Introduction

Human natural killer (NK) cells play an important role in immune defense and reproduction (1–3). They respond to alterations in healthy cells caused by cellular stress, infections, and malignancy by killing these aberrant cells and producing cytokines and chemokines (2). In the context of pregnancy, NK cells, the most abundant lymphocyte at the maternal fetal interface, are not killers. Instead, they play a constructive role by interacting with and responding to fetal trophoblast cells. Data suggest that a secretory response by NK cells promotes remodeling of the maternal vasculature to support a successful pregnancy (4–6).

NK cells express an array of receptors to sense their environment. Integration of signals from both activating and inhibitory receptors that interact with ligands on target cells determine NK cell responses (7). The killer-cell Ig-like receptors (KIR), most of which recognize HLA class I molecules, regulate NK cell activation through expression of both inhibitory and activating KIR family members. KIR2DL4 is a unique member of the KIR family whose expression, unlike other KIR, is not limited to a subset of NK cells. KIR2DL4 has two Ig-like domains (termed D0 and D2) and activates a secretory response after engagement by a soluble agonist antibody or by soluble HLA-G (8). KIR2DL4 resides in early endosomes and binds soluble HLA-G to stimulate a broad transcriptional response, including an IFN-independent interferon stimulated gene (ISG) response in primary NK cells that may protect NK cells from pathogens (9). These responses are relevant in the context of early pregnancy where HLA-G expression by fetal trophoblast cells is controlled spatially and temporally as they invade the maternal decidua following implantation of the embryo (5).

Despite evidence of an endosomal signaling pathway initiated by a KIR2DL4–HLA-G interaction (10), direct binding of HLA-G to KIR2DL4 protein has not been reported. Even though KIR2DL4 is required for responses to HLA-G and an agonist antibody to KIR2DL4 is sufficient for activation (11, 12), the requirement for a coreceptor for HLA-G mediated activation of NK cells was not excluded. A recent study reported a strong correlation in the transcriptional response of primary NK cells stimulated with soluble HLA-G with their response to the agonist monoclonal antibody #33 (mAb #33), consistent with direct engagement of KIR2DL4 with HLA-G (9). To examine how KIR2DL4 interacts with HLA-G, we chose a random mutagenesis approach and mapped residues in the extracellular domain that are required for a physical and functional interaction with HLA-G. Several independent mutations implicated three cysteine residues in the D0 domain that controlled binding of HLA-G and endocytosis of KIR2DL4. We provide evidence for an allosteric disulfide switch between these three cysteines that regulates both endocytosis and ligand binding.

Results

Random mutagenesis of KIR2DL4 to map its interaction with HLA-G

Random mutagenesis of a cDNA fragment that encodes the KIR2DL4 extracellular domain was performed using a low-fidelity PCR reaction calibrated to generate approximately 1 to 2 base change per 1000 bp (Fig. S1A). HA-tagged KIR2DL4 with single or double amino acid changes (n=293) were transfected individually in 293T cells and examined for their ability to bind HLA-G (Fig. 1A and S1B). Transfected cells were incubated for 2 hours at 37°C with Alexa-594 coupled to an anti-HA tag antibody and with soluble HLA-G coupled to Alexa-647 (Fig. S1B). The distribution of KIR2DL4 and of HLA-G was determined by confocal microscopy (Fig. 1B). As shown previously (12), soluble HLA-G was endocytosed along with KIR2DL4 into endosomes. As controls, soluble HLA-E coupled to Alexa-647 was not bound nor internalized into KIR2DL4⁺endosomes, and HLA-G-647 was not internalized in 293T cells expressing an HA-tagged 2B4 (CD244) receptor (Fig. 1B). Out of 216 clones with interpretable images of distribution at the plasma membrane or in intracellular vesicular compartments (vesicles) 153 (71%) had a phenotype that resembled wild-type KIR2DL4, leaving 63 mutants (29%) with distinct non-wild-type phenotypes (Fig. 1C).

Figure 1:

The main objective was to identify KIR2DL4 mutants that did not bind HLA-G and were otherwise in vesicles, similar to wild-type KIR2DL4 in the absence of HLA-G (12). Out of the 63 mutants, 37 failed to bind HLA-G. Only 7 of those 37 were in intracellular vesicles (Fig. 1C, first panel, Dataset S1). The more common phenotype of KIR2DL4 mutants that did not bind HLA-G (30 out of 37) was an even distribution at the plasma membrane (Fig. 1C, second panel). Mutations with that phenotype were distributed throughout the D0 and D2 Ig domains (Dataset S1), consistent with internalization into vesicles that is controlled by the KIR2DL4 extracellular domain.

Two distribution patterns were observed among the 26 mutants that bound HLA-G. Six mutants displayed a hybrid phenotype: colocalization of HLA-G with KIR2DL4 in vesicles, like wild-type, but with some KIR2DL4 remaining the plasma membrane (Fig. 1C, fourth panel), suggesting that internalization of these 6 mutants was facilitated by HLA-G. To test this, these mutants were examined in the absence of HLA-G. Results showed that, indeed, they remained at the plasma membrane (Fig. S1C). The other 20 mutants were detected at the plasma membrane along with HLA-G (Fig. 1C, right most panel). Overall, these results showed that 50 out of 63 mutants with a non-wild type phenotype (Dataset S1) failed to internalize, reinforcing the evidence that KIR2DL4 internalization is controlled by the extracellular domain (13).

