Hypoimmunogenic HLA-E Single Chain Inhibits Alloreactive Immune Responses

Autologous chimeric Ag receptor (CAR) T cells have transformed the treatment paradigm for patients with hematological malignancies. Yet, the production of these CAR-T cells is lengthy, expensive, and there is an inherent risk of manufacturing failure due to poor T cell fitness that may result from advanced disease and prior lines of therapy (1–3). Thus, CAR-T cells derived from healthy allogeneic donors is an attractive solution that could reduce patient wait time to treatment and cost by standardizing a scalable manufacturing process (4). However, HLA mismatch between an allogeneic donor and recipient patient challenges the clinical feasibility of allogeneic CAR-T cells. For example, disruption of the endogenous TCR in allogeneic T cells is necessary to prevent life-threatening graft-versus-host disease (5–7), but retention of non-self HLA molecules renders these cells vulnerable to immunologic rejection by the recipient (8), which may serve to limit their in vivo persistence. Consequently, additional genome engineering is required to facilitate clinical investigation of allogeneic CAR-T cells.

The elimination of surface HLA class I and HLA class II expression by inactivating β₂-microglobulin (B2M) and HLA class II transcriptional activator (CIITA), respectively, renders allogeneic CAR-T cells invisible to HLA-mismatched T cells (8, 9). However, the absence of HLA class I triggers missing-self recognition and lysis by NK cells (10). Previous attempts to restore NK cell tolerance to HLA class I–deficient T cells include genetic disruption of NK cell activating ligands (11) and adhesion molecules (12) or forced expression of an inhibitory ligand (11, 13–17). The latter approach necessitates that the inhibitory ligand is nonimmunogenic insofar as not to induce alloreactive T cell or Ab responses, and that the corresponding inhibitory receptor is widely expressed on NK cell subsets.

HLA-E is an attractive NK cell inhibitory ligand because this molecule is minimally polymorphic (18) and binds the heterodimeric inhibitory receptor NKG2A/CD94 (19, 20), which is expressed by ∼50% of circulating NK cells (21, 22). Similar to other HLA class I molecules, HLA-E requires noncovalent interactions with B2M and a cognate peptide ligand to correctly assemble and traffic to the cell surface, so it, as well, would not be expressed in B2M-deficient cells. Hence, previous attempts to restore HLA-E expression have required constructing a single-chain HLA-E (scHLA-E) molecule that fused together the HLA-E H chain, B2M, and loading peptide using intervening flexible linkers (23). After forced expression in xenografts (23, 24) and allogeneic human T cells or pluripotent stem cells (13–15, 25), scHLA-E has been shown to mitigate NK cell–dependent lysis.

In this study, we systematically optimized the scHLA-E molecule to improve cell surface expression and NK cell inhibitory capacity while minimizing immunogenic potential. To do so, we structurally modified the scHLA-E chassis by replacing the HLA-E transmembrane and cytoplasmic regions with that of the CD4 (4) coreceptor. We then functionally compared scHLA-E⁴ molecules fused to eight independent VL9 epitopes derived from the signal sequences of common HLA class Ia alleles and HLA-G. scHLA-E molecules that comprised VL9 epitopes from HLA-A and HLA-C allotypes, as well as disulfide trap mutations to anchor the loading peptide, maximized NK cell inhibition and minimally stimulated NKG2C⁺ NK cells. Lastly, our disulfide trap HLA-E⁴ (dtHLA-E⁴) variants induced HLA-E–specific CD8⁺ T cell responses, which necessitated introducing mutations into the α3 domain of HLA-E H chain to abrogate CD8-dependent T cell recognition. Taken together, these findings describe a hypoimmunogenic dtHLA-E⁴ molecule that effectively averts NK cell–driven rejection of HLA class I-deficient T cells.

HLA-E single-chain molecules were constructed in the following N- to C-terminal orientation: B2M signal peptide, VL9 loading peptide, (G₄S)₃ linker, B2M chain, (G₄S)₄ linker, and HLA-E*01:03 H chain. scHLA-E⁸ and scHLA-E⁴ molecules retained the HLA-E*01:03 extracellular domain but used the transmembrane domain and truncated cytoplasmic regions from CD8α or CD4, respectively. The VL9 epitopes VMAPRTLFL (v1), VMAPRTLLL (v2), VMAPRTLVL (v3), VMAPRTLIL (v4), VMAPRALLL (v5), VTAPRTVLL (v6), VMAPRTVLL (v7), and VTAPRTLLL (v8) were derived from signal sequences of HLA-G*01:01, HLA-A*01:01, HLA-A*02:01/HLA-C*02:01, HLA-C*01:02, HLA-C*07:01, HLA-B*35:01, HLA-B*07:02, and HLA-B*13:01, respectively. A comprehensive list of HLA class I alleles comprising these VL9 epitopes can be found elsewhere (26). scHLA-A*02 molecules were constructed by fusing the B2M signal peptide, B2M chain, (G₄S)₄ linker, and HLA-A*02:01 H chain. Disulfide trap scHLA-E⁴ molecules contained mutations within the (G₄S)₃ linker (GCGGS(G₄S)₂) and HLA-E*01:03 H chain (Y84C). Hypoimmunogenic scHLA-A*02 and dtHLA-E^4v4 molecules contained D227K, T228A, and A245V mutations within the α3 domain of the HLA H chain. All amino acid sequences are reported in Supplemental Table I. Each scHLA-E molecule was separated by a T2A linker to methotrexate-resistant dihydrofolate reductase (27). All gene fragments were codon optimized (IDT) and then custom synthesized and cloned into a self-inactivating lentiviral vector (GenScript).

