|Home | About | Journals | Submit | Contact Us | Français|
Certain cell surface receptors engage ligands expressed on juxtaposed cells, but also ligands on the same cell. The structural basis for trans versus cis binding is not known. Here, we show that Ly49 natural killer (NK) cell receptors bind two MHC class I (MHC-I) molecules in trans when the two ligand-binding domains are back-folded onto the long stalk region. In contrast, dissociation of the ligand-binding domains from the stalk and their reorientation relative to the NK cell membrane allow monovalent binding of MHC-I in cis. The distinct conformations (back-folded and extended) define the structural basis for cis–trans binding by Ly49 receptors and explain the divergent functional consequences of cis versus trans interactions. Further analyses identified specific stalk segments that were not required for MHC-I binding in trans, but were essential for inhibitory receptor function. These data identify multiple distinct roles of stalk regions for receptor function.
Cell surface receptors allow cell-to-cell communication by interacting with ligands expressed on other cells (trans interaction). Certain receptors can also engage ligands expressed on the same cell (cis interaction) (Held and Mariuzza, 2008). The molecular basis for cis versus trans binding is currently unknown. Here we have investigated the structural requirements for cis–trans interactions by Ly49 receptors, which allow natural killer (NK) cells to detect diseased cells.
NK cells are activated upon encounter with most normal cells. However, activation signaling is interrupted when inhibitory NK cell receptors engage major histocompatibility complex class I (MHC-I) molecules on other cells. This dual receptor system enables NK cells to detect virally infected or transformed cells with reduced levels of MHC-I molecules, i.e. cells which fail to trigger inhibitory MHC-I receptors. In this case NK cell activation can proceed and bring about target cell lysis, which is known as “missing-self” recognition.
Inhibitory MHC-I receptors include human leukocyte immunoglobulin (Ig)-like receptors (LILRs) and their orthologues, mouse paired Ig-like receptors (PIRs), killer Ig-like receptors (KIRs) (human), C-type lectin-like Ly49 receptors (mouse), and CD94–NKG2A receptors (human and mouse) (Lanier, 2005). Engagement of these receptors by MHC-I on prospective target cells leads to the recruitment of phosphatases such as SHP-1 (Burshtyn et al., 1996; Nakamura et al., 1997) and the adaptor molecule Crk (Peterson and Long, 2008), which counteract NK cell activation. Additionally, KIR and Ly49 receptors “educate” developing NK cells, i.e. they establish and maintain functional competence of NK cell activation pathways (Anfossi et al., 2006; Chalifour et al., 2009; Fernandez et al., 2005; Kim et al., 2005).
In addition to the inhibitory interaction with MHC-I on juxtaposed membranes, members of the Ly49 and LILRB/PIR receptor families can also bind MHC-I molecules in the plane of the same membrane (in cis) (Doucey et al., 2004; Masuda et al., 2007; Scarpellino et al., 2007). While cis-associated PIR-B delivers tonic inhibition signals (Masuda et al., 2007), there is no evidence that cis engagement of Ly49s results in inhibitory signalling (Doucey et al., 2004). Rather, a significant fraction of Ly49 receptors becomes unavailable for trans interaction (Back et al., 2007). Consequently, cis interaction reduces the inhibitory capacity of Ly49 and lowers the threshold at which the NK cells activation exceeds inhibition. In addition, cis interaction of Ly49A is required for NK cell education (Chalifour et al., 2009).
It is not known how cell surface receptors like Ly49 can bind MHC-I expressed in cis and trans, and why the two types of interactions have distinct functional outcomes. Ly49s are homodimeric type II glycoproteins, with each chain composed of a ligand-binding C-type lectin-like domain, termed the natural killer receptor domain (NKD), connected by a stalk of approximately 70 residues to the transmembrane and cytoplasmic domains (Deng and Mariuzza, 2006; Natarajan et al., 2002). Crystal structures of Ly49–MHC-I complexes have shown that Ly49s engage MHC-I at a broad cavity beneath the peptide-binding platform formed by the α1, α2 and α3 domains, and β2-microglobulin (β2m) (Dam et al., 2003; Deng et al., 2008; Tormo et al., 1999). Because trans and cis interactions utilize the same binding site (Doucey et al., 2004), Ly49 receptors must likely reverse the orientation of their ligand-binding domains relative to the NK cell surface in order to bind MHC-I in trans versus cis. If so, the exceptionally long stalk region of Ly49s, for which there is currently no structural information, may be crucial to support two orientations of the NKDs. In addition, structures of Ly49–MHC-I complexes revealed two modes of MHC-I engagement. In the Ly49C–H-2Kb complex (Dam et al., 2003; Deng et al., 2008), the Ly49C homodimer engages H-2Kb bivalently, such that each NKD makes identical interactions with MHC-I to form a symmetrical, butterfly-shaped assembly. By contrast, in the Ly49A–H-2Dd complex (Tormo et al., 1999), the Ly49A dimer contacts H-2Dd asymmetrically, with only one of its two subunits binding a single MHC-I molecule. The very different modes of MHC-I engagement are based on different geometries of the Ly49C and Ly49A dimers. The Ly49C dimer adopts an “open” conformation, while the Ly49A dimer adopts a “closed” conformation. Steric clashes preclude the closed Ly49A dimer from binding simultaneously two MHC-I molecules. However, a nuclear magnetic resonance study of unbound Ly49A revealed that in solution the receptor exists predominantly in the “open” state and that this form of Ly49A can bind two MHC-I molecules (Dam et al., 2006). Collectively, these data raise the possibility that cis–trans interactions are mediated by distinct Ly49 conformations (Held and Mariuzza, 2008).
