BCG therapy downregulates HLA-I on malignant cells to subvert antitumor immune responses in bladder cancer

Patients with high-risk, nonmuscle-invasive bladder cancer (NMIBC) frequently relapse after standard intravesical bacillus Calmette-Guérin (BCG) therapy and may have a dismal outcome. The mechanisms of resistance to such immunotherapy remain poorly understood. Here, using cancer cell lines, freshly resected human bladder tumors, and samples from cohorts of patients with bladder cancer before and after BCG therapy, we demonstrate 2 distinct patterns of immune subversion upon BCG relapse. In the first pattern, intracellular BCG infection of cancer cells induced a posttranscriptional downregulation of HLA-I membrane expression via inhibition of autophagy flux. Patients with HLA-I–deficient cancer cells following BCG therapy had a myeloid immunosuppressive tumor microenvironment (TME) with epithelial-mesenchymal transition (EMT) characteristics and dismal outcomes. Conversely, patients with HLA-I–proficient cancer cells after BCG therapy presented with CD8+ T cell tumor infiltrates, upregulation of inflammatory cytokines, and immune checkpoint–inhibitory molecules. The latter patients had a very favorable outcome. We surmise that HLA-I expression in bladder cancers at relapse following BCG does not result from immunoediting but rather from an immune subversion process directly induced by BCG on cancer cells, which predicts a dismal prognosis. HLA-I scoring of cancer cells by IHC staining can be easily implemented by pathologists in routine practice to stratify future treatment strategies for patients with urothelial cancer.


Patient cohort and clinical end points
A cohort of 27 patients with non-metastatic, primary high-grade NMIBC was included in this study. Patients who acquired BCG resistance defined as tumor progression to muscle-invasive tumor (MIBC) post-BCG were included. Clinical follow-up included cystoscopy and urine cytology every 3 months for the first 2 years, then every 6 months.
Tumor samples were obtained by TURBT or radical cystectomy and provided formalinfixed and paraffin-embedded (FFPE). Only initial high-grade tumors with a visible, clearly identifiable, and disease-free muscularis propria were included in this study. Tumor samples obtained pre-BCG at baseline and post-BCG at time of disease progression was required for enrollment. Importantly, the FFPE blocks were selected from the first TURBT to avoid post-surgical inflammatory lesions in bladder tissue wound repair (Supplemental Figure 6). Tumor samples were initially reviewed by two pathologists (C.R. and J.A.).
Clinicopathological and demographic data were collected from patient records at Hôpital Foch and are shown in Supplemental Table 5. Overall survival (OS) was defined as the time between the TURBT and the date of death (any cause). Cancer-specific survival (CSS) was defined as the time between the first TURBT and the date of death related to bladder cancer. For subjects without documentation of death, OS and CSS were censored on the last date the patient was known to be alive. Distant metastasis free survival (DMFS) was defined as the time between the first TURBT and the date of documented distant metastasis. For patients without documentation of distant metastasis, DMFS was censored on the last date the patient was known to be without metastasis.

Fresh tumor samples
All the tumors were collected within 1hr from surgery and stored in complete medium at 4°C from 12hrs to 20hrs prior to dissociation. Clinicopathological and demographic data were collected from patient records at Hôpital Foch and are shown in Supplemental Table   1. Primary bladder tumors were freshly mechanically and enzymatically dissociated using

BCG reconstitution
ImmuCyst BCG is made from a live-attenuated strain of Mycobacterium bovis. The bacilli are lyophilized (freeze-dried) and are viable upon reconstitution. It contains 81 mg (dry weight) of BCG and 150 mg monosodium glutamate. Each vial of ImmuCyst is reconstituted with 3 mL of sterile, preservative-free saline solution. The reconstituted dose contains approximately 10.5 ± 8.7 x 10 8 colony-forming units (CFU). For clinical use, the reconstituted material from the vial is further diluted in an additional 50 mL of sterile, preservative-free saline solution to a final volume of 53 mL for instillation into the bladder.
Therefore, the final concentration in the bladder is approximately 2 x 10 7 CFU/mL.

Ex vivo bladder tumor stimulation assay
The experimental protocol was adapted from Jacquelot et al (37). We performed an ex vivo tumor stimulation assay providing from 12 fresh human bladder tumors. Freshly dissociated cells were seeded in 96-wells plate and incubated in complete medium (RPMI 1640 (GIBCO Life Technologies, ref: 31870-025)

Cell lines
Murine bladder cancer cell line MB49 was originally generated by 7,12dimethylbenz[a]anthracene (DMBA) in vitro exposure to bladder epithelium from male C57BL/lcrf mice (38). The UPPL bladder cancer cell line was obtained from Prof William Y. Kim of the University of North Carolina. UPPL1541 was originally generated from spontaneously arising bladder tumors in Upk3a-Cre ERT2 ; Trp53 L/L ; Pten L/L ; Rosa26 LSL-Luc mice (39). All the cell lines have been tested and found negative for Mycoplasma contamination.

BCG co-culture
Urothelial cancer cells were plated a day prior to infection in antibiotic-free media to reach 80-90% confluence on the day of infection. Lyophilized ImmuCyst Bacillus Calmette-Guérin (BCG) (Connaught sub-strain, 81 mg at 10.5 ± 8.7 10 8 CFU/mL) was reconstituted within PBS 1X as recommended for clinical use. BCG was co-incubated in antibiotic-free media to achieve a multiplicity of infection (MOI) of 10:1 or 5:1 as reported. Plates were incubated at 37°C for the indicated time and then washed with 1XPBS, detached using trypsin, resuspended in complete medium. After passage on 70μm filter and centrifugation, cells were resuspended in 1XPBS for analysis by flow-cytometry.

