host receptor for rosetting requires further confirmation. HS in present on the luminal surface of microvascular endothelial cells (albeit at a much lower density than on basolateral surfaces) (de Agostini et al., 1990;Stoler-Barak et al., 2014), therefore interactions between IE and endothelial HS (Vogt et al., 2003;Adams et al., 2014) are physiologically relevant and are likely to contribute to cytoadherence and sequestration in vivo.
Despite the uncertainty on the precise role of HS as an erythrocyte rosetting receptor, heparin and other sulphated glycoconjugate compounds have clear potential as adjunctive therapies for severe malaria due to their rosette-disrupting effects Rogerson et al., 1994;Rowe et al., 1994;Kyriacou et al., 2007). There are reports of successful heparin treatment in severe malaria (Rampengan, 1991) but its use is not recommended due to a high incidence of bleeding complications (World Health Organisation, 1986). As an alternative, Wahlgren and coworkers have developed a low anti-coagulant heparin derivative, Sevuparin, that reverses rosetting and cytoadherence in some P. falciparum isolates (Leitgeb et al., 2011;Saiwaew et al., 2017) and also blocks merozoite invasion . Sevuparin has been shown to be safe in adults with uncomplicated malaria , but has not yet been tested in severe malaria patients.
The evidence for CS as a rosetting receptor is minimal. One report shows that rosetting in the P. falciparum line TM284 was partially inhibited by soluble CS and by chondroitinase enzyme treatment of erythrocytes, and that several clinical isolates showed reduced rosetting in the presence of CS (Barragan et al., 1999). However, other studies have found no effect of CS on rosetting in a variety of culture-adapted lines and clinical isolates (Rogerson et al., 1994;Rowe et al., 1994). There is also no convincing evidence that CS is found on the surface of normal human erythrocytes. Overall, current data do not support a role for CS in rosetting.

CD36
The membrane glycoprotein CD36 is a scavenger receptor for oxidized lipoproteins and a fatty acid translocase (Silverstein and Febbraio, 2009). It is expressed on a variety of cell types including monocytes, macrophages, platelets, microvascular endothelial cells and adipocytes (Silverstein and Febbraio, 2009), and at low levels on erythrocytes . The binding of PfEMP1 (group B and C variants) to CD36 on microvascular endothelial cells plays a major role in P. falciparum sequestration (Baruch et al., 1996;Robinson et al., 2003). Almost all P. falciparum isolates bind to CD36, and increased CD36 binding Ochola et al., 2011) and predominant expression of group B and C PfEMP1 (Kraemer and Smith, 2006;Kyriacou et al., 2006) are associated with uncomplicated malaria.
The role of CD36 in rosetting is less clear. Anti-CD36 mAbs are capable of disrupting rosettes in a single culture-adapted parasite line, Malayan Camp , but not in a wide range of other laboratory lines or clinical isolates Rowe et al., 2000;Niang et al., 2014). The PfEMP1 variants identified as parasite rosetting ligands (Rowe et al., 1997;Vigan-Womas et al., 2011;Ghumra et al., 2012) are mostly of the group A type, which do not bind to CD36 (Robinson et al., 2003).
Intriguingly, while CD36 deficiency is fairly common in African populations, large-scale genetic studies have shown that CD36 polymorphisms do not influence severe malaria risk (Fry et al., 2009). There is some evidence that interaction between IEs and CD36 may benefit the host, as CD36 may contribute to innate immune clearance of IEs and platelet-mediated parasite death (McGilvray et al., 2000;McMorran et al., 2012;Cabrera et al., 2014). Overall, it is unlikely that CD36 is a clinically significant rosetting receptor or a useful therapeutic target in severe malaria (Cabrera et al., 2014).