To elucidate the basis for cis versus trans engagement of MHC-I by Ly49s, we determined the crystal structure of Ly49L in the absence and presence of a significant portion of the stalk region. This information was used to construct, and experimentally test, models of cis and trans Ly49–MHC-I complexes. This study, which represents the first comprehensive structure–function analysis of receptor interactions in cis versus trans, highlights how stalk regions can play essential roles in the function cell surface receptors.
The structure of Ly49L NKD was determined to 2.0 Å resolution (Table S1, Figure 1A). Like other Ly49 NKDs (Dam et al., 2003; Deng et al., 2008; Tormo et al., 1999), the Ly49L NKD adopts a fold consisting of two α-helices (α1 and α2) and two anti-parallel β-sheets (β0, β1, β5 and β2, β2′, β3, β4), including four intramolecular disulfide bonds (Figure 1A,D). The two monomers in the asymmetric unit form a dimer that closely resembles the Ly49A dimer (Figure 1B,E) (Tormo et al., 1999).
Superposition of Ly49L NKD on other Ly49 NKD structures (Deng et al., 2008; Tormo et al., 1999) yielded root-mean-square (r.m.s.) differences of 0.9 Å, 1.2 Å and 2.2 Å for Ly49A, Ly49G and Ly49C, respectively, suggesting that Ly49L is more closely related to Ly49A or Ly49G than to Ly49C. Indeed, although Ly49L is an activating receptor, it is structurally more similar to the inhibitory receptor Ly49A (or Ly49G) than is Ly49A to Ly49C, another inhibitory receptor. We therefore conclude that activating and inhibitory Ly49s are distinguished not by the structure of their NKDs, but by whether their cytoplasmic regions contain ITIM motifs or are associated with DAP12, respectively. The main structural difference between Ly49L and Ly49C is in loop L3, which contacts MHC-I in the Ly49A–H-2Dd and Ly49C–H-2Kb complexes and is a key determinant of MHC binding specificity (Deng et al., 2008). This loop is continuous in Ly49L, Ly49A and Ly49G (Figure 1A,B), but is interrupted by an α-helix (α3) in Ly49C (Figure 1C,F).
At the NK cell surface, Ly49 receptors exist as homodimers (Natarajan et al., 2002). In the crystal, the Ly49L dimer shows a similar subunit arrangement as Ly49A in the Ly49A–H-2Dd complex (Tormo et al., 1999), which we term the “closed” conformation, since the α2 helices of the NKDs are juxtaposed across the dimer interface (Figure 1A,B). In addition, the two subunits interact through strand β0, creating an extended anti-parallel β-sheet, with six main chain–main chain hydrogen bonds linking the strands (Figure S1A). This conformation is distinct from the “open” conformation of Ly49C and Ly49G (Dam et al., 2003; Deng et al., 2008), in which the α2 helices make no contacts (Figure 1C).
Previous structural studies of Ly49 receptors have not provided any information on the stalk region connecting the NKD to the cell membrane (Held and Mariuzza, 2008). To visualize the stalk, and define its orientation relative to the NKD, we expressed various versions of the extracellular portion of several Ly49s, including Ly49A, Ly49C and Ly49L. However, only the extracellular portion of Ly49L (Ly49L-EC; residues 79–265), comprising the NKD and most of the stalk, could be crystallized. The structure was determined to 2.5 Å resolution (Table S1, Figure 2A). The four monomers in the asymmetric unit (designated A–D) form two homodimers (AB and CD). Superposition of the four monomers gave r.m.s. deviations in α-carbon positions of 0.5–0.8 Å, indicating close similarity. Accordingly, the following description of Ly49L-EC is based on monomer A, and that of the Ly49L-EC homodimer on monomers A and B, unless stated otherwise.
Although SDS-PAGE analysis of dissolved crystals confirmed the presence of an intact stalk, we could only detect electron density for the C-terminal 41 residues of the 68 stalk residues in the expressed protein, implying mobility of the N-terminal portion. In each Ly49L-EC monomer, the stalk is composed of an α-helix (α3S; residues Arg111–Lys132) and a loop (LS; residues Thr133–Gly144) connecting the helix to the NKD (Figure 2). However, rather than projecting from the NKD as might be expected for a stalk region, the Ly49L stalk is back-folded onto the NKD. In the Ly49L-EC dimer, the stalks are linked by an N-terminal disulfide bond (Cys110–Cys110) (Figure 2A). The α3S helices of the stalks lie in long grooves at the junction between NKD subunits (Figure 2B), such that each helix makes numerous interactions with both NKDs (Table S2). By contrast, the α3S helices do not contact each other, except at their N-termini (Figure 2C). Hence, the α3S helices may be regarded as two blades of an open pair of scissors, whose pivot is the Cys110–Cys110 disulfide.