In vitro BCG re-stimulation assay
This protocol originally described to study trained immunity in human monocytes has been adapted from Bekkering et al (40).

Chromogenic Immunohistochemistry (IHC)
The main steps for chromogenic IHC are described below. Sections were deparaffinized in xylene. Antigen retrieval was performed using ultra cell conditioning 1 (CC1) buffer for 36 minutes at 95°C. Sections were incubated with a primary antibody (see table below) during 1hr at room temperature. Amplification was achieved using an UltraView universal DAB detection kit. Revelation using 3,3'diaminobenzidine as chromogen was applied to sections. Nuclear counterstaining was performed with Hematoxylin II and bluing reagent.
Coverslip was applied with a permanent mounting medium.

Image analysis
Image analysis was done on manually selected (exclusion of areas of necrosis, preparation artifacts) regions of interest (ROI). As these regions were large, they were divided into blocks of pixels processed individually and stitched at the end. The method combined watershed segmentation on DAB staining and color and morphological characteristics to retrieve automatically CD3 + or CD8 + cells. The program exports the number of CD3 + or CD8 + cells and the tissue areas in µm² for each analyzed ROI. DABstained nucleus is automatically detected using their IHC spectral properties in manually selected regions of interest. The routine scores (low, medium, and high) each nucleus on its intensity. In our cases, low classification corresponded to false positive staining and was discarded.

NanoString gene expression profiling
Formalin-fixed paraffin-embedded tumor specimens with sufficient bladder tumor area were selected for RNA extraction (n=12). Only paired bladder samples pre-and post-BCG therapy were used for this study. Macrodissection of selected tumor areas followed by RNA extraction using High Pure FFPET RNA Isolation Kit -Roche Life Science CodeSet. For cancer cell analysis, cell sorting was performed 24hrs after BCG co-culture.
Next, RNA was extracted using the Direct-zol RNA MiniPrep Kit (Zymo Research). The samples were stored at -80°C. Isolated RNA was hybridized with the NanoString nCounter IO360 Panel Human Panel CodeSet and quantified using the nCounter Digital Analyzer. Data were processed with nSolver Analysis Software (NanoString) using the Advanced Analysis module.

Transcriptomics (3'Tag-Seq)
In contrast to traditional RNA-seq, which generates sequencing libraries for the whole transcript, 3′ Tag-Seq only generates a single initial library molecule per transcript, complementary to 3′ end sequences. In consequence, for human samples the restriction to a small part of the transcripts reduces the number of sequencing reads required, has exceptionally low background noise, as well as insensitivity to RNA sample quality variations, in particular FFPE tumor samples.

Bioinformatics analysis
To improve the statistical power of the analysis, only genes expressed in at least one sample (CPM >= 0.1) were considered. A qval threshold of <= 0.05 and a minimum fold change of 1.2 were used to define differentially expressed genes. Pathway enrichment analysis -GSEA Gene list from the differential analysis was ordered by decreasing log2 fold change. Gene set enrichment analysis was performed by clusterProfiler:GSEA function using the fgsea algorithm.

ImageStream analysis
For in vitro infection, BCG was labelled with calcein after co-incubation in complete medium for 30 min at 37°C 5%CO2 protected from light. Urothelial cancer cells were coincubated with calcein-labeled BCG (MOI 10:1) for 24hrs or incubated with IFNg (1.10 3 U/mL) or RMPI 10% heat-inactivated FBS. Cells were harvested using trypsin, washed twice with 1xPBS, and stained with surface antibodies. Hoeschst was added 15 min before image acquisition.

Quantification of autophagy in U2OS cells (image analysis)
Images were processed and segmented with R using the EBImage package (available on the Bioconductor repository https://www.bioconductor.org), the MorphR and the MetaxpR package (both available at https://github.com/kroemerlab). First the nuclear region was defined by using a polygon mask based on the fluorescent signal of Hoechst 33342, which allowed for the further segmentation of cells and for evaluating morphological parameters such as the size of the nucleus and the Hoechst signal intensity. Extending from this nuclear region of interest (ROI), the cytoplasmic region was identified based on the diffuse GFP signal present in the biosensor cells expressing GFP-LC3. Cytoplasmic regions of higher GFP fluorescens were detected to quantify GFP-LC3 puncta formation. For the assessment of autophagic flux the LC3 puncta were additionally detected based on the RFP signal.

Quantification of autophagy in U2OS cells (data analysis)
Data extracted from image analysis were further analyzed with the R software. First pyknotic nuclei, dead cells and debris were excluded from the data set, based on intensity and size parameters. GFP-LC3 aggregation was evaluated by the number/surface area of GFP dots per cell; data were normalized using negative control (untreated condition) to obtain a fold change. Autophagic fluxes were assessed using the surface of autophagosomes compared to the total surface of aggregates (autophagosomes and autophagolysosomes). A linear regression between the surface of dots present in both channels (autophagosomes only) and the total surface of dots (autophagosomes and autophagolysosomes) was calculated on control data (negative and positive control data), and a flux inhibition score was computed by calculating the distance of each data point to the regression line. Results were then evaluated based on inhibition score and the total surface of aggregates.            Supplemental Table 5. Clinical characteristics of patients from our cohort of bladder tumors with BCG acquired resistance (n=27)