Most bladder cancers are poorly responsive to immune checkpoint blockade (ICB). With the need to define mechanisms of de novo resistance, including contributions from the tumor microenvironment (TME), we used single-cell transcriptional profiling to map tumor-infiltrating lymphocytic and myeloid cells in 10 human bladder tumors obtained from patients with a history of smoking either with or without previous ICB. Human datasets were qualitatively compared with single cell datasets from the BBN carcinogen-induced mouse model of bladder cancer, which was poorly responsive to PD-L1 blockade. We applied an established signature of acquired ICB resistance to these human and murine datasets to reveal conservation in EMT and TGFβ ICB resistance signatures between human–mouse stromal and myeloid cells. Using TCGA transcriptional datasets and deconvolution analysis, we showed that patients with a history of smoking and bladder tumors high in M2 macrophage tumor content had a significantly worse survival outcome compared with nonsmokers who were M2 high. Similarly, BBN-induced tumors were high in M2 macrophage content and contained exhausted T–NK cells, thereby modeling the identified TCGA patient subpopulation. The combined targeting of TGFβ + PD-L1 reverted immune cell exclusion and resulted in increased survival and delayed BBN-induced tumor progression. Together, these data support a coordinated role for stromal and myeloid cell populations in promoting de novo resistance to PD-L1 blockade, particularly in patients with a history of smoking.

Significance:

Most patients with bladder cancer do not respond to ICB targeting of the PD-L1 signaling axis. Our modeling applied a de novo resistance signature to show that tumor-infiltrating myeloid cells promote poor treatment response in a TGFβ-dependent mechanism.

Therapies targeting T-cell function have changed the clinical management of non–muscle-invasive, invasive, and metastatic bladder cancer. This is well exemplified with PD-1 and PD-L1 blocking agents that have achieved effective responses for cisplatin-ineligible or cisplatin-resistant metastatic bladder cancer patients (1–4). However, only about 15% to 30% of patients will experience a clinically durable response (5). Tumor mutational burden as well as the molecular status of infiltrating immune cells in the tumor microenvironment (TME) can regulate the response to immune checkpoint blockade (ICB) in human cancers (3). Bladder cancers are frequently infiltrated with immune cells and inflammatory inducing cells such as myeloid-derived suppressor cells (MDSC) and macrophages. Myeloid cells can play an important role in modulating tumor progression and treatment response in multiple types of cancer (6, 7). Moreover, both M2 macrophages and MDSCs are postulated to be a leading cause of therapeutic resistance to ICB (8, 9). Functional studies have also shown that targeting of such myeloid cells can enhance the effectiveness of checkpoint inhibitors in cisplatin-resistant cell lines (10).

For these reasons, defining the transcriptional landscape of immune cell populations including tumor-infiltrating myeloid cells (TIM) and tumor-infiltrating lymphocytes (TIL) is needed to understand how immune cells impact the response to ICB. In our recent studies, we devised a predictive resistance signature for PD-1 ICB based on EMT-stroma core genes traced to cancer associated fibroblasts (11, 12). Despite these data, there remains a poor functional understanding of the immune cell landscape and contributions toward de novo resistance to ICB. We sought to determine whether tumor-infiltrating myeloid cells present in a bladder cancer model could provide mechanisms to explain the frequent occurrence of de novo resistance to PD-L1 ICB.

In this study, we assessed treatment-naïve human and murine model systems (13) to identify coordinate EMT-stromal and TGFβ resistance signatures present both in stromal and tumor-infiltrating myeloid cells. This “dual” resistance mechanism supports poor response to PD-L1 blockade and T-cell exclusion in tumors. We further established that BBN induced primary tumors, and derived transplants are heavily infiltrated with M0-M2 myeloid cells and models a patient population having poor survival outcome and limited response to single-agent PD-L1 inhibition. Finally, we show that cotargeting of PD-L1 and the TGFβ signaling axis results in markedly enhanced tumor inhibition and survival in mice.

Human sample acquisition

In this study, we analyzed 10 human primary bladder cancer samples (Table 1). Samples were obtained from patients who underwent either transurethral resection bladder tumor (TURBT) or radical cystectomy. All patients had localized disease, which was taken during the time of diagnostic resection.

Table 1.

Clinical and demographic information for human bladder tumors.

LabelSpecimen typeClinical stageGenderPT ageSmoking history
BC_#155_Tumor_CD45+ TURBT cT2N0M0 59 Yes (30 pack per year) 
BC_#157_tumor-CD45pos TURBT cT0N0M0 73 Yes (28 pack per year, quit 1994) 
BC_#158_tumor_CD45pos TURBT cT3N0M0 66 Yes (20 pack per year, quit 2017) 
BC_#166_tumor-tumor_CD45+ TURBT cTaN0M0 83 Yes (30 pack per year, quit 1993) 
#167 CD45+ Tumor TURBT ycT4bN3M0 67 Yes (1 pack per year, quit 1994) 
#168 CD45+ Tumor Cystectomy pT3bN0M0 82 Yes (60 pack per year, quit unknown) 
BC_#170_CD45pos TURBT cT1N0M0 86 Yes (20 pack per year, quit 1970) 
BC_#171_CD45pos TURBT ypT4N2M0 68 No (never smoker) 
357_Bladder_1A_05_04_2018/ TURBT cT2N0M0 87 Yes (unknown packs, quit 1982) 
359_Bladder_2A_05_04_2018/ TURBT cT2N0M0 55 Yes (80 pack per year, quit 2016) 
LabelSpecimen typeClinical stageGenderPT ageSmoking history
BC_#155_Tumor_CD45+ TURBT cT2N0M0 59 Yes (30 pack per year) 
BC_#157_tumor-CD45pos TURBT cT0N0M0 73 Yes (28 pack per year, quit 1994) 
BC_#158_tumor_CD45pos TURBT cT3N0M0 66 Yes (20 pack per year, quit 2017) 
BC_#166_tumor-tumor_CD45+ TURBT cTaN0M0 83 Yes (30 pack per year, quit 1993) 
#167 CD45+ Tumor TURBT ycT4bN3M0 67 Yes (1 pack per year, quit 1994) 
#168 CD45+ Tumor Cystectomy pT3bN0M0 82 Yes (60 pack per year, quit unknown) 
BC_#170_CD45pos TURBT cT1N0M0 86 Yes (20 pack per year, quit 1970) 
BC_#171_CD45pos TURBT ypT4N2M0 68 No (never smoker) 
357_Bladder_1A_05_04_2018/ TURBT cT2N0M0 87 Yes (unknown packs, quit 1982) 
359_Bladder_2A_05_04_2018/ TURBT cT2N0M0 55 Yes (80 pack per year, quit 2016) 

Patient blood was drawn independently from the surgical procedure. This study was conducted under approval of the Mount Sinai School of Medicine IRB (10–1180, William OH) and written consent from all subjects. Resected tumor sample were examined by a genitourinary pathologist who confirmed the pathologic staging and tumor content. Patient information is included in Table 1.

Ethics approval and consent to participate

Tumor specimens utilized to generate single-cell RNA sequencing data were derived from patients participating in project #10–1180 approved by the Icahn School of Medicine Institutional Review Board. Written informed consent was obtained from participants. The research conformed to the Declaration of Helsinki.

Animal experiments

All studies conducted under the approved IACUC protocol LA13–00060.

Primary mouse tumor model

FVB/NJ mice (JAX, 001800) were treated with the 0.05% OHBBN carcinogenic in drinking water for 14 weeks followed by 4 weeks of progression. Tumors at this stage consisted of invasive disease and heterogeneous patterns of pathology and lineage composition. Age and sex matched tumors were used for transcriptomic analysis in this study. IACUC guidelines for proper animal husbandry and experimental use were followed (protocol LA13–00060).

Subcutaneous mouse tumor model

To generate tumor lines, primary tumors were dissociated to the single-cell level and transplanted to sex matched FVB donor mice by SQ injection below the skin in 100% Cultrex or Matrigel.

Tissue preparation

Mouse tumors were resected and dissociated to the single-cell level in a solution of RPMI-10% FBS (Gibco) + 1% type 1 collagenase (Gibco). The dissociate was mechanically dissociated through a 16 G needle (5–10 passes) and then a 21.5 G needle (5–10 passes). Human tumors were dissociated using Human Tumor Dissociation Kit Reagents (Miltenyi Biotec) and a GentleMACS instrument (Miltenyi Biotec) with program hTDK_2_37°. Resultant single-cell suspensions were filtered through a series of 100, 70, and 50 μm nylon strainers and resuspended in cold DPBS for FACS staining. Cells were stained with an antibody cocktail including a viability dye (FVD-Blue, Invitrogen) and CD45-BV510 (Clone 2D1, BioLegend). Cells to be sorted were subsequently resuspended in FACS buffer (DPBS containing 2 mmol/L EDTA and 1% BSA) and live, CD45+ singlets sorted into 100% FBS on a FACS Aria (BD Biosciences). For human bladder tumors 171 and 172, tissue was dissociated as described but CD45+ and CD45 fractions were obtained using CD45 TIL microbeads (Miltenyi Biotec) prior to sequencing.