Packaging plasmids encoding HIV Gag/Pol (pALD-GagPol), HIV Rev (pALD-Rev), and VSV glycoprotein (pALD-VSV-G) were purchased from Aldevron and used to generate lentiviral particles. The self-inactivating transfer plasmid and packaging plasmids were transfected into HEK293T cells using Lipofectamine 2000 (Life Technologies). Lentivirus supernatant was collected at 24 h posttransfection and then clarified through a 0.45-µm filter. Filtered lentivirus supernatant was mixed with PEG-it virus precipitation solution (System Biosciences) according to the manufacturer’s instructions and stored at 4°C overnight, after which the virus solution was concentrated for 30 min at 1500 × g, 4°C. The supernatant was aspirated, and the virus pellet was resuspended in 600 μl of ImmunoCult-XF T cell expansion medium (STEMCELL Technologies) and stored at −80°C. Lenti-X GoStix Plus (Takara) was used to quantify lentivirus titer.

Adult human leukopaks were purchased from HemaCare (Charles River) and then T cells were isolated using the StraightFrom Leukopak CD4/CD8 MicroBead kit (Miltenyi Biotec) following the manufacturer’s instructions and cryopreserved in CS10 (STEMCELL Technologies). T cells were thawed and resuspended at 1 × 10⁶ cells/ml in expansion medium consisting of ImmunoCult XF T cell expansion medium, 2 mM GlutaMAX, 25 mM HEPES buffer, and 1% penicillin-streptomycin, and 5% CTS immune cell SR (serum replacement) (Thermo Fisher Scientific). Expansion medium was supplemented with 10 ng/ml IL-7 and IL-15 (BioLegend). T cells were stimulated with ImmunoCult human CD3/CD28/CD2 T cell activator (STEMCELL Technologies) following the manufacturer’s protocol and incubated at 37°C, 5% CO₂, and 95% humidity. Forty-eight hours following stimulation, T cells were washed with PBS and resuspended in P3 buffer (Lonza) at 5 × 10⁷ cells/ml. T cells were then electroporated using the Lonza 4DNucleofector system, program DH-102. The electroporation reaction consisted of 2 µg of mRNA encoding ABE8.20m and 1 µg of B2M-specific single-guide RNA (sgRNA) from IDT (protospacer sequence, 5′-CTTACCCCACTTAACTATCT-3′) per 1 × 10⁶ cells. T cells were recovered in expansion medium at 1 × 10⁶ cells/ml and were transduced with lentiviral vector supernatant (∼100 ng of p24). T cell donors were transduced with independent lots of lentiviruses. Four days later, T cell cultures were divided and left untreated or treated with 220 nM methotrexate (MilliporeSigma). Medium was exchanged every second day to adjust cell concentration to 5 × 10⁵ cells/ml until day 13 when cells were cryopreserved in CS10.

Cells were stained in 50 µl of PBS containing 2% FBS and 2 mM EDTA with anti-human Abs: CD3 (OKT3), CD2 (RPA-2.10), CD4 (OKT4), CD8 (SK1), CD56 (5.1H11), CD107a (H4A3), HLA-ABC (W6/32), HLA-E (3D12), HLA-A2 (BB7.2), B2M (A17082A), CD86 (IT2.2), NKG2A (S19004C), and NKG2C (S19005E). Intracellular cytokines were detected using cell fixation and cell permeability kits (Invitrogen) per the manufacturer’s instructions using anti-human TNF-α (MAb11) (BioLegend). Surface-bound NKG2A and NKG2C recombinant proteins were detected by incubating T cells with biotinylated human NKG2A&CD94 protein, Fc Avitag and biotinylated human NKG2C&CD94 Protein, His Avitag (Acro Biosystems) followed by staining with fluorophore-labeled streptavidin (BioLegend). Flow cytometry data were acquired on the MACSQuant analyzer 16 (Miltenyi Biotec) and analyzed using FlowJo software v10 (Tree Star). All analyses were performed on live cells within the lymphocyte gate (forward scatter area versus side scatter area) following doublet exclusion (side scatter height versus side scatter width) and staining negative for fixable viability dye eFluor 780 (eBioscience).

PBMCs were obtained from human leukopaks by Ficoll separation. To generate effector cells, CD8⁺ T cells were isolated from PBMCs (donor 1) using a CD8 Microbead kit (Miltenyi Biotec). To generate stimulator cells, CD3⁺ cells were depleted from PBMCs of an HLA-mismatched donor (donor 2) using a CD3 Microbead kit (Miltenyi Biotec). The CD3-positive fraction from donor 2 was then used to generate B2M knockout (B2M^KO) T cells and unmodified T cells. All cell isolations were performed following the manufacturer’s protocols. Effector CD8⁺ T cells (donor 1) were cultured at a 1:1 ratio with stimulator CD3-depleted PBMCs (donor 2) for 1 wk in expansion medium supplemented with 300 U/ml IL-2 (Sartorius). Afterward, these activated effector CD8⁺ T cells were labeled with CellTrace Violet (Thermo Fisher Scientific) according to the manufacturer’s protocol, and then restimulated with a 1:1 ratio of B2M^KO or unmodified T cells (donor 2) at 20:1, 10:1, 5:1, 2.5:1, 1.25:1 or 0:1 E:T ratios. After 48 h, the assay was analyzed by flow cytometry to detect the change in the frequency of B2M^KO and unmodified T cells. Flow cytometry gating strategy follows viable lymphocytes/CD3⁺/CellTrace Violet⁻/with/without B2M.

