Peptide occupancy governs HLA-E checkpoint interactions enabling VISTA recognition of destabilized HLA-E conformers

Abstract

Human leukocyte antigen E (HLA-E) plays a central role in immune regulation across homeostasis, infection, cancer, and transplantation. Canonically, HLA-E presents leader sequence-derived VL9 peptides to engage CD94/NKG2 receptors, while alternative peptides disrupt this interaction. HLA-E also interacts with LILRB1/2 and the checkpoint molecule VISTA, but the role of peptide occupancy in these interactions remains unclear. Here, we demonstrate that HLA-E engages VISTA when peptide occupancy is absent or suboptimal, whereas interaction with LILRB1/2 is peptide independent. Both receptors recognize peptide-deficient or weakly stabilized HLA-E conformers. In contrast, stable peptide loading prevents VISTA binding, indicating that peptide occupancy governs receptor engagement. These findings support a model in which HLA-E exists along a spectrum of conformational states defined by peptide stability, enabling selective interaction with distinct immune checkpoints. This work identifies VISTA as a receptor for destabilized HLA-E conformers and reveals a previously unrecognized layer of immune regulation.

VISTA recognizes peptide-deficient HLA-E, revealing peptide occupancy as a conformational switch for checkpoint receptor selectivity distinct from VL9 loaded HLA-E recognition by NKG2A and peptide independent binding by LILRB1/II.

Introduction

HLA-E is a non-classical major histocompatibility complex (MHC) class I molecule with roles in both innate and adaptive immunity^1,2. HLA-E presents peptides to T cell receptors (TCRs), as well as interacts with CD94/NKG2A and CD94/NKG2C when presenting canonical peptides derived from signal sequences of other MHC class I molecules (VL9 peptides)³. The interaction of HLA-E with CD94/NKG2A inhibits NK cell and CD8+ T cell activation. In the absence of VL9 peptides, HLA-E no longer engages with the CD94/NKG2A inhibitory or CD94/NKG2C activating receptors. HLA-E also exhibits binding to LILRB1 and LILRB2 similarly to other MHC class I molecules⁴.

HLA-E exists in several partially overlapping conformations, or “conformers,” with structural differences demonstrated by antibody binding and X-ray crystallography^5,6. HLA-E heavy chain forms a complex with beta-2-microglobulin (β2m) and a peptide. Individual peptides stabilize the structure of HLA-E with β2m to varying degrees, indicated by variable complex half-lives. Additionally, the crystal structure of HLA-E shows distinct differences when VL9 peptides are loaded in the peptide-binding groove compared to pathogenic peptides. Furthermore, there is evidence for surface expression of HLA-E in the absence of signal sequence-derived peptides⁵, although such peptide-deficient conformers are generally less stable and may represent transient or peptide-receptive states rather than dominant steady-state species.

A recent paper by Shilts et al. screened interactions between leukocyte surface proteins to determine unexplored binding partners⁷. One of the previously undiscovered interactions detected by their screen was between HLA-E/β2m and the extracellular domain (ECD) of VISTA. HLA-E and VISTA are both expressed on multiple cell types^8–12. HLA-E is expressed at low levels on most nucleated cells, with higher levels on endothelial cells and immune cells^2,13. VISTA is commonly described as an immune checkpoint molecule on activated T cells, but is broadly expressed; including on naïve T cells, some dendritic cells, and myeloid cells^14,15. VISTA also has several identified binding partners, including VSIG3, PSGL-1, and LRIG-1^16–18. Several anti-VISTA antibodies are currently in clinical development for cancer immunotherapy¹⁹.

Given the previously described peptide-dependent interactions of HLA-E and the introduction of a potential interaction with VISTA, we aimed to validate this interaction and determine whether it is peptide-dependent. We further compared this interaction to that of HLA-E with LILRB1/2 and CD94/NKG2A by ELISA based binding studies and antibody blocking using multiple antibody clones against HLA-E. We modeled the interaction using AlphaFold3 to predict the binding site of HLA-E with VISTA. We also characterized whether this interaction was conserved by murine homolog Qa-1^b. Finally, we evaluated the expression of both HLA-E and VISTA across normal human tissues to determine possible biological contexts of this interaction.

Results

VISTA does not interact with HLA-E loaded with canonical VL9 peptide

The interaction between VISTA and HLA-E described by Shilts et al. showed an affinity of 38 μM by surface plasmon resonance (SPR)⁷. Given the low affinity predicted for the interaction, we utilized HLA tetramer constructs, for increased avidity, to determine whether the VISTA and HLA-E interaction was peptide-dependent (Table 1). We first compared HLA-E loaded with either a canonical peptide derived from the HLA-G leader sequence (VL9; VMAPRTLFL), or a non-canonical peptide derived from heat shock protein60 (hsp60; QMRPVSRVL)¹. We evaluated protein-binding interactions between HLA-E tetramers and VISTA ECD in real time using an optical label-free detection system, which measures analyte binding to a plate-bound ligand through changes in refractive index. HLA-E + VL9 peptide interacted strongly with CD94/NKG2A, an interaction which was lost with HLA-E complex loaded with hsp60 peptide, in favor of interaction with VISTA (Fig. 1a, b). Binding of αVISTA monoclonal antibody to VISTA was used as a positive control. HLA-E with hsp60 peptide demonstrated binding to VISTA, with a binding curve similar to the positive control, which remained stable post-washing.

