All the members of this family of proteins are structurally characterized by the presence of a highly conserved core composed of four (eight in annexin A6) homologous domains of about 70 amino acids showing a similar three-dimensional structure highly conserved throughout annexin evolution [6]. Inside each domain, a sequence of 17 amino acid residues is found which presents an even higher degree of conservation. Annexin A5 was the first member of this family to be crystallized and its structure to be studied by X-ray diffraction [6,7]. Since then, several other annexins have been crystallized, and all of them present an almost identical three-dimensional arrangement in the protein core. The overall shape of these molecules is that of a slightly bent disc (Figure 1A) where the four repeated domains are arranged around a central hydrophilic hole (Figure 1B); this hole could be responsible for the voltage-dependent calcium channel activity reported for several annexins when bound to membranes (A1, A2, A5–A7, B12) [8–10]. Each domain contains a four α-helix bundle (helices A, B, D & E) organized in a cylindrical way and capped by a fifth α-helix (C) that is located on the concave surface of the annexin molecule (Figure 1A; helices are labeled in Domain III). Calcium ions bind on the convex side of the molecule. There are up to three possible coordination sites per domain, the so-called Type II or “AB” site being the one that shows higher affinity for calcium. In this location, calcium binds to carbonyl oxygens in the loop connecting the A and B helices and to a bidentate carboxyl group from a glutamic or aspartic acid residue located around 40 residues downstream in the loop connecting helices D and E (Figure 1).

In contrast to the annexin core, which shows a clear homology, sequence analysis of annexins shows the great diversity acquired by their N-terminal extension during evolution. These domains are variable both in length and amino acid sequence. In human annexins, the N-terminus region ranges from a few residues to 200 or more amino acids and, based on the length of this region, they can be classified into three groups. The first one includes those members with a short extension below 20–21 amino acids, such as annexins A3, A4, A5, A6, A8, A10 and A13a. A second group includes annexins with an intermediate N-terminus, up to 55 residues, such as annexins A1, A2, A9 and A13b. Finally, a third group can be established with annexins that possess a long N-terminus with more than 100 amino acid residues (annexins A7 and A11).

Even though the N-terminus of annexins is significantly smaller than the C-terminal protein core, it is essentially involved in the structural and functional peculiarities of each member of this superfamily of calcium-binding proteins [13,14]. The N-terminal extension contributes significantly to the stability of the overall structure; it is also essential for self-association and mediates the interaction with other proteins, mainly from the S100 family (also calcium binding proteins but with the EF-hand motif). Moreover, post-translational modifications of this region (phosphorylation, acetylation, myristoylation, proteolysis, etc.) modify the structure of key regions of the protein core, even if they are located at the opposite side of the molecule. Consequently, the N-terminal domain of these proteins has to be considered as the main regulatory element of both annexin structure and function.

Crystallographic data have revealed that the N-terminal domain determines the structural arrangement of protein regions located on the opposite concave region of the annexin molecule binding together Domains I and IV [6,12,15–18], at least in annexins with a short N-terminal domain such as annexin A5. Interestingly, using circular dichroism and fluorescence spectroscopy, we have found that an annexin A5 mutant with a truncated N-terminus also presents a more solvent-exposed IIIAB loop [19]. Thus, it seems that the interaction of the N-terminus with Domain IV in the concave side is transmitted to the loop located in Domain III at the opposite side of the molecule, as the truncation of the N-terminus forces a conformational rearrangement of the loop. Moreover, phosphorylation of Thr6 in annexin A4, that involves the displacement of the N-terminus, also impairs its ability to induce vesicle aggregation [20]. A similar effect is observed in annexin A1 after phosphorylation of Ser27; however, it does not affect binding to membranes and lateral self-association [21].