Sequencing

Sequencing was done on a Nextseq 500 platform and analyzed using Cell Ranger V2. Single-cell data were filtered to exclude putative doublet populations, dead or mitochondrial high cells (Supplementary Fig. S1A). The numbers of total cells sequenced ranged from 807 to 1,927 for 10 human samples and 1,354 to 8,839 for 3 mouse tumors. The quality of cells was assessed by the number of genes detected, or percentage of sequence reads mapped to mitochondrial genes (Supplementary Fig. S1). Specific cell numbers before and after QC for CD45-negative and CD45-positive cell sorts were determined (Table 2).

Table 2.

Cell numbers, mean reads, and genes detected for individual human and mouse bladder tumors.

SampleObjectNumber of cells before QCNumber of cells after QCMean reads per cellMedian genes per cellNumber of readsReads mapped to genomeTotal genes detected
Human samples 155P 4,361 3,593 62,259 704 271,515,268 93.00% 20,377 
 157P 3,918 3,566 76,424 612 299,432,183 85.70% 20,483 
 158P 3,251 3,141 92,361 441 300,268,312 75.60% 18,603 
 166P 5,510 4,354 96,315 1,311 530,698,400 91.50% 21,842 
 167P 6,334 5,807 50,118 1,135 317,450,411 93.90% 20,838 
 168P 1,927 1,738 164,029 1,161 316,085,195 92.50% 18,867 
 170P 8,070 5,032 36,760 851 296,659,112 93.30% 22,505 
 171P 5,046 4,645 49,865 815 251,621,339 93.10% 19,440 
 1B 3,707 3,586 59,353 953 220,025,068 93.40% 18,366 
 2B 4,110 3,883 68,551 1,076 281,747,526 94.70% 19,899 
 170N 9,394 4,321 26,423 1,246 248,221,580 94.40% 22,497 
 171N 3,129 2,234 93,055 804 291,172,061 95.00% 20,239 
 1A 4,181 1,484 55,471 2,262 231,926,343 94.90% 21,774 
 2A 808 288 355,782 2,204 287,472,159 93.30% 19,719 
Mouse samples 4950P 1,354 1,338 67,453 1,160 91,331,658 83.60% 16,559 
 8524P 3,433 2,624 62,764 3,021 215,470,312 86.10% 18,965 
 8525P 8,839 4,938 74,283 2,498 656,593,068 85.90% 19,898 
 3928T 5,626 5,618 20,445 2,168 115,026,980 74.80% 16,682 
 4950N 3,134 3,104 29,741 1,753 93,209,670 82.50% 18,558 
 8524N 6,962 5,825 27,448 3,052 191,095,329 90.60% 20,043 
 8525N 6,877 5,143 29,067 2,857 199,899,368 88.10% 20,200 
SampleObjectNumber of cells before QCNumber of cells after QCMean reads per cellMedian genes per cellNumber of readsReads mapped to genomeTotal genes detected
Human samples 155P 4,361 3,593 62,259 704 271,515,268 93.00% 20,377 
 157P 3,918 3,566 76,424 612 299,432,183 85.70% 20,483 
 158P 3,251 3,141 92,361 441 300,268,312 75.60% 18,603 
 166P 5,510 4,354 96,315 1,311 530,698,400 91.50% 21,842 
 167P 6,334 5,807 50,118 1,135 317,450,411 93.90% 20,838 
 168P 1,927 1,738 164,029 1,161 316,085,195 92.50% 18,867 
 170P 8,070 5,032 36,760 851 296,659,112 93.30% 22,505 
 171P 5,046 4,645 49,865 815 251,621,339 93.10% 19,440 
 1B 3,707 3,586 59,353 953 220,025,068 93.40% 18,366 
 2B 4,110 3,883 68,551 1,076 281,747,526 94.70% 19,899 
 170N 9,394 4,321 26,423 1,246 248,221,580 94.40% 22,497 
 171N 3,129 2,234 93,055 804 291,172,061 95.00% 20,239 
 1A 4,181 1,484 55,471 2,262 231,926,343 94.90% 21,774 
 2A 808 288 355,782 2,204 287,472,159 93.30% 19,719 
Mouse samples 4950P 1,354 1,338 67,453 1,160 91,331,658 83.60% 16,559 
 8524P 3,433 2,624 62,764 3,021 215,470,312 86.10% 18,965 
 8525P 8,839 4,938 74,283 2,498 656,593,068 85.90% 19,898 
 3928T 5,626 5,618 20,445 2,168 115,026,980 74.80% 16,682 
 4950N 3,134 3,104 29,741 1,753 93,209,670 82.50% 18,558 
 8524N 6,962 5,825 27,448 3,052 191,095,329 90.60% 20,043 
 8525N 6,877 5,143 29,067 2,857 199,899,368 88.10% 20,200 

Data preprocessing

R package Seurat (14, 15) was used to process both mouse and human scRNA-seq data. Cells with more than 2,500 (5,000) genes detected were removed from human (mouse) scRNA-seq samples to reduce doublet contamination. Different cutoffs were used for human and mouse data to keep the percentage of doublets (6.5%) similar between human and mouse. In addition, cells with percentage of mitochondrial genes higher than 5% (10%) were removed from human (mouse) data. Again, the different cutoffs were chosen to keep the percentage of possible dying cells removed (10%) to be similar between human and mouse. Gene counts were normalized and log-transformed using Seurat. Batch effect among samples is removed by Seurat MNN algorithm (14, 15).

Broad immune cell clustering and annotation

The top 2000 most variable genes were scaled with the effects of total UMI count and percentage of mitochondrial genes regressed out using Seurat, from which principal components (PC) were calculated. The top 30 PCs were selected for SNN clustering (16). The number of PCs were decided by Seurat function ScoreJackStraw (14, 15). To choose the optimal number of clusters, we altered the resolution parameter and manually checked the resulting clusters and their top marker genes. The optimal cluster number was chosen where the top marker genes showed most consistency with the canonical markers (see below). Each cell cluster was manually annotated by comparing the top marker genes for that cluster and canonical markers for different immune cells: CD3/CD4/CD8 (T-cell markers), NKG7 (NK-cell marker), LYZ (monocyte/macrophage marker), CD1C (DC marker), MS4A1 (B-cell marker), MZB1 (plasma cell marker), MS4A2 (mast cell marker), IL3RA (pDC marker), CD24 (neutrophil marker). Nonimmune cell clusters including erythrocytes (marked by HBB), platelets (marked by GP9), epithelial (marked by KRT19), endothelial (marked by PLVAP), and fibroblasts (marked by MMP2) were removed from further analysis. The cell clusters were visualized as UMAP plots (15).

Refined subset clustering within monocyte/macrophage and T–NK clusters

The cells within the clusters of monocytes/macrophages from the previous immune cell type clustering procedure were extracted for further analysis. The top 2000 most variable genes within these monocytes/macrophages were used for calculating PCs and clustering following the same procedure as described above. The cell clusters were annotated as monocytes by highly expressed S100A4/S100A6/S100A8/S100A9 or as macrophages by highly expressed C1QA/C1QB/C1QC. Different monocytes or macrophage subsets were annotated by top overexpressed genes for each cluster as calculated by Seurat function of FindAllMarkers (14, 15). Similarly, cells within T–NK cell clusters were further clustered into subsets using the above procedure. T-cell subsets were annotated according to the classical cell markers, including CCR7 (naïve), LAG3 (exhausted) and FOXP3 (regulatory), or top overexpressed genes in each subset.

Immune cell type proportion

For each human and mouse sample, the proportion of each broad immune cell types were calculated with nonimmune cells being excluded. To test the difference of the cellular proportion between human and mouse, Wilcoxon rank-sum test and Student t test were used.