The KIR2DL4 D0 domain includes 3 cysteines at positions 10, 28, and 74. Among the 7 mutants that did not bind HLA-G but resided in vesicles, two included a mutation at Cys74, such as C74W (Fig. 1C). This suggested that KIR2DL4 with a Cys10-Cys28 disulfide bond, as seen in the KIR2DL4 crystal structure (14), does not bind HLA-G but is otherwise distributed in vesicles like wild-type KIR2DL4 (12). Conversely, out of the 6 mutants with a hybrid phenotype, namely distribution at the plasma membrane and in vesicles but colocalization with HLA-G in vesicles only, one had a mutation at Cys10 (C10R). This observation showed that KIR2DL4 with the alternate bond, Cys28-Cys74, is detectable at the plasma membrane in the absence of HLA-G (Fig. S1C, left). This C10R mutant was at the plasma membrane only, demonstrating that its internalization was facilitated by soluble HLA-G. Conversely, we tested if KIR2DL4 mutants retained at the plasma membrane with bound HLA-G, such as P48S (Fig. 1C), could internalize on their own. In the absence of HLA-G, mutant P48S was in vesicles (Fig. S1C, right), showing that, in this case, HLA-G binding caused retention at the plasma membrane. Several targeted mutations were generated to confirm these results. The phenotypes of mutants C74W and C10R (Fig. 1C) were reproduced with targeted mutations C74R and C10S, respectively (Fig. S1D). These mutations also supported the conclusion that KIR2DL4 with a Cys28-Cys74 bond is detectable at the plasma membrane in the absence of HLA-G. Conversely, HLA-G prevented the internalization of KIR2DL4 mutants, such as P48S which otherwise endocytosed spontaneously.

An allosteric disulfide bond in KIR2DL4

The impact of cysteine mutations on the localization of KIR2DL4 led us to compare its sequence with that of other D0 domains, in KIR2DL5 and the KIR3D family (Fig. 2A). The cysteine at position 10 of KIR2DL4 is absent in these other KIRs, which have a leucine at that position. The single disulfide bond in those D0 domains is a canonical Ig domain trans-beta sheet bond (15), comparable to the Cys28-Cys79 bond in D1 Ig domains (Fig. 2B). Of the three potential disulfide bonds (Cys10-Cys28, Cys28-Cys74, and Cys10-Cys74), the one visible in the KIR2DL4 crystal structure (14) is Cys10-Cys28, leaving a free Cys at position 74 (Fig. 2C).

Figure 2:

Disulfide bonds that contribute to protein folding and stabilization tend to be inert and are referred to as structural bonds (15). Labile disulfide bonds that can switch to a reduced form and back to the oxidized bond regulate protein function or localization and are referred to as allosteric bonds (15). Such bonds may be stressed and prone to cleavage (16). Formation of disulfide bonds may be reversed in reducing conditions or by the enzymatic activity of protein disulfide isomerases (PDI). Examination of the Cys10-Cys28 bond in the D0 domain of KIR2DL4, visible in the crystal structure (14), revealed that the second cysteine, Cys28, was partially exposed to solvent (Fig. 2B), compatible with a propensity for thiol disulfide exchange. By comparison, the disulfide bond in the KIR3DL1 D0 domain (Fig. 2B) is not solvent accessible, according to the crystal structure (17). Given that KIR2DL4 may be regulated by an allosteric disulfide bond, we examined the configuration of the Cys10-Cys28 disulfide (Fig. 2C). Among the 20 possible disulfide bond conformations, as determined by the geometry of the five dihedral angles that describe the cysteine residue, three of them are characteristic of allosteric bonds (15, 18). The Cys10-Cys28 bond in the KIR2DL4 structure has an archetypal –RHstaple allosteric conformation (Fig. 2D), which is characterized by high tensile pre-stress due to stretching of the S-S bond and neighboring α angles (19).

Disulfide bonds in KIR2DL4

As cysteine mutations in the D0 domain of KIR2DL4 had an impact on internalization and cellular distribution, we determined what disulfide bond out of the three possible (Cys10-Cys28, Cys28-Cys74, and Cys10-Cys74) (Fig. 3A) existed in recombinant KIR2DL4 produced in insect cells. Cysteine pairings for disulfide bonds in recombinant KIR2DL4 were mapped using disulfide-linked peptides and mass spectrometry. The conserved Cys123-Cys172 bond in domain D2 (Fig. 3A) served as a reference. In wildtype KIR2DL4, peptides linked by disulfides Cys10-Cys28, Cys10-Cys74, Cys28-Cys74 from D0 and Cys123-Cys172 from D2 were identified and quantified (Fig. 3A and B; Fig. S2). Cys10-Cys28 represented the most abundant bond in D0 and as expected, was abundant in the C74S mutant but absent in the C10S mutant (Fig. 3B). The recovery of the reference Cys123-Cys172 bond in the D2 domain was lower (approximately one third) but, as expected, its abundance was virtually identical in the 3 isoforms of KIR2DL4 (wild-type, C10S, and C74S). The Cys28-Cys74 bond in D0 was detected at a lower abundance than Cys10-Cys28, and as expected, was more abundant in the C10S mutant (Fig. 3B). The Cys10-Cys74 bond was detected at a low abundance and was completely absent in the C10S and the C74S mutants (Fig. 3B). In conclusion, wild-type purified KIR2DL4 existed in both configurations, Cys10-Cys28 and Cys28-Cys74, representing independent isoforms distinguished by their disulfide bond in the D0 domain Fig. 3B).