Parts of loop LS, which links α3S to the NKD, are disordered in monomers B and D in the asymmetric unit of the Ly49L-EC crystal. Moreover, LS adopts significantly different conformations in monomers A and C, where it is completely ordered (r.m.s. deviation of 2.5 Å for 12 α-carbon pairs). Thus, the LS loop, along with the Cys110–Cys110 pivot, may confer flexibility to the stalk region at specific points that could be important for Ly49 function. Based on these results, we carried out sequence comparisons and secondary structure predictions of Ly49 stalk regions (residues 110–144) (Figure S2A). This analysis revealed that Ly49L likely exemplifies other Ly49s, both in the structure of the stalk region and its disposition relative to the NKD. First, residues 111–132 are predicted to be α-helical in all cases, in excellent agreement with the Ly49L-EC crystal structure where they compose helix α3S. Second, residues 133–144 are predicted as random coil in all Ly49s, again consistent with the Ly49L-EC structure in which they form the flexible LS loop. Third, Cys110 is conserved in 21 of 23 Ly49 family members (the sole exceptions, Ly49F and Ly49S, contain serine at this position), suggesting that the α3S helices are typically disulfide-linked at their N-termini. Fourth, examination of the Ly49L-EC structure showed that NKD-contacting residues of the stalk region are well conserved among Ly49s, as are the NKD residues with which they interact. For example, of the 15 α3S residues that contact the Ly49L NKD, 11 are identical in Ly49A and one is substituted conservatively (Lys/Arg122). In addition, a number of interaction pairs are strictly, or highly, conserved across the entire Ly49 family, including NKD Trp149–α3S Thr131, NKD Tyr152–α3S Glu/Asp123, NKD Glu188–α3S Glu/Asp123, NKD Phe191–α3S Leu/Ile120, and NKD Phe191–α3S Leu/Ser116 (Table S2). Hence, it appears probable that stalk regions can generally back-fold onto Ly49 NKD dimers as observed in the Ly49C-EC structure (Figure 2).
The most notable difference between the Ly49L NKD and Ly49L-EC structures, apart from the presence of the stalk region in the latter, is the relative orientation of the NKD subunits composing the two homodimers. As indicated above, the Ly49L NKD dimer adopts a closed conformation (Figure 1A). By contrast, the Ly49L-EC dimer assumes an open conformation, such that the α2 helices do not interact (Figure 2A). The closed conformation is incompatible with a back-folded stalk, since the α3S helices of the stalks occupy the gap between the α2 helices of the NKDs, effectively opening up the Ly49L dimer by ~20° compared to the closed state. Concomitantly, the register between β0 strands in the Ly49L4 dimer interface is shifted by four residues, reducing the number of main chain–main chain hydrogen bonds between the strands from six to four (Figure S1A,B). Thus, at least some Ly49s can adopt both open and closed states, which may correlate with trans and cis binding to MHC-I (see below).
A critical question is whether the open dimer observed in the Ly49L-EC structure, with its back-folded stalk region, is compatible with MHC-I recognition. Because the MHC specificity of Ly49L is unknown, we cannot measure ligand binding. However, the dimerization mode of Ly49L-EC closely resembles that of Ly49C in the Ly49C–H-2Kb complex (Dam et al., 2003), with an r.m.s. difference in α-carbon positions of 1.71 Å between the two Ly49 dimers. Least-squares superposition of the Ly49L-EC and Ly49C–H-2Kb structures showed that Ly49L, like Ly49C, could engage two MHC-I molecules, with some potential contacts, but no significant steric collisions, between the back-folded stalks and MHC-I (Figure 3A).
We propose a model for cis–trans engagement of MHC-I by Ly49s that integrates all the available crystal structures (Dam et al., 2003; Deng et al., 2008; Tormo et al., 1999). Because trans and cis interactions utilize the same binding site beneath the peptide-binding platform of MHC-I (Doucey et al., 2004), Ly49 receptors must drastically reorient their NKDs relative to the NK cell membrane to bind MHC-I in trans versus cis. It is most likely that the exceptionally long stalk regions of Ly49s provide the requisite flexibility. In the modeled Ly49–MHC-I complex constructed using the Ly49L-EC and Ly49C–H-2Kb structures (Figure 3A), the N-termini of the stalks point in a direction completely opposite from the C-termini of the MHC-I molecules. In this view, the two MHC-I stand on the target cell surface at the bottom and the Ly49 dimer reaches MHC-I from an opposing NK cell above, to which it is tethered via back-folded stalks. Accordingly, we propose that the Ly49L-EC structure represents the conformation Ly49s assume to bind MHC-I in trans. Conversely, cis binding would require the stalks to assume an extended conformation that orients the NKDs with their N-termini towards the NK cell (Figure 3B). In contrast to the trans interaction, in which one Ly49 dimer binds two MHC-I molecules (Figure 3A), this model predicts that cis engagement of both NKDs by MHC-I is unlikely, due to the orientation that binding of one MHC-I molecule would impose on the Ly49 dimer (Figure 3B). If so, the bivalent Ly49C–H-2Kb and monovalent Ly49A–H-2Dd complexes would exemplify trans and cis recognition, respectively.