Comparison of expression profiles of cell subsets

To compare the gene expression profile of T–NK subsets between human and mouse, we followed procedures like those used previously (17). Briefly, a list of variable genes across T–NK subsets (defined below) were obtained for human and mouse datasets separately. Assuming gij annotated the average expression (before log-transformation) of gene i in cell set j, we sorted gij for each gene among cell sets of interests (i.e., T–NK cells):gi1<gi2<…<giN, and those genes with max (giN/giN-1, …, gi3/gi2, gi2/gi1) > 1.2 were defined as variable genes across these cell sets. The gij was then scaled gene-wisely across T–NK cells for human and mouse dataset separately. Common variable genes between human and mouse datasets were used to cluster and compare the expression profiles of T–NK cell sets. In heatmaps, hierarchical clustering was used with Pearson correlation coefficient as similarity measurement and complete linkage as the clustering method. The comparison of macrophage/monocyte expression profiles between human and mouse were conducted in a similar manner.

Pathway activation analysis

Pathway activation is described by module score, calculated by R package Seurat function AddModuleScore. One module score is calculated for each of the pathways and each of the samples/clusters. To compare the activation of different samples, rank-sum test is performed on module scores.

TCGA data acquisition

The bladder Urothelial Carcinoma transcriptome profiling FPKM dataset (TCGA-BLCA) were downloaded from The Cancer Genome Atlas (https://portal.gdc.cancer.gov/projects/TCGA-BLCA, updated in July 2019). The TCGA-BLCA dataset included 408 patients, 430 samples, 3 replicates, for a total of 433 transcriptome profiling data, with 19 samples from normal solid tissues and 414 samples from primary solid tumors. After CIBERSORT deconvolution (18), samples with P values >0.1 were further excluded, leaving 165 samples. Clinical data were obtained from TCGA, using “days to last follow-up or death combined” as the number of survival days. After excluding samples with “not available,” “discrepancy,” or negative survival days, 159 samples were included to generate final survival curves.

M1/M2 characteristics of clusters of macrophages

M1 or M2 characteristics genes are set to be M1 module or M2 module, respectively. Module scores were calculated by R package Seurat function AddModuleScore for each of the clusters. M1 and M2 module scores were compared by a rank-sum test on each of the clusters to determine whether this cluster was M1 or M2 macrophages.

Immunohistochemical analysis of immune cell localization

To assess immune cell numbers and localization within bladder tumors, consecutive tissue sections were stained using conventional immunohistochemistry methods and DAB exposure. Stained slides were then digitally scanned and imported to Adobe Photoshop for conversion to pseudo coloring and image overlays. Channel merged images were then imported to QuPath software where cells were counted with respect to 4 defined regions of equal spacing with respect to the lesion center. Cell numbers were determined for each region by subtracting values from the preceding regions. For example, Region 3 cell numbers = Region 3 ROI counts – Region 2 ROI counts.

In vivo treatment studies

BBN-induced cancer lines were harvested from primary bladder tumors, dissociated to single cells, and expanded in vitro over 2 passages in DMEM with 10% FBS plus 5 μg/mL human insulin (Gibco), 50 ng/mL hEGF (Peprotech), and A83 Alk inhibitor (Tocris) on culture plates coated with 0.1 mg/mL PureCol (Advanced BioMatrix). Subconfluent cells were collected, counted, and resuspended in 80% Cultrex (R&D Systems) at a concentration of 1 × 105 cells per 100 μL SQ injection. C57BL/6 mice (The Jackson Laboratory) aged 8 to 10 weeks were used for tumor cell implants. When tumor masses were 150 to 200 mm3, mice were treated with IgG control antibodies (clone MOPC-21, 10 mg/kg/dose), anti–PD-L1 (mouse IgG1 clone 10F.9G2, 10 mg/kg/dose), anti-Tgfβ (mouse IgG1 clone 1D11.16.8, 10 mg/kg/dose), and the combination of anti–PD-L1 with anti-Tgfβ treatment. Antibodies were delivered 3 times per week with tumors being measured 2 times per week by caliper. When tumors reached IACUC limits of 1,500 mm3 mice were euthanized.

Inferred CNV from scRNA profiles

R package was used to infer CNV from human/mouse scRNA CD45 cells. To infer CNV profiles of epithelial, fibroblasts, and myofibroblasts endothelial were set to be reference for inferring. Ward clustering criterion was used for clustering.

Data availability statement

The mouse scRNA-seq data have been deposited in the GEO database under the accession code GSE146137: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE146137.

The human scRNA-seq data have been deposited in the GEO database under the accession code GSE211388: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?Acc=GSE211388.

For questions regarding all other data, please contact D.J. Mulholland (david.mulholland@mssm.edu).

Single-cell RNA-seq of ICB-naïve human and mouse bladder tumors

We compared the transcriptional landscape of the TMEs in human and mouse bladder tumors. For this, we obtained fresh bladder tumors from 10 patients including TURBTs (n = 9) and a cystectomy (n = 1; Table 1). All tumors were invasive and were resected from patients between 55 and 87 years of age (median 73) with 9/10 having a history of smoking and 7/10 having no history of ICB. Of the 10 tumors assessed, one cT0 tumor was a restaged TURBT from a non–muscle-invasive bladder cancer. To generate mouse bladder tumors, FVB/NJ mice were exposed to the BBN carcinogen for 14 weeks, followed by 4 weeks of progression and subsequent tumor resection (n = 3 primary tumors, n = 1 S.Q. transplant tumor; refs. 13, 19–21). Cell processing and QC analysis were completed as described in Materials and Methods and as in previously investigations (13).

Identification of a conserved EMT-stromal and TGFβ resistance signatures in ICB-naïve human and mouse bladder tumors

Expression of TGFβ is associated with poor clinical outcome in several tumor types accompanied by immunosuppression, increased tumor progression, and EMT (22–25). To assess this signature in scRNA-seq datasets, we performed clustering analysis to identify distinct cell types in CD45-negative fractions in human tumors with (Pt 171) and without (Pt 1a, 2a, and 170) previous immunomodulatory therapies alongside untreated mouse bladder tumors. In both species, we identified four cell types including epithelial, endothelial, fibroblasts, and myofibroblasts (Fig. 1A). Cell types were annotated by highly expressed marker genes including KRT (epithelial), PLVAP (endothelial), MMP2 (fibroblasts), and RGS5 (myofibroblasts; Fig. 1B). Both human and mouse epithelia showed high expression of PSCA-Psca, KRT7-Krt7, and KRT19-Krt19 (Supplementary Fig. S2A and S2B). Cell counts for human and mouse violin plots are shown in Supplementary Fig. S3A–S3D.

Figure 1.

Transcriptional profiling of the TME in human and mouse bladder tumors. A, UMAP clusters of human and mouse CD45-negative cells. B, Cell fractions in individual human and mouse CD45-negative samples. C, Violin plots showing expression distributions of EMT-stromal resistance genes in human and mouse tumor cell types. D, Expression distributions of checkpoint inhibitor biomarker genes and Tgfβ signaling markers.

Figure 1.

Transcriptional profiling of the TME in human and mouse bladder tumors. A, UMAP clusters of human and mouse CD45-negative cells. B, Cell fractions in individual human and mouse CD45-negative samples. C, Violin plots showing expression distributions of EMT-stromal resistance genes in human and mouse tumor cell types. D, Expression distributions of checkpoint inhibitor biomarker genes and Tgfβ signaling markers.

Close modal

Our previous analysis defined an EMT-stromal core gene signature that associates with progression and ICB acquired resistance as a function of cell intrinsic molecular subtyping (11, 12). Thus, we compared untreated and ICB-treated patient tumors for expression of these gene signatures. Analysis of these cells showed high EMT-stromal core genes in the TME as compared with tumor epithelia (Fig. 1C). We observed EMT-stromal core genes to be highly expressed both in pretreated and treated tumor datasets. In untreated mouse samples, core genes were also highly expressed with patterns like human tumors including high CALU-Calu and EMP3-Emp3 but low LRP1-Lrp1 and FOXC2-Foxc2 (Fig. 1C and D).

To distinguish between normal stromal cells and cancerous cells having an EMT signature, we conducted CNV analysis (copy number variation) in human and mouse datasets. In doing so, we aimed to determine whether tumor cells with an EMT signature (high CNV) are not purely a consequence of normal stromal cell contamination (low-negative CNV). Interestingly, in human datasets, fibroblasts–myofibroblasts did now show CNVs, suggesting low EMT. Differently, mouse tumors did show CNVs in fibroblasts–myofibroblasts albeit to a lesser extent than in tumor epithelia. In addition to expression of these predictive signature genes, we also assessed the expression of PD-L1 and components of the TGFβ signaling axis. In comparison to the ICB-naïve human samples assessed in this study, mouse tumors demonstrated markedly higher levels of Tgfb1, Tgfbr2, and Arid1a in myofibroblasts, fibroblasts, and endothelia (Fig. 1D). To determine whether resistance gene expressions were intrinsic to the bladder microenvironment or tumor cell autonomous in nature, we established a transplantable S.Q. BBN-induced bladder cancer model. Although the stromal cells in BBN-induced primary and S.Q. tumors differed in expression levels of Lama2 and Foxc2, we observed robust expression of Tgfb1, Tgfbr2, and Arid1a in both models, suggesting conservation of resistance signatures (Fig. 1D). These data also validate the use of the S.Q. implant model of bladder cancer for therapeutic analysis. Together, these data show that in the ICB-naïve CD45-negative tumor populations, human and mouse tumors present with a high EMT-stromal and TGFβ ICB resistance signatures.