Figure 3:

The redox state of cysteines in KIR2DL4 purified from insect cells was then determined by differential cysteine alkylation and mass spectrometry (Fig. 3C; Table S1). Unpaired or reduced cysteines were first labelled with the cysteine alkylator ¹²C-IPA. Disulfide bonds were then reduced by dithiothreitol followed by cysteine alkylation with ¹³C-IPA. The fraction of reduced disulfide bonds was measured from the relative ion abundance of peptides containing ¹²C-IPA or ¹³C-IPA (Fig. 3C). By this approach, the relative abundance of reduced cysteines in wild-type KIR2DL4 was 32.3%, 4.5%, and 74.3% for Cys10, Cys28, and Cy74 respectively (Fig. 3D). Based on these measurements, the calculated abundance of Cys10-Cys28 and Cys28-Cys74 was 67.7% and 25.7%, respectively. This data demonstrated that these two disulfide forms exist in the D0 domain.

Next, we tested if the Cys10-Cys28 and Cys28-Cys74 disulfide pairs existed in KIR2DL4 obtained from human cells, using the same protocol as in Fig. 3C and 3D with the following modifications. Unpaired cysteines on intact 293T-2DL4-gfp cells were first labelled with ¹²C-IPA, followed by cell lysis, pull-down of KIR2DL4 via the GFP tag, and a second labelling with ¹²C-IPA prior to elution. Eluted proteins were separated by gel electrophoresis and those with an approximate mass of 75 kD were isolated. Disulfide bonds were then reduced and the new unpaired cysteines alkylated, this time with ¹³C-IPA (Fig. 3C). The redox states of cysteines were quantified by mass spectrometry analysis. The fraction of reduced disulfide bonds was calculated from the relative ion abundance of peptides containing ¹²C-IPA and ¹³C-IPA (Fig. 3C). The Cys10-Cys28 and the Cys28-Cys74 forms of KIR2DL4 were detected at a 48.15% and 35.04% relative abundance, respectively (Fig. 3E). A KIR2DL4 in the Cys10-Cys74 configuration was not detected, consistent with the very low abundance of reduced Cys28 in recombinant KIR2DL4 (Fig. 3D). The structural Cys123-Cys172 bond in the D2 Ig domain, was present at an 82.96% relative bonded state, equivalent to the total bonded state of the D0 domain (48% + 35%). We concluded that wild-type KIR2DL4 expressed in human cells exists in both disulfide forms (Cys10-Cys28 and Cys28-Cys74), supporting the possibility of an allosteric transition between the two.

Protein disulfide isomerase controls endocytosis of KIR2DL4

Given that the disulfide-bonded Cys28 in the KIR2DL4 crystal structure is partially exposed to solvent (Fig. 2B), we tested if it was sensitive to cleavage by thiol isomerases. Recombinant KIR2DL4 isolated from insect cells was incubated with the three reduced thiol isomerases PDI (PDIA1), Erp57 (PDIA3), and thioredoxin (TXN). Significant reduction of the Cys10-Cys28 bond but not the Cys28-Cys74 bond was observed in the presence of PDI, but not the other two isomerases (Fig. 3F). Therefore, reduction of Cys10-Cys28 by PDI revealed a selectivity at two levels. One is that the Cys28-Cys74 was resistant, suggesting that the Cys10-Cys28 bond is more accessible to PDI. Second, the lack of reduction by the two other isomerases, Erp57 and Thioredoxin, suggested that accessibility of the Cys10-Cys28 bond to these isomerases is limited.

So far, our results, including data obtained with an unbiased mutagenesis screen, pointed to a major role of disulfide bonds in the D0 domain for KIR2DL4 trafficking and to regulation by the activity of protein disulfide isomerase. Several PDI inhibitors were used to examine the cellular distribution of HA-tagged KIR2DL4 in 293T cells incubated with soluble HLA-G coupled to a fluorophore. Two types of inhibitors were used. The first consisted of two thiol blockers that interact irreversibly with free and exposed thiol groups (DTNB and pCMPS). In untreated cells, KIR2DL4 was found mostly in endocytic vesicles together with soluble HLA-G (Fig. 1B and Fig. 4, first panels). In transfected cells with a high level of KIR2DL4 expression, some KIR2DL4 was detected at the plasma membrane. With increasing concentrations of DTNB (Fig. 4A) or pCMPS (Fig. 4B), the proportion of cells exhibiting KIR2DL4 at the plasma membrane increased. Uptake of HLA-G was sensitive to DTNB and, to a lesser extent, with pCMPS.