Since the biological relevance of bivalent and monovalent MHC-I binding was unclear, we determined the stoichiometry of Ly49–MHC-I complexes on cell surfaces. Because the MHC-I specificity of Ly49L is not known, we used the highly homologous Ly49A receptor, which binds H-2Dd (Dd) but not Db or Kb molecules, both in trans (Karlhofer et al., 1992) and in cis (Doucey et al., 2004). To determine the stoichiometry of Ly49A–Dd cis complexes in situ, we exposed C1498 transfectants to the cell impermeable cross-linker Bis(Sulfosuccinimidyl)suberate (BS3). Following Ly49A immunoprecipitation (i.p.) and SDS-PAGE under reducing conditions, we detected a predominant band of 80–90 kD (Figure 4A), which corresponds to the Ly49A homodimer (Figure S3A) and an additional band of 140–150 kD, which included both Ly49A (Figure 4A) and Dd (Figure 4B). A corresponding complex of 140–150 kD was observed following Dd i.p. (Figure S4A) but was absent when Ly49A cells did not co-express Dd or in the absence of BS3 treatment (Figure 4A). The molecular weight of 140–150 kD suggests a complex of one Ly49A homodimer (80–90 kD) (Figure S3A) linked to a single Dd heavy chain (45 kD) with or without the β2m light chain (10 kD) (Figure S4C). Importantly, there was essentially no evidence for Ly49A–Dd complexes of >150 kD, which would be expected if one Ly49A homodimer associated with two Dd molecules (predicted size ~200 kD). We conclude that Ly49A and Dd are constitutively associated on living cells and that cis complexes consist of an Ly49A homodimer associated with a single Dd molecule, in agreement with the structural model (Figure 3B).
The stoichiometry of Ly49A–Dd trans binding was assessed using soluble Ly49A (Figure 5A). Soluble Ly49A dimers, obtained by fusing the Fc region of human IgG1 to most of the extracellular portion of Ly49A (Figure S5A), specifically bound Dd transfected C1498 cells (Figure 5B). After cross-linking soluble Fc-Ly49A to Dd transfectants and Fc-Ly49A i.p. we detected Dd-containing complexes of ~200 and ~100 kD (Figure 4C). These correspond to complexes of one Fc–Ly49A dimer (~110 kD) (Figure S5A) associated with two Dd molecules (90–110 kD), and of a single Fc–Ly49A chain (~55 kD) (Figure S5A) linked to a single Dd heavy (with or without the β2m light chain (45–55 kD) (Figure S5A). There was essentially no evidence for complexes of ~160 kD, which would be expected if a Fc–Ly49A homodimer (~110 kD) was cross-linked to a single Dd molecule (45–55 kD). These data indicate that trans binding (by soluble Ly49A dimers) occurred in a bivalent fashion, whereas cis binding was monovalent.
The Ly49L-EC structure revealed that the NKDs make numerous contacts with the α3S helices of the stalk (Figure 2, Table S2), and that many of these interactions are probably conserved across the Ly49 family (see above). If the Ly49L-EC structure indeed represents a trans-binding conformation (Figure 3A), disruption of NKD–α3S interactions should interfere with trans, but not cis, binding. In addition, the long Ly49 stalk should be essential for trans binding, to allow the Ly49 NKDs to fold back onto the stalk, but less critical for cis binding, as the NKDs are dissociated from the stalk (Figure 3). To test these predictions, we deleted known and/or predicted α-helical segments (αS) from the Ly49A stalk (Figure 5A, Figure S2A,B). The resulting Ly49A stalk deletion variants were stably introduced into C1498 (H-2b) cells, where they were detected using Dd tetramers and Ly49A mAb JR9 (Table 1) or A1 (not shown), which indicated proper folding of the NKDs.
A cell-cell adhesion assay was used to determine whether the Ly49A stalk deletion variants bound membrane-anchored Dd. Wild type Ly49A mediated efficient adhesion to Dd+ cells (>50% conjugates) (Figure 5D, Table 1). This was based on a specific interaction since conjugate formation was inefficient in the absence of Ly49A or Dd or both (~10% conjugates) (Figure 5D, Table 1). Compared to wild type, the deletion of the α3S segment from the Ly49A stalk (Δα3S) significantly reduced adhesion (Table 1). This was not likely due to the reduced length of the stalk, since adhesion was efficient when the α1S or the α2S segments were deleted (Δα1S and Δα2S) (Table 1). A need for α3S for trans binding was confirmed using soluble Ly49A: While soluble receptors that contained α3 S (Δα1S and Δα1-2S) stained Dd-expressing cells (Figure 5B), there was no binding in the absence of α3S (Δα1S Δ α3S) (Figure 5B). Importantly, this Δα3S receptor reacted with Ly49A mAb (Figure S5B), indicating that the NKDs were properly folded. We conclude that the α3S segment plays a role for trans binding, consistent with a requirement for back-folding of the NKDs onto α3S.
When the Ly49A stalk was shortened further, by deleting two α-helical segments (Δα1-2S or ΔαS2-3S), cell-cell adhesion was completely lost (Table 1). However, the soluble version of the Δα1-2S receptor efficiently stained Dd+ cells (Figure 5B). These data suggest that it the attachment of the Δα1-2S to a cell membrane precludes ligand binding. This outcome is expected if trans recognition is mediated by a receptor conformation in which the NKDs are back-folded onto the stalk. In this case, steric clashes with the cell membrane prevent MHC-I binding.