Table 1.

HLA alleles and peptide sequences tested

MHC Allele	Peptide Sequence	Abbreviation	Source	Monomer	Tetramer
HLA-E*01:03	No Peptide	np		✓
HLA-E*01:03	VMAPRTLFL	VL9	HLA-G*01 leader sequence	✓	✓
HLA-E*01:03	VMAPRTLVL	VMA9	HLA-A*02, A*23, A*24, A*25, A*26, A*3402, A*43, A*66, A*69		✓
HLA-E*01:03	VMAPRTLIL	V21	HLA-C*01, C*03, C*04, C*05, C*06, C*0801-03, C*12, C*14, C*16, C*1702; Human cytomegalovirus UL40	✓
HLA-E*01:03	VMAPRALLL	V41	HLA-C*07, C*18	✓
HLA-E*01:03	VMPPRTLLL	V45	HLA-A*80	✓
HLA-E*01:03	QMRPVSRVL	hsp60	60 kDa Heat shock protein	✓	✓
HLA-E*01:03	SQAPLPCVL	BZLF1	EBV (BZLF1)		✓
HLA-E*01:03	ILPCRIKQI	HIV	HIV Envelope	✓	✓
HLA-E*01:03	RLPAKAPLL	Mtb44	Mycobacterium tuberculosis		✓
HLA-E*01:03	VMAPRSLLL	CMV-01	Human cytomegalovirus UL40	✓
HLA-A*02:01	RLNMFTPYI	Ct	Chlamydia trachomatis		✓
HLA-A*11:01	IVTDFSVIK	EBNA3b	Epstein-Barr virus (EBV) (EBNA3b)		✓
HLA-B*07:02	LPFDRTTVM	NP	Influenza nucleoprotein (NP)		✓
HLA-B*35:01	DANDIYRIF	pp65	Cytomegalovirus (CMV) antigen pp65		✓

Fig. 1. HLA-E with suboptimal peptide occupancy binds to VISTA extracellular domain.

a Binding of tetrameric HLA complexes (20 μg/mL) and αVISTA antibody (5 μg/mL) to VISTA, PD-1, or CD94/NKG2A (5 μg/mL). Curves represent a 5 min baseline, 1 hour binding, and 40 min post-wash. Data below zero is due to refractive index shifts when analytes are added. The mean of 5 replicate experiments with 3 technical replicates per condition is shown without smoothing. b Differences in resonance shifts at 10 min post-wash. Mean $\pm$SD is shown. Statistics calculated by one-way ANOVA with Šidáks multiple comparisons test (***p < 0.001, ****p < 0.0001).

Highly polymorphic proteins, such as classical HLA molecules, were not evaluated in the SAVEXIS screen which first identified the HLA-E/VISTA interaction⁷. We therefore also included a peptide-loaded classical HLA-A2 in the label-free detection assay. HLA-A2 tetramer loaded with Chlamydia trachomatis derived peptide (HLA-A2+Ct) did not show binding to VISTA, or the structurally similar protein, PD-1 (Fig. 1a,b).

By comparing resonance shift at 10 min post-wash of tetramers on VISTA, CD94/NKG2A, and negative control (PD-1), we observed binding of HLA-E + hsp60 to VISTA was significantly higher than HLA-E + hsp60 on PD-1 and HLA-E + VL9 on VISTA (Fig. 1b). No difference was seen in HLA-E + VL9 interaction with negative control versus VISTA. These results suggest that the presence of a canonical VL9 peptide in the peptide-binding pocket of HLA-E prevents interaction with VISTA.

VISTA and CD94/NKG2A interactions with HLA-E depend on loaded peptide

After demonstrating that HLA-E tetramer binding was stable post-washing, we further evaluated HLA binding to VISTA and other ligands by fluorescence-based ELISA. We used PE-labeled HLA-E, HLA-A, and HLA-B tetramers loaded with an array of peptides. ELISA results showed a clear dose-dependent interaction of VISTA with HLA-E complexes loaded with hsp60 peptide, BZLF1 peptide derived from EBV, or an HIV envelope peptide (Fig. 2a). In contrast, no measurable interaction was detected with HLA-E loaded with either of two canonical (VL9 or VMA9) peptides or with Mtb44 peptide. As expected, CD94/NKG2A interacted only with HLA-E loaded with canonical peptides (Fig. 2b). Based on previous X-ray crystallography analyses, the structure of Mtb44 loaded into HLA-E and thermal stability of this conformer is comparable to VL9-loaded complexes⁶. We observed that both VL9- and Mtb44 loaded complexes of HLA-E were incompatible with VISTA engagement.