The crystal structure of an intermediate annexin with its complete N-terminus in the absence of calcium was first determined for annexin A1 [22]. In this protein, as expected from sequence homology to other crystallized annexins and from the structure of the truncated form, the N-terminus runs first through the concave side of the molecule from Domain I to Domain IV as an extended coil. This region is constituted by amino acids 28 to 41, which are preceded by two consecutive amphipathic α-helices (Figure 2). The second helix includes amino acids 18 to 27 and interacts with Domain IV. It is directed towards Domain III where it turns in position 18. The most N-terminal helix (residues 2–17) is inserted into Domain III and displaces Helix IIID, which in turn becomes an extended loop over the convex side of the molecule. It has been shown that the interaction of annexin A1 with phosphatidylserine (PS) bilayers increases the N-terminal degradation susceptibility and the accessibility of disulfide bridges [23]. Calcium binding to annexin A1 induces a conformational rearrangement mainly consisting of the refolding of Helix IIID and the release of the N-terminal α-helix from the protein core, as deduced from the crystal structure of the entire protein in the presence of calcium [24]. The release of the N-terminus allows the binding of S100A11 in a calcium-dependent fashion. In this sense, the N-terminus would act as a regulatory region that modulates the possible interactions of the protein core with either membranes or other proteins.

In annexins with a long or intermediate N-terminus, some α-helical stretches can be predicted. These α-helical sub-domains could be involved in the interaction with other proteins from the S100 family, these associations being essential for their function. In annexin A11, with a long unordered N-terminus, residues 45–62 are predicted to form an amphipathic α-helix, and it is tempting to suggest a similar interaction between this helix and the protein core through Domain III. However, we have shown by fluorescence acrylamide quenching that the unique tryptophan residue in Position 23 is completely exposed to the solvent suggesting that this region does not interact with the protein core [25].

As previously mentioned, annexins are mainly characterized by their ability to reversibly interact with membranes in a calcium-dependent manner. Although this is true for the vast majority of the members of this family of proteins, some members, such as mammalian annexin A9, do not bind calcium but still are able to interact with biological membranes through different mechanisms. Human annexin A5 was the first annexin crystallized in 1990 [6,26]. From then on, numerous crystal structures have been reported and are available through the Protein Data Bank. These structures have allowed an extensive analysis of the calcium binding sites in annexins. Figure 1 shows the structure of rat annexin A5 saturated with calcium (PDB 1A8A); ten calcium ions are coordinated with specific residues in the convex membrane-binding surface of the protein. In fact, calcium-binding measurements with annexins in solution indicate that these proteins may bind up to 10–12 Ca²⁺ ions, although with different affinities [27–29]. Differences in the stoichiometry of the binding have been associated with the ability of annexins to establish trimers upon membrane binding or not: Trimer-forming annexins A5 and B12 bind around 12 mol of Ca²⁺/mol of protein whereas nontrimer-forming annexins A1 and A2 show a lower stoichiometry (3–4 mol of Ca²⁺/mol of protein). In addition, the A5/B12 group of annexins binds to bilayers in the liquid-crystal phase but not in the gel phase whereas the opposite occurs in the A1/A2 group, suggesting that there is a complementarity between the spacing of the Ca²⁺-binding sites and the spacing of the phospholipid head groups in the bilayers [28]. In any case, most annexins contain two to four Type II (“AB”) high-affinity Ca²⁺-binding sites located in the loop between Helices A and B. In addition, annexins may contain two additional calcium binding sites with lower affinity, named Type III, located at the N-terminus of Helix B (“B” site) or in the loop between Helices D and E (“DE” site) (Figure 1). Mutagenesis studies indicate that the “AB” sites are required for attachment to membranes, while the B or DE sites are not sufficient for membrane binding, although their presence increases the binding affinity [30,31].

Each of the calcium-binding sites shows a structural pattern in which the Ca²⁺ ion is coordinated by seven oxygen atoms forming a more or less regular pentagonal bipyramid. Five oxygens lie in the equatorial plane and two in axial positions. In the main “AB” calcium-binding sites, five oxygens belong to the protein: A bidentate carboxylate group from an aspartic or glutamic acid side chain located 40 residues downstream of a conserved glycine present in the interhelical AB loop (equatorial plane) and three oxygens from carbonyl groups form peptides bonds within the loop. The remaining two oxygens belong to water molecules in the absence of phospholipids, or to phosphate groups from the polar heads of two phospholipid head groups.