We next addressed whether cis binding showed a corresponding dependence on the length of the stalk and/or on the presence of α3S. Cell-cell adhesion mediated by wild type Ly49A was considerably reduced when Ly49A was co-expressed with Dd (Figure 5D, Table 1). Cis Dd masks a large fraction of wild type Ly49A and this prevents cis-engaged Ly49A from interacting with Dd in trans (Back et al., 2007). Similar to wild type, Dd expression in cis reduced adhesion by Ly49A variants, which lacked a single αS element (Δα1S, Δα2S or Δα3S) (Table 1), indicating that all these variants bind Dd in cis. Corresponding experiments using Δα1-2S or Δα2-3S receptors were not informative, as these variants did not mediate specific adhesion.
Cis binding was further tested by comparing Dd tetramer binding to Ly49A before and after disrupting MHC-I complexes (Doucey et al., 2004; Scarpellino et al., 2007). A brief exposure of live cells to an acidic buffer disrupted trimolecular MHC-I complexes, as judged by the complete loss of the β2m light chain from the cell surface (Figure 5E). Acid treatment of control (H-2b) transfectants did not alter Ly49A detection using mAb or Dd tetramer (Figure 5E), suggesting that Ly49A is not acid-sensitive and is not significantly masked by Kb or Db molecules. Co-expression of Ly49A and Dd significantly reduced Dd tetramer binding and this was reversed by acid treatment (Figure 5E, Table 1) (Doucey et al., 2004; Scarpellino et al., 2007). Likewise, acid stripping considerably improved Dd tetramer binding to all Ly49A receptor variants lacking a single α-helical stalk segment (Table 1). Acid treatment also increased tetramer binding to Ly49A Δα1-2S and Δα2-3S receptors (Table 1), indicating that very short Ly49A variants were masked by cis Dd. The physical association of Dd with Ly49A variants lacking one, and even two, αS elements was confirmed using co-i.p. (Figure 5F and data not shown). Thus, very short Ly49A variants, which retain a single αS segment, can bind Dd expressed in cis but not in trans. This outcome is expected if trans binding requires back-folded and cis binding extended Ly49 conformations, respectively (Figure 3). Consistent with this interpretation, α3S is not essential for cis binding suggesting that back-folding of the NKD is not required for cis interaction.
A prominent feature of the Ly49L-EC structure is the 12-residue LS loop, which links α3S to the NKD (Figure 2). This apparently flexible loop may allow the NKD to back-fold onto the stalk for trans binding to MHC-I (Figure 3A) or it may be needed for the NKD to assume an extended conformation for cis binding (Figure 3B) or it may contribute to both trans and cis interactions. To distinguish among these possibilities, we deleted the LS loop (aa 128–137; ΔLS) (Figure S2B) from full-length Ly49A. The ΔLS receptor mediated efficient conjugation with Dd+ cells (Table 1) and a soluble version of this receptor efficiently stained Dd+ cells (Figure 5C), indicating that the LS loop is not essential for trans binding. Based on the Ly49L-EC structure, the α-carbons of Thr127 and Gly138 are estimated to be 9.5 Å apart in wild type Ly49A. This distance should allow Thr127 and Gly138 to be connected in the ΔLS mutant through minor structural adjustments of residues Gly138–Lys140, without disturbing the back-folded conformation. While trans binding was intact, the ΔLS receptor failed to bind Dd in cis (Table 1). These data suggest that LS allows Ly49A to adopt an extended conformation needed to engage MHC-I on the same cell.
To determine whether the role of LS is mediated by a specific amino acid sequence, we replaced the wild type loop with a glycine-rich sequence (G-LS) of equal length (Figure S3B), which should confer a high degree of flexibility to the NKDs. In contrast to removal of the loop, its replacement diminished adhesion to Dd+ cells (Table 1). In addition, staining of Dd+ cells using a soluble version of the G-LS receptor was very inefficient (Figure 5C), although the fusion protein reacted with Ly49A mAb (Figure S5C,D), implying proper folding of the NKDs. Even though the deletion of LS was compatible with trans binding, its replacement interfered with trans binding. This indicated that trans binding requires a specific amino acid sequence. Despite impaired trans binding, the G-LS variant showed a significant, albeit limited, ability to bind Dd in cis (Table 1). Therefore, a glycine-rich loop can rescue ΔLS with respect to cis binding, indicating that the LS loop needs to provide spacing and/or flexibility to allow cis binding. Overall, the data show that the LS loop region is essential for cis but not for trans binding.
Our mutational analyses identified elements of the Ly49A stalk, notably αS1, αS2 and LS, that were not required for Ly49A–Dd trans binding. To address whether these elements were needed for receptor function, we introduced the respective stalk deletion mutants into the rat NK cell line RNK (Figure 6A). RNK lines produced IFNγ upon stimulation with CD161 (NKRP1) mAb (Figure S6). Co-crosslinking of CD161 and Ly49A (using mAb A1, which binds the Ly49A NKD) inhibited IFNγ production by RNK cells expressing wild type Ly49A as well as those expressing stalk mutant receptors (Figure S6). This demonstrated that the mutant receptors are competent to transduce inhibitory signals.