Figure 4:

Incubation with rutin (quercetin-3-rutinoside), a selective inhibitor of PDI that does not inhibit other extracellular thiol isomerases present in the vasculature (20), had a similar effect as thiol blockade (Fig. 4C). Blocking of PDI with a specific antibody inhibited internalization of KIR2DL4 but not as efficiently as thiol blockers (Fig. 4D). Thus, blocking the activity of PDI with inhibitors and PDI blockade with an antibody prevented endocytosis of KIR2DL4 and reduced the binding of HLA-G to KIR2DL4.

Disulfide bonds in KIR2DL4 regulate binding to HLA-G and endocytosis

To further evaluate the role of alternative disulfide bonds in the D0 domain and for a comparison of the same cysteine substitutions, each one of the three cysteines in the D0 domain was replaced with a serine. The C10S and the C28S mutants remained mostly at the plasma membrane (Fig 5A), indicating that a Cys28-Cys74 configuration or a Cys10-Cys74 configuration (if it were to exist) did not internalize into vesicles. Conversely, the C74S mutant, which is limited to a Cys10-Cys28 bond, was distributed in intracellular vesicles nearly as much as wild-type KIR2DL4 (Fig. 5A).

Figure 5:

Considering that HLA-G bound to KIR2DL4 in a Cys28-Cys74 configuration, several mutants of Cys10 were tested for their ability to bind HLA-G and internalize. Three substitutions of Cys10 in HA-KIR2DL4 (C10R, C10S and C10A) transfected into 293T cells were able to bind HLA-G and to internalize as well as wild-type (Fig. 5B). There was very little internalization of HA-KIR2DL4 into vesicles without concomitant uptake of HLA-G (Fig. 5B), as was seen with wild-type KIR2DL4 (Fig. 1B). Therefore, KIR2DL4 with a Cys28-Cys74 disulfide bond behaved as wild-type as far as HLA-G binding and transport to intracellular vesicles was concerned but failed to internalize on its own. A summary of cysteine mutants and their impact on KIR localization and interaction with HLA-G is shown in Table 1.

Table 1:

The X-ray crystal structure of a KIR2DL4 that had been produced and purified from human cells (14) revealed an unexpected cross-strand Cys10-Cys28 bond in D0, rather than the canonical cross-β sheet Cys28-Cys74 (Fig. 5C). By comparison, AlphaFold2, a deep learning based prediction tool to model a protein’s 3D structure from its amino acid sequence (21, 22), predicted a KIR2DL4 structure with a Cys28-Cys74 bond (Fig. 5C), supporting the notion that KIR2DL4 can exist in both configurations and consistent with an allosteric transition between two disulfide bonded states.

The agonist mAb #33 and the antibody MAB2238 to KIR2DL4 were tested for binding with the three different Cys mutations. As control, an antibody to the HA-tag fused to the extracellular N-terminus of KIR2DL4 was used (Fig. 5D, Fig. S2). Neither mAb #33 nor MAB2238 bound to the C28S mutant, likely to be misfolded due to the lack of a Cys10-Cys74 disulfide bond, as shown by mass spectrometry. While they both bound to the C74S mutant, limited to a Cys10-Cys28 bond, only MAB2238 bound to the C10S mutant, limited to a Cys28-Cys74 bond (Fig. 5D). Therefore, mAb #33 does not recognize KIR2DL4 in a Cys28-Cys74 configuration and binds selectively to the Cys10-Cys28 disulfide bonded state (Fig. 5D and Fig.S3).

HLA-G binding is controlled by an allosteric disulfide in the KIR2DL4 D0 domain

Alpha-Fold2 (21, 22) predicted a KIR2DL4 structure in the Cys28-Cys74 disulfide-bonded form (Fig. 6A), rather than the Cys10-Cys28 configuration in the crystal structure (14). This is consistent with the two configurations of KIR2DL4 that we identified biochemically using differential cysteine labeling and mass spectrometry (Fig. 3E). An overlay of these two structures, the PDB 3WYR structure (purple) and the AlphaFold prediction (blue), revealed differences in the D0 domain. Besides the difference in the disulfide bond, a loop that follows the β-strand C with two prolines at position 46 and 48 was clearly different (Fig. 6A, blue arrow, and Fig. 6B). This loop is located near the D2 domain (Fig. 6A) at a point similar to the “elbow” formed by domains D1 and D2 of KIR2D receptors and where HLA-C binds (23, 24). The different orientation of the P46 side chain in the two structures is not due to proline isomerization, as prolines P46 and P48 are in a trans configuration in both structures. The conformation predicted by AlphaFold2 for KIR2DL4 in the Cys28-Cys74 bonded form showed a 5.4 Å distance between the side chain of P46 with that of the crystal structure, and a 5.3 Å distance for the side chain of V45 (Fig. 6B).