Figure 1C shows a detail of the “AB” calcium-binding site in Domain III of rat annexin A5 indicating the residues involved in coordination as well as one glycerophosphoserine (GPS) moiety that resembles the head group of PS, the main lipidic ligand of annexins in biological membranes. The “B” site is close to the “AB” site and collaborates in PS binding, as one of the oxygen atoms of the carboxyl group of serine from PS is coordinated with the calcium ion at this position. The annexin AB loop conformation is almost identical to that observed in crystals saturated only with calcium or with calcium and bound GPS. No other phospholipid, besides PS, has oxygens that may coordinate to the calcium at the “B” site. Moreover, the serine α-amino group appears to stabilize the interaction of the phosphoserine head by establishing a hydrogen bond with the hydroxyl group of a conserved threonine residue (T¹⁸⁷ in Domain III of rat annexin A5; Figure 1C). All these stabilizing interactions can explain why PS is the “preferred” phospholipid ligand of annexins, as this calcium binding seems to be specifically designed to fit the serine head group. In fact, the crystal structure of this annexin with bound glycerophosphoethanolamine (resembling phosphatidylethanolamine (PE); PDB file 1A8B) shows that this head group is only coordinated with the “AB” site, but no additional stabilizing interactions occur [12].

The GPS molecule shown in the crystal structure of rat annexin A5 (Figure 1C) represents the so-called “apical” PS molecule that binds to the “AB” site with high affinity [32], and that is present in the four domains of almost all vertebrate annexins. In addition, the calcium ion bound to this site may coordinate with the phosphate group of an additional PS molecule in the “equatorial” binding site. This additional site involves not only the coordination with calcium, but also six or seven protein ligands from residues at the C-terminus of Helix A, the CD loop, and Helix D. This consensus sequence is found with slight variations in all vertebrate annexins, but only in Domains I and II, not being present in Domains III and IV [32]. The existence of this “equatorial” PS binding site was determined by molecular modeling, as radiocrystallography or nuclear magnetic resonance (NMR) hardly provide essential information at the atomic resolution when a phospholipid bilayer is concerned. Mutation of this consensus sequence in annexin A5 did not suppress calcium or PS binding to the “AB” site, but required more calcium to compensate for the loss of PS binding in the equatorial site.

The affinity of the “B” site for calcium is quite low as only three oxygens from the protein contribute to calcium coordination: A bidentate carboxyl group and a carbonyl group from a peptide bond (Figure 1C). In the absence of PS, four positions are occupied by oxygens from water molecules, but when PS is present, one of the water molecules is replaced by a carboxyl oxygen from the phosphoserine head group. The establishment of tertiary annexin-Ca²⁺-PS complexes strongly increases the affinity of annexins towards calcium possibly due to the replacement of solvent molecules by oxygens from the phospholipid polar head. In fact, crystallographic data as well as solution measurements are consistent with an increase in affinity of two or three orders of magnitude [33–35].

A third possible phospholipid head group may bind to each annexin domain via the “DE” calcium-binding site. This additional calcium ion is coordinated with four oxygens from the loop between Helices D and E: A bidentate carboxyl group from the lateral chain of an aspartic or glutamic acid at the beginning of Helix E, and two carbonyl oxygens from peptide bonds within the loop. The remaining three coordinations are occupied by water in the absence of phospholipids (Figure 1C). In the presence of phospholipids, two water molecules are replaced by oxygens from the polar head group, as suggested by molecular dynamics simulation [36].

All in all, up to three phospholipid polar head groups can interact with each domain of the annexin core structure. Additional stabilizing interactions can be established between the convex surface of the annexin molecules and phospholipid bilayers. Figure 1D shows the molecular surface of rat annexin A5 indicating the hydrophobic (yellow) and hydrophilic (blue) regions. When calcium is bound to the “AB” sites, hydrophobic lateral chains are exposed in the otherwise mainly hydrophilic convex surface. These hydrophobic residues are highly conserved in the AB loops and may contribute to the interaction with bilayers by establishing van der Waals forces with the hydrophobic acyl chains of phospholipids. However, these residues do not penetrate deeply into the bilayer as observed from the behavior of the fluorescence spectra of the tryptophan residue located in the AB loop of Domain III of several vertebrate annexins [35,37].