Next we tested function of mutant Ly49A receptors using cellular targets. Parental RNK cells killed untransfected and Dd+ YB2/0 target cells (Figure 6B). Introduction of wild type Ly49A into RNK cells strongly inhibited lysis of Dd+ targets (Figure 6B), as shown previously (Nakamura et al., 1997). As expected, lysis inhibition was not observed with the very short Ly49A stalk variants ΔαS1-2 (not shown) or ΔαS2-3 (Figure 6B), which did not bind Dd in trans (Table 1). Unexpectedly, lysis inhibition was also not observed when a single α-helical segment was deleted from the Ly49A stalk (ΔαS1 or ΔαS2) (Figure 6B,D), despite the fact that such variants mediated efficient adhesion to Dd+ cells (Table 1) and that ΔαS1 was signaling competent (Figure S6). To determine whether deficient function was due to a reduced length of the stalk or some other role of α1S and/or α2S, we replaced α1S with sequence from α2S (α1S -> α2S). We also replaced the α1-2S region with heterologous sequence from the CD72 stalk which, like the Ly49A stalk, is predicted to be α-helical (α1-2S -> CD72) (Figure S2B). Both replacements resulted in efficient lysis inhibition (Figure 6C,D). Corresponding results were obtained with RMA target cells expressing the NKG2D ligand H60 (not shown). We conclude that an optimal length of the stalk, and not some other function of the α1S or α2S segments, is crucial for lysis inhibition by Ly49A.
We also tested lysis inhibition by loop region modified Ly49A variants. The G-LS receptor did not inhibit lysis (Figure 6B), which was expected based on its reduced trans binding capacity (Table 1). Unexpectedly, the ΔLS variant did also not inhibit lysis (Figure 6B). Despite the fact that the LS loop is not required for MHC-I binding (Table 1) or inhibitory signaling (Figure S6), it is essential for the inhibitory function of Ly49A.
To further address the basis for defective function of certain stalk modified receptors, we investigated the redistribution of Ly49A to the immunological synapse, which correlates with inhibitory function (Back et al., 2007). Like synapses formed by primary NK cells (Back et al., 2007), wild type Ly49A accumulated at the site of contact between RNK and Dd-transfected but not untransfected target cells (Figure 6E). Even though the inhibitory capacity of the ΔαS1 and ΔαS1 receptors was impaired (Figure 6B,D), they redistributed efficiently to the immune synapse (Figure 6E). Conversely, we did not observe an obvious accumulation of the ΔLS receptor at the NK cell synapse (Figure 6E), which is in agreement with the deficient inhibitory function of this mutant (Figure 6B). Notwithstanding the adhesive function of this variant (Table 1), low amounts of ΔLS receptor were detectable at the contact zone (Figure 6E). Corresponding data were obtained when inspecting cellular interfaces between Ly49A and Dd transfected C1498 cells (not shown). Staining for an epitope tag present in the intracellular portion ruled out the possibility that ΔLS was somehow inaccessible to Ly49A mAb at the cellular interface (not shown). The LS loop is thus essential for post-ligand binding events, which allow the redistribution of receptor-ligand complexes to the NK cell synapse and mediate inhibitory receptor function.
Due to technical difficulties associated with crystallizing cell surface receptors bearing long stalks (e.g. CD8), X-ray crystallographic studies of such receptors have largely been restricted to their globular ligand binding domains. However, there is increasing evidence that stalk regions may have specialized roles in receptor function, besides simply connecting the ligand binding domains to the membrane. One striking example is the stalk region of the T cell co-receptor CD8, which undergoes developmentally programmed O-glycan modification (Moody et al., 2001). Decreased glycosylation of the CD8 stalk during thymic development is associated with increased affinity for MHC-I and improved T cell signaling, thereby affecting T cell selection. However, the basis for differential binding is not understood, as no three-dimensional structural information is available for the CD8 stalk. Likewise, the structural basis for cis versus trans recognition of MHC-I by inhibitory immunoreceptors such as Ly49s was unknown prior to this study (Held and Mariuzza, 2008).
Here we provide structural, biochemical and functional evidence that cis–trans interactions of Ly49 NK cell receptors with MHC-I are mediated by two distinct receptor conformations. Based on the Ly49L-EC structure, we proposed that NKDs must back-fold onto the α3S region of the stalk to mediate trans interaction. This model was validated in several independent ways, including deletion of the entire α3S segment from the Ly49A stalk. This resulted in greatly reduced trans binding of Dd by the Δα3S variant in a cell-cell adhesion assay, an effect that was not observed when the α1S or α2S segment was removed. In addition, trans binding by membrane-anchored (but not soluble) Ly49A required the presence of at least two αS segments, in agreement with our structural model in which steric clashes with the NK cell membrane would preclude trans binding of MHC-I to a back-folded receptor lacking α1S and α2S (Figure 3A). We further showed that (soluble) Ly49A could bind two membrane-bound Dd molecules in trans, as predicted by the model. These findings indicate that trans binding is mediated by a receptor conformation in which the NKDs are back-folded onto the stalk. This model also explains early domain swap experiments, which indicated that the specificity of Ly49C was altered when the C-terminal part of its stalk was replaced with that of Ly49A (Brennan et al., 1996). We propose that the predominant (or default) conformation of Ly49 receptors in the absence of cis MHC-I is back-folded, since Ly49A on NK cells from H-2b mice efficiently binds Dd in trans and strongly inhibits effector function (Doucey et al., 2004; Karlhofer et al., 1992).