Infiltrating myeloid cells in ICB-naïve bladder tumors have high expression of resistance signature genes

CD45-positive cells in 10 human and 4 mouse pooled tumors were partitioned into cell clusters and compared on a qualitative and quantitative level (Fig. 2A; Supplementary Fig. S4). Human cell type-specific markers were based on whole transcriptome profiles of FACS-sorted subpopulations, while markers for mice were derived from datasets from the IMMGEN consortium (26). Based on highly expressed genes, immune cell subsets were annotated as: T cells (CD3E, Cd3e), NK cells (NKG7, Nkg7), macrophages + monocytes (LYZ, Lyz2), classic DCs (CCR7, Ccr7), pDCs (TCF4, Tcf4), mast cells (MS4A2, Ms4a2), B cells (MS4A1, Ms4a1), plasma cells (MZB1, Mzb1), and neutrophils (CD24A, Cd24; Fig. 2B). Thus, we identified all major immune cell lineages including myeloid (mast cells, neutrophils, classic DCs, and plasmacytoid DCs) and lymphoid cells (T cells, NK cells, B cells, and plasma cells). Major immune populations were compared for their relative abundance in human and mouse tumors (human = black, mouse = red dots; Supplementary Fig. S4B–S4C). In all human bladder tumors examined, T–NK cells were the predominant immune cell types while in mouse, monocyte, and macrophage populations predominated. In the human tumors examined, we measured considerable patient-to-patient variation in the composition of immune cell types. For example, we detected the presence of T–NK “high” tumors (patients 1b, 168, 2b, 155, 166, 167, >75% of CD45+ cells in composition) and T–NK “low” tumors (patients 158, 170, 171, 157, <55% CD45+). Of 10 PTs, we identified 3 tumors (170, 171, 157) with comparable macrophage–monocyte fractions to mouse tumors (Fig. 4A and B). Thus, BBN-induced primary mouse tumors most closely represent patient tumors with high myeloid cell infiltration.

Figure 2.

Mouse BBN bladder tumors are enriched for M0-M2 myeloid cells and model a subpopulation of patients with bladder cancer having poor clinical outcome. A, Clusters of CD45-positive cells in human and mouse tumors. B, Immune cell fractions in individual human and mouse tumors. C, Integrated UMAP analysis of monocyte–macrophage populations in human and mouse tumors. D, Association of M2 macrophage infiltration with overall survival in TCGA patients stratified by smoking history. E, Heatmaps showing expression of M1 and M2 genes for human and mouse monocytes-macrophages. See Supplementary Fig. S10.

Figure 2.

Mouse BBN bladder tumors are enriched for M0-M2 myeloid cells and model a subpopulation of patients with bladder cancer having poor clinical outcome. A, Clusters of CD45-positive cells in human and mouse tumors. B, Immune cell fractions in individual human and mouse tumors. C, Integrated UMAP analysis of monocyte–macrophage populations in human and mouse tumors. D, Association of M2 macrophage infiltration with overall survival in TCGA patients stratified by smoking history. E, Heatmaps showing expression of M1 and M2 genes for human and mouse monocytes-macrophages. See Supplementary Fig. S10.

Close modal

Comparative annotation of human and mouse myeloid transcriptionally defined clusters

When we assessed immune cell proportions in primary human bladder scRNA-seq datasets we detected no significant differences in landscapes between those having previous ICB (PT 158, 167, 171) and those with no previous systemic treatment including ICB (1b, 168, 2b, 155, 166, 170, 157; Fig. 2A and B). Thus, to obtain a qualitative signature for myeloid cells, we focused on purified CD45+ populations isolated from pretreated tumors to generate macrophage–myeloid cell clustering for human (Fig. 2C, top) and mouse (Fig. 2C, bottom). In doing so, we identified 13 cell clusters in human bladder tumors (Supplementary Fig. S5A–S5F) that were annotated using established gene signatures for monocyte and macrophage populations (27), and merged 8 functional clusters including 3 monocytes clusters with high expression of monocyte signature genes (VCAN, FCGR3A, CXCL8), 4 macrophage clusters (CD163, C1QC, CXCL11, SPINK1) and one myeloid-T cell CD3-positive cluster (CD3; Fig. 2C, left). For mouse tumors, we merged 15 clusters into 10 functional clusters including 3 as monocytes (S100a9, TremI4, Vcan), 5 as macrophages (Cd74, Cdh1, Ctsd, Gpnmb, Lyve1), one myeloid DC (Ccr7, Cd74, MhcII), and one myeloid-T cell Cd3 positive (Fig. 2C, right).

M2 macrophages associate with reduced survival probability in ICB-naïve bladder cancer patients

Using the TCGA bladder cancer dataset consisting of 430 samples (414 tumor, 19 normal), from 408 patients, we applied CIBERSORT (https://cibersort.stanford.edu) to associate the composition of major immune cell types with overall patient survival. Plots of root square mean error (RSME) versus P value and correlation versus P value are shown including thresholds for P < 0.05 (red dashed line) and P < 0.1 (blue dashed line; Supplementary Fig. S1B). We determined that in primary tumor samples whose cell compositions could be accurately estimated, 136 samples had a threshold of P < 0.05 and 165 samples at P < 0.1 (Supplementary Fig. S1B). To understand the clinical implications of mouse bladder tumors that are high in macrophage content, we assessed patient survival probability in the TCGA dataset stratified by levels of M0, M1, and M2 macrophage content. Using total patient samples, we determine that inferior patient survival associated significantly with higher infiltration of M0 (P = 0.05) and M2 (P = 0.0002) but not M1 macrophages (P = 0.58; Fig. 2D; Supplementary Fig. S6).

Because smoking is the primary risk factor for bladder cancer development, we assessed the impact of high versus low macrophage subtypes stratified by patients who were classified as either nonsmokers or smokers (Supplementary Fig. S6B–S6D). The breakdown of patients stratified by smoking status is shown in violin plot format (Supplementary Fig. S6A). In this analysis, we determined that the prognostic importance of any macrophage subpopulation status in nonsmokers was not significant at the P = 0.05 level. However, in patients who are current smokers or reformed smokers, patients with high M2 macrophages showed significantly reduced survival probability as compared with those with low M2 macrophages (Supplementary Fig. S6C and S6D). Thus, these TCGA data show that in ICB-naïve patients, tumors high in M0 and M2 macrophages have reduced survival as compared with M0-M2 low tumors particularly in patients with a history of smoking.

Protumor M2 macrophages predominate ICB-naïve murine BBN tumors

To further understand the function of the identified myeloid clusters, we assessed human and mouse cell clusters based on expression of genes that could define monocyte–macrophage populations as either M1 (e.g., CXCL9, IDO1) or M2 (e.g., CD163, CCL18; refs. 28–30). We compared M1 and M2 module scores for each cluster based on expression module genes using the rank-sum test (Fig. 2E). Human FCGR3A monocytes, CXCL11 macrophages, and myeloid T cells were more M1-like (with higher M1 module scores than M2 module scores) while VCAN, CD163, C1QC, and CXCL8 defined cell clusters were more M2-like. In mouse, except for TremI4-defined monocytes, all populations were M2-like (Fig. 2E). These data show that while human samples have a mixed M1-M2 signature, M2-like macrophages predominated in untreated BBN bladder tumors. With these data, we investigated the association of infiltrating myeloid and lymphocytic cells on overall survival.

