Urine scRNAseq reveals new insights into the bladder tumor immune microenvironment

Bladder cancers (BCs), the majority of which are diagnosed pathologically as urothelial carcinoma (UC), represent a significant health burden. BC is among the 10 most common cancers globally, with around 550,000 new cases annually (Richters et al., 2020). BC is also the most expensive cancer to treat, costing the healthcare system $4 billion each year (Park and Hahn, 2014).

Recently, efforts to accelerate research and clinical trials investigating immunotherapies for BC have gathered momentum. BC has a high tumor mutational burden, facilitating neoantigen generation and indicating a potential response to immunotherapy (Alexandrov et al., 2013). BC’s amenability to immunotherapy is evidenced by the fact that the oldest Federal Drug Administration (FDA)–approved cancer immunotherapy was for BC; in 1990, intravesical instillation of Bacillus Calmette-Guérin (BCG), a live attenuated strain of Mycobacterium bovis, was FDA approved for treatment of non-muscle invasive BC (NMIBC) and has remained standard of care. Despite being used for decades, BCG’s precise therapeutic mechanisms remain unclear (Morales et al., 1976; Eidinger and Morales, 1976). While most patients respond to BCG, 30–45% have local recurrence, and 10–15% progress to a higher mortality risk state (Fernandez-Gomez et al., 2008; Cambier et al., 2016). For patients who progress to metastatic urothelial cancer (mUC), five anti-PD-1/PD-L1 immune checkpoint blockade (ICB) monoclonal antibody therapies were FDA approved over the last decade (Katz et al., 2017; Sharma et al., 2017; Balar et al., 2017; Bellmunt et al., 2017; Patel et al., 2018; Powles et al., 2017). Unfortunately, only 15–25% of mUC patients respond to ICB, underscoring a need to improve outcomes. The recent 2023 approval of enfortumab vedotin with ICB for advanced-stage BC (Powles et al., 2021) has reinvigorated the search for alternative approaches to treat resistant disease and more rapidly and non-invasively monitor for recurrence.

A unique feature of BC is the convenience of urine for assessing disease. Urine contains tumor cells, proteins, and nucleic acids that can be non-invasively profiled for BC detection and recurrence. Existing FDA-approved urine-based clinical tests for BC diagnosis and follow-up are narrowly focused on cancer cells (Planz et al., 2005; Varella-Garcia et al., 2004; Pode et al., 1999; O’Sullivan et al., 2012; Moonen et al., 2005). However, preliminary work has shown BC urine can contain immune cells (Wong et al., 2018). Since longitudinal tumor biopsies can be invasive, profiling urine-derived cells (UDCs) during immunotherapy could generate insight into the pretreatment tumor microenvironment (TME) associated with sensitivity and resistance to immunotherapy, while also advancing understanding of treatment-induced immune modulation.

Investigations of immune UDCs in BC have been limited and biased to date. A prior study compared exhaustion marker expression on CD4⁺ and CD8⁺ T cells across BC urine, tumor, and blood, and found urine lymphocytes were more similar to those from tumor than blood (Wong et al., 2018). However, the study was limited by predetermined flow cytometry antibody panels, which reduced the depth and breadth of cell phenotypes that could be profiled. While informative, this study was restricted to T cell phenotyping and did not characterize other features or cells found in the TME.

There have been limited studies monitoring BC patients’ response to immunotherapy. In 1991, a flow cytometry analysis of UDCs from NMIBC patients was conducted at various timepoints after BCG instillation, finding granulocytes, monocytes, macrophages, and T and B cells (de Boer et al., 1991). Unfortunately, at the time of the study, technologies were limited, and more granular subsets were not resolved. Recently, mass cytometry analysis was performed on BC UDCs 3–7 days after BCG instillation, revealing a recruitment of granulocytes and monocytes (Castellano et al., 2022), but comprehensive, unbiased transcriptomic-based profiling of BC UDCs during BCG or other immunotherapies have yet to be performed. This has left a need to conduct an unbiased investigation of cells in BC patient urine in general and in the context of immunotherapy.

Single-cell RNA-sequencing (scRNAseq) of UDCs has been proven to be a method to analyze cells from the genitourinary (GU) tract. The first UDC scRNAseq was performed in lupus nephritis patients (Arazi et al., 2019). A healthy donor (HD) urine atlas was also created from urine pooled from 12 adults to obtain sufficient cells for sequencing. HD UDCs comprised tubular cells, urothelial cells, podocytes, neutrophils, monocytes, dendritic cells (DCs), and T cells (Wang et al., 2021). Since these studies, UDC scRNAseq was performed for diabetes (Abedini et al., 2021), acute kidney injury (Cheung et al., 2021), and focal segmental glomerulosclerosis (Fu and Campbell, 2021). These studies demonstrated the value of UDC scRNAseq in studying renal pathology. Until the present study, UDC scRNAseq has not yet been extended to BC.