Figure 6:

AphaFold3, a predictor of the structure of protein complexes, revealed a potential interaction between HLA-G and KIR2DL4 in a Cys28-Cys74 configuration (Fig. 6C; Fig. S4A). Contacts with HLA-G were apparent in D0 at Val45 and Pro48, two residues within the predicted loop that distinguishes the two disulfide bonded states of KIR2DL4 (Fig. 6C). Val45 and Pro48 were predicted to form bonds with Met76 and Glu19 of HLA-G, respectively (Fig. 6C). Met76 is unique to HLA-G, as it is different in other human HLA-I (25). Position 76 is a Glu in HLA-B, which contacts KIR3DL1 at Ala167 in the D1 domain (17). Several predicted interactions occurred in the more conserved D2 domain, present in all KIR2D and KIR3D receptors. Four amino acids within a 7-amino acid long stretch in HLA-G (R145 to N151) could provide contacts with 6 amino acids in the D2 domain of KIR2DL4. Arg145 in HLA-G contacts Ser128 and Asp130 of KIR2DL4, and Lys146 contacts Ser179 in the C28-C74 disulfide bonded KIR2DL4 (Table S2). In addition, Ala149-Ala150-Asn151 in HLA-G are predicted to contact the 3 consecutive residues, Leu99-Tyr100-Glu101 (Table S2), which is conserved in the D2 domains of KIR2DL4, KIR2DL1, and KIR3DL3.

In the case of a Cys10-Cys28 configuration, as seen in the crystal structure, the V45 to P48 residues are predicted to be too distant from the Met76 and Glu19 of HLA-G, which contact KIR2DL4 in a Cys28-Cys74 configuration (Fig. 6C and D; Fig. S4A and B). Notably, the structure predicted by AlphaFold for a C74R mutant (limited to a Cys10-Cys28 disulfide bond) is very similar to that of the KIR2DL4 crystal structure and is also too distant from HLA-G for an interaction (Fig. 6E; Fig. S4C). Consistently, the structure of a C10R mutant (limited to a Cys28-Cys74 bond) is compatible with similar contacts with HLA-G as those predicted by AlphaFold for a wild-type KIRDL4 in a Cys28-Cys74 configuration (Fig. 6F; Fig. S4D). The conformational change in the Pro46-Pro48 loop, which clearly distinguishes the two forms of KIR2DL4, the Cys10-Cys28 from the Cys28-Cys74 configurations, is in the portion of the D0 domain sequence with the widest divergence from D0 domains of other KIR. In the nine amino acid stretch G44 to N52 (GVPVPELYN), only P48 is shared with 3DL1 and 3DL2 sequences.

The predicted D2 domain contacts with HLA-G, which are similar to those of KIR2DL1 with HLA-Cw4, are unaffected by the disulfide switching, as seen in the overlay of D2 domains for all 4 structures, the crystal structure (3WYR) and AlphaFold predictions of wild-type, C74R, and C10R KIR2DL4 (Fig. 6G). Thus, the D2 domain contacts (Table S2, Fig. S3) are preserved regardless of the disulfide bond usage in the D0 domain. These results support the interpretation that allosteric changes induced by a switch from a Cys10-Cys28 disulfide bond to a canonical Cys28-Cys74 bond in KIR2DL4 changes the orientation of a distant loop in the D0 domain, such that it could now contact HLA-G (Movie S1).

Discussion

Transcription of HLA-G is turned on in fetal extravillous trophoblast (EVT) cells just as they reach the fetal-maternal interface during early pregnancy, concomitant with high transcription of several matrix metalloprotease genes (5), including MMP2, which can promote shedding of soluble HLA-G at the plasma membrane (26, 27). The abundant NK cells in the decidua are distinct from maternal blood NK cells (28) and have their own transcriptional signature, including high expression of KIR2DL4 (5). As expected for cellular processes that support human reproduction, the encounter of EVT cells with decidual NK cells at the maternal-fetal interface is well orchestrated (5, 29). High transcription of KIR2DL4 in decidual NK cells is matched by high and selective expression of HLA-G in EVT cells. Uptake of soluble HLA-G into cells depends on KIR2DL4 and both colocalize in Rab5⁺ early endosomes from where signaling occurs (12). As a strong ligand for the inhibitory receptor LILRB1, which is selectively expressed by the dNK1 subset of decidual NK cells, HLA-G at the plasma membrane can provide protection to EVT cells. Another protective role of HLA-G occurs by the release of a peptide from its signal sequence that provides a strong ligand for the inhibitory receptor NKG2A (KLRC1) via presentation by HLA-E (30–32). Therefore, inhibitory signaling through LILRB1–HLA-G and NKG2A–HLA-E interactions protect EVT from NK killing. However, signaling from endosomes triggered by the interaction of soluble HLA-G with KIR2DL4 is protected from inhibition at the cell surface and induces secretory responses that support vascular remodeling and fetal growth.