Human and mouse myeloid cell subpopulations are enriched for TGFβ signaling and EMT-stromal ICB resistance signature genes. Functional similarity was assessed among myeloid cell clusters by calculating correlations between average profiles of individual clusters (Fig. 3A; Supplementary Fig. S7). When comparing clusters by alignments we identified at least 6 sets of commonly over expressed genes including human and mouse myeloid T cells (sets 1 and 2), mouse S100a9 monocytes and human CXCL8 monocytes (set 3), mouse Ctsd macrophages and human SPINK macrophages (set 4), mouse Cd74 macrophages and human CXCL11 macrophages (set 5) as well as mouse Gpnmb macrophages with human C1QC-CD163 macrophages (set 6) (Supplementary Fig. S7). We further defined the expression of Cd274 and Tgfb1 in the identified myeloid clusters (Kruskal–Wallis, P < 2.2e−16; Fig. 3B). These data identify the presence of highly coexpressed genes between pretreated human and mouse monocyte–macrophage populations. To further understand which genes of the TGFβ hallmark signaling pathway associate with individual myeloid populations, we identified outlier populations in both human (n = 10) and mouse (n = 4). Specifically, we identified genes within the Tgfβ hall mark signaling pathway with more restricted expression in certain myeloid subpopulations. For example, in mouse, high Serpine1, Thbs1, and Eng expression occurred in myeloid cells only while Id1 and Id3 gene expression was low in all mouse CD45 positive cells. In human tumors, high ENG and THBS1 expression was also observed in myeloid cells. In the CD45-negative compartment, we also observed conserved and restricted gene expression patterns (Supplementary Fig. S8). Because CD163 is highly expressed in mouse TAMs, we examined infiltrating myeloid and lymphocyte cells for Tgfβ and PD-L1 expression.

Figure 3.

Identification of TGFβ- and EMT-stroma–dependent ICB resistance signatures in human and mouse tumor-infiltrating myeloid cells. A, Heatmaps showing expression similarities between monocytes–macrophages derived from human and mouse bladder tumors. B, Volcano plots showing expression of PD-L1 and Tgfβ1 in functionally defined macrophage–monocytes populations. C, Volcano plots showing checkpoint inhibitor biomarker genes and Tgfβ signaling markers in myeloid cell populations. D, Volcano plots showing EMT-stroma resistance marker expression in myeloid cell populations.

Figure 3.

Identification of TGFβ- and EMT-stroma–dependent ICB resistance signatures in human and mouse tumor-infiltrating myeloid cells. A, Heatmaps showing expression similarities between monocytes–macrophages derived from human and mouse bladder tumors. B, Volcano plots showing expression of PD-L1 and Tgfβ1 in functionally defined macrophage–monocytes populations. C, Volcano plots showing checkpoint inhibitor biomarker genes and Tgfβ signaling markers in myeloid cell populations. D, Volcano plots showing EMT-stroma resistance marker expression in myeloid cell populations.

Close modal

Interestingly, we observed high expression of Tgfβ and PDL-1 (CD274) in all human and mouse monocyte–macrophages clusters identified. CXCL8 mainly expressed in human monocytes and macrophages (there is no CXCL8 ortholog in mouse) while CXCL13–Cxcl13 expression was low in both human and mouse cells (Fig. 3C). To further understand the contribution of tumor infiltrating immune cells with de novo ICB resistance, we assessed EMT-stromal core genes, which we previous established to be predictive of poor response (11, 12). We detected high expression of FLNA, VIM, and EMP3 in all human bladder samples (Fig. 3D). Similar trends were observed in mouse immune cells including high levels of Flna, Fn1, Lrp1, Vim, Ecm1 and Emp3, and Calu (Fig. 3D). These modeling data suggest that a pronounced TGFβ signature in untreated animals and patient samples may associate with de novo resistance to PD-L1 blockade.

Myeloid cell–infiltrated mouse bladder tumors are poorly infiltrated with CD8 T cells

The stromal-EMT ICB resistance signatures that we established include vimentin. Thus, to consider whether high levels of vimentin in the BBN mouse model associates with immune cell exclusion or PD-L1 resistance, we immunostained primary and transplant (S.Q.) BBN tumor samples for CD3+ T cells, Nk1.1+ NK cells, and CD19+ B cells.

While CD3+ T cells could infiltrate the parenchyma of BBN tumors (Fig. 4A and B), we also detected many pockets or aggregates of immune cells between the muscle wall (lamina propria) and tumor parenchyma that frequently coincided with regions showing high levels of vimentin (Fig. 4A). To quantitate levels of infiltrating immune cells, tumors were resected, formalin fixed, and processed for immunostaining using consecutive tissue section stained for T, NK, and B cells. Immuno stains were visualized using DAB exposures that were digitally scanned and converted to a monochrome signal. Unique pseudo colors were then applied to each DAB exposure in Adobe Photoshop imaging software followed by image super imposition (Fig. 4B). Independent images were then assessed using QuPath software counting for (i) absolute immune cell numbers and (ii) localization of each immune cell type with respect to the distance from the lesion center. Using concentric contour lines within the lesion, we defined 4 regions spaced roughly equally but with increasing distance from the lesion perimeter (Fig. 4B). NK and B cells were found predominantly near the lesion perimeter (region 1) as defined by histologic examination, while CD3+ T cells could be detected in significantly higher numbers in regions 1 and 2 but failed to efficiently locate to regions 3 and 4 where they occupied less than 10% and 5% of total T cells, respectively (Fig. 4B; n = 10–15 regions/treatment). These data support that in ICB-naïve tumors T–NK cells are predominantly located at the tumor periphery.

Figure 4.

TGFβ signaling drives exclusion of T–NK cells and de novo resistance to PD-L1 treatment in BBN mouse bladder tumors. A, Distribution of vimentin and CD3+ T cells in primary BBN tumors shown at low (left) and insets (right). B, Overlayed immunostains stains of a primary BBN-induced tumor showing CD3+ T cells, NK1.1+ NK cells, and CD19+ B cells as a pseudo-color overlay. The degree of immune cell infiltration was scored using regions 1 to 4 and is based on roughly equal distances from the lesion center to the lesion periphery. C, Conservation of gene expression for immune cell exhaustion markers between mouse and human. D, Survival of mice implanted with S.Q. BBN tumors and treated with Tgfβ and PD-L1–neutralizing antibodies, alone, or in combination (log rank P values). E, Tumor progression analysis of BBN-induced bladder tumors treated with Tgfβ and PD-L1–neutralizing antibodies, alone, or in combination (**, P < 0.05; ***, P < 0.005). F, Percentage of immune cells found in regions 1 to 4 of B in BBN mouse tumors. G, Effect of antibody treatments on the distribution of NK and CD3+ T cells in BBN mouse tumors. A, Bar, 500 μm, 100 μm (inset). B, Bar, 200 μm.

Figure 4.

TGFβ signaling drives exclusion of T–NK cells and de novo resistance to PD-L1 treatment in BBN mouse bladder tumors. A, Distribution of vimentin and CD3+ T cells in primary BBN tumors shown at low (left) and insets (right). B, Overlayed immunostains stains of a primary BBN-induced tumor showing CD3+ T cells, NK1.1+ NK cells, and CD19+ B cells as a pseudo-color overlay. The degree of immune cell infiltration was scored using regions 1 to 4 and is based on roughly equal distances from the lesion center to the lesion periphery. C, Conservation of gene expression for immune cell exhaustion markers between mouse and human. D, Survival of mice implanted with S.Q. BBN tumors and treated with Tgfβ and PD-L1–neutralizing antibodies, alone, or in combination (log rank P values). E, Tumor progression analysis of BBN-induced bladder tumors treated with Tgfβ and PD-L1–neutralizing antibodies, alone, or in combination (**, P < 0.05; ***, P < 0.005). F, Percentage of immune cells found in regions 1 to 4 of B in BBN mouse tumors. G, Effect of antibody treatments on the distribution of NK and CD3+ T cells in BBN mouse tumors. A, Bar, 500 μm, 100 μm (inset). B, Bar, 200 μm.