In this study, we unbiasedly analyzed the spectrum of UDCs present in BC urine and compared UDCs to matched tumor and peripheral blood mononuclear cells (PBMC) using scRNAseq. We identified frequencies and transcriptional profiles of several immune subsets across urine, tumor, and PBMC of nine BC patients. We performed scRNAseq on longitudinal urine from a patient undergoing intravesical BCG therapy to monitor therapy-induced changes to the local microenvironment. We finally compared our BC findings with HD urine from a public dataset (Wang et al., 2021). Overall, we demonstrate BC UDCs better reflect the tumor immune microenvironment than blood. Our results implicate that UDCs have potential application to non-invasively interrogate the bladder tumor immune microenvironment before and during therapy.

We collected urine, tumor, and blood from nine BC patients on the day of their tumor resection by transurethral resection of bladder tumor (TURBT) or cystectomy. Clinical characteristics of this cohort are presented in Table 1. There were eight men and one female, and five had NMIBC while four had MIBC. All were naïve to chemotherapy, immunotherapy, or radiation except for Patient 5, who previously received and was resistant to BCG therapy, and Patient 7, for whom we obtained specimens before and after neoadjuvant chemotherapy. Peripheral blood and cystoscope- or catheter-collected urine were obtained in the operating room before tumors were resected. Urine was collected from all nine patients, paired tumor tissue from eight patients, and matched blood from five patients (Fig. 1 A). For Patient 7, samples were obtained before (thereby referred to as Patient 7A) and after chemotherapy (Patient 7B). Single-cell suspensions from urine, tumor, and blood were analyzed using a droplet-based scRNAseq pipeline (10× Genomics) (Zheng et al., 2017a).

We obtained 25–100 ml of urine from each patient, ranging in cellularity (Table 1). In parallel, we collected a similar amount of voided urine from 10 HDs receiving their annual physicals. Notably, BC urine had significantly higher cellularity than HD urine (P value <0.0001, Wilcoxon rank sum test). On average, BC urine contained 770,054 live cells/ml as opposed to 3,408 live cells/ml in HD urine, a striking 226-fold difference (Fig. 1 B).

We obtained sequencing data from 63,493 UDCs from our patient cohort. The sequencing depth and quantity of cells for each sample are in Table S1. After applying our quality control, including deletion of cells exhibiting over 15% mitochondrial content, comprising dead cells, 40,342 cells remained with 29,081 genes detected across all UDCs (McGinnis et al., 2019). We integrated cells across specimens using the Seurat analysis pipeline (Satija et al., 2015) and unbiasedly clustered them with a resolution of 0.5, resulting in 18 clusters we annotated using published markers (Fig. 1 C and Fig. S1 A).

Across patients’ UDCs, we found eight non-immune and nine immune cell clusters. Non-immune UDCs included urothelial cells (UPK2, UPK1A, KRT18) (Habuka et al., 2015), neuronal cells (NNAT, NPY) (Shinde et al., 2016), endothelial cells (PLVAP, SPARCL1) (Guo et al., 2016), myofibroblasts (ACTA2, COL1A1, FN1) (Rockey et al., 2013), and prostate epithelial cells (KLK3, KRT1). Immune-related UDCs included a cluster of monocytes and macrophages (LYZ, CD14, MRC1, CD68), CD4⁺ T cells (CD3D, CD4), CD8⁺ T cells (CD3D, CD8A), neutrophils (CXCL8, S100A8, FCGR3B, CEACAM8) (Zilionis et al., 2019), natural killer cells (NK; NKG7, FCGR3A, PRF1), B cells (CD19, CD79A), DCs (CD1C, CLEC9A), plasma cells (CD79A, IGHG1), mast cells (CPA3, TPSAB1) (Jiang et al., 2020; Maaninka et al., 2013), and plasmacytoid DCs (pDC; TCF4, LILRA4) (Gerhard et al., 2021) (Fig. 1, C and D). We confirmed the presence of these major immune subsets in BC urine using flow cytometry (Fig. S1 B). Cell-type clusters were not patient-specific, indicating shared profiles across patients (Fig. 1 E). However, comparing the composition of UDCs across patients revealed considerable heterogeneity (Fig. 1 F). These data suggest that BC urine contains a diversity of cells likely shed from tumor into urine.

To compare UDCs with cells from other tissues, we integrated tumor cells, UDCs, and PBMC across patients. In addition to UDCs, there were 39,367 PBMCs and 28,880 tumor cells with sequencing depths in Table S1. After applying the same quality control, this resulted in 106,587 cells with 30,686 genes across cells. These cells were clustered with a resolution of 0.5, resulting in 22 subsets (Fig. 2 A and Fig. S2 A).

