NKG2A Is a Therapeutic Vulnerability in Immunotherapy Resistant MHC-I Heterogeneous Triple-Negative Breast Cancer

Abstract

Despite the success of immune checkpoint inhibition (ICI) in treating cancer, patients with triple-negative breast cancer (TNBC) often develop resistance to therapy, and the underlying mechanisms are unclear. MHC-I expression is essential for antigen presentation and T-cell-directed immunotherapy responses. This study demonstrates that TNBC patients display intratumor heterogeneity in regional MHC-I expression. In murine models, loss of MHC-I negates antitumor immunity and ICI response, whereas intratumor MHC-I heterogeneity leads to increased infiltration of natural killer (NK) cells in an IFNγ-dependent manner. Using spatial technologies, MHC-I heterogeneity is associated with clinical resistance to anti-programmed death (PD) L1 therapy and increased NK:T-cell ratios in human breast tumors. MHC-I heterogeneous tumors require NKG2A to suppress NK-cell function. Combining anti-NKG2A and anti-PD-L1 therapies restores complete response in heterogeneous MHC-I murine models, dependent on the presence of activated, tumor-infiltrating NK and CD8+ T cells. These results suggest that similar strategies may enhance patient benefit in clinical trials.

Significance: Clinical resistance to immunotherapy is common in breast cancer, and many patients will likely require combination therapy to maximize immunotherapeutic benefit. This study demonstrates that heterogeneous MHC-I expression drives resistance to anti-PD-L1 therapy and exposes NKG2A on NK cells as a target to overcome resistance. This article is featured in Selected Articles from This Issue, p. 201.

Figure 1.

Low tsMHC-I expression is associated with a lack of benefit to immunotherapy-treated metastatic (m) TNBC. A, mIF for HLA-A/B/C (MHC-I) and panCK (merged and single channel example shown) was performed on 84 baseline/pretherapy archived biopsy samples collected from patients with metastatic TNBC in a randomized phase II clinical trial of carboplatin ± atezolizumab. Scale bar, 20 µm. B, Progression-free survival was compared using the log-rank test between carboplatin ± atezolizumab metastatic TNBC patients with the highest tsMHC-I expression (top 33%) and those with the lowest tsMHC-I expression (bottom 33%). Confidence intervals for each arm are represented as red or green dotted lines and median survival for each arm are plotted as black dotted line.

Figure 2.

TNBCs display high tumor-specific MHC-I expression heterogeneity. A, mIF for HLA-A/B/C and panCK was performed on 295 breast tumors in tissue microarray format (n = 227 ER⁺; 9 HER2⁺/ER⁺; 7 HER2⁺/ER⁻; 52 TNBC) to obtain single-cell resolution MHC-I expression. The circle plot is a summary of the patients’ HLA-A/B/C expression. Each line in the center represents single patient tumors comprised of individual tumor cell HLA-A/B/C expression plotted as a dotted plot. Spikes are colored by the tumor's mean MHC-I intensity. The inner layer of the circle reflects the coefficient of variation of each tumor. The outer layer of the circle is labeled by breast cancer subtype. B, Total sample HLA-A/B/C IF intensity across breast cancer subtypes. C, Coefficient of variation of HLA-A/B/C expression within breast tumors. D, Hartigan's dip test was applied to determine whether the distribution of tsMHC-I expression is unimodal or multimodal. ****, P < 0.0001; **, P < 0.01; ns, not significant. Bonferroni-corrected t test was used for statistical comparisons. Box plots display the median, 25th, and 75th quantiles.

Figure 3.

Enforcing heterogeneity in MHC-I expression reshapes NK-cell infiltration. A, Flow cytometry histogram showing MHC-I (H2D^d) expression on live parental EMT6 and EMT6 B2M KO cells. B, Parental, MHC-I^HET, and B2M KO EMT6 tumor cells were injected subcutaneously in BALB/c mice. Mice were treated at 7-day intervals with IgG antibody or anti–PD-L1 antibody. Graphs show combined tumor growth and survival curves (n = 10 to 15 per group). C, Schematic representing MHC-I^HET tumor model implanted into the mammary fat pad of BALB/c mice. MHC-I^HET represents a 50% MHC-I null tumor. Created in Biorender licensed by Vanderbilt University Medical Center. D and E, Calculated z-score sum of NK and T-cell–related genes in untreated parental, MHC-I^HET, and B2M KO EMT6 tumors CD45⁺ microbead fractionated (n = 4 per group). F and G, Flow cytometry analysis of tumor-infiltrating CD3⁺ T cells and NK cells from untreated parental, MHC-I^HET, and B2M KO EMT6 tumors. CD45⁺ tumor-infiltrating leukocytes were bead isolated. F, CD3 T cells are gated on live CD45⁺ and CD3⁺. n = 9 per group. G, NK cells are gated on live CD45⁺, NKp46⁺, and CD3⁻ (n = 8 per group). H, Representative NK-cell and CD3⁺ T-cell flow cytometry for CD45⁺ microbead-fractionated EMT6 tumors. Analyzed by ANOVA followed by Tukey post hoc test and log-rank (Mantel–Cox) tests. *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001; bars, mean ± SEM.