Fig. 2. HLA-E binds to immune checkpoints in a peptide-dependent manner.

a Binding of serial dilutions of PE-labeled tetrameric HLA complexes to 1.25 μg/mL of plate-bound (a) VISTA, (b) CD94/NKG2A, (c) LILRB1, and (d) LILRB2. Background fluorescence (tetramer in blank wells) was subtracted. Bars represent mean ± SD of $\ge$3 replicates per condition.

Next, we assessed whether VISTA could recognize HLA-E in the absence of a bound peptide. We generated a multimeric VISTA construct, using Klickmer® technology, bound to monomeric peptide-free HLA-E (Figure S1). HLA-E loaded with other VL9 canonical peptides (V21, V41, V45) did not interact with VISTA, but showed binding to CD94/NKG2A and LILRB1/2 (Figure S1). All HLA-E conformers tested with VL9 peptides, pathogen-derived peptides, or refolds lacking peptide interacted with LILRB1 and LILRB2 (Fig. 2c, d and Figure S1). Neither VISTA nor CD94/NKG2A showed significant interaction with several peptide-loaded HLA-A or HLA-B tetramers by ELISA (Fig. 2a, b). HLA-E with Mtb44 peptide did not exhibit an interaction with CD94/NKG2A or VISTA. Together, these findings demonstrate that VISTA selectively engages HLA-E when peptide occupancy is absent or suboptimal, consistent with the recognition of less stable HLA-E conformers.

VISTA interacts with HLA-E at physiological pH

VISTA has been shown to exhibit pH-dependent interactions, particularly with PSGL-1¹⁷. The VISTA and PSGL-1 interaction occurs preferentially under acidic pH, such as in tumor microenvironments. We therefore tested whether the interaction of HLA-E and VISTA was similarly influenced by pH. Using buffers adjusted to physiological (7.4) or acidic (6.0) pH, we evaluated protein interactions using fluorescent detection of PE-labeled tetramers bound to immobilized ligands. We observed that pH 6.0 reduced the observed interaction of HLA-E + BZLF1 and HLA-E + HIV on VISTA from pH 7.4 (Figure S2). HLA-E + VMA9 interaction with CD94/NKG2A, as well as HLA-E + HIV interaction with LILRB1, were maintained at both pH 6.0 and 7.4. This pH sensitivity was unique to VISTA-associated conditions, indicating that the VISTA and HLA-E interaction is favored under normal physiologic pH, but is unlikely in acidic environments. Our results indicate this change is independent of peptide and likely reflects protonation-dependent changes on VISTA rather than alterations of the HLA-E complex.

HLA-E monoclonal antibodies differentially block binding

To test whether VISTA binding to HLA-E occurs at distinct binding sites from other checkpoint receptors, we assessed whether two distinct αHLA-E monoclonal antibodies were able to block the interaction with LILRB1/2, VISTA, or CD94/NKG2A in ELISA. The anti-HLA-E clone 3D12 epitope is in the alpha-3 domain of HLA-E (amino acids 219–224), whereas clone 4D12 has been shown to bind in the alpha 2-1 helix, adjacent to the C terminus of the presented peptide²⁰. We observed that both 3D12 and 4D12 monoclonal antibodies inhibited HLA-E tetramer interactions with VISTA, whereas they did not block HLA-E + VL9 interaction with CD94/NKG2A (Figure S3). Notably, addition of 3D12 antibody appeared to enhance the binding of CD94/NKG2A to HLA-E + VL9 in a dose-dependent manner. 4D12 induced a modest but significant increase in CD94/NKG2A and LILRB1/2 binding to HLA-E + VL9. 3D12 also blocked LILRB1/2 binding to most HLA-E constructs, whereas 4D12 exerted only minor and variable effects.

Binding site of VISTA on HLA-E

We then utilized AlphaFold3 (Google DeepMind) to predict the binding site of HLA-E and VISTA²¹. Utilizing the sequences of HLA-E heavy chain, β2 m, and the extracellular domain (ECD) of VISTA, the model predicted two loops of VISTA to interact within the peptide-binding cleft of HLA-E (Fig. 3a, b). We further evaluated the amino acids involved in the predicted interaction using AlphaBridge and PyMol (Fig. 3b, c)²². The primary loop of the predicted binding site on VISTA contains 3 histidines, which can contribute to pH-dependent interactions (Fig. 3e). These three histidines are also implicated in supporting PSGL-1 binding¹⁷. To confirm these results, we repeated the AlphaFold3 prediction with the addition of VL9, BZLF1, or HIV Env peptide sequences (Fig. 3d and Figure S4). As expected, peptide sequences showed binding within the HLA-E binding pocket. Furthermore, there were no predicted interacting residues between VISTA and HLA-E in these iterations containing peptide. These results suggest that the presence of a stably bound peptide in the HLA-E binding pocket precludes binding to VISTA.