According to our structural model for cis–trans interactions, cis binding of MHC-I requires Ly49 to assume an extended conformation in which the α3 segments of the stalk do not contact the NKDs (Figure 3B). Indeed, we showed that α3S, which is essential for binding in trans, was not required for binding in cis, providing evidence that the NKDs are disengaged from the stalk during cis interactions. Further, when the stalk region was progressively shortened, trans binding was impaired before cis binding, as expected if cis binding is mediated by an extended receptor. Moreover, we showed that, on living cells, the Ly49A–Dd cis complex consists of an Ly49A homodimer associated with a single Dd molecule, as predicted by the model (Figure 3B). Collectively, these results provide strong experimental evidence that cis and trans interactions are mediated by extended and back-folded conformations, respectively, of the Ly49 receptor.
Mutational analysis of the flexible LS loop, which links α3S to the NKD, produced insights into Ly49 function beyond those predicted by the structural model. Although deletion of LS did not interfere with trans binding, the loop was required for cis binding, most likely by permitting Ly49 to switch from the back-folded to extended conformation. This particular function of LS was sequence-independent, since the loop could be replaced by a glycine-rich sequence. Unexpectedly, even though the LS loop was not needed for MHC-I binding in trans or the transduction of inhibition signals, LS was nevertheless essential for lysis inhibition. Typically, the engagement of cell surface receptors induces intracellular signaling via two principle mechanisms: ligand-induced structural changes in the receptor and/or oligomerization of receptor–ligand complexes (Schwartz et al., 2002). We previously showed that Ly49C in complex with H-2Kb retains nearly the same structure as the free receptor (Deng et al., 2008), arguing against MHC-induced conformational changes in Ly49 as a signaling mechanism. On the other hand, zinc-dependent multimerization of KIR is thought to promote the formation of clusters of KIR and HLA-C molecules at the NK cell immune synapse, resulting in inhibitory function (Boyington et al., 2000). Similarly, Ly49 signaling may depend on clustering of ligand-engaged Ly49 dimers. Although it is not immediately obvious how lateral association of Ly49–MHC-I complexes might occur, our data suggest that LS could contribute to clustering, since the loop deficient receptor failed to accumulate at the immune synapse. In the modeled bivalent Ly49–MHC-I trans complex (Figure 3A), the LS loop is on the exterior of the complex, fully exposed to solvent, such that it could potentially mediate lateral interactions with other cell surface molecules or adjacent bivalent Ly49–MHC-I trans complexes. A analogous ligand-induced, receptor-mediated lateral association is essential for the activation of the epidermal growth factor receptor (EGFR) (Ogiso et al., 2002).
The analysis of receptors lacking a single αS segment (Δα1S and Δα2S) revealed an additional discrepancy between ligand binding and inhibitory receptor function. In contrast to deletion of the LS loop, receptors with a shortened stalk did accumulate efficiently at the immune synapse. Lysis inhibition was restored when the α1S or the entire α1-2S region was replaced with a heterologous stalk. This demonstrated that a minimal length of the stalk, and not some other function of the α1S or α2S segments, was critical for inhibitory Ly49 receptor function. It is thought that inhibitory receptors must be co-engaged with activating NK cell receptors in submicroscopic clusters in order to antagonize activation signaling (Treanor et al., 2006). According to the concept of size-based segregation of cell surface molecules (Choudhuri et al., 2005; Velikovsky et al., 2007), activating and inhibitory receptor–ligand pairs should fit into the narrow synaptic cleft of 100–150 Å between the NK cell and the target cell membrane. Activating and inhibitory interactions may thus become mutually exclusive when the Ly49 stalk is too short. For example, when the activating NKG2D receptor engages its H60 ligand (which was the case when RNK cells were used against RMA H60 targets), the shortened inhibitory receptor may be unable to reach across to Dd molecules on target cells. Consequently, lysis would not be inhibited. Conversely, when the shortened inhibitory receptor interacts with ligand (consistent with physical interaction), the activating receptor–ligand pair may not fit into the narrowed synaptic cleft between the two juxtaposed membranes. Such a microcluster would not contribute to activation signaling.
We propose that other receptors acting in both cis and trans undergo large structural rearrangements analogous to those of Ly49. Similar to Ly49, cis–trans binding by LILRB2 and PIR-B receptors shows broad MHC-I specificity (Masuda et al., 2007) and is thus likely mediated by the same binding site. If so, the ligand binding D1 and D2 domains of LILRB2 would need to reverse direction with respect to the effector cell surface to bind MHC-I in cis. This would require that LILRB2 bends back on itself and adopts a horseshoe-shaped configuration of the four Ig domains (D1–D4). Such a large reversal implies considerable flexibility in the segment connecting the D2 with the D3 domain. Consistent with this possibility, the four N-terminal Ig-like domains of the Drosophila Dscam protein indeed assume a horseshoe arrangement, which is made possible by a five-residue hinge between D2 and D3 (Meijers et al., 2007). In the case of PIB-B, which has two additional membrane proximal Ig-like domains compared to LILRB2, the D4–D5 or D5–D6 connecting segments might provide further flexibility. Therefore, the ability to reverse orientation may be a general feature of receptors binding in cis and trans, involving at least two structurally distinct strategies: exceptionally long stalk regions in the case of Ly49s and very flexible, or multiple, interdomain hinges in the case of LILRB2 and PIR-B.