Close modal

EMT-stroma and Tgfβ resistance signatures associate with high checkpoint expression in human and mouse T–NK cells

To understand the implications of Tgfβ and EMT-stroma de novo resistance signatures on T-cell function, we assessed biomarkers of T-cell exhaustion, transcription factors related to exhaustion, and epigenetic regulators of exhaustion. Major observations include the conservation in expression of CD8+ and CD4+ T-cell exhaustion markers, signatures across NK and Tregs, whereas naïve CD4+ T cells showed less correlation (Fig. 4C). FoxP3, IL2RA, and IKZF2 were identified as biomarkers for regulatory T cells based on scRNA-seq studies (31). In our datasets, these genes were also highly expressed in human CD4 T regs while in mouse, Il2ra, Foxp3, and Ikzf2 were less specific to regulatory cells but also found in exhausted CD8 T cells (Supplementary Fig. S9A and S9B). Regarding biomarkers for exhausted T cells, LAG3, PDCD1, HAVCR2 (TIM-3), and TIGIT are identified in scRNA-seq studies (31). T-cell exhaustion related transcription factors include NFAT (NFATC1), TOX, TCF1 (TCF7), and TBX21 (T-BET; ref. 32). Furthermore, TOX/TOX2 serve an essential role in the epigenetic modeling of exhausted T cells (32). In our human scRNA-seq data, expression of LAG3 was specifically expressed in exhausted T cells, while the expression of PDCD1, HAVCR2, NR4A, and TCF7 was not specific to exhausted T cells. TOX2 expression was specific to exhausted CD4 T cells, while the expression of TIGIT and BATF was specific to regulatory T cells, instead of exhausted T cells (Supplementary Fig. S9B). In our mouse datasets, Lag3 was specific to exhausted CD8+ T cells while other exhaustion genes were not specific to exhausted T cells. The expression of NFATC1 and TBX21 in human T cells and, Nfatc1 and Tox mouse T cells is low (shown as scaled raw data; Supplementary Fig. S9B). Despite these trends, we also observed heterogeneity in PDCD1, HAVCR2, CTLA4, and CD274 (PD-L1) expression among human tumors (Supplementary Fig. S10A), a finding compatible with the observed wide range in the relative numbers of immune cell types between individual tumors (low average expression = 166, high average expression = tumor 168, 1b, Kruskal–Wallis test P values <1e−15 for PD-L1 and CTLA4). These data indicate that the BBN mouse model, like human bladder cancer, is composed of immune exhausted NK and T cells. In conclusion, LAG3/Lag3 is the primary exhausted T-cell marker common to our human and mouse scRNA-seq data.

CD274 (PDL-1) is a major biomarker for ICB response in BBN-induced bladder tumors

Mouse tumors clustered together based on high CD274 (PD-L1) expression, although in other instances could cluster more closely with human specimens. Examples of such heterogeneity include (i) similarity of tumors H168 and H1b with respect to PDCD1 and LAG-3 expression (rank-sum test, P < 1e−50); (ii) human tumor H168 with high CTLA4 expression (rank-sum test P < 1e−50); (iii) mouse tumor M8524 with high HAVCR2 expression; and (iv) mouse tumors M8524 and M4950, clustering more closely to other human tumors than to the remaining mouse tumor (M8525; Supplementary Fig. S10B). These data further support that CD274 is a major signaling pathway in the BBN mouse model but that clinical samples have heterogeneity in checkpoint marker expression including CD274.

Cotargeting of Tgfβ and PD-L1 in BBN bladder tumors results in significantly reduced tumor progression

To test whether Tgfβ signaling promotes resistance to PD-L1 inhibition, we conducted single and combination neutralizing antibody studies using transplants derived from primary BBN tumors. To circumvent the variability in tumor growth kinetics associated with the autochthonous BBN models, we have developed BBN transplant models amenable for treatment studies (13). Using implants of total tumor cells (cancer stroma), we established S.Q. implants in immune competent mice. Once tumors measured 100 mm3, mice were treated with IgG isotype control, PD-L1, Tgfβ, and combined PD-L1 + Tgfβ antibodies (IP, 2× wk., 10 mg/kg) and continued until tumors reached IACUC limits. Mice treated with single-agent IgG or Tgfβ mAbs showed rapid tumor formation. However, those with combined PD-L1 and Tgfβ antibody treatment demonstrated significant extension of survival (IgG = 21d, Tgfβ = 26d, PD-L1 = 33d, Tgfβ + PD-L1 = 43d) and hazard ratios (HR) of 19.57 (IgG:PD-L1, P = 0.0007), 4.79 (IgG:Tgfβ, P = 0.038), 39.57 (IgG:Tgfβ+PD-L1, P < 0.0001), 4.078 (PD-L1:Tgfβ+PD-L1, P = 0.047), and 10.84 (Tgfβ:Tgfβ+PD-L1, P = 0.0065; Fig. 4D). P values were calculated by the log-rank test. Tumor growth showed markedly delayed kinetics for combination mAb treatment, which, in some instances, showed a complete regression (Fig. 4E). We further measured infiltrating immune cells in regions 1 to 4 and observed that most T–NK cells resided on the tumor periphery (Fig. 4F); however, in the presence of antibody treatment, infiltration was detected including to regions 3 + 4 (where *, P < 0.05; **, P < 0.005; Fig. 4G). These data show that the inhibition of Tgfβ expression and function enhance PD-L1 targeting and associate with significantly delayed tumor progression of mouse BBN tumors mediated by increased T–NK cell infiltration.

In this study, we completed a transcriptomic analysis of the tumor microenvironment of human and mouse muscle-invasive bladder tumors (MIBC). We assessed the heterogeneity in stromal and immune cell populations found in tumors that had received no previous ICB. This led to the identification of a conserved gene signature that may explain the presence of de novo resistance to PD-L1 blockade that occurs in a significant portion of patients with bladder cancer.

These conclusions were supported experimentally by the following components of our study: (i) qualitative and quantitative transcriptional analysis using purified epithelial, stromal, and immune cell populations isolated from ICB-treated and ICB-naïve tumors; (ii) identification of a stromal transcriptional signature based on high expression of Tgfβ; (iii) identification of M2 myeloid cells high in Tgfβ and PD-L1 expression, which (iv) model a patient population with a history of smoking and poor survival outcome; and (v) that targeting of Tgfβ sensitizes myeloid cell–enriched mouse tumors to PD-L1. These experiments allowed for the identification of a convergent mechanism of de novo ICB resistance involving Tgfβ–EMT signaling from stromal and myeloid TME cell subpopulations.

Human bladder cancer is a heterogeneous disease both at the histologic and DNA mutational level (33). Thus, it was not surprising that the content of immune cell populations varied in the patient samples assessed including tumors having a range of infiltration T–NK cells. Using TCGA patient data and our scRNAs-seq human datasets, we identified samples with high levels of myeloid cells, which recapitulated expression of resistance signatures that we have previously shown to be present in patients with acquired resistance to ICB (11, 34, 35). Our deconvolution analysis of TCGA bulk transcriptomic data showed that patients high in M0-M2 cells have markedly reduced survival compared to patients having low M0-M2 cell numbers. Interestingly, our analysis also identified significant differences in survival when comparing never smokers versus patients with a history of smoking. Specifically, patients who are M2 high with a smoking history have a significantly reduced life expectancy. The OHBBN carcinogen is found in tobacco (36–38) and, thus, serves as an excellent model for smoking induced bladder cancer. Moreover, because BBN-induced tumors have similar mutational and histopathologic landscapes with human BLCAs (39), experimental studies concerning neoantigen dependent immunomodulatory therapies can be translated to human disease with greater ease than with preclinical tumor models with a nonmutational genetic landscape.

Mouse BBN tumors showed pronounced Tgfβ and EMT-stromal resistance signatures, which associated with poor response to ICB single-agent treatment. These signature data are compatible with the prevalence of stromal components in primary BBN tumors including vimentin, fibronectin as well as the general exclusion and peripheral aggregation of T cells. We showed that BBN tumors were refractory to single-agent PD-L1 monoclonal antibody treatment, a finding that is consistent with recent studies showing that BBN tumors are poorly responsive to PD-1 inhibition (40). Using combination Tgfβ1 and PD-L1 monoclonal antibody treatment, we observed tumor subcutaneous tumor growth inhibition accompanied by increase infiltration of CD8+ T cells and NK cells. These data are also consistent with observations from other studies demonstrating the association of TGFβ1 high tumor that are immune cell excluded and poorly responsive to atezolizumab (41). Indeed, studies in pancreas have defined a role for the stroma in preventing anticancer drugs from effectively targeting cancer cells (11, 12) including GEM models, where modulation of the stromal tumor component can reverse primary resistance to ICB mediated by increased infiltration of CD8+ T cells (42).

Previous studies have showed that TGFβ1 signaling can regulate the response to atezolizumab and that combined treatment with TGFβ1 and PD-L1 blocking agents is more effective at increasing T-cell infiltration and delaying tumor growth (41). These studies, however, did not consider the contributions of infiltrating myeloid cells, which may be related to the observation that myeloid high tumors are only present in a portion of bladder cancers. In other solid tumor types, Tgfβ signaling has been demonstrated to reduce CD8 T-cell function as exemplified in a murine model of thymoma (43). In a murine model of melanoma, combinatorial treatment using PD-1 and TGFβ kinase inhibition resulted in cooperative increases in CD8+ T-cell function and tumor inhibition (44).