Figure 4.

NK cells in MHC-I heterogeneous tumors are enriched with Itga2 and IL2 signaling-related genes. A–C, Single-cell RNA sequencing performed on untreated parental and MHC-I ^HETEMT6 tumors. A, UMAP of tumor-infiltrating NK cells using SingleR. B, Differentially expressed genes compared between the NK cells in parental and MHC-I^HET EMT6 tumors (colored genes all show absolute log FC > 1 and P < 1e−5). C, IL2-STAT5 single-sample gene set enrichment analysis (ssGSEA) was performed on the NK cells taking individual cells as the sample (P < 0.0001). D, Untreated parental and MHC-I^HET EMT6 tumors were CD45⁺ tumor-infiltrating leukocytes bead isolated. RNA was extracted and utilized for NanoString gene expression using the PanCancer Immune Pathways codeset (>700 immune-related genes). Analyzed by Student t test (n = 4). E, Flow cytometry analysis of CD49b expression tumor-infiltrating NK cells (CDp46⁺) from untreated parental and MHC-I^HET EMT6 tumors. Mean shown with SEM. Analyzed by Student t test. n = 3–4. *, P ≤ 0.05; **, P ≤ 0.01.

Figure 5.

Combined targeting of T and NK cells restores immunotherapy responses in heterogeneous MHC-I murine mammary tumors. A, Flow cytometry analysis of Qa-1 expression on MHC-I^HET tumors. Gated on live CD45⁻ cells. B, Flow cytometry analysis of NKG2A expression on the surfaces of NK and CD8⁺ T cells in the spleen and MHC-I^HET EMT6 tumors. NK cells are gated on live CD45⁺, NKp46⁺, and CD3⁻. CD3⁺ T cells are gated on live CD45⁺ and CD3⁺. Student t test was used for comparison (n = 5). C, MHC-I^HET EMT6 tumor cells were injected subcutaneously in BALB/c mice. Mice were treated at 7-day intervals with IgG antibody, anti–PD-L1 antibody, anti-NKG2A antibody, or a combination of anti-NKG2A and anti–PD-L1 antibodies. Graphs show combined tumor growth and survival curves and analyzed by ANOVA followed by Tukey post hoc test and log-rank (Mantel–Cox) tests, respectively (n = 5–10 per group). D, Flow cytometry histogram showing MHC-I (H2D^b) expression on live E0771 cells after stimulation in vitro for 48 hours with 50 ng/mL of IFNγ. E, Flow cytometry analysis of tumor-infiltrating NK and T cells from untreated E0771 and MHC-I^HET EMT6 tumors where CD45⁺ tumor-infiltrating leukocytes were bead isolated. NK cells are gated on live CD45⁺, NK1.1⁺, and CD3⁻. T cells are gated on live CD45⁺, CD3⁺ (n = 4–5 per group). F, E0771 tumor cells were injected subcutaneously in C57BL/6 mice. Mice were treated at 7-day intervals with IgG antibody, anti–PD-L1 antibody, anti–NKG2A antibody, or a combination of anti-NKG2A and anti–PD-L1 antibodies. Graphs show combined tumor growth and survival curves and were analyzed by ANOVA followed by Tukey post hoc test and log-rank (Mantel–Cox) tests, respectively (n = 5–10 per group). G, MHC-I^HET EMT6 or E0771 tumor cells were injected subcutaneously in BALB/c mice or C57BL/6 mice, respectively. Three days after tumor implantation, mice were treated with IgG, anti-CD8α, anti-NK1.1, or anti-aGM1 antibodies. Mice were treated at 7-day intervals with IgG antibody, anti–PD-L1 antibody, anti–NKG2A antibody, or a combination of anti-NKG2A and anti-PD-L1 antibodies. Graphs show combined survival curves and were analyzed by log-rank (Mantel–Cox) tests. Significance shown is nondepleted IgG control compared with both anti-CD8 and anti-aGM1 treated mice. n = 5 to 10 per group. H, Flow cytometry analysis of MHC-I^HET tumor MHC-I expression after treatment with IgG, anti-PD-L1, anti-NKG2A, or a combination of both anti-NKG2A and anti–PD-L1. EMT6 tumors resected at endpoint. Tumor cells are gated on live CD45⁻, H2D⁺. n = 5. I–K, Flow cytometry analysis of tumor-infiltrating NK and CD3⁺ T cells in MHC-I^HET EMT6 tumors treated IgG antibody, anti–PD-L1 antibody, anti-NKG2A antibody, or a combination of anti-NKG2A and anti–PD-L1 antibodies day 3 and day 7 after treatment. CD45⁺ tumor-infiltrating leukocytes were bead isolated. I, NK cells are gated on live CD45⁺ and CD3⁻ NKp46⁺. Analyzed by ANOVA followed by Tukey post hoc test. n = 3–5 per group. J, Flow cytometry analysis of tumor-infiltrating NK-cell degranulation in MHC-I^HET EMT6 tumors. NK cells are gated on live CD45⁺ and CD3⁻ NKp46⁺, CD107⁺. n = 3–5 per group. K, Flow cytometry analysis of tumor-infiltrating CD3⁺ T-cell degranulation in MHC-I^HET EMT6 tumors. NK cells are gated on live CD45⁺ and CD3⁻ NKp46⁺ (n = 3–5 per group). Analyzed by ANOVA followed by Tukey post hoc. *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001; ns, not significant.