NK cells were isolated from human leukopaks (Charles River) using a StraightFrom Leukopak REAlease CD56 MicroBead kit (Miltenyi Biotec) or CD56 MicroBeads (Miltenyi Biotec). NK cells were cultured for 3 d in expansion medium supplemented with 300 U/ml IL-2 and 5 ng/ml IL-15. Afterward, NK cells were cultured at different ratios with unmodified or B2M^KO allogeneic T cells. B2M^KO T cells engineered to express scHLA-E molecules were methotrexate (MTX) selected prior to assay setup. Unmodified and B2M^KO allogeneic T cells were combined at an equal ratio before adding NK cells to measure the relative change in frequency of B2M^KO T cells compared with unmodified T cells in the absence of NK cells. NK cell–dependent lysis of B2M^KO T cells was measured 48 h later by flow cytometry using the following formula: NK cell–dependent lysis (%) = 100 – [100 × (% survival in the presence of NK cells/% survival in the absence of NK cells)]. Flow cytometry gating strategy follows viable lymphocytes/CD56⁻/CD3⁺/CD4⁺CD8⁺/with/without HLA class I identified by HLA-ABC, B2M, or HLA-A2 staining. NK cell degranulation and TNF-α production were measured by culturing 1 × 10⁵ NK cells with 1 × 10⁵ B2M^KO or unmodified T cells, or alone (media only control). Anti-CD107a Ab was added at the start of stimulation followed by the addition of 1× monensin solution and 1× brefeldin A (BioLegend) 1 h later. Cells were incubated for 6 h total before Ab staining and analysis by flow cytometry. Flow cytometry gating strategy follows viable lymphocytes/CD3⁻/CD56⁺/CD107a⁺ or TNF-α⁺, or viable lymphocytes/CD3⁻/CD56⁺/NKG2A⁺ or NKG2C⁺/CD107a⁺ or TNF-α⁺.

Artificial APCs (aAPCs) were generated by transducing 5 × 10⁵ K-562 cells (American Type Culture Collection) with lentivirus encoding CD86. The resulting K.86 cell line recovered in R10 medium comprising RPMI 1640 supplemented with 2 mM GlutaMAX, 25 mM HEPES buffer, 1% penicillin-streptomycin, and 10% FBS (Thermo Fisher Scientific). After recovery, K.86 cells were transduced with lentiviral vectors encoding individual dtHLA-E⁴ or scHLA-A*02 molecules, and then expanded in R10. Cell surface expression of CD86 and B2M were detected on these aAPC lines by flow cytometry. To perform the HLA immunogenicity assays, aAPCs were treated with 10 µg/ml mitomycin C (MMC) (MilliporeSigma) for 3 h at 37°C, and then washed twice to remove residual MMC. T cells were thawed and mixed with MMC-treated aAPCs at a 2:1 ratio in expansion medium supplemented with 300 U/ml IL-2 and incubated at 37°C, 5% CO₂, and 95% humidity. At 7 d poststimulation, samples were analyzed for T cell proliferation and were restimulated with the appropriate aAPC and K.86 cell lines at a 1:1 ratio in the presence of an anti-CD107a Ab. One hour later, 1× monensin solution and 1× brefeldin A were added to the culture and then incubated at 37°C for 5 h before Ab staining and analysis by flow cytometry. The percentage of responding T cells was calculated by subtracting background production after stimulation with K.86 cells. Flow cytometry gating strategy follows viable lymphocytes/CD3⁺/CD8⁺ or CD4⁺/CD107a⁺ or TNF-α⁺.

Genomic DNA samples from B2M^KO T cells were prepared to determine on-target genomic editing efficiency as previously described (28). Briefly, ∼1 × 10⁶ cells were lysed using QuickExtract DNA extraction solution (Lucigen) following the manufacturer’s instructions. Two microliters of genomic DNA was added to a 25-µl PCR containing Q5 high-fidelity DNA polymerase (New England Biolabs) and 0.5 µM forward and 0.5 µM reverse primers. Forward and revers primer sequences are 5′-TGTCTTTCAGCAAGGACTGGTCTTTCTA-3′ and 5′-GACTCATTCAGGGTAGTATGGCCATAGA-3′, respectively. Products were then amplified using Illumina barcoding primer pairs, and then the amplicon was purified using solid phase reversible immobilization beads (Beckman Coulter) and quantified using a NanoDrop spectrophotometer (Thermo Fischer Scientific). Barcoded products were sequenced using an Illumina MiSeq instrument according to the manufacturer’s protocol.

Statistical comparisons of paired samples were performed using a two-sided nonparametric Wilcoxon matched-pairs signed rank test or Friedman test followed by Dunn’s test for multiple comparisons. Comparisons of unmatched samples were performed using a nonparametric Wilcoxon rank sum test. All statistical analyses were performed using GraphPad Prism version 9.3.0 (GraphPad).

Recognition of mismatched HLA class I alleles expressed on allogeneic CAR-T cells by the patient immune system may result in rapid clearance by cytotoxic CD8⁺ T cells. However, genome editing approaches to disrupt HLA class I expression in allogeneic CAR-T cells could alleviate recipient CD8⁺ T cell recognition, thereby improving their in vivo lifespan and therapeutic window (Fig. 1A). Thus, we leveraged an adenosine base editor (ABE8.20m) that when complexed with a sgRNA could disrupt B2M expression by inactivating a conserved splice donor site (exon|GT-intron) via a transition mutation (A·G) (Fig. 1B). To do so, primary human T cells were activated and then electroporated with mRNA encoding ABE8.20m and a B2M-specific sgRNA. This pairing achieved a mean on-target genomic editing efficiency of 95.5% (Fig. 1C), corresponding to a profound loss of HLA class I cell surface expression (Fig. 1D). The resulting B2M^KO T cells were invisible to alloreactive CD8⁺ T cells from an unrelated donor, whereas unmodified T cells were eliminated (Fig. 1E, 1F). Although disrupting B2M in allogeneic T cells prevented rejection by HLA-mismatched CD8⁺ T cells, in turn the absence of HLA class I expression could activate NK cells by missing self-recognition (Fig. 1A). As a result, B2M^KO T cells induced degranulation by NK cells (Fig. 1G) and were rapidly eliminated after in vitro culture with NK cells (Fig. 1H, 1I). These data indicate that the status of HLA class I expression on allogeneic T cells dictates their susceptibility to either T cell– or NK cell–mediated responses, and, importantly, additional genome engineering approaches are required to avert NK cell–driven allorejection.