Fig. 3. AlphaFold3 and AlphaBridge predict VISTA binding site to HLA-E.

a Binding predicted by AlphaFold3 as visualized in PyMOL (b) Zoom view on predicted binding region of VISTA (blue) with two loops predicted to interact (yellow, green) in the peptide-binding pocket of the HLA-E heavy chain (gray) in the absent or suboptimal peptide occupancy. c AlphaBridge prediction of binding residues between VISTA to HLA-E. Outer circle colors show each chain. Chain A: HLA-E heavy chain (HC), chain B: b2m, chain C: ECD VISTA, D: VL9 peptide d) only. Inner circle colors indicate the AlphaFold predicted local distance difference test (pLDDT) which measures confidence in the placement of each atom in its local environment (Dark blue: very high = 90–100, light blue: high= 70-90, yellow: low = 50–70, orange: very low <50). Bridges within the circle show interaction interfaces, with thicker curves indicating adjacent residues involved in the predicted interaction. ed Residues identified by PyMOL as binding to HDVT and HHH loop are colored in green and yellow, respectively. Underlined residues were identified as involved in binding by both AlphaBridge and PyMOL. Residues in blue were identified in AlphaBridge only.

Stable peptide loading impairs VISTA binding

We hypothesized that due to the unstable nature of non-VL9 peptides loaded into HLA-E, that both monomeric and tetrameric HLA-E proteins possibly contain a heterogenous mix of both peptide-loaded and empty conformers. Given the prediction by AlphaFold3 that two loops of VISTA interact with HLA-E within the peptide-binding groove, we compared binding of empty (no peptide) HLA-E with those same monomers loaded with peptide. We observed that addition of a 10:1 molar ratio of pathogen-derived or VL9 peptide to HLA-E monomers significantly decreased binding to VISTA (Fig. 4a). We also observed that loading of a VL9 or a VL9-mimicking CMV peptide into our peptide-deficient HLA-E construct allowed binding of the monomer to CD94/NKG2A, validating that our monomer was of functional conformation to correctly load and present peptides for ligand interactions. These results strengthen our observations that VISTA preferentially engages peptide-deficient or weakly stabilized HLA-E conformers.

Fig. 4. Stable peptide loading in HLA-E prevents VISTA binding.

a, b Binding of PE-labeled tetramers of VISTA, LILRB2, or CD94/NKG2A to HLA-E monomers either without peptide (HLA-E_np) or peptide-loaded (10:1 molar ratio of peptide:HLA-E added before using in assay). Mean ± SD of$\ge$3 technical replicates from a representative experiment is shown. Binding of peptide-loaded monomers was compared to HLA-E without peptide by one-way ANOVA using Dunnett’s multiple comparisons test (**p < 0.01, ****p < 0.0001). b Melting points (T_m) of various refolded HLA-E monomers either with or without peptide.

Thermal stability of HLA-E:peptide complexes

We evaluated the thermal stability of HLA-E when empty or with VL9, HIV, and hsp60 peptides. Peptide-deficient HLA-E exhibited the lowest melting point (T_m) which was only incrementally higher when HIV or hsp60 peptide was loaded (Fig. 4b). As previously investigated, VL9 peptides show the highest thermal stability of the HLA-E complex^6,23. Our results reflect previous reports that pathogen-derived peptides in HLA-E exhibit lower melting points, which corresponds to shorter complex half-lives. This instability may indicate the presence of a heterogenous mixture of peptide-loaded and peptide-deficient or weakly stabilized conformers of HLA-E when refolded with weak binding peptides, consistent with recent findings that MHC-E molecules exhibit reduced intrinsic stability and strong dependence on peptide loading for durable surface expression²⁴.

VISTA tetramer binding to K562-E cells is lost with excess VL9 peptide loading, but not with HIV peptide

To evaluate cellular binding of VISTA to HLA-E, we utilized a genetically engineered cell line, K562 expressing HLA-E (K562-E). We incubated these cells overnight at 26^o C to increase the stability of any expressed peptide-free HLA-E^20,25. Cells were either unpulsed or exposed to excess HIV or V25 peptide. VISTA tetramer showed binding to both unpulsed and HIV-pulsed K562-E, whereas binding was eliminated with V25 (VMAPRTLFL) peptide loading (Figure S5a). We also observed CD94/NKG2A tetramer binding in all peptide conditions, indicating the potential for endogenous canonical peptide loading in spite of the lack of classical HLA surface expression. CD94/NKG2A tetramer staining was drastically increased by the loading of excess V25 peptide and was only slightly decreased by the addition of HIV peptide (Figure S5a, b). K562-E cells showed substantial staining with HLA-E antibody clone 3D12, confirming surface expression (Figure S5c).

VISTA/MHC-E interaction is not conserved in mice

We next evaluated whether murine VISTA interacts with the murine ortholog of HLA-E, Qa-1^b. We performed an ELISA using Klickmers of HLA-E or Qa-1b, either without peptide or with VL9 or Qa-1 determinant modifier peptide(Qdm), respectively. We observed no detectable binding of murine VISTA with empty Qa-1^b (Figure S5a). We further evaluated the interaction of HLA-E with cynomolgus or murine VISTA. We observed cross-species reactivity with cynomolgus VISTA and HLA-E but not murine VISTA (Figure S5b). The interaction of VISTA with HLA-E therefore does not appear to be conserved in mice but is conserved between HLA-E and cynomolgus VISTA.