To address the question of how KIR2DL4 interacts with HLA-G, a random mutagenesis approach was chosen to generate a library of mutant KIR2DL4 that were then tested for their ability to endocytose soluble HLA-G. This analysis revealed that binding of HLA-G and uptake of soluble HLA-G into endosomes was regulated by a trio of cysteine residues in the first Ig domain of KIR2DL4, called D0. In addition to the disulfide bond typical of Ig domains (Cys28-Cys74), such as the D0 domain of KIR2DL5 and KIR3DL1, a third cysteine is present at position 10 in KIR2DL4. The only KIR2DL4 structure reported so far has this atypical Cys10-Cys28 bond, leaving a free cysteine at position 74 (14). Our mutants of Cys74 revealed that a KIR2DL4 with a Cys10-Cys28 bond as the only option, was mainly in endosomes of transfected cells, like wild-type KIR2DL4, but neither bound HLA-G nor induced endocytosis of soluble HLA-G. Alternatively, mutations at Cys10, which left the D0 domain with a single disulfide bond (Cys28-Cys74), failed to internalize, leaving KIR2DL4 at the plasma membrane. Furthermore, in contrast to mutations at Cys74, mutants without Cys10 remained at the plasma membrane and their internalization into endosomes was induced by soluble HLA-G. Thus, proper KIR2DL4 function depends on a switch from one disulfide bond to another, keeping Cys28 as a common partner for either Cys10 or Cys74. We found that the Cys10-Cys28 bond has a –RHstaple conformation, which is a signature of allosteric bonds having an inherent high bond stress that predisposes it to cleavage (19). Here, we have shown that KIR2DL4 in human cells exists in both disulfide configurations, and that the allosteric Cys10-Cys28 bond is reduced in vitro by the thiol disulfide oxidoreductase PDI. As expected, if PDI activity is required to maintain a pool of Cys28-Cys74 KIR2DL4 at the plasma membrane (by reducing Cys10-Cys28 bonds), thiol blockade with membrane impermeable DTNB did inhibit HLA-G binding.

AlphaFold predictions of KIR2DL4 folding and interaction with HLA-G showed that a switch from Cys10-Cys28 to a Cys28-Cys74 disulfide bond would cause an allosteric shift in a distant loop in D0 and bring the Val45 to Pro48 loop in a position that permits interactions with Met76 and Glu19 of HLA-G. The lack of reported detectable binding of purified KIR2DL4 with HLA-G in vitro (14) may be explained by the incompatibility of a Cys10-Cys28 bonded KIR2DL4 with HLA-G binding.

The regulation of a KIR2DL4–HLA-G interaction by an allosteric disulfide that controls protein conformation provides yet another example of disulfide bonds serving as switches in protein function. Allosteric disulfides regulate diverse biological processes, such as hemostasis (tissue factor, platelet fibrinogen receptor integrin αIIb-β3, and von Willebrand factor), and viral infection (gp120 of HIV-1) (15). Allosteric disulfide bonds specifically targeted by extracellular PDI at the cell surface have been identified in proteins such as GP1bα, tissue factor (F3), αvβ3 integrin during thrombosis, and HIV-1 envelope gp120 when bound to CD4 during viral entry (33, 34).

KIR2DL4, the only KIR family member expressed by all human NK cells, and highly transcribed in decidual NK cells, joins the list of proteins whose function is controlled by a labile disulfide bond acting as a switch, but here in the context of early pregnancy. In response to fetal HLA-G, NK cells endocytose soluble HLA-G bound to KIR2DL4, which signals from endosomes for a secretory response that leads to a sustained secretion of factors that support vascular remodeling and fetal development (4, 13). In addition, NK cells induce an ISG response upon interacting with HLA-G which protect NK cells from infections in early pregnancy (9). In this unique environment at the maternal fetal interface, the allosteric disulfides may act as a functional switch to regulate KIR2DL4 localization, binding to HLA-G and signaling to enable NK cells to support a successful pregnancy.

Materials and Methods

Cells and reagents

HEK293T cells (hereafter called 293T) were obtained from the American Type Culture Collection (ATCC) and cultured in Iscoves Modified Dulbecco’s medium containing 10% fetal calf serum (FCS). KIR2DL4 and receptor 2B4 (CD244) were tagged at the N-terminus with the short YPYDVPDYA amino acid sequence (HA tag) present in the pDisplay expression vector. 293T–2DL4-gfp cells were generated as described (12). Briefly, 293T–2DL4-gfp cells were made by transfecting 293T cells using LipofectAMINE 2000, with a KIR2DL4 cDNA fused to green fluorescence protein (gfp) in the vector pBABE (puro) CMV⁺ using SalI and BamH1 sites. Transfected cells were selected and maintained in Iscove’s medium containing puromycin at a concentration of 2 mg/ml. Recombinant HLA-G and HLA-E were obtained from the Immune Monitoring Services, Fred Hutchinson Cancer Center (Seattle, WA). Briefly, MHC heavy chain protein and β2-microglobulin (β2m) were produced in E. coli and purified from inclusion bodies. The HLA-G heavy chain and b2m proteins were refolded in vitro in the presence of peptide KGPPAALTL into a monomeric peptide-MHC complex. For HLA-E, refolding was done in the presence of peptide VMAPRTLFL. Monomers of the refolded p-MHC were purified by FPLC using size exclusion chromatography and concentrated to 1 mg/mL. Soluble HLA-G and HLA-E were labeled using the Alexa Fluor 647 Protein Labeling Kit (Cat #: A20173, Molecular Probes, ThermoFisher). mAb#33 was produced in our laboratory (11) and MAB2238 was obtained from RnD Systems.

