HLA-E and NKG2A Mediate Resistance to BCG Immunotherapy in Non-Muscle-Invasive Bladder Cancer

Abstract

Bacillus Calmette-Guérin (BCG) is the first-line therapy for high-grade non-muscle-invasive bladder cancer (NMIBC), yet many patients experience recurrence due to immune evasion. We identify HLA-E and NKG2A as mediators of adaptive resistance involving chronic activation of NK and T cells in BCG-unresponsive tumors. Prolonged IFN-γ exposure enhances HLA-E and PD-L1 expression on recurrent tumors, accompanied by the accumulation of NKG2A+ NK and CD8 T cells. HLA-E^high tumor cells preferentially cluster near CXCL12-rich stromal regions with dense effector cell presence, underscoring a spatially segregated tumor architecture. Although cytotoxic lymphocytes retain effector potential, their activity is restrained by HLA-E/NKG2A and PD-L1/PD-1 pathways located in their immediate neighborhood within the bladder tumor microenvironment. These data reveal a spatially organized immune escape program that limits anti-tumor immunity. Our findings support dually targeting NKG2A and PD-L1 checkpoint blockade as a rational, bladder-sparing strategy for patients with BCG-unresponsive NMIBC.

Keywords: BCG-unresponsive; Immunotherapy; NK cells; Non-muscle-invasive bladder cancer.

Figure 1:. HLA-E increases on tumor cells upon BCG treatment and associates with proximity with NKG2A⁺ NK cells and T cells

(A-D) IHC was performed on bladder tumor and adjacent non-involved tissues from BCG-naïve and BCG-unresponsive NMIBC patients. (A) Representative digital pathology analyses identifying tumor, adjacent tissue along with exposed areas of glass to be excluded from subsequent analyses. (B) Representative density map showing gradient of HLA-E tumor expression. (C) Summary analysis of frequency of tumor cells that are HLA-E-bright and HLA-E-dim/negative in BCG-naïve (n=17) and BCG-unresponsive (“unresp.”, n=24) NMIBC tumors. The p-value was obtained using an independent two-sided t-test. (D) Summary analysis of frequency of HLA-E^bright epithelial cells in tumor and adjacent, non-involved bladder tissues from BCG-unresponsive patients with NMIBC. The p-value was obtained using a paired t-test. Lines show matching samples from a same donor. (E-J) IHC was performed on NMIBC tumors (n=41 tumors). (E) Representative digital pathology analysis on one BCG-naïve and one BCG-unresponsive NMIBC tumor section highlighting nuclear expression of DAPI and identification of CD3⁺NKG2A⁺ T cells, CD3-NKG2A⁺ NK cells and tumor cells with bright or dim/negative expression of HLA-E. (F) Frequencies of NKG2A⁺ cells in BCG-naïve (n=17) and BCG-unresponsive (“unresp.”, n=24) tumors. (G) Frequencies of NK cells within the NKG2A⁺ cell compartment in BCG-naïve (n=17) and BCG-unresponsive (n=24) tumors. (H) Proximity analysis measuring the cell distance from CD3⁻ NKG2A⁺ NK cells or CD3⁺NKG2A⁺ T cells and tumor cells, depending on the tumor cell expression of HLA-E. P-values were assessed via independent two-sided t-test. (I-J) Interactions between HLA-E and NKG2A were profiled using an immunofluorescence-based proximity ligation assay in bladder tumors from NMIBC patients (n=10). (I) Representative staining in one BCG-naïve (left) and one BCG-unresponsive (right) patient. (J) Summary analysis of the interactions between HLA-E and NKG2A in BCG-naïve (n=6) and BCG-unresponsive (n=4) tumors. The p-value was assessed via independent two-sided t-test.

Figure 2:. BCG therapy induces IFN- secretion, that increases HLA-E and PD-L1 expression levels on tumor cells

(A) Olink protein analysis was performed on urine supernatants of patients undergoing BCG therapy. IFN- concentrations were compared between dose 1 and dose 6 of the first induction cycle (left panel) and between non-evidence of disease (absence of tumor) and recurrence cases of BCG-treated patients (right panel). Lines show matching samples from a same patient. P-values were obtained with paired (left) or unpaired (right) t-tests. (B) Flow cytometry was performed on primary CD45⁻ tumor cells from NMIBC patients and immortalized bladder tumor lines ex vivo and upon 24 hour stimulation with rhIFN- . Representative (left) and summary (right) expression of PD-L1 and HLA-E in triplicate experiments. (C-D) Since-cell RNA sequencing was performed on ex vivo bladder tumors (n=3 BCG-naïve, n=4 BCG-unresponsive). (C) UMAP clustering analysis. Each color represents a cluster. (D) Distribution of the IFNG^high cells across all clusters.