VISTA and HLA-E are in interaction proximity across normal human tissues

To begin to determine biological relevance of the VISTA and HLA-E interaction, we screened a panel of 60 normal human tissue types using both immunofluorescent staining and an in-situ proximity ligation assay (isPLA). We detected HLA-E and VISTA expression in the majority of profiled normal tissues by immunofluorescence, including spleen, thymus, placenta, small intestine, and colon (Fig. 5). We went on to test for proximity ligation of HLA-E and VISTA in these tissues, to determine the possibility of interaction. We saw that VISTA and HLA-E showed proximity ligations in many tested normal tissues, particularly in tissues with known immunoregulatory function. The signal from HLA-E and VISTA proximity ligation was quantified in each tissue core by isPLA+ area (μm²) per 100 cells (-<8, 8>+<25, 25<++>50, +++>50) (Fig. 5 and Figure S6). The proximity ligation assay gave distinct results for each antibody pair and shows the possibility of interactions of HLA-E with NKG2A and VISTA in many of the same tissue compartments with immune infiltration.

Fig. 5. In-situ proximity ligation of VISTA and HLA-E in normal tissues.

First column: Detection of HLA-E (green), VISTA (pink), and nuclei/DAPI (blue) staining in normal human TMA cores. Representative tissue samples are shown. Second column: In-situ proximity ligation signal (isPLA + , yellow) of HLA-E and VISTA across multiple tissues from normal human TMA. Third column: isPLA (yellow) of HLA-E and NKG2A in sequential TMA slides. Area of positive isPLA signal per 100 cells in each core is quantified as -<8 μm², + 8–25 μm², ++ 25–75 μm². Representative tissue areas are shown. Scale bars equal 20 µm for IF panels and 10 µm for isPLA panels.

Discussion

We have validated the interaction between VISTA and HLA-E and demonstrated that this interaction is dependent on the peptide presented by HLA-E. In the presence of the canonical VL9 peptides, HLA-E engages the CD94/NKG2 receptor. However, when peptide occupancy is absent, suboptimal, or dynamically regulated, HLA-E can engage the ECD of VISTA. The observation that HLA-E interacts with different checkpoint molecules dependent on loaded peptide, and in the absent or suboptimal peptide occupancy, suggests that other interaction partners exist, as HLA-E has often been studied under the context of complex-stabilizing VL9 peptides. Recent work by another group shows an interaction of HLA-E with Stabilin 1/Stabilin 2 in absent or suboptimal peptide occupancy, although in the context of a mutationally stabilized HLA-E^26,27. Importantly, HLA-E conformers in our study did not contain any stabilizing mutations. We conclude that HLA-E interacts with distinct immunoregulatory molecules, potentially indicating context-dependent functions governed by its peptide repertoire in homeostatic or pathological contexts, as well as interaction with VISTA and LILRB1/2 in a peptide-deficient state.

Importantly, we do not propose that peptide-free HLA-E represents the dominant steady-state form on all cells. Rather, our data support that VISTA recognizes a less stable, peptide-receptive or weakly loaded conformer state of HLA-E that becomes accessible under specific biological conditions. Emerging evidence indicates that canonical MHC-E/HLA-E ligand presentation is not solely dictated by the presence of leader peptides, but is tightly regulated by peptide loading efficiency, intracellular trafficking, and inflammatory signaling. In the murine Qa-1b system, canonical Qdm peptide presentation requires inflammatory cues and a functional peptide-loading pathway⁴. In human systems, recent work demonstrates that VL9/HLA-E surface expression is similarly regulated and is not uniformly abundant across cell types, with myeloid cells uniquely displaying VL9/HLA-E even in the absence of inflammatory stimulation²⁸. Together, these findings support a model in which canonical VL9/HLA-E complexes dominate under certain conditions, but do not saturate all HLA-E molecules across all cellular contexts.

In the murine Qa-1b system, canonical Qdm peptide presentation requires inflammatory cues and an intact peptide-loading pathway⁴. In human systems, VL9/HLA-E surface expression is similarly regulated and varies across cell types, indicating that canonical peptide occupancy is context dependent rather than universally present²⁸. Simultaneously, VISTA is expressed on naïve T cells and has a role in peripheral T cell tolerance¹⁴. This information, along with our data showing proximity ligation in normal human tissues, gives evidence for a role of the interaction between peptide-free HLA-E and VISTA at homeostasis.