OHBBN-induced tumors contain a high content of monocytes and macrophages, which suggests that the Tgfβ signaling axis may function as a critical mechanism for tumor progression and immune suppression. Our data show that while LAG3, PD-1, and CTLA4 are highly expressed in mouse CD8+ T cells, PD-L1 (CD274) levels are low. Conversely, myeloid cells were high expressors of CD274 (PD-L1), further supporting a potential functional role for TAMs in tumor progression and response to checkpoint inhibitors. Interestingly, previously preclinical studies have demonstrated that APC expression of PD-L1, but not cancer cells, mediates response to checkpoint therapy further underscoring the utility of myeloid cell enriched models for the study of ICB response (45). The efficiency of the Tgfβ and PD-L1 targeting combination is underscored by the fact that, like certain clinical bladder tumors, BBN mouse tumors have Tgfβ being produced by both the TME and infiltrating myeloid cells.

Limitations to our study and areas of future investigation include further understanding the impact of smoking history and relationship with infiltrating myeloid cells. Patient smoking history includes variables of smoking frequency, duration of smoking (years), and the impact of being a reformed smoker versus continuous smoker. Thus, to further understand associations between exposure to tobacco containing carcinogens and changes in myeloid cell function, additional clinical datasets should be assessed. From a preclinical modeling perspective, variable durations of BBN exposure may be considered as well as variable periods of recovery where tumors are not exposed to the carcinogen. Future areas of investigation also include understanding the effects of tumor infiltrating myeloid cells on tumor progression, therapeutic resistance, and T–NK cell targeting (28, 29). Indeed, previous studies have shown that TAMs impede CD8+ T cells from infiltrating tumors cells (46), supporting that therapeutic targeting of infiltrating myeloid cells in the context of ICB should be explored. Further areas of interest include understanding the potential for plasticity and polarization changes in myeloid cell subpopulations as a function of progression and therapy.

Together, our study defines a preclinical model and de novo resistance mechanisms to PD-L1 treatment and a patient population having myeloid cell-high bladder tumors and poor clinical outcome.

M.D. Galsky reports grants and personal fees from BMS, Merck, Genentech, and AstraZeneca, personal fees from Pfizer, EMD Serono, SeaGen, Janssen, NuMab, Dragonfly, Glaxo Smith Kline, Basilea, Urogen, Rappta Therapeutics, and Alligator outside the submitted work, as well as a patent for compositions and methods for treating cancer, overcoming PD-1/PD-L1 blockade resistance, and determining resistance to checkpoint inhibitor treatment pending. O. Elemento reports other support from OneThree Bio, personal fees and other support from Volastra Therapeutics, other support from Owkin, personal fees from Pionyr Immunotherapeutics, other support from Freenome, and personal fees from Champions Oncology during the conduct of the study. B.M. Faltas reports serving in a consulting role for QED therapeutics and Boston Gene; serving on the advisory board at Merck, Immunomedics/Gilead, QED therapeutics, ant Guardant; patent royalties from Immunomedics/Gilead; honoraria from Urotoday; and research support from Eli Lilly. No disclosures were reported by the other authors.

H. Yu: Conceptualization, data curation, software, formal analysis, investigation, writing–original draft. J.P. Sfakianos: Resources, validation, visualization. L. Wang: Formal analysis, supervision. Y. Hu: Data curation, software, formal analysis, methodology. J. Daza: Resources, investigation. M.D. Galsky: Resources, writing–review and editing. H.S. Sandhu: Supervision. O. Elemento: Data curation, software, formal analysis, supervision. B.M. Faltas: Methodology, writing–review and editing. A.M. Farkas: Conceptualization, resources, funding acquisition. N. Bhardwaj: Conceptualization, resources, funding acquisition. J. Zhu: Formal analysis, supervision, methodology, writing–original draft, project administration, writing–review and editing. D.J. Mulholland: Conceptualization, resources, data curation, formal analysis, supervision, investigation, project administration.

Human scRNA sequencing was funded by NIH R01 CA201189 (N. Bhardwaj), T32 AI007605 (A.M. Farkas), T32 AI007647 (A.M. Farkas), and a Translational Team Science Award from the Department of Defense, CA181008 (N. Bhardwaj, A.M. Farkas, J.P. Sfakianos). The Tisch Cancer Institute (D.J. Mulholland). We acknowledge the Human Immune Monitoring Center at Mount Sinai. We acknowledge the Human Immune Monitoring Core at Mount Sinai.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.

Note: Supplementary data for this article are available at Molecular Cancer Therapeutics Online (http://mct.aacrjournals.org/).