Glycophorin C (GYPC; GPC; CD236)
GYPC is a red cell membrane glycoprotein that carries the Gerbich blood group antigens (Jaskiewicz et al., 2018). It is a P. falciparum invasion receptor bound by the merozoite protein EBA-140/BAEBL (Maier et al., 2003;Mayer et al., 2006). Recently, two studies have suggested that GYPC is a rosetting receptor for both P. falciparum (Niang et al., 2014) and P. vivax (Lee et al., 2014;Niang et al., 2014). Niang et al. showed that the rosetting of a 3D7-derived P. falciparum laboratory strain (5A-R+) was partially inhibited by a GYPC mAb (clone Ret40f) and by soluble recombinant GYPC (Niang et al., 2014). Furthermore, cultured GYPC knockdown erythrocytes failed to rosette, providing strong evidence that GYPC is an essential rosetting receptor for 5A-R+ parasites (Niang et al., 2014). Other parasite lines or clinical isolates were not tested, therefore the wider role of GYPC in P. falciparum rosetting was not determined. Lee et al. (2014) focussed mainly on P. vivax, but also assessed the ability of GYPC mAb fragments to inhibit rosette formation in ten P. falciparum clinical isolates from Thailand. A significant decrease in rosetting was reported with GYPC mAb BRIC 4, although the reduction in the median rosette frequency was small (from 11.5 to 5.5%), and was based on a single count for each isolate with no replication (Lee et al., 2014). The biological significance of these results is difficult to assess, given the low starting rosette frequencies and inherent variation in the rosetting assay. Lee et al. also used a different definition of rosetting to all previous studies, defining a rosette as an IE binding one or more uninfected erythrocytes. The usual definition requires the binding of two or more uninfected erythrocytes, which helps to identify genuine cell-cell interactions and avoid spurious identification of rosettes due to close packing of cells under the coverslip during microscopy.
For P. vivax, Lee et al. showed that the GYPC mAb reduced the median rosette frequency from 30 to 22% when tested on 11 Thai 6 Fiona McQuaid and J. Alexandra Rowe isolates, and that GYPC knockdown cultured erythrocytes formed rosettes poorly compared to GYPC-positive control cells (median rosette frequency 6.2 vs. 35.4% in controls, tested on three isolates). Plasmodium falciparum isolates were not tested with the GYPC knockdown erythrocytes. If GYPC is a rosetting receptor, it is possible that the 'Gerbich-negative' blood group type, which is common in Melanesian populations (Patel et al., 2001), might influence rosetting. As part of a screen of null blood group erythrocytes with five high-rosetting P. falciparum culture-adapted parasite lines, Rowe et al. (1997) tested two donors with the Gerbich-negative blood group (formed by deletion of exon 3 of the GYPC gene on chromosome 2, giving a truncated protein with altered glycosylation). Gerbich-negative erythrocytes formed rosettes normally with the five parasite lines tested. Goel et al. also report normal rosetting of Gerbich-negative erythrocytes from two donors (Goel et al., 2015). The true null phenotype for GYPC, called the Leach phenotype (which arises due to the deletion of exon 3 and exon 4, encoding the transmembrane and cytoplasmic domains, respectively) is rare and has not been tested in rosetting assays to our knowledge.
Taking into account all existing evidence, further investigation of a wider range of parasite lines is needed to determine whether GYPC is an important host receptor for both P. falciparum and P. vivax rosetting.
Glycophorin A (GYPA, GPA, CD235a) GYPA is a highly-expressed erythrocyte surface glycoprotein that carries the MNS blood group antigens. It is known to be a receptor for P. falciparum erythrocyte invasion (Sim et al., 1994), and polymorphisms in GYPA are associated with resistance to severe malaria (Band et al., 2015;Leffler et al., 2017).
There is some limited evidence to suggest that GYPA may have a role in rosetting. Parasites of the strain FCR3S1.2 transfected with a specific RIFIN gene formed rosettes that were largely dependent on blood group A (Goel et al., 2015). However, rosetting of the RIFIN-transfected parasites was significantly reduced with GYPA null cells from blood group O and B donors, whereas blood group A GYPA null erythrocytes formed rosettes normally. These data suggest that GYPA may have an accessory role for RIFIN-mediated rosetting in the absence of the A antigen (Goel et al., 2015), although whether this applies to rosetting in nongenetically manipulated parasites in unknown.
Despite the above positive evidence, there are no other data supporting a role for GYPA in rosetting. GYPA mAb fragments had no inhibitory effect on rosetting in ten P. falciparum and 11 P. vivax clinical isolates (Lee et al., 2014), and a GYPA mAb did not inhibit 3D7 5A-R+ rosettes (Niang et al., 2014). Furthermore, GYPA null erythrocytes (MkMk cells, lacking both GYPA and glycophorin B) formed rosettes with five cultureadapted P. falciparum lines (Rowe et al., 1997). Overall, existing evidence does not support a major role for GYPA in rosetting, but as with GYPC, further investigation is needed.

New receptors and new approaches
None of the receptors described above fully account for the adhesion interactions between infected and uninfected erythrocytes, and it is likely that other host rosetting receptors remain to be identified. There is evidence to suggest that these unknown host receptors are carbohydrates or protease-resistant proteins, because uninfected group O erythrocytes treated with trypsin and other proteases are still able to form rosettes Rowe et al., 1994).
In order to progress rosetting research, alternative methods are needed. Rosetting experiments with GYPC and CR1 knockdown cultured human red cells derived from CD34+ haematopoetic stem cells have been performed (Lee et al., 2014;Niang et al., 2014), using lentiviral transduction of short hairpin RNA (Bei et al., 2010). However, these cultured erythrocytes have a short life-span, limiting their usefulness. The development of immortalized erythroid lines (Kurita et al., 2013;Kanjee et al., 2017;Trakarnsanga et al., 2017;Scully et al., 2019) may overcome this limitation. Nevertheless, attention must be paid to the subtle but real differences between mature erythrocytes and these, still relatively immature, immortalized CD34+ derived cells (Wilson et al., 2016;Dankwa et al., 2017;Trakarnsanga et al., 2017). CRISPR-Cas9 technology (Doudna and Charpentier, 2014) has led to an explosion in the ability to genetically manipulate multiple cell types, including erythrocyte precursors and immortalized haematopoietic lines (Song et al., 2015;Kanjee et al., 2017;Hawksworth et al., 2018;Chung et al., 2019;Scully et al., 2019), potentially giving the opportunity to generate multiple knockout lines for rosetting research. A consistent supply of knockout erythrocytes would allow large-scale screens for new rosetting receptors using cells as close to their normal physiological form as possible, raising exciting prospects for future work.

Conclusions
Of the rosetting receptors described over the past 30 years, only the blood group A trisaccharide has been authenticated by a variety of methodological approaches from a range of different investigators. For all other potential rosetting receptors, the evidence remains fragmentary (Table 1) and further research is needed (Table 2). Recent technical advances in genetic manipulation of red cell precursors and immortalised lines should enable reverse genetic studies to bring further clarity to this biologically important topic.