For the purposes of our work, we utilized CD94/NKG2A heterodimer as a positive control for HLA-E + VL9 interaction. We did not investigate peptide dependency of CD94/NKG2C compared to CD94/NKG2A, as this has been previously described²⁹. One limitation of this study is investigation using HLA-E*01:03 without evaluating HLA-E*01:01. These alleles differ at amino acid position 107, arginine (*01:01) or glycine (*01:03) located on a loop between β-strands in the α2 domain of the heavy chain³⁰. The *01:03 allele is expressed at higher levels than *01:01³¹. Additionally, stability measurements show that HLA-E*01:03 complex has a significantly higher melting temperature over HLA-E*01:01 independent of bound peptide. Assays in our study were performed primarily at room temperature, with the exception of ResoSens which reagents and plate during reading were kept at 30 °C. Temperatures lower than 37 °C favor the stability of peptide-free HLA-E complex²⁵. Our results show the T_m of peptide-deficient HLA-E monomer to be 41 °C, which is higher than previous reports, potentially due to differences in preparation protocols. We believe that weak binding peptides, leading to HLA-E complex with shorter half-lives, generate a heterogenous mixture of peptide-loaded and empty conformers. Accordingly, VISTA binding was only observed in the context of weak binding peptides supporting the concept that transient or peptide-receptive HLA-E conformations exist when optimal peptide stabilization is not achieved. Recent work²⁴ further demonstrates that MHC-E molecules exhibit reduced intrinsic stability and require peptide loading for durable surface expression, supporting the existence of transient or weakly stabilized conformers. We therefore propose that VISTA selectively engages this dynamically regulated subset of HLA-E molecules rather than a globally peptide-free population. At present, an antibody specific to peptide-deficient HLA-E does not exist, which would allow for isolation or direct identification of these peptide-free conformers.

We utilized two different antibody clones to HLA-E, 3D12 and 4D12, of which differences in epitope have been previously described²⁰. Because both 4D12 and 3D12 antibodies interfered with VISTA binding, despite recognizing separate epitopes of HLA-E, we were unable to make any hypotheses about interaction region from these results. However, these antibodies may alter the conformational stability of the HLA-E complex, thereby enhancing or disrupting its interactions with different ligands³². Critical amino acids for 4D12 binding are proposed to be 142, 144, and 151²⁰. AlphaFold3 results show binding of the primary VISTA loop (residues 117-124 on VISTA) to discontinuous epitope of HLA-E residues 7, 63–66, 70–73, 99, 159, 163, and 167. We propose that 4D12 binding may sterically hinder binding of VISTA to the peptide-binding groove of HLA-E.

In the context of cellular interactions of HLA-E and VISTA, we did not see substantial binding of VISTA tetramer to the surface of K562-E cells. Several reasons may contribute to this. First, K562-E are most likely presenting endogenous peptides, as evidenced by the clear binding of CD94/NKG2A tetramer even in unpulsed cells. Also, surface staining with tetramers would only account for trans membrane interactions and would not account for potential cis or intracellular interactions. VISTA is known to interact with several of its ligands in cis, and additional models are needed to investigate this possibility with HLA-E³³. Additionally, HLA-E and VISTA also exist as soluble proteins, and this may be relevant to the function of their interaction and explain the modest level of membrane interaction we observed^13,34. Moreover, attempts to manipulate peptide occupancy by acid stripping were not informative, likely because such treatments reduce the overall stability of surface HLA-E/β2m complexes rather than selectively enriching a stable peptide-free conformer. This interpretation is consistent with the intrinsic instability of peptide-deficient MHC-E conformers described in recent work²⁴.

We show that differences in pH did not affect HLA-E interactions with LILRB1/2 and CD94/NKG2A. In contrast, HLA-E tetramer interactions with VISTA were diminished at lower pH. VISTA is known to interact with at least one of its other ligands in a pH-dependent manner¹⁷. We found that in contrast to VISTA’s known interaction with PSGL-1, which occurs in acidic conditions, the interaction with HLA-E was favored at pH 7.4. The predicted model of binding site of VISTA to HLA-E includes three histidine residues, where the imidazole side chain becomes protonated at acidic pH while neutral at physiologic pH³⁵. Interestingly, the three histidines of VISTA identified to be involved in binding to HLA-E have also been shown to be necessary, but not sufficient for binding to PSGL-1¹⁷. Future experiments involving mutation of these residues are necessary to confirm their involvement in HLA-E binding.

We also observed that Qa-1^b, the murine ortholog of HLA-E, failed to interact with murine VISTA. A limitation of this assay was being performed at room temperature, similarly to assays performed with the human proteins. This may limit our ability to detect an interaction that occurs optimally at physiologic temperature (37°C). Interestingly, VISTA is the most conserved of the B7 family of proteins with 76% sequence identity between mouse and human³⁶ while human HLA-E and mouse Qa-1 share 73% homology². The primary sequence of human VISTA predicted to interact with HLA-E (VEIRHHHS) is conserved in cynomolgus macaques but differs from the mouse VISTA (VIELKNHHP). This lends support for the importance of this sequence in binding to HLA-E, although this will need to be experimentally validated. These interspecies differences will make evaluation of the function of this interaction in mouse models more challenging.