Annexins are mainly cytosolic proteins that bind preferentially to the major phospholipidic components of the intracellular leaflet, PS and PE, although they exhibit a marked preference for PS over PE [2,38]. In addition, annexins may bind other phospholipids at physiological pH, such as the negatively charged ones phosphatidic acid (PA), phosphatidylglycerol (PG) or phosphatidylinositol (PI). Some annexins also show further, more specific interactions with membrane lipids; for example, annexins A2 and A8 bind phosphatidylinositol-4,5-bisphosphate (PIP₂) [39–41]. In general, the net charge of the polar head of phospholipids seems to be important for recruiting annexins to the membrane. However, the molecular interaction of the lipid head group with specific protein residues strongly influences the affinity with which the lipids bind and to which site they do. This could explain why PS binds more tightly than PI to annexins even though they have the same negative charge at physiological pH, as well as why uncharged PE binds to annexins whereas phosphatidylcholine (PC) does not.

Annexins do not require all four domains from the protein core to interact with phospholipid bilayers. In fact, different annexins show preferences for one domain or another in order to establish the interactions with membranes. On this idea, mutational analysis has suggested that Domain I in annexin A5 plays a prominent role in the lipid-binding process, although the rest of the Ca²⁺ and phospholipid binding sites collaborate to stabilize the interaction. Interestingly, annexin A1 lacks the consensus Type II or “AB” calcium binding site at Domain I; moreover, this sequence is lost in all repeats of annexin A9, in Domains I and IV of annexin A2, Domains I, III and IV of annexin A10 and in Domain III of the C-terminal tetrad of annexin A6 [42]. In addition, the free Ca²⁺ concentrations required to initiate phospholipid binding differ markedly among different annexins and different phospholipid head groups. They range from 20 μM for the binding of annexin A5 to PS-containing liposomes, through submicromolar Ca²⁺ concentrations for the PS- and PA-binding of annexins A1 and A2, to less than 100 nM for the interaction of the annexin A2-S100A10 complex with PS-containing membranes [2,38]. This indicates that the annexin family as a whole can respond to a large spectrum of Ca²⁺ concentration changes induced by different stimuli with intracellular translocation from the cytosol to the membrane. Moreover, individual annexins are specifically designed to react only to stimuli of certain amplitude. Among vertebrate annexins, the only member that does not interact with PS-containing liposomes at submillimolar calcium concentrations is annexin A9 which lacks the Type II consensus calcium and phospholipid binding sites [43] (no data are available regarding annexin A10 that lacks three of the four binding sites). It is worth mentioning that around one third of annexin domains in the eukaryotic annexin superfamily has lost the Type II calcium binding domains. However, many of these exposed regions present a novel K/H/RGD motif (that may also appear at the loop between Domains II and III), that may have a functional role in annexin membrane interactions probably via receptor targeting [42]. Interestingly, annexin A9 (as well as annexins A1 and A2) possesses two such novel motifs in Repeats III and IV that may be responsible for the potential membrane associated physiological functions of annexin A9 [44,45].

Although calcium-dependent phospholipid binding is the main characteristic of the annexin superfamily of proteins, additional lipid binding mechanisms have been described in some annexins that do not require, at least directly, calcium bridging. Among others, the main parameter regulating the calcium-independent binding is the pH value. It has been shown in vitro that annexins A5 and A13b undergo a conformational change at mildly acidic pH (midpoints around pH 4.1 and 5.8, respectively) that resembles that observed after calcium binding (the so-called “open conformation” with exposure of the tryptophan residue located at the AB loop of Domain III) [37,46,47]. These pH values can be reached inside the cell as it is well known that pH can decrease around 1.6 units in the proximity of the membrane in regions rich in anionic phospholipids as PS [48]. The N-terminal half of annexin A6 (highly similar to annexin A5) also undergoes a conformational change at acidic pH, but it involves a partial denaturation of the protein core with exposure of hydrophobic regions [49]. This conformational change seems to be an intermediate state for the calcium-independent binding of annexin A5 to PS-liposomes at pH 4.0 [50], and induces leakage from PS vesicles [51]. Interaction with acidic phospholipid vesicles at slightly low pH has also been described for annexin A4, which induces leakage from these vesicles [52], and for annexin A6, suggesting that this insertion is a prerequisite for the formation of calcium channels [53]. A pH-driven calcium-independent insertion into PS-rich monolayers has also been observed in annexin A1 [54] and has been suggested for annexin A5 [55]. These observations led to postulate that annexins could penetrate into the bilayer, in contrast to the preferred location at the vesicle surface at neutral pH, possibly adopting different conformations under the two conditions.