To protect B2M^KO T cells from allorecognition, we constructed an scHLA-E molecule that could inhibit NK cell–mediated lysis through binding to NKG2A/CD94 (Fig. 2A). Four types of scHLA-E molecules were designed for our experiments (Fig. 2B). The scHLA-E dimer (D) comprised an HLA-E*01:03 H chain covalently fused to the signal peptide and mature chain of B2M via an intervening (G₄S)₄ linker, enabling the scHLA-E^D to bind endogenous peptide ligands for proper assembly and presentation. The scHLA-E trimer (T) contained an additional N-terminal (G₄S)₃ linker fused to a peptide ligand from the signal sequence of HLA-G*01 (VMAPRTLFL). In addition, we constructed scHLA-E^T molecules where the HLA-E transmembrane and cytoplasmic domains were replaced with those of CD8α (8) or CD4 (4), except to prevent undesired signaling, the cytoplasmic regions were truncated before the CXCP motif required for interactions with Lck (29–31). In this way, we hypothesized that surface expression of scHLA-E⁸ and scHLA-E⁴ may be greater than either construct containing the HLA-E cytoplasmic tail, which facilitates rapid recycling of endogenous HLA-E molecules (32).

scHLA-E molecules were cloned into a bicistronic lentiviral vector containing methotrexate-resistant dihydrofolate reductase (mDHFR) (27) separated by an intervening T2A peptide. After lentiviral transduction, each scHLA-E molecule was successfully detected on the cell surface of B2M^KO T cells by flow cytometry (Fig. 2C). Forced expression of each scHLA-E molecule substantially improved the expression level of HLA-E on transduced B2M^KO T cells over the endogenous level of HLA-E in unmodified T cells (Fig. 2D). Moreover, the inclusion of the VMAPRTLFL epitope stabilized longitudinal expression of the scHLA-E^T over the scHLA-E^D (Supplemental Fig. 1A), but this effect was nullified when the CD8α transmembrane domain and cytoplasmic region replaced their HLA-E counterparts (Supplemental Fig. 1A). The scHLA-E⁴ molecule exhibited similar surface expression levels as the scHLA-E^T at 4 d posttransduction, but thereafter the scHLA-E⁴ sustained a higher level of steady-state expression at latter timepoints of manufacturing (Fig. 2D, Supplemental Fig. 1A), including in both CD4⁺ and CD8⁺ T cells (Supplemental Fig. 1B). Furthermore, B2M^KO T cells expressing the scHLA-E⁴ construct exhibited a 2-fold increase in fluorescence intensity of bound recombinant NKG2A/CD94 protein relative to scHLA-E^T-positive T cells (Fig. 2E, 2F). These data suggest that the scHLA-E⁴ molecule is most likely to sustain adequate cell surface expression, which may confer B2M^KO T cells greater long-term protection against NK cell–mediated lysis.

Given that the scHLA-E⁴ molecule exhibited high-level cell surface expression, we carried this construct forward into functional assays. To do so, we leveraged expression of the mDHFR cassette to render scHLA-E⁴-positive B2M^KO T cells resistant to lymphotoxic concentrations of MTX, whereas nontransduced B2M^KO T cells were eliminated (Supplemental Fig. 1C). In this way, MTX selection facilitated evaluation of the scHLA-E⁴ in the absence of nontransduced B2M^KO T cells within the same culture that in turn could stimulate NK cells. Importantly, MTX treatment of B2M^KO T cells did not impact their susceptibility to alloreactive NK cell responses, as the frequency of degranulating and TNF-α⁺ NK cells (Supplemental Fig. 1D), as well as the magnitude of NK cell–dependent lysis (Supplemental Fig. 1E), was unchanged from stimulation with untreated B2M^KO T cells. We then MTX treated B2M^KO T cells transduced with scHLA-E⁴ and cultured these cells with NK cells from unrelated donors. Notably, the scHLA-E⁴ potently decreased the frequency of NKG2A⁺ and total NK cells that degranulated (Fig. 2G) and produced TNF-α (Fig. 2H). The ability of scHLA-E⁴ to decrease the frequency of total responding NK cells relative to nontransduced B2M^KO T cells correlated with the prevalence of NKG2A expression (Fig. 2I), which could explain the divergent extent to which scHLA-E⁴ decreases the susceptibility of B2M^KO T cells to NK cells from unique donors. Furthermore, the scHLA-E⁴ molecule mitigated in vitro NK cell–driven lysis (Fig. 2J) relative to stimulation with nontransduced B2M^KO T cells. Taken together, inclusion of the CD4 transmembrane and cytoplasmic components into scHLA-E^T improved surface expression, resulting in a molecule that inhibits NK cell reactivity against B2M^KO T cells.