Random Mutagenesis Screen

Random mutagenesis of the KIR2DL4 sequence spanning amino acids 1-208, consisting of the Ig domains D0 and D2 and a portion of the stem was carried out using the Diversify PCR Random Mutagenesis kit (Clontech, Cat#630703) according to manufacturer’s instructions. HA-tagged KIR2DL4 expressed in pDisplay served as the DNA template. The primers used at 10 mM for PCR were as follows: Sense: CCTGCTATGGGTACTGCTGCTCTGGGTTCC; Antisense: GAAGGGGAGGATGGTGAAGAGGATGATGGC. Mn2⁺ (80 mM) and dGTP (40 mM) was used to obtain the desired mutation rate of 1 to 2 mutations per kb of DNA. Thermal cycling conditions consisted of denaturation at 94° for 30 sec, followed by annealing at 94°C for 30 sec and extension at 68°C for 1 min for a total of 25 cycles. Mutated KIR2DL4 was cloned back into pDisplay, and colonies were sequenced to select for single or double amino acid mutations. Each mutant (1 mg) was then transiently transfected into 293T cells using LipofectAMINE 2000 according to manufacturer’s instructions. After 36 h, cells were trypsinized and plated into 8 well glass chambers. Cells were plated at a density of 7 ξ 10⁴ cells/well in eight-well coverglass slides (Nunc Lab-Tek, Thermo Fisher Scientific) 18 h prior to loading. Cells were then loaded with anti-HA Ab coupled to Alexa 594 (Biolegend) (0.03 mM) and HLA-G coupled to Alexa 647 (0.04 mM) for 2 h at 37°C. Cells were then washed in PBS and viewed by microscopy using a Zeiss 780 confocal microscope.

Confocal microscopy

Image acquisition was performed using a LSM 780 confocal laser scanning microscope (Zeiss) fitted with a LD C-Apochromat 40ξ/1.1 W Corr objective. Cells were incubated at 37°C in 5% CO₂ for the entire acquisition duration. The HFT 543 nm and 633 nm main beam splitters were used to detect signals from fluorophores Alexa-Fluor 594 and 647, respectively. Parameters were set to acquire at a fixed pixel density of 512 ξ 512, 16-bit pixel depth, and pinhole size of 100 mm. Images were processed and colocalization coefficients were calculated using the Zeiss ZEN software. Quantification of receptor and ligand localization was performed as follows. Transfected cells expressing fluorescently tagged proteins were viewed in a series of multiple fields using the LSM 780. Cells in each field were counted and categorized based on the location and ligand colocalization status of the expressed receptor. Counts were averaged over 2 to 4 experiments, and the number of cells counted per sample ranged from 100 ≤ n ≤ 430. In a typical transfection with wild-type KIR2DL4, approximately 80% of the transfected cells had clear signals for internalized HLA-G and KIR2DL4 in the same endosomal compartments, and the remaining 20% had low but detectable 2DL4 at the plasma membrane but no HLA-G. Endosomes positive for 2DL4 but without an HLA-G signal were noted in less than 2% of the cells.

Thiol Blockade and PDI Inhibition Assays

In eight-well coverglass slides, 293T cells transfected with HA-2DL4 were preincubated in Iscoves media containing DTNB (5,5′-Dithiobis 2-nitrobenzoic acid; Sigma), pCMPS (para-chloromercuriphenylsulfonate; Carbosynth), Rutin (Sigma), or anti-PDI antibody RL90 (Thermo Fisher) at varying concentrations for 1 h at 37°C. Anti-HA-594 and HLA-G-647 were then added and incubated for 2 h in the continued presence of inhibitor at 37°C. Cells were then washed twice with PBS and analyzed by confocal microscopy as described above.

Quantification of cysteine redox states

GFP-tagged KIR2DL4 expressed in 293T cells was immunoprecipitated from cells using ChromTek-GFP Trap Agarose (Proteintech). Free thiols were labelled with 2-iodo-N-phenylacetamide (¹²C-IPA) followed by cell lysis, immunoprecipitation of 2DL4-GPF and second ¹²C-IPA labelling of proteins prior to elution. 2DL4-GFP immunoprecipitated from cells were analyzed on SDS-PAGE. Protein band corresponding to 2DL4-GFP (75 kDa) was detected from a single eluate prepared from cells and the gel area corresponding to 75 kDa was excised, reduced with dithiothreitol (DTT), and reduced thiols alkylated with ¹³C-IPA. The redox states of cysteines were quantified by mass spectrometry analysis. Reactions were stopped by adding 5% (v/v) formic acid, and peptides were eluted from the gel slices with 5% formic acid and 50% (v/v) acetonitrile.