The introduction of site directed spin labeling for the study of calcium-independent annexin-membrane interactions has allowed further advances in understanding the molecular mechanisms by which annexins interact with acidic bilayers. These studies have been carried out with annexin B12 from Hydra by engineering protein mutants with specific derivatized cysteines with a paramagnetic nitroxide chain. These experiments revealed that annexin B12 inserts into the lipid bilayer after undergoing a profound structural reorganization [56–60]. Electron paramagnetic resonance analysis of the loop between Helices D and E in Domain II showed that this region refolded and formed a continuous amphipathic α-helix after calcium-independent binding to membranes at mildly acidic pH. At pH 4.0, this helix assumed a transmembrane topography, while at pH around 5.0–5.5, it was peripheral and approximately parallel to the membrane; this form was reversibly converted into the transmembrane helix by lowering the pH and returned to the surface upon increasing pH [61]. These observations suggest the presence of a proton-dependent switch in annexins that harbors the information to induce membrane insertion. This insertion could explain some of the physiological properties of these proteins, such as calcium channel activity, and could also underlie its pathway of secretion.

Annexin A13 deserves a special mention regarding calcium-independent binding to membranes. This protein is the founder and most ancient member of mammalian annexins [62]. A short “a” isoform was first identified as a gut-specific annexin highly similar to annexin A5 [63]. Later on, an alternative splicing form with an insertion of 41 residues at the N-terminus was described: The “b” isoform [64]. Another peculiar feature of this annexin is the presence of an N-terminal sequence that leads to the N-myristoylation of the terminal glycine residue. This post-translational modification is unique among vertebrate annexins and is present in both isoforms allowing the direct interaction of these proteins with biological membranes in a calcium-independent manner. Moreover, the myristoylated form of the A13b isoform interacts better with PC and vesicles with a raft-like composition than with acidic phospholipids [37]. In addition, annexin A13 can interact with acidic phospholipid bilayers in the presence of calcium as described for other members of the annexin superfamily. Although the core of the long A13b isoform is quite similar to that of annexin A5, its requirements for calcium binding are higher (around 10-fold) probably due to the interaction of the relatively long N-terminus with the protein core [37]. On the other hand, a truncated form of annexin A13b lacking the first 48 amino acid residues (quite similar to the annexin A13a isoform) presented a behavior almost identical to annexin A5 regarding calcium-dependent phospholipid binding [37].

In addition to their ability to interact with membranes, several vertebrate annexins are able to mediate vesicle aggregation, among them annexins A1, A2, A4, A6 and A7, whereas others, such as annexins A3, A5, A11 and non-myristoylated A13b, do not promote aggregation [2,25,37,65]. Interestingly, although mammalian annexin A5 lacks the ability to aggregate vesicles, we have reported that its highly similar chicken ortholog induces aggregation of vesicles containing acidic phospholipids even at low protein and/or calcium concentrations [66]. Chicken annexin A5 shows a high sequence and structural similarity with mammalian annexins absent in chicken, such as annexins A3 and A4. Thus, some of the physiological functions exerted by these proteins may be carried out by chicken annexin A5, even those that could require calcium-dependent membrane aggregation.

Vesicle aggregation is negatively regulated by phosphorylation of the N-terminal domain at serine/threonine and tyrosine residues [13,20,67,68] or by proteolysis of the N-terminal extension [69]. In addition, the aggregation process may finally lead to fusion processes as described for annexin A7 regarding its involvement in exocytosis of lung surfactant, catecholamines, and insulin [70]. Phospholipid composition and calcium sensitivity for this aggregation activity significantly varies among individual annexins [2,38,65]. Moreover, the lipid composition and organization regulates the Ca²⁺-sensitivity of certain annexins for membrane bridging by modulation of membrane fluidity [71].