Finally, proximity ligation analysis of HLA-E and VISTA in normal tissues, revealed a higher level of interaction in normal tissues than VISTA and one if its recently described ligands, LRIG1¹⁶. We did not evaluate the cellular localization of HLA-E and VISTA. However, VISTA and HLA-E can be cell surface expressed, soluble, or enriched in late and recycling endosomes^34,37–39. HLA-E and VISTA being expressed on multiple cell types and in various cellular and cell-free compartments complicates the development of assays to assess the functional outcome of the interaction. As of yet, we lack a model or optimal assay to remove the effects of alternative binding partners.

Given our results, we have considered plausible biological functions for VISTA binding to HLA-E. One possibility is that VISTA binding in the HLA-E peptide-binding groove could stabilize peptide-deficient HLA-E until appropriate peptide is available for loading. The lack of binding at low pH could be functional to allow VISTA to release HLA-E in the endosome or lysosome for peptide loading. A lack of high-affinity peptides has been shown as a major limiting factor for HLA-E surface expression³⁸.VISTA may contribute to immune regulation by stabilization of the HLA-E complex when canonical peptides are in short supply, such as in the absence of inflammation. Consistent with these findings, we propose that VISTA functions as a sensor or regulator of peptide-receptive HLA-E conformers, enabling immune modulation in contexts where canonical peptide loading is limited, transient, or dynamically regulated. This model positions HLA-E as a dynamically regulated checkpoint hub, where peptide occupancy and conformational stability dictate selective engagement of distinct immune receptors.

Herein, we have established and validated the interaction between peptide-deficient or weakly stabilized HLA-E conformers and VISTA by protein assays and in silico modeling. We detected proximity ligation of HLA-E and VISTA in multiple normal human tissues. This interaction occurred at physiologic pH and is conserved in a primate model. Limitations of our study were the inability to detect interaction with murine proteins, and at this point, we can only surmise at the functional relevance of the interaction. Future work will focus on determining the outcome of the HLA-E and VISTA interaction at a cellular level. Given that both HLA-E and VISTA are immune checkpoints, it is essential to characterize the function of their interaction at homeostasis and in the context of cancer, autoimmunity, and infection.

Materials and methods

Recombinant proteins

Biotinylated recombinant human VISTA protein was obtained from Sino Biological (Paoli, PA, USA). Protein was quality checked in-house by endotoxin testing and analytical size-exclusion chromatography. Biotinylated CD94/NKG2A and VSIG3 were purchased from ACROBiosystems (Newark, DE, USA). Biotinylated LILRB1, LILRB2, PD-L1, and recombinant human VISTA for confirmation of results were purchased from R&D Systems (Minneapolis, MN, USA). HLA-E monomer refolds (biotinylated and unbiotinylated) were obtained from Resonant Sensors, Inc (RSI; Arlington, TX, USA). Tetrameric HLA-E conjugated to PE were purchased from RSI and Baylor MHC Tetramer Core Facility (Houston, TX, USA).

Label-free detection assay

Biotinylated ligands or control proteins were dilute to 5 mg/mL in assay buffer (PBS with 0.1% (w/v) BSA (MilliporeSigma; Burlington, MA, USA) and 0.1% Tween-20 (Sigma-Aldrich; St. Louis, MO, USA)) and immobilized on NeutrAvidin-coated Bionetics label-free microarray plates (Fisher Scientific; Waltham, MA, USA) at 30 °C. Unbound neutravidin sites were blocked with excess biotin (Thermo Fisher; Waltham, MA, USA). Antibody, protein, or HLA tetramers were added to the plate at specified concentrations in 0.1% BSA in PBST. Binding was monitored for 1 h in the ResoSens Ultra-Mab Pro instrument with integrated ResoVu software (RSI). At least 3 replicates per condition were performed in each experiment. Analyte binding was recorded, then plate was washed 3x before reading post-wash for 40 min. Resonance shift values were quantified in ResoVu software by subtraction of ligand-only background well values.

ELISA

All reagent additions and washes were performed in 0.1% BSA (Millipore Sigma) in PBS (Thermo Fisher) with 0.1% Tween-20 (Sigma-Aldrich). Biotinylated target protein monomers were added to Pierce Streptavidin Coated High Binding Capacity White 96-well plates with SuperBlock Blocking Buffer (Thermo Fisher). Additional blocking was performed with 5% milk. APC-conjugated Klickmers (Immudex; Copenhagen, Denmark) were made according to manufacturer’s instructions using biotinylated recombinant proteins. Tetramers or Klickmers® were added at the indicated concentrations and incubated for 30 min-1 h with gentle shaking. Wells were washed three times prior to reading fluorescence on a SpectraMax i3x plate reader (Molecular Devices; San Jose, CA, USA) using optimized endpoint fluorescence settings ((PE: ex.530/em.575), APC: ex.615/em.675). For pH-dependent binding assessment, buffer was adjusted to pH 6.0 or 7.4. Data was plotted in GraphPad Prism version 10.1.2. Data shown is the mean of at least 3 technical replicates per experiment and at least 2 experimental replicates.