Peptides were desalted using Ziptip (Millipore), dried and reconstituted in 0.1% formic acid. Using a Thermo Fisher Scientific Ultimate 3000, peptides were injected and resolved on a 35cm × 75 mm C18 reverse phase analytical column with integrated emitter using a 2-35% acetonitrile over 22 min with a flow rate of 300 nL/min. The peptides were ionized by electrospray ionization at +2.0 kV. Tandem mass spectrometry analysis was carried out on a Q-Exactive Plus mass spectrometer (Thermo Fisher Scientific) using HCD fragmentation. The data-dependent acquisition method acquired MS/MS spectra of the top 5 most abundant ions at any one point during the gradient. The data was searched using Mascot (Matrix Science) against Uniprot database. Search parameters were as follows: precursor tolerance of 6 ppm and product ion tolerances of ± 0.4 Da. Oxidized Met, pyro-Glu and pyro-Gln were selected as variable modifications with full chymotryptic cleavage of up to three missed cleavages. To calculate ion abundance of peptides, extracted ion chromatograms were generated using the XCalibur Qual Browser software (v2.1.0; Thermo Scientific). The area was calculated using the automated peak detection function built into the software. Abundance of disulfide-linked peptides was expressed as a ratio relative to disulfide-linked peptide between Cys123-Cys172 in the D2 domain. Disulfide bond analysis was performed using the Disulfide bond Analysis Tool.

Recombinant KIR2DL4 protein (WT, C10S and C74S) were produced in insect cells and purified by His-tag affinity chromatography. The extracellular domain of KIR2DL4*001 (residues 1-195) was subcloned into the pAC insect expression vector in frame with an N-terminal 8× HisTag and a 3C protease cleavage site. The KIR2DL4 expression vector was then transfected with BestBac 2.0 linearized DNA (Expression Systems) into Sf9 cells to generate P1 virus. Virus was amplified by shaking culture in Sf9 cells prior to expression. KIR2DL4 was then expressed in BTI-Tn-5B1-4 insect cells (High Five) and purified using Ni-NTA affinity purification followed by Superdex 200 size exclusion into 10 mM HEPES, 150 mM NaCl, pH 7.2. Fractions containing the purified protein were verified by SDS-PAGE. Recombinant thiol isomerases (PDI, ERp57 and thioredoxin) were expressed in E. coli and purified by His-tag affinity chromatography. The thiol isomerases were then reduced with final 10 mM dithiothreitol for 30 min at 25°C and desalted into PBS using a Zeba spin column (Themo Fisher Scientific). KIR2DL4 (3 mg) was treated with either a 2- or a 10-fold molar excess of each thiol isomerase for 30 min at 25°C. Free thiols in 2DL4 were alkylated with 5 mM 2-iodo-N-phenylacetamide (¹²C-IPA) and the protein was resolved on non-reducing SDS-PAGE, stained with Bio-Safe Coomassie stain and destained in deionized water. Bands corresponding to 2DL4 were excised, dried, incubated with 40 mM dithiothreitol and washed. The fully reduced proteins were alkylated with 5 mM ¹³C-IPA. The gel slices were washed and dried before digestion of proteins with 12 ng/mL of chymotrypsin (Roche) in 25 mM NH4CO2 containing 10 mM CaCl2 overnight at 25°C. Peptides were eluted from the slices with 5% formic acid, 50% acetonitrile. Peptides were desalted using Ziptips (Merck Millipore). Liquid chromatography, mass spectrometry and data analysis were performed as described (35, 36). The fraction of reduced disulfide bond was measured from the relative ion abundance of peptides containing ¹²C-IPA and ¹³C-IPA. To calculate ion abundance of peptides, extracted ion chromatograms were generated using the XCalibur Qual Browser software (v2.1.0; Thermo Fisher Scientific). The area was calculated using the automated peak detection function built into the software. The ratio of ¹²C-IPA and ¹³C-IPA alkylation represents the fraction of the cysteine in the population that is in the reduced state (35).

Site directed mutagenesis

Engineered mutations of KIR2DL4 in the pDisplay vector were carried out using the QuikChange site directed mutagenesis kit (Stratagene) according to the manufacturer’s instructions. Mutations were verified by sequencing. Stable transfections of cysteine mutants of KIR2DL4 were carried out using Lipofectamine 2000 according to the manufacturer’s instructions and selected using 1 mg/ml puromycin. Transfected cells were sorted by flow cytometry to generate transfectants with uniform expression of the HA-tag.

Data availability

All data generated or analyzed during this study are included in the manuscript and supporting files

Supplementary figures and tables

Fig. S1:

Fig. S2

Fig. S3

Fig. S4.

Table S1.

Table S2.

Acknowledgements

The contributions of the NIH authors were made as part of their official duties as NIH federal employees, are in compliance with agency policy requirements, and are considered Works of the United States Government. However, the findings and conclusions presented in this paper are those of the authors and do not necessarily reflect the views of the NIH or the U.S. Department of Health and Human Services.

Additional information

Author Contributions

S.R., P.J.H. and E.O.L. designed research; S.R., G.M.M., J.C., J.L, S.M., and K.N. performed research; S.R., G.M.M, J.C., J.L., P.J.H. and E.O.L. analyzed data; S.R. and E.O.L. wrote the paper; S.R., J.C., P.J.H., E.O.L. edited the paper; E.J.A., P.D.S., P.J.H, and E.O.L. acquired funding.

Funding

HHS | NIH | NIDA | Intramural Research Program (IRP) (ZIA AI000525)

• Eric Long

• Peter D Sun

HHS | NIH | National Institute of Allergy and Infectious Diseases (NIAID) (RO1 AI170952)

• Erin J Adams

Government of New South Wales

• Phillip J Hogg

Additional files

Dataset S1

Video 1