In summary, based on crystal structure determinations, mutational analysis and functional assays, we have shown that the Ly49 stalk region mediates several essential functions. First, the stalk allows Ly49 receptors to reorient their NKDs relative to the NK cell membrane in order to engage MHC-I in trans (back-folded conformation) or cis (extended conformation). Second, the stalk provides sufficient distance between the NKDs and the NK cell membrane for trans binding of MHC-I, which requires a minimum of two αS segments. Third, the stalk permits optimal spacing between the NK cell and target cell membranes for productive Ly49–MHC-I interactions at the immune synapse, which requires all three αS segments. Fourth, the LS loop is essential for efficient accumulation of Ly49–MHC-I complexes at the NK cell synapse, which is necessary for inhibitory function. Together, these results demonstrate that stalk regions can have multiple specialized roles in receptor function, well beyond simply attaching the ligand-binding domains to the cell membrane.
The extracellular portion of Ly49L from mouse strain C3H (Ly49L-EC; residues 79–265), comprising the NKD and most of the stalk region, was cloned into the vector pT7-7 (Novagen). The protein was expressed in Escherichia coli as inclusion bodies, solubilized in 6 M guanidine, and folded in vitro by dilution into 30% (volume/volume) glycerol, 0.4 M arginine, 3 mM reduced glutathione, and 0.8 mM oxidized glutathione. Ly49L was purified with sequential MonoS and Superdex 75 HR columns (Amersham Biosciences). The NKD of Ly49L (residues 138–265) was prepared similarly. Details on crystallization, data collection, structure determination, and secondary structure predictions are provided in Supplemental Experimental Procedures.
The Ly49A stalk variants described herein (Figure S2B) were generated and expressed using procedures described in the Supplementary Methods. Acid treatment of stable transfectants (Doucey et al., 2004) and cell-cell adhesion assays (Back et al., 2007) were described before.
Cells were surface stained with mAbs to H-2Dd (34-2-12), Ly49A (JR9-318) or β2m (Ly-m11) (BD Pharmingen). Binding of soluble Fc–Ly49A fusion proteins was revealed using PE-conjugated anti-human IgG (Jackson Lab). Dd-HIV-mouse β2m tetramers were produced as described before (Scarpellino et al., 2007). Three-color flow cytometry was performed using a FACScan flow cytometer (Becton Dickinson) and Flowjow (Tree Star, Ashland, OR) software for data evaluation.
C1498 transfectants (107) were incubated for 30 min with 0.25 mM of BS3 (Bis(Sulfosuccinimidyl) suberate (BS3) (Pierce)) in 1 ml of PBS at 4 °C. In some experiments, Dd transfectants cells were reacted with soluble Fc–Ly49A (2 μg/107 cells) prior to cross-linking using decreasing concentrations of BS3 (0.25, 0.05 and 0.01 mM). Excess cross-linker was quenched by washes in TBS. The cells were lysed at 4 °C in Tris buffer (50 mM Tris pH 7.5, 150 mM NaCl, 1% NP40) and lysates were immunoprecipitated using VSV-agarose or Protein G Sepharose (Sigma). Immunoprecipitates were separated using reducing SDS-PAGE, transferred to nitrocellulose membranes and analysed using anti-HA (3F10, Roche) or anti-VSV (P5D4, Sigma) Western blot as described (Doucey et al., 2004).
RNK cells expressing Ly49A were mixed with YB2/0 Dd-YFP cells at a 1:1 ratio, centrifuged and incubated for 8–10 min at 37 °C. After resuspending, conjugates were adhered for 1–2 min to poly-L or poly-D lysine coated slides before fixation for 10 min using cold acetone. Ly49A was stained using mAb JR9-318 followed by Alexa 568-conjugated goat-anti mouse IgG Ab (Molecular Probes). Confocal microscopy and image quantification have been described before (Back et al., 2007).
All p values were determined using a two-tailed student's t-test with equal sample variance. Data sets were considered significantly different when p<0.05.
Coordinates and structure factors for Ly49L NKD and Ly49L-EC have been deposited in the Protein Data Bank under accession codes 3G8K and 3G8L, respectively.
This work was supported in part by grants from the Swiss National Science Foundation and Oncosuisse to W.H. and by the National Institutes of Health (AI47990 to R.A.M.). E.L.M. is supported by the Fogarty International Center (TW007972). We are grateful to P. Guillaume (Ludwig Institute) for tetramers, F. Lévy (Ludwig Institute) for MHC-I Abs, S.K. Anderson (National Cancer Institute) and A.P. Makrigiannis (Clinical Research Institute of Montreal) for Ly49L cDNA, and Howard Robinson (Brookhaven National Synchrotron Light Source) for X-ray data collection.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.