The same epithelial, stromal, and immune cells identified in the UDC scRNAseq were present in this pan-tissue object, with the addition of cells allowing the distinction of CD14⁺ monocyte (LYZ, FCN1, CD14), CD16⁺ monocyte (LYZ, FCN1, FCGR3B), macrophage (MRC1, CD163, CD68), and CD4⁺ regulatory T cell (CD4⁺ T_REG; CD3D, CD4, FOXP3, IL2RA/CD25) clusters. Parsing by tissue revealed tumor and urine contained epithelial and stromal subsets underrepresented in blood (Fig. 2 B and Fig. S2 B). The epithelial cell clusters detected in the blood were likely circulating tumor cells, though we did not have non-malignant tissue with normal cells or DNA sequencing to determine their malignancy with certainty. Importantly, parsing cells by tissue revealed BC urine and tumor contain analogous cell types and frequencies. This was consistent when cell types were parsed by tissue for each individual patient as we found no significant differences in frequencies of the clusters between tumor and urine (Fig. S2 C). Overall, these findings suggest UDCs may accurately reflect the cellular composition of the TME.

To assess whether immune UDCs recapitulate the TME, we subset immune populations to compare across tissue. We began with CD4⁺ T cells, the most prevalent immune cell. We integrated the CD4⁺ T cell clusters and removed contaminating cells to obtain 25,538 CD4⁺ T cells. Reclustering resolved 11 clusters comprising two naïve CD4⁺ T cell subsets expressing CCR7, TCF7, and SELL (CD4T_{NAIVE_1} and CD4T_{NAIVE_2}: log₂FC = 0.3–0.7) (Szabo et al., 2019); two effector memory subsets expressing CRIP1, ITGB1, and S100A11 (CD4T _{EM_1} and CD4T _{EM_2}: log₂FC = 0.3–0.7) (Sade-Feldman et al., 2018); a CD4⁺ T cell terminally differentiated cluster expressing GZMA and CCL5 (CD4T_EMRA: log₂FC = 1.5, 1.1); two activated CD4⁺ T cell clusters expressing CD69 (CD4T_ACT: log₂FC = 0.9, 1.8); two CD4⁺ T_REG subsets co-expressing FOXP3, TIGIT, and IL2RA/CD25 (CD4T_REST-CM: log₂FC = 0.9–2.15); a cytotoxic CD4⁺ T subset co-expressing PRF1, GZMB, GNLY, and IFNG described to mediate anti-tumor cytotoxicity in BC (CD4T_CYTOXIC: log₂FC = 1.4, 3.5, 3.0, 2.3) (Oh et al., 2020); and a third activated CD4⁺ T cell cluster (CD4T_{ACTIVATED_3}) upregulating TNFAIP3, expressed after T cell receptor (TCR) signaling with a role in autophagy (Matsuzawa et al., 2015), and ubiquitin protein USP9Y, suggesting these cells were undergoing autophagy after TCR stimulation (Fig. 2 C and Fig. S3 A).

We next compared frequencies of CD4⁺ T cell subsets across tissue. While most clusters were detected across tissues, the CD4T_CYTOTOXIC and CD4T_{ACTIVATED_3} clusters were exclusively in tumor and urine (Fig. 2, D and E). Interestingly, the composition of urine CD4⁺ T cells represented that of the tumor with no significant differences in frequencies across patients. However, blood contained proportionally more CD4T_{NAIVE_1}, CD4T_{EM_1}, and CD4T_{EM_2} and less CD4T_ACT1 and CD4T_CYTOTOXIC clusters than tumor and urine (Fig. S3 B). Our data supports prior reports of cytolytic and activated CD4⁺ T cells in BC tumors (Oh et al., 2020) and their presence in urine.

To further evaluate CD4⁺ T cells, we examined the expression of various functional markers. We found urine and tumor CD4⁺ T cells more highly expressed inhibitory markers: CTLA4, HAVCR2, LAG3, PDCD1, TIGIT, and TOX compared to blood. The B cell chemoattractant CXCL13, associated with T follicular helper cells and tertiary lymphoid structures (TLSs) (Bindea et al., 2013), was also more highly expressed in urine and tumor CD4⁺ T cells (Fig. 2 F). Conversely, circulating CD4⁺ T cells more highly expressed TCF7, indicating T cell stemness (Willinger et al., 2006; Zhao et al., 2022). Upon assessing which clusters expressed these genes, the CD4T_CYTOTOXIC cluster, present in urine and tumor but not blood, expressed the highest levels of CXCL13 and these inhibitory markers, followed by the CD4⁺ T_REG clusters (Fig. S3 A). These data confirm tumors contain more suppressive and exhausted CD4⁺ T cells than blood, and this is reflected in urine.