Figure 3:. HLA-E^high tumors are in proximity with cytolytic effector cells in the bladder tumor micro-environment

(A-G) Spatial transcriptomics sequencing (ST-seq) analysis was performed on NMIBC tumors (n=8). (A) UMAP visualization of the ST-clusters. Each color represents one cluster. (B) Relative composition of the ST-clusters across immune, stromal and tumor cell subtypes (columns) as defined by scRNAseq analysis. The size of each bubble indicates its relative enrichment and the shading of the surrounding boxes represents the significance by FDR (corrected p-values). (C) Relative enrichment of each ST-cluster in BCG-naïve (N=4) and BCG-unresponsive (N=4) tumors. (D) Distribution of each cluster in BCG-naïve and BCG-unresponsive tumor specimens. Each color represents one cluster. (E) Representative ST-seq images showing the distribution of ST-clusters in one BCG-naïve (left) and one BCG-unresponsive (right) NMIBC tumor specimens. Each color represents one cluster. (F) Representative ST-seq images from one BCG-naïve (left) and one BCG-unresponsive (right) NMIBC tumor specimens showing proximity analyses of HLA-E^low and HLA-E^high tumor cells as well as NK cells, CD8 T cells, and Tregs alone or in combination. Each dot represents one cell type. (G) Summary comparisons of proximity of HLA-E^low and HLA-E^high tumor cells to NK cells, CD8 T cells and Tregs. p-values were determined using two-sided Wilcoxon test.

Figure 4:. Geographic organization and cellular interactions in the tumor segregate the microenvironment through local chemotactic hubs while fueling tumor growth

(A-D) Single-cell RNA sequencing was performed on ex vivo bladder tumors (n=3 BCG-naïve, n=4 BCG-unresponsive) and tumor cells selected for further analyses. (A) UMAP visualization of bladder tumor cells from unsupervised clustering. Each color represents one cluster. (B) Average expression of HLA-E per cluster (left) and distribution of the bladder tumor cell clusters in BCG-naïve and BCG-unresponsive NMIBC tumors (right). (C) Pathway analysis of Hallmark gene networks that are significantly differentially expressed on HLA-E^high and HLA-E^low bladder tumor cells. (D) Differential expression in HLA-E^high and HLA-E^low tumors of key genes of interest. (E-L) Spatial transcriptomics sequencing (ST-seq) was performed on ex vivo bladder tumors (n=4 BCG-naïve, n=4 BCG-unresponsive) (E) Representative ST-seq image from one NMIBC tumor highlighting the proximity of stromal cells and tumor cells according to HLA-E expression in the tumors and CXCL12 expression in the stromal cells. (F) Summary graph of the proximity between HLA-E^low/high tumors and CXCL12^high spots. (G) Average expression in all cell populations of genes of interest that are upregulated in neutrophils and their ligands/receptors. (H) Representative ST-seq image from one NMIBC tumor highlighting the proximity of NK and CD8 T cells to neutrophils. (I) Summary graph of the neutrophils deconvolution scores in spots that are neighbors of NK/CD8T cells vs other cells. (J) Average expression in all cell populations of genes of interest that are upregulated in mReg DCs and their ligands/receptors. (K) Representative ST-seq image from one NMIBC tumor highlighting the proximity of NK and CD8 T cells to mReg DCs. (L) Summary graph of the mReg DCs deconvolution scores in spots that are neighbors of NK/CD8T cells vs other cells

Figure 5:. Bladder tumors are enriched with KLRC1^high NK and CD8 T cells that offer helper and cytolytic effector functions.

(A-I) Single-cell RNA sequencing was performed on ex vivo bladder tumors (n=3 BCG-naïve, n=4 BCG-unresponsive). Further analyses were then performed on selected NK cells or CD8 T cells. (A) UMAP visualization of NK cells from the bladder tumors, showing the Groups 1–6 clusters defined by Netskar et al (B) Distribution of the Groups 1–6 NK clusters in BCG-naïve and BCG-unresponsive tumors (C) For each of the six NK clusters, violin plot displaying KLRC1 expression (top) and list of key genes of interests (bottom) (D) Differential transcriptomic signature between KLRC1^high and KLRC1^low NK cells. (E) UMAP visualization of CD8 T cells from the bladder tumors, showing eight clusters from unbiased cluster analysis (F) Distribution of the eight CD8 T cell clusters in BCG-naïve and BCG-unresponsive tumors (G) For each of the eight CD8 T cell clusters, violin plot displaying KLRC1 expression (top) and list of key genes of interests (bottom) (H) Differential transcriptomic signature between KLRC1^high and KLRC1^low CD8 T cells. (I) Heatmap displaying the expression of key genes of interest between Groups 1–6 NK cells and KLRC1^high/low CD8 T cells.

Figure 6:. NKG2A and PD-L1 blockade increase NK and CD8 T cell-mediated antitumor activity in BCG-unresponsive patients

(A-C) Tumor-infiltrating lymphocytes (TILs) were expanded in vitro from BCG-unresponsive bladder tumors (n=4) during 13 days in the presence of IL-2 and CD3/CD28/CD2 T cell activator, prior to a 6-hour co-culture with K562 cell lines. NK and CD8 T cell functions were then assessed using flow cytometry. “WT” K562: Wild-type (HLA-E⁻) K562 ; “E+” K562: HLA-E-induced K562; “L1+E+” K562: HLA-E induced K562 that were stimulated with IFN- to induce PD-L1 expression. (A) Representative and (B) summary expression of IFN- and CD107a by NK cells and CD56⁺ CD8 T cells after co-culture with HLA-E^−/+ PD-L1^−/+ K562 cell lines in the presence or absence of anti-PDL1 and/or anti-NKG2A antibodies.