HLA-E binds nonamer peptides derived from the signal sequences of HLA-A, HLA-B, HLA-C, and HLA-G (33, 34). However, only a subset of HLA class I signal sequence variants are efficiently processed to generate epitopes that bind HLA-E and engage NKG2A/CD94 on NK cells (26). In this study, we used the scHLA-E⁴ chassis to evaluate whether eight different VL9 epitopes from common HLA-A, HLA-B, and HLA-C allotypes and HLA-G stabilize scHLA-E⁴ expression and impact its ability to inhibit NK cells (Fig. 3A). The scHLA-E⁴ variants exhibited differential cell surface expression levels, where molecules comprising the VMAPRTLFL (v1), VMAPRTLLL (v2), VMAPRTLIL (v4), and VMAPRTVLL (v7) epitopes were highly expressed at near-equivalent levels (Fig. 3B, 3C). In contrast, scHLA-E⁴ molecules fused to HLA-B allotype-derived VL9 epitopes with threonine at position 2 (2T), VTAPRTVLL (v6) and VTAPRTLLL (v8), expressed poorly on the cell surface (Fig. 3B, 3C), which is congruent with reports indicating that these peptides bind HLA-E with lower affinity (20).

We next determined the inhibitory potential of each scHLA-E⁴ variant to attenuate NK cell reactivity to B2M^KO T cells. Nontransduced B2M^KO T cells stimulated NKG2A⁺ NK cells responses, but importantly, each scHLA-E⁴ variant considerably decreased the frequency of CD107a⁺ (Fig. 3D, 3E) and TNF-α⁺ (Fig. 3D, 3F) NKG2A⁺ NK cells to levels comparable to stimulation with unmodified T cells. Moreover, each scHLA-E⁴ molecule exerted a substantial inhibitory effect such that the frequencies of responding NK cells were also reduced at the population level (Fig. 3G, 3H). Notably, the scHLA-E⁴ variants comprising the VL9 epitopes v1, v4, and v7 demonstrated enhanced inhibitory potential relative to other peptide variants (Fig. 3E, 3G). Interestingly, the scHLA-E⁴ variants comprising v1, v2, v4, and v7 peptides sustained the highest surface expression (Fig. 3B, 3C), but unlike the rest, the scHLA-E^4v2 molecule did not exhibit enhanced inhibitory potential compared with the other variants (Fig. 3E, 3G). Lastly, all scHLA-E⁴ molecules, except scHLA-E^4v8, attenuated NK cell–driven lysis of B2M^KO T cells, even reducing lysis by as much as 50% on average relative to nontransduced B2M^KO T cells (Fig. 3I). The scHLA-E⁴ variants presenting peptides from HLA-A (v3) and HLA-C (v4) allotypes prevented NK cell–dependent lysis of B2M^KO T cells to a greater extent than scHLA-E^4v2 and scHLA-E^4v8. Taken together, these data indicate that selection of a VL9 epitope can substantially impact both the surface expression and inhibitory capacity of scHLA-E⁴ molecules.

We further modified the scHLA-E⁴ variants by incorporating mutations that stabilize the VL9 peptides into the HLA-E binding groove to improve expression and inhibitory potency. To do so, we introduced cysteine residues into the flexible (G₄S)₃ linker (GCGG(G₄S)₂) and tyrosine-84 (Y84C) of the HLA-E H chain (Fig. 4A). The resulting disulfide trap has been demonstrated to anchor relatively weak binding peptides and prolong cell surface half-life of scHLA class Ia trimers (35). The incorporation of the disulfide trap into each scHLA-E⁴ variant improved cell surface expression over their respective scHLA-E⁴ counterparts (Fig. 4B). Of note, dtHLA-E^4v6 and dtHLA-E^4v8 molecules, which comprised the VL9 epitope with the 2T substitutions, exhibited a 3- and 4-fold increase in surface expression, respectively (Fig. 4B). We then evaluated whether dtHLA-E⁴ variants could mitigate NK cell responses against B2M^KO T cells. Indeed, each dtHLA-E⁴ variant reduced the frequency of CD107a⁺ and TNF-α⁺ NKG2A⁺ NK cells (Fig. 4C, 4D) and total NK cells (Supplemental Fig. 2A, 2B), as well as mitigated NK cell–driven lysis of B2M^KO T cells (Fig. 4E). We previously showed that VL9 epitopes modulate the inhibitory potential of scHLA-E⁴ molecules (Fig. 3E, 3G, 3I), but now, we failed to detect any substantial difference in potency among the dtHLA-E⁴ variants (Fig. 4C, 4D), although B2M^KO T cells expressing dtHLA-E⁴ molecules bound to v3 and v4 peptides exhibited the lowest degree of NK cell–dependent lysis (Fig. 4E).

Given that the dtHLA-E⁴ variants exhibited comparable abilities to inhibit NK cells, we assessed whether the inclusion of the disulfide trap improved inhibitory activity over their scHLA-E⁴ counterparts. Inclusion of the disulfide trap mutations largely improved the ability of scHLA-E⁴ molecules to reduce the frequency of CD107a⁺ and/or TNF-α⁺ NKG2A⁺ NK cells (Fig. 4F, 4G) and total NK cells (Supplemental Fig. 2C, 2D) and reduce NK cell–dependent lysis of B2M^KO T cells to a greater extent than their scHLA-E⁴ counterparts (Fig. 4H). Strikingly, the addition of the disulfide trap substantially enhanced the function of scHLA-E⁴ molecules comprising v2, v6, and v8 epitopes, which previously underperformed relative to the other scHLA-E⁴ molecules (Fig. 3E, 3G, 3I). However, the inclusion of the disulfide trap did not similarly augment NK cell inhibition of high-activity molecules, such as scHLA-E^4v7 (Fig. 4F–H). Collectively, these data indicate that introducing the disulfide trap can enhance both the surface expression and inhibitory potency of scHLA-E⁴ variants comprising different VL9 epitopes. This effect was emphasized in VL9 epitopes from HLA-B allotypes with the 2T substitution, which is consistent with the ability of the disulfide trap to anchor low-affinity peptides.