To unbiasedly assess for differences across tissue, we conducted differential gene expression analysis of CD4⁺ T cells. We found only two significantly differentially expressed genes (DEGs) between urine and tumor CD4⁺ T cells (Wilcoxon test, P value <0.05; log₂FC > |1|) (Fig. 2 G). Only IFITM1, an interferon-stimulated gene, was upregulated in tumor, and ADIRF, encoding adipogenesis regulatory factor, was upregulated in urine. The lack of DEGs between tumor and urine CD4⁺ T cells supports their phenotypic similarity.

However, comparing tumor to blood CD4⁺ T cells, tumor CD4⁺ T cells upregulated 94 DEGs (Fig. 2 H). Among these were RGS1, a tissue-residency marker, as well as CD69 and TNFAIP3, both associated with T cell activation. When we performed gene set enrichment analysis (GSEA) using Ingenuity Pathway Analysis, these DEGs were involved in pathways related to heat stress, oxidative stress, unfolded protein response, autophagy, and hypoxia, all suggesting that metabolic stressors of the TME impact tumor T cells more than circulating ones (Fig. 2 I). There were conversely 10 DEGs upregulated in blood versus tumor CD4⁺ T cells including TCF7, indicating better effector capacity.

Comparing urine to blood CD4⁺ T cells, we discovered 101 DEGs, of which the majority overlapped with tumor versus blood DEGs including RGS1, CD69, and TNFAIP3 (Fig. 2 H and Fig S3 C). DEGs upregulated in urine CD4⁺ T cells were involved in similar pathways as those from tumor, indicating UDCs are impacted by similar stressors of the tumor niche (Fig. 2 K).

We also characterized CD4⁺ T cells using functional gene signatures in Table S2 (Azizi et al., 2018; Chtanova et al., 2005; Smith-Garvin et al., 2009; Glimcher et al., 2004; Schietinger et al., 2012; Wherry and Kurachi, 2015; Wherry, 2011; Best et al., 2013) and found that transcriptionally, circulating CD4⁺ T cells had higher cytolytic effector capacity scores than urine and tumor CD4⁺ T cells. Conversely, urine and tumor CD4⁺ T cells had higher anergy, activation, glucose deprivation, and hypoxia scores. The elevated expression of signatures related to tumor metabolic perturbation in urine and tumor CD4⁺ T cells provides further support that UDCs pass through the TME (Makino et al., 2003; Benita et al., 2009; Ho et al., 2015) (Fig. S3 D). While we can only postulate biological function from gene signatures, overall, we found urine CD4⁺ T cells transcriptionally more similar to those from tumor than blood.

We next assessed CD8⁺ T cells. The 8,784 CD8⁺ T cells clustered into 12 clusters. Six clusters were found in all three tissues: an activated CD8⁺ T effector memory (T_{EM_ACT_1}) cluster expressing CCL4, CD69, and IFNG (log₂FC = 1.0, 1.2, 1.3); a resting CD8⁺ T_EM cluster expressing GZMK and KLRG1 (CD8T_{EM_REST}: log₂FC = 0.7, 0.4) (Szabo et al., 2019; Andreatta et al., 2021); two CD8⁺ terminally differentiated effector cell clusters (T_EMRA) expressing cytotoxic markers GZMB, GZMH, NKG7, and PRF1 (CD8T_EMRA: log₂FC = 0.5–2.1) (Szabo et al., 2019); and two CD8⁺ T naïve clusters expressing IL7R, LTB, and TCF7 (CD8T_NAÏVE: log₂FC = 0.4–1.6) (Deng et al., 2020) (Fig. 3 A and Fig. S4 A).

However, there were six tissue-resident and exhausted CD8⁺ T cell clusters exclusive to tumor and urine and absent in blood. These included two exhausted CD8⁺ T resident memory (T_{RM_EX}) subsets upregulating the tissue-residency markers CXCR6 and ITGAE (log₂FC = 1–1.5) (Szabo et al., 2019), the tumor-reactive marker CXCL13 (log₂FC = CD8T_{RM_EX_1}: 1.5, CD8T_{RM_EX_2}: 0.3) (Workel et al., 2019), and exhaustion markers LAG3 and HAVCR2/Tim-3 (log₂FC = 1.3–1.7) (Carmona et al., 2020). There were also two activated CD8⁺ T_RM clusters expressing tissue-specific RGS1 (Zheng et al., 2017b; Doedens et al., 2016; Tang et al., 2023) and IFNG (CD8T_{RM_ACT}: log₂FC = 1.3–1.7) and a resting CD8⁺ T_RM cluster expressing RGS1, ITGA1, and GZMK (CD8T_{RM_REST}: log₂FC = 1.0, 0.9, and 0.8) and lacking activation-related genes. Finally, there was an exhausted CD8⁺ T cell (T_EX) cluster expressing CTLA4, HAVCR2, and LAG3 (log₂FC = 0.4–0.9) (Fig. 3, B and C; and Fig. S4 A). The