How do annexins induce membrane aggregation? Several mechanisms have been proposed and are depicted in Figure 3: (A) self-association/dimerization of annexins bound to separate membranes; (B) establishment of heterotetramers formed by two annexins and two S100 proteins; or (C), the appearance of a second phospholipid-binding site after the interaction of specific annexins with membranes via the calcium-dependent type II sites located on the convex surface of these annexins.

Cryo-electron microscopy of aggregated lipid vesicles in the presence of wild-type annexin A4 shows a separation compatible with two layers of membrane-bound annexin A4 [20], in accordance with the first mechanisms described for vesicle aggregation. We have also demonstrated that chicken annexin A5 induces vesicle aggregation via establishment of protein dimers between molecules on different vesicles through the concave surfaces based on the kinetic analysis of the aggregation process together with cross-linking experiments at low-membrane occupancy and inhibition of vesicle aggregation by blockage of the concave annexin surface with heparin tetrasaccharide. Moreover, the essential role of the N-terminus of this protein for aggregation was put forward using chimeras with human annexin A5 (that does not aggregate vesicles) and truncation mutants [66].

Annexins A1 and A2 seem to be able to induce membrane aggregation via different mechanisms in which the N-terminal domain also plays an important role: (i) membrane bridging through heterotetramers (or octamers) with proteins of the S100 family (S100A11 and S100A10) [72,73], (ii) formation of dimers via interaction through the N-terminal domains of monomeric annexins, as suggested for annexin A4 and chicken annexin A5 [74], or (iii) interaction of one molecule with two adjacent membranes simultaneously [24,75,76]. Annexin A2 is more prone to form heterotetramers due to a constitutive dimeric active conformation of S100A10 under physiological conditions. However, at mildly acidic pH, this protein does not undergo the proposed structural rearrangement described for annexin B12 but experiments conformational changes that allow interaction with membranes in a calcium-independent manner. This leads to vesicle aggregation via exposure of a secondary hydrophobic binding site that allows membrane bridging as a monomer [76]. On the other hand, annexin A1 interacts with S100A11; this protein requires calcium binding to adopt an active conformation to form heterotetramers with annexin A1. The crystal structures of annexin A1 in the absence [22] and presence of calcium [24] (Figure 2) suggest a calcium dependent relocation of the N-terminal helices. Upon calcium binding, the N-terminal domain is exposed acquiring a very flexible conformation, as confirmed by molecular dynamics [77]. In the proximity of phospholipid membranes, the N-terminal domain refolds into α-helical structures that allow favorable direct interactions with the membrane providing a peripheral membrane anchor for vesicle aggregation [75].

Additional membrane aggregation mechanisms have been suggested for annexin A6. As we have previously commented, this protein is unique in the annexin superfamily as it consists of two annexin core modules connected by a 40 amino acid loop. Crystallographic data reveal that the two annexin cores are tilted with respect to each other in such a way that none of the two cores can bind to a membrane with its convex surface parallel to the membrane [78]. However, phosphorylation of a threonine residue in the connecting loop seems to increase the flexibility of the molecule, as deduced from the crystal structure and biochemical analysis of the phosphorylation-mimicking mutant T356D [79]. In fact, electron microscopic studies of membrane-bound annexin A6 2D crystals have determined that the two annexin cores reorient and become coplanar upon membrane binding [80,81]. More recently, and based on quartz crystal microbalance with dissipation monitoring, cryo-electron microscopy and atomic force microscopy techniques, a new model for the association of this annexin with membranes has been proposed. Authors suggest that at low calcium concentration (<150 μM), annexin A6 binds to membranes via the two coplanar annexin cores and cannot induce vesicle aggregation. However, at higher calcium concentration (~2 mM), a conformational switch allows annexin A6 to bind two adjacent phospholipid membranes due to the appearance of a dormant phospholipid binding sequence in the interconnecting loop, adopting a conformation similar to that observed in the crystal structures [82].