HLA-E binds to the activating receptor NKG2C/CD94, albeit with lower affinity than NKG2A (36), so we investigated whether dtHLA-E⁴-driven activation of NKG2C⁺ NK cells could nullify the protective benefit of these molecules. We found that B2M^KO T cells expressing dtHLA-E^4v1 bound recombinant human NKG2C/CD94 protein greater than endogenous HLA-E by unmodified T cells (Fig. 5A), and each dtHLA-E⁴ variant markedly increased the frequency of degranulating NKG2C⁺ NK cells over stimulation with nontransduced B2M^KO T cells (Fig. 5B, 5C). The dtHLA-E^4v1 molecule induced the greatest NKG2C⁺ NK cell degranulation, whereas in comparison the dtHLA-E⁴ molecules comprising v3, v4, and v5 epitopes from HLA-A and HLA-C allotypes exhibited lower activation of NKG2C⁺ NK cells (Fig. 5C). Importantly, inclusion of the disulfide trap did not induce greater NKG2C⁺ NK cell degranulation compared with their matched scHLA-E⁴ variants (Supplemental Fig. 2E). Despite each dtHLA-E⁴ driving activation of NKG2C⁺ NK cells, all variants previously mitigated NK cell–driven lysis of B2M^KO T cells (Fig. 4E), indicating that NKG2C⁺ NK cells were too infrequent to counter the inhibitory effect afforded by these molecules. Indeed, NKG2C expression among those donors (termed NKG2C^low donors) was <10% (0.25–6.02%), whereas NKG2A expression exceeded 50% (58.2–73.2%) of total NK cells (Fig. 5D, 5E). We next identified NKG2C^high donors in which the frequency of NKG2C expression surpassed 10% (15–77.9%) of total NK cells (Fig. 5D, 5E) and then determined which dtHLA-E⁴ variants could avert rejection of B2M^KO T cells. In contrast to our previous findings, now only dtHLA-E⁴ molecules comprising v3, v4, and v5 epitopes reduced NK cell–dependent lysis of B2M^KO T cells by NKG2C^high donors (Fig. 5F). Collectively, these findings support utilizing a dtHLA-E⁴ variant fused to either VMAPRTLVL (v3), VMAPRTLIL (v4), or VMAPRALLL (v5) to maximize NK cell inhibition through NKG2A engagement and decrease activation of NKG2C⁺ NK cells.

Given that HLA-E–restricted CD8⁺ T cell responses against self and nonself VL9 epitopes have been identified from human donors (37–39), we hypothesized that supraphysiologic cell surface expression of dtHLA-E⁴ variants could elicit recipient CD8⁺ T cell responses to HLA-E–bound peptide ligands. To examine the immunogenic potential of dtHLA-E⁴ variants, we first generated aAPCs by transducing K-562 cells that are devoid of endogenous HLA expression (40) with lentiviruses encoding the CD86 costimulatory ligand and each dtHLA-E⁴ variant (Supplemental Fig. 3A, 3B). Next, T cells from a panel of human donors were stimulated with aAPCs, but we did not detect an increase in the frequency of responding CD8⁺ T cells after subtracting the signal from stimulation with K-562 cells expressing CD86 alone (K.86) (Fig. 6A). This finding suggests that the population of pre-existing HLA-E–specific T cells may be too infrequent for quantification, so we then leveraged the aAPC platform to expand these low-frequency precursors in vitro. One week after primary stimulation, these T cells were restimulated with their respective aAPCs and K.86 control cells, and notably we detected a substantial increase in the frequency of responding CD107a⁺ and TNF-α⁺ CD8⁺ T cells (Fig. 6B, 6C), but not CD4⁺ T cells (Supplemental Fig. 3C, 3D). These findings are indicative of functional dtHLA-E⁴–specific CD8⁺ T cell responses, which could limit the in vivo persistence of allogeneic B2M^KO T cells expressing dtHLA-E⁴.

One approach to disrupt CD8⁺ T cell recognition is to abrogate the CD8/HLA interaction by introducing mutations into critical CD8 contact sites within the α3 domain of the HLA class I H chain (41, 42). To demonstrate this, we generated aAPCs expressing scHLA-A*02 molecules comprising either the wild-type HLA-A*02 H chain (scA2) or the hypoimmunogenic α3 domain mutations D227K, T228A, and A245V (hypoA2) (Supplemental Fig. 3A, 3B). T cells from HLA-A*02⁺ and HLA-A*02⁻ donors were then stimulated with each aAPC and K.86 cell line. As expected, aAPC.scA2 cells induced CD8⁺ T cell expansion mainly driven by T cells from mismatched HLA-A*02 donors (Fig. 6D). In contrast, T cells stimulated by aAPC.hypoA2 cells failed to expand to the same extent, and instead exhibited near equivalent persistence to T cells activated by K.86 control cells (Fig. 6D). CD4⁺ T cells did not expand in response to either aAPC, confirming that this response is mediated by HLA class I–restricted CD8⁺ T cells (Supplemental Fig. 3E). Importantly, the hypoA2 molecule did not restimulate responses by HLA-A*02-specific CD8⁺ T cells that were previously primed by aAPC.A2 cells (Fig. 6E, 6F). These data indicate that α3 domain mutations are sufficient to abrogate CD8-dependent T cell responses in the context of HLA-A*02.