1.
Balar
AV
,
Castellano
D
,
O'Donnell
PH
,
Grivas
P
,
Vuky
J
,
Powles
T
, et al
.
First-line pembrolizumab in cisplatin-ineligible patients with locally advanced and unresectable or metastatic urothelial cancer (KEYNOTE-052): a multicentre, single-arm, phase 2 study
.
Lancet Oncol
2017
;
18
:
1483
92
.
2.
Balar
AV
,
Galsky
MD
,
Rosenberg
JE
,
Powles
T
,
Petrylak
DP
,
Bellmunt
J
, et al
.
Atezolizumab as first-line treatment in cisplatin-ineligible patients with locally advanced and metastatic urothelial carcinoma: a single-arm, multicentre, phase 2 trial
.
Lancet
2017
;
389
:
67
76
.
3.
Rosenberg
JE
,
Hoffman-Censits
J
,
Powles
T
,
van der Heijden
MS
,
Balar
AV
,
Necchi
A
, et al
.
Atezolizumab in patients with locally advanced and metastatic urothelial carcinoma who have progressed following treatment with platinum-based chemotherapy: a single-arm, multicentre, phase 2 trial
.
Lancet
2016
;
387
:
1909
20
.
4.
Sharma
P
,
Callahan
MK
,
Bono
P
,
Kim
J
,
Spiliopoulou
P
,
Calvo
E
, et al
.
Nivolumab monotherapy in recurrent metastatic urothelial carcinoma (CheckMate 032): a multicentre, open-label, two-stage, multi-arm, phase 1/2 trial
.
Lancet Oncol
2016
;
17
:
1590
8
.
5.
Alfred Witjes
J
,
Lebret
T
,
Comperat
EM
,
Cowan
NC
,
De Santis
M
,
Bruins
HM
, et al
.
Updated 2016 EAU guidelines on muscle-invasive and metastatic bladder cancer
.
Eur Urol
2017
;
71
:
462
75
.
6.
Shu
Y
,
Cheng
P
.
Targeting tumor-associated macrophages for cancer immunotherapy
.
Biochim Biophys Acta Rev Cancer
2020
;
1874
:
188434
.
7.
Duan
Z
,
Luo
Y
.
Targeting macrophages in cancer immunotherapy
.
Signal Transduct Target Ther
2021
;
6
:
127
.
8.
Meyer
C
,
Cagnon
L
,
Costa-Nunes
CM
,
Baumgaertner
P
,
Montandon
N
,
Leyvraz
L
, et al
.
Frequencies of circulating MDSC correlate with clinical outcome of melanoma patients treated with ipilimumab
.
Cancer Immunol Immunother
2014
;
63
:
247
57
.
9.
Weide
B
,
Martens
A
,
Zelba
H
,
Stutz
C
,
Derhovanessian
E
,
Di Giacomo
AM
, et al
.
Myeloid-derived suppressor cells predict survival of patients with advanced melanoma: comparison with regulatory T cells and NY-ESO-1- or melan-A-specific T cells
.
Clin Cancer Res
2014
;
20
:
1601
9
.
10.
Takeyama
Y
,
Kato
M
,
Tamada
S
,
Azuma
Y
,
Shimizu
Y
,
Iguchi
T
, et al
.
Myeloid-derived suppressor cells are essential partners for immune checkpoint inhibitors in the treatment of cisplatin-resistant bladder cancer
.
Cancer Lett
2020
;
479
:
89
99
.
11.
Wang
L
,
Saci
A
,
Szabo
PM
,
Chasalow
SD
,
Castillo-Martin
M
,
Domingo-Domenech
J
, et al
.
EMT- and stroma-related gene expression and resistance to PD-1 blockade in urothelial cancer
.
Nat Commun
2018
;
9
:
3503
.
12.
Chang
SS
.
Re: EMT- and stroma-related gene expression and resistance to PD-1 blockade in urothelial cancer
.
J Urol
2019
;
202
:
458
.
13.
Sfakianos
JP
,
Daza
J
,
Hu
Y
,
Anastos
H
,
Bryant
G
,
Bareja
R
, et al
.
Epithelial plasticity can generate multi-lineage phenotypes in human and murine bladder cancers
.
Nat Commun
2020
;
11
:
2540
.
14.
Butler
A
,
Hoffman
P
,
Smibert
P
,
Papalexi
E
,
Satija
R
.
Integrating single-cell transcriptomic data across different conditions, technologies, and species
.
Nat Biotechnol
2018
;
36
:
411
20
.
15.
Stuart
T
,
Butler
A
,
Hoffman
P
,
Hafemeister
C
,
Papalexi
E
,
Mauck
WM
3rd
, et al
.
Comprehensive integration of single-cell data
.
Cell
2019
;
177
:
1888
902
.
16.
Xu
C
,
Su
Z
.
Identification of cell types from single-cell transcriptomes using a novel clustering method
.
Bioinformatics
2015
;
31
:
1974
80
.
17.
Zilionis
R
,
Engblom
C
,
Pfirschke
C
,
Savova
V
,
Zemmour
D
,
Saatcioglu
HD
, et al
.
Single-cell transcriptomics of human and mouse lung cancers reveals conserved myeloid populations across individuals and species
.
Immunity
2019
;
50
:
1317
34
.
18.
Newman
AM
,
Steen
CB
,
Liu
CL
,
Gentles
AJ
,
Chaudhuri
AA
,
Scherer
F
, et al
.
Determining cell type abundance and expression from bulk tissues with digital cytometry
.
Nat Biotechnol
2019
;
37
:
773
82
.
19.
Condeelis
J
,
Pollard
JW
.
Macrophages: obligate partners for tumor cell migration, invasion, and metastasis
.
Cell
2006
;
124
:
263
6
.
20.
Freedman
ND
,
Silverman
DT
,
Hollenbeck
AR
,
Schatzkin
A
,
Abnet
CC
.
Association between smoking and risk of bladder cancer among men and women
.
JAMA
2011
;
306
:
737
45
.
21.
Nagao
M
,
Suzuki
E
,
Yasuo
K
,
Yahagi
T
,
Seino
Y
.
Mutagenicity of N-butyl-N-(4-hydroxybutyl)nitrosamine, a bladder carcinogen, and related compounds
.
Cancer Res
1977
;
37
:
399
407
.
22.
Lin
RL
,
Zhao
LJ
.
Mechanistic basis and clinical relevance of the role of transforming growth factor-beta in cancer
.
Cancer Biol Med
2015
;
12
:
385
93
.
23.
Massague
J
.
TGFbeta in cancer
.
Cell
2008
;
134
:
215
30
.
24.
Calon
A
,
Lonardo
E
,
Berenguer-Llergo
A
,
Espinet
E
,
Hernando-Momblona
X
,
Iglesias
M
, et al
.
Stromal gene expression defines poor-prognosis subtypes in colorectal cancer
.
Nat Genet
2015
;
47
:
320
9
.
25.
Flavell
RA
,
Sanjabi
S
,
Wrzesinski
SH
,
Licona-Limon
P
.
The polarization of immune cells in the tumour environment by TGFbeta
.
Nat Rev Immunol
2010
;
10
:
554
67
.
26.
Jojic
V
,
Shay
T
,
Sylvia
K
,
Zuk
O
,
Sun
X
,
Kang
J
, et al
.
Identification of transcriptional regulators in the mouse immune system
.
Nat Immunol
2013
;
14
:
633
43
.
27.
Gentles
AJ
,
Newman
AM
,
Liu
CL
,
Bratman
SV
,
Feng
W
,
Kim
D
, et al
.
The prognostic landscape of genes and infiltrating immune cells across human cancers
.
Nat Med
2015
;
21
:
938
45
.
28.
Long
KB
,
Collier
AI
,
Beatty
GL
.
Macrophages: key orchestrators of a tumor microenvironment defined by therapeutic resistance
.
Mol Immunol
2019
;
110
:
3
12
.
29.
Yang
M
,
McKay
D
,
Pollard
JW
,
Lewis
CE
.
Diverse functions of macrophages in different tumor microenvironments
.
Cancer Res
2018
;
78
:
5492
503
.
30.
Azizi
E
,
Carr
AJ
,
Plitas
G
,
Cornish
AE
,
Konopacki
C
,
Prabhakaran
S
, et al
.
Single-cell map of diverse immune phenotypes in the breast tumor microenvironment
.
Cell
2018
;
174
:
1293
308
.
31.
Guo
X
,
Zhang
Y
,
Zheng
L
,
Zheng
C
,
Song
J
,
Zhang
Q
, et al
.
Global characterization of T cells in non-small-cell lung cancer by single-cell sequencing
.
Nat Med
2018
;
24
:
978
85
.
32.
Zeng
Z
,
Wei
F
,
Ren
X
.
Exhausted T cells and epigenetic status
.
Cancer Biol Med
2020
;
17
:
923
36
.
33.
Lavallee
E
,
Sfakianos
JP
,
Mulholland
DJ
.
Tumor heterogeneity and consequences for bladder cancer treatment
.
Cancers
2021
;
13
:
5297
.
34.
Wang
L
,
Sebra
RP
,
Sfakianos
JP
,
Allette
K
,
Wang
W
,
Yoo
S
, et al
.
A reference profile-free deconvolution method to infer cancer cell-intrinsic subtypes and tumor-type-specific stromal profiles
.
Genome Med
2020
;
12
:
24
.
35.
Wang
L
,
Sfakianos
JP
,
Beaumont
KG
,
Akturk
G
,
Horowitz
A
,
Sebra
RP
, et al
.
Myeloid cell-associated resistance to PD-1/PD-L1 blockade in urothelial cancer revealed through bulk and single-cell RNA sequencing
.
Clin Cancer Res
2021
;
27
:
4287
300
.
36.
Cohen
SM
.
Urinary bladder carcinogenesis
.
Toxicol Pathol
1998
;
26
:
121
7
.
37.
Mirvish
SS
.
Role of N-nitroso compounds (NOC) and N-nitrosation in etiology of gastric, esophageal, nasopharyngeal and bladder cancer and contribution to cancer of known exposures to NOC
.
Cancer Lett
1995
;
93
:
17
48
.
38.
Bonfanti
M
,
Magagnotti
C
,
Bonati
M
,
Fanelli
R
,
Airoldi
L
.
Pharmacokinetic profile and metabolism of N-nitrosobutyl-(4-hydroxybutyl)amine in rats
.
Cancer Res
1988
;
48
:
3666
9
.
39.
Fantini
D
,
Glaser
AP
,
Rimar
KJ
,
Wang
Y
,
Schipma
M
,
Varghese
N
, et al
.
A carcinogen-induced mouse model recapitulates the molecular alterations of human muscle invasive bladder cancer
.
Oncogene
2018
;
37
:
1911
25
.
40.
Kerzeli
IK
,
Lord
M
,
Doroszko
M
,
Elgendy
R
,
Chourlia
A
,
Stepanek
I
, et al
.
Single-cell RNAseq and longitudinal proteomic analysis of a novel semi-spontaneous urothelial cancer model reveals tumor cell heterogeneity and pretumoral urine protein alterations
.
PLoS One
2021
;
16
:
e0253178
.
41.
Mariathasan
S
,
Turley
SJ
,
Nickles
D
,
Castiglioni
A
,
Yuen
K
,
Wang
Y
, et al
.
TGFbeta attenuates tumour response to PD-L1 blockade by contributing to exclusion of T cells
.
Nature
2018
;
554
:
544
8
.
42.
Zhao
J
,
Xiao
Z
,
Li
T
,
Chen
H
,
Yuan
Y
,
Wang
YA
, et al
.
Stromal modulation reverses primary resistance to immune checkpoint blockade in pancreatic cancer
.
ACS Nano
2018
;
12
:
9881
93
.
43.
Thomas
DA
,
Massague
J
.
TGF-beta directly targets cytotoxic T cell functions during tumor evasion of immune surveillance
.
Cancer Cell
2005
;
8
:
369
80
.
44.
Zhao
F
,
Evans
K
,
Xiao
C
,
DeVito
N
,
Theivanthiran
B
,
Holtzhausen
A
, et al
.
Stromal fibroblasts mediate anti-PD-1 resistance via MMP-9 and dictate TGFbeta inhibitor sequencing in melanoma
.
Cancer Immunol Res
2018
;
6
:
1459
71
.
45.
Tang
H
,
Liang
Y
,
Anders
RA
,
Taube
JM
,
Qiu
X
,
Mulgaonkar
A
, et al
.
PD-L1 on host cells is essential for PD-L1 blockade-mediated tumor regression
.
J Clin Invest
2018
;
128
:
580
8
.
46.
Peranzoni
E
,
Lemoine
J
,
Vimeux
L
,
Feuillet
V
,
Barrin
S
,
Kantari-Mimoun
C
, et al
.
Macrophages impede CD8 T cells from reaching tumor cells and limit the efficacy of anti-PD-1 treatment
.
Proc Natl Acad Sci U S A
2018
;
115
:
E4041
E50
.

Supplementary data