Figure 6.

Tumor microenvironmental interferon gamma is required for NK-cell recruitment to tumors lacking MHC-I expression. A, Flow cytometry staining for intracellular IFNγ. B2M KO EMT6-IFNγ cells were treated with increasing doxycycline (Dox) concentrations (0, 62.5, 150, 250, 500, and 1,000 ng/mL) and incubated for 48 hours. B, IFNγ secretion after doxycycline treatment was measured in supernatant and detected by ELISA. Data represent mean ± SEM. C,B2M KO EMT6-IFNγ cells were treated with 62.5 ng/mL of doxycycline and collected after 48 hours. The supernatant containing secreted IFNγ was placed on parental EMT6 cells to induce PD-L1 expression. D, Flow cytometry analysis of B2M KO dox-inducible IFNγ EMT6 tumor-infiltrating NK cells and CD8 T cells with or without doxycycline. NK cells are gated on live CD45⁺, NKp46⁺, and CD3⁻. T cells are gated on live CD45⁺, CD3⁺. Normalized to tumor weight (n = 4). Analyzed by Student t test. E, Doxycycline inducible IFNγ B2M KO EMT6 tumor cells were injected subcutaneously in BALB/c mice. Mice were given water with or without doxycycline and treated at 7-day intervals with IgG antibody or anti-NKG2A antibody. Graphs show combined tumor growth and individual tumor sizes on day 28 after tumor inoculation (n = 4–10 per group). Data analyzed by ANOVA followed by Tukey post hoc test. F, Flow cytometry analysis of MHC-I^HET tumor-infiltrating NK cells from MHC-I^HET tumors grown in wild-type or Ifng^−/− BALB/c mice. G, Flow cytometry analysis of MHC-I^HET tumor-infiltrating NK cells from mice treated with either IgG or anti-CD8α depletion antibody. *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001; ns, not significant.

Figure 7.

Tumor-specific MHC-I expression is associated with the regional NK-cell and T-cell infiltration. A, mIF for HLA-E and panCK (merged and single channel example shown) was performed on 93 TNBC tumor microarrays. B, The expression HLA-E on tumor cells positively correlates with NK-cell and CD8 T-cell abundance. P = 0.00078, Spearman test. Log₁₀ normalization was performed on HLA-E intensity, CD8 T-cell counts, and NK-cell counts to improve visualization. C, Representative mIF staining for DAPI, panCK, HLA-A/B/C, and CD8 or CD56. Cell identity, spatial coordinates, and HLA-A/B/C expression were recorded for subsequent analysis of the tumor microenvironment. Tumor cells were further clustered with the DBSCAN algorithm to capture spatial tumor islands and the islands were labeled based on their MHC-I expression. Scale bar, 10 µm. D, CD8 T cells were clustered with their nearest tumor island to form a CD8–tumor pair. Box plot showing the average of adjusted CD8–tumor pair count per patient; ****, P < 0.0001; ***, P < 0.001; Bonferroni-corrected paired t test, ROI = 154. E, NK cells were clustered with their nearest tumor island to form NK–tumor pair. F, Box plot showing average of adjusted NK–tumor pair count per patient. *, P < 0.05, Bonferroni-corrected paired t test, ROI = 69. Scale bar, 10 µm. G, Patients from the NCT03206203 clinical trial were classified as HLA-A/B/C^het region dominant (labeled as abundant) or HLA-A/B/C^het region low (labeled as low). PFS was compared using the log-rank test between the two cohorts within carboplatin + atezolizumab arm (n = 41, P = 0.00071) and carboplatin arm (n = 39, P = 0.14). Confidence intervals for the corresponding patients were represented as yellow or blue dotted line and median survival for each cohort is plotted as a black dotted line. H, Normalized NK: CD8 cell ratio was calculated based on gene expression-imputed CD8 and NK cell abundance. *, P < 0.05, Wilcoxon test. All the box plots capture the median, and 25th and 75th quantile.