Amino acid residues critical for CD8 binding to HLA class I are well conserved, and indeed, the α3 domain of HLA-E*01:03 contains the amino acids D227, T228, and A245 (Fig. 6G). Thus, we introduced these mutations into dtHLA-E^4v4, which previously exhibited potent NK cell inhibitory potential, and then engineered aAPCs to express this molecule (Supplemental Fig. 3A, 3B). Importantly, the resulting hypoHLA-E^4v4 substantially reduced the frequency of CD107a⁺ and TNF-α⁺ CD8⁺ T cells that were previously primed with aAPCs expressing the nonmutated dtHLA-E^4v4 variant (Fig. 6H, 6I). Moreover, the incorporation of α3 mutations did not interfere with the ability of the dtHLA-E^4v4 molecule to inhibit NK cell–dependent lysis of B2M^KO T cells (Fig. 6J). Taken together, these data describe the development of hypoHLA-E⁴ that is capable of evading CD8-dependent T cell responses while retaining NK cell inhibitory capacity.

In this study, we systematically constructed a scHLA-E molecule that exhibited improved surface expression and enhanced NK cell inhibitory function, as well as decreased the risk of immunogenicity. Our scHLA-E comprised the CD4 transmembrane domain and truncated cytoplasmic region, as well as disulfide trap mutations to anchor an HLA class I signal peptide-derived VL9 epitope, and, lastly, mutations in the α3 domain of HLA-E to abrogate CD8 coreceptor binding. Forced expression of a hypoimmunogenic dtHLA-E⁴ molecule in B2M^KO T cells substantially mitigated NK cell and CD8⁺ T cell reactivity. These findings support the utility of this molecule to increase the therapeutic potential of allogeneic CAR-T cells by extending their in vivo persistence.

NK cell activity is governed by the balance of activating and inhibitory signals, and attempts to restore tolerance to HLA class I–deficient cells have overexpressed inhibitory molecules, including HLA class Ia killer Ig-like receptor (KIR) ligands (43), HLA-G (11, 13), CD47 (16, 17), PD-L1 (16), and HLA-E (11, 13–15). However, the corresponding immune checkpoint receptors for HLA-G (LILRB1), CD47 (SIRPα), and PD-1 (PD-L1) are restricted to minor subpopulations of NK cells and/or require cytokine stimulation to induce meaningful expression at steady state (44–46), which challenges their ability to broadly inhibit NK cells. Likewise, forced expression of an HLA class Ia single-chain molecule may be insufficient to inhibit the total NK cell population given that NK cells display a variegated expression pattern of inhibitory KIRs (47, 48). The extensive genetic diversity of KIRs influences both KIR expression level and frequency of KIR⁺ NK cells (49). As a result, individual NK cells may only express a subset of available KIRs that can sense allogeneic cells with altered expression of HLA class I or that do not express the appropriate KIR ligand. Given that NKG2A inversely correlates with KIR expression (50, 51), NK cell repertoires characterized by low KIR expression or KIR-negative NK cells are reported to rely on NKG2A expression for functional competence (52). In these individuals, forced expression of a scHLA-E molecule may be more effective at inhibiting NK cells than in individuals with KIR-dominated repertoires. Notably, NK cell repertoires are likely skewed following lymphodepleting chemotherapy, which is required therapy prior to the infusion of CAR-T cells. Indeed, NK cells are first to recover after lymphodepleting chemotherapy (53, 54), and data following hematopoietic stem cell transplant indicate that early repopulating NK cells are predominately immature and ∼90% are NKG2A⁺ (55). These findings imply that targeting the NKG2A inhibitory pathway with an HLA-E–based strategy could inhibit most rebounding NK cells following conditioning, thereby enabling maximal CAR-T cell expansion, which is a correlate of long-term remission in patients with hematological malignancies (56).

To restore HLA-E expression in B2M-disrupted T cells, it was necessary to construct a scHLA-E^T molecule that couples the HLA-E H chain, B2M L chain, and peptide ligand. Previous reports indicated that the scHLA-E^T in this format exhibited short cell surface half-life and stability (23). This phenomenon may be due to the cytoplasmic tail of HLA-E, which promotes intracellular accumulation and rapid internalization relative to HLA class Ia molecules (32). We found that replacing the HLA-E transmembrane and cytoplasmic regions with the corresponding domains from CD8α decreased surface expression relative to scHLA-E^T, which may reflect improper assembly due to the propensity of the CD8α transmembrane domain to homodimerize (57). In contrast, the scHLA-E⁴ molecule sustained the highest level of surface expression, which could be attributed to the truncated cytoplasmic region because critical serine residues were omitted that when phosphorylated mediate CD4 coreceptor internalization (58, 59). We suspect that sustaining high-level surface expression of scHLA-E⁴ will be necessary to maximize long-term protection of HLA class I–deficient T cells from NK cell–mediated lysis.

We then functionally characterized the scHLA-E⁴ molecule bound to the HLA-G*01 signal peptide-derived epitope VMAPRTLFL. However, polymorphisms within HLA class Ia signal peptides yield alternative VL9 epitopes that bind HLA-E and influence its cell surface stability and affinity for NKG2X/CD94 receptors (26). Indeed, scHLA-E⁴ VL9 variants were differentially expressed on the surface, and molecules bound to HLA-B allotype-derived epitopes with a 2T substitution, VTAPRTVLL (v6) and VTAPRTLLL (v8), exhibited the lowest surface expression. Notably, disulfide trap mutations improved the expression of all dtHLA-E⁴ VL9 variants, particularly for weak binding HLA-B allotype-derived epitopes. These findings are congruent with previous observations that disulfide bond engineering generates robust scHLA complexes by anchoring low-affinity epitopes into the HLA binding groove, thereby resisting displacement from high-affinity competitor peptides (35). Moreover, all dtHLA-E⁴ variants, except dtHLA-E^4v5 and dtHLA-E^4v7, exhibited greater NK cell inhibitory capacity than their scHLA-E⁴ counterparts, suggesting that surface stabilization of these molecules augmented their inhibitory potential.