Antibody blocking

HLA-E tetramers were incubated with specified concentration of αHLA-E, Clone 3D12 (BioLegend; San Diego, CA, USA) or Clone 4D12 (MBL; Tokyo, Japan) antibody for 30 min on ice prior to addition to ELISA.

Peptide loading

Peptides (Genscript; Piscataway, NJ, USA) were added to peptide-deficient HLA-E monomers (RSI) at a molar ratio of 10:1 and incubated for 10 min on ice. For cellular assays, HIV or VL9 peptide were added to cells at a concentration of 50 uM in cell culture media for 1 hr at 37 °C and then cells were washed before proceeding to assay.

Thermal stability assay

Assay was performed as in Barber et al. 2022²³, except that data was collected on a CFS Opus 96 Real-time PCR System (Bio-Rad; Hercules, CA, USA) and analyzed in CFX Maestro version 5.3.022.1030.

Cell lines

K562-E cells were obtained from Professor Thorbald van Hall, Leiden University Medical Center. Cells were cultured in IMDM media (ATCC; Manassas, VA, USA) with 1 μg/mL Puromycin (Millipore Sigma; Burlington, MA, USA) in a 37 °C/5% CO₂ humidified incubator. Incubator temperature was reduced to 26 °C for 24 h before assay.

Flow cytometry

Cells were plated at 2e5 cells/well in a V-bottom plate and washed with PBS. Zombie nIR (Biolegend; San Diego, CA, USA) was added at 1:1000 for 10 min at RT. Cells were washed and TruStain FcX (Biolegend) was added at 2 μL/well for 10 min on ice. Tetramers were made by adding biotinylated proteins at a 6:1 excess molar ratio to Streptavidin PE (Biolegend) and incubated for 30 min at RT. Tetramers were added to cells at 0.2 μg/test and incubated for 30 min on ice. Cells were washed 3x with 2% FBS in PBS. Data was acquired on a CytoFLEX SRT (Beckman Coulter, Indianapolis, IN, USA) and analyzed in FlowJo v. 10.8.1 (Becton Dickinson, Ashland, OR, USA).

Immunofluorescent staining and in-situ proximity ligation assay

Normal human tissue microarrays were purchased from Super Bio Chips (Seoul, Korea). Slides were baked at 60 °C for 1 hour, followed by xylene dewaxing and ethanol gradient rehydration. Antigen heat-induced epitope retrieval was performed in pH 6 sodium citrate buffer at 100 °C for 20 min. Primary antibodies for HLA-E (Clone MEM:E/02) and VISTA (Clone RM503) were incubated overnight at 4 °C. For proximity ligation, NaveniFlex Tissue MR Atto647N kit (Navinci; Uppsala, Sweden) was used according to manufacturer’s instructions. Slides were mounted with Prolong Diamond antifade mounting media with DAPI (Invitrogen; Carlsbad, CA, USA). Slides were imaged on Akoya Phenoimager (Akoya Biosciences; Marlborough, MA, USA) at 40× magnification. Data was analyzed in QuPath 0.5.1.

Statistics and reproducibility

Statistical analyses and graphing were performed using GraphPad Prism (version 8.2.2). For comparisons among three or more groups, one-way ANOVA followed by Šidáks or Dunnett’s multiple comparisons test was used. P values less than 0.05 were considered statistically significant. Specific statistical tests used for each experiment are indicated in the corresponding figure legends. Sample sizes (n) represent independent biological replicates unless otherwise indicated. Bars represent mean ± SD of ≥3 replicates per condition. Experiments were repeated independently to confirm reproducibility.

Inclusion and ethics statement

This study was conducted through collaborative research involving researchers from multiple institutions in accordance with relevant institutional and national guidelines. All authors contributed to the study and approved the final manuscript.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Conceptualization: C.R.W., J.W. Methodology: C.R.W., S. Ghaffari, G.O., S.Gimlin, K.U., J. W. Formal analysis: C.R.W., J.W. Investigation: C.R.W. Resources: N.K., J. W. Data curation: C.R.W. Writing-original draft: C.R.W. Writing- Review and editing: C.R.W., S.Ghaffari., C.W., M.D., S.Gimlin. Visualization: C.R.W., S.Ghaffari, Supervision: N.K., J.W. Project administration: N.K., J.W. Funding acquisition: J.W.

Peer review

Peer review information

Communications Biology thanks the anonymous reviewers for their contribution to the peer review of this work. Primary Handling Editor: Mengtan Xing. A peer review file is available.

Funding

Funding provided by Boehringer-Ingelheim International GmbH and The Univeristy of Texas at Arlington

Data availability

All numerical data underlying the graphs and charts in the manuscript are provided in SupplementaryData. All other data supporting the findings of this study are available from the corresponding author upon reasonable request.

Code availability

This paper dose not report original code.

Competing interests

J.W. was a consultant for Boehringer-Ingelheim International GmbH at the time of this study. S. Ghaffari was a consultant for Resonant Sensors Inc. The remaining authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s42003-026-10440-4.