The NK cell donors from our initial assays were biased toward NKG2A expression, but HLA-E also binds the NKG2C/CD94 heterodimer (20, 36) and may stimulate adaptive NKG2C⁺ NK cells. Adaptive NK cells expand in response to human CMV infection (60), meaning some individuals may harbor a sizable frequency of NKG2C⁺ NK cells that could nullifying the protective benefit of dtHLA-E⁴ molecules. Indeed, in NKG2C^high NK cell donors the dtHLA-E⁴ variant comprising the VMAPRTLFL (v1) epitope drove substantial activation of NKG2C⁺ NK cells, whereas epitopes derived from HLA-A and HLA-C allotypes, VMAPRTLVL (v3), VMAPRTLIL (v4), and VMAPRALLL (v5), minimized the magnitude of their responses. These data substantiated previous findings (26) and indicate that HLA class Ia signal peptide-derived VL9 epitopes influence NKG2X/CD94 engagement.

Although HLA-E is minimally polymorphic, we suspected that forced expression of dtHLA-E⁴ VL9 variants may induce detrimental CD8⁺ T cell–driven responses. For example, HLA-E–specific T cells may recognize, on allogeneic cells, HLA-E bound to nonself peptides (39), including the human CMV UL40 VL9 mimic epitopes VMAPRTLLL (v2), VMAPRTLVL (v3), and VMAPRTLIL (v4) (37, 38). Moreover, it is possible to activate self-reactive T cells by providing abundant stimulation (61), which suggests that supraphysiologic cell surface expression of dtHLA-E⁴ molecules presenting VL9 epitopes of host origin could also stimulate autoreactive CD8⁺ T cells. To this end, we detected in vitro HLA-E–specific T cell responses to aAPCs expressing dtHLA-E⁴ VL9 variants, which necessitated additional protein engineering to mitigate their immunogenicity. CD8 serves a critical role in the generation of HLA class I–restricted T cell responses, particularly when T cells display low-affinity autoreactive or alloreactive TCRs (62–65). Indeed, disruption of the CD8–HLA interaction using anti-CD8 nondepleting Abs has ameliorated pathogenesis of autoimmune type 1 diabetes (66–68), suggesting that a similar approach could prevent priming of dtHLA-E⁴–specific CD8⁺ T cells. Thus, we introduced mutations (D227K/T228/A245V) into the α3 domain of HLA-E. These mutations have been previously described to abolish CD8 binding to HLA class Ia molecules (41, 42), and when incorporated into dtHLA-E^4v4 these mutations substantially reduced CD8⁺ T cell reactivity without compromising its ability to inhibit NK cell–mediated lysis.

Notably, if the TCR–dtHLA-E⁴ interaction is of sufficient affinity, it may obviate CD8 coreceptor dependence (64, 69), thereby rendering these allospecific T cells insensitive to the α3 domain mutations. TCRs have been reported to interact with HLA-E bound to self-peptides with micromolar affinity; however, these T cells are hyporesponsive, as their activity is regulated by the expression of HLA-C–specific inhibitory KIRs (70). In our setting, it is conceivable that the activation of CD8-independent HLA-E–reactive T cells could eliminate allogeneic cells expressing dtHLA-E⁴, especially upon B2M disruption, which abrogates expression of inhibitory KIR ligands. Thus, it may be necessary to include a compensatory approach to inhibit recipient T cell–mediated allorejection. We recently demonstrated that short-course immunosuppressant therapy comprising rapamycin or tacrolimus permitted the engraftment and expansion of HLA-mismatched CAR-T cells in lymphoreplete humanized mice (71). Genetic inactivation of FKBP1A was necessary to render CAR-T cells drug-resistant, and this strategy could be combined with multiplex base editing of TCR and B2M to enable treatment with allogeneic CAR-T cells expressing dtHLA-E⁴.

dtHLA-E⁴ variants bound to VMAPRTLVL (v3), VMAPRTLIL (v4), or VMAPRALLL (v5) substantially protected B2M^KO T cells from lysis by NK cells that were biased for either NKG2A or NKG2C expression. As well, the insertion of mutations into the HLA-E α3 domain was necessary to mitigate the likelihood of stimulating host CD8⁺ T cell responses. Although we evaluated these dtHLA-E⁴ variants in allogeneic T cells, forced expression of these molecules may also facilitate the stable engraftment of pluripotent stem cells (25), as well as allogeneic and xenogeneic tissues (23, 24, 72). However, it may be necessary to enhance resistance to NK cell–driven rejection, particularly against cytolytic NKG2A⁻KIR⁺CD56^dim NK cell subsets. To accomplish this, we could envision combining expression of a hypoHLA-E⁴ molecule with multiplex base editing of NK cell adhesion (12) and activating ligands (11, 73), which have demonstrated the ability on their own to inhibit NK cell alloreactivity. Collectively, these findings demonstrate the systematic development and comprehensive in vitro characterization of hypoimmunogenic dtHLA-E⁴ molecules to increase the therapeutic potential of allogeneic cell therapies by overcoming allorejection.

All authors were employees of Beam Therapeutics when the work was conducted and are shareholders in the company. Beam Therapeutics has filed patent applications based on this work. Funding was provided by Beam Therapeutics.

We thank J. Decker and the Lab Automation team at Beam Therapeutics for next-generation sequencing support. We acknowledge W. Schmidt, N. Mastrangelo and E. Chiang for facilitating patent filings based on this work. Some figures were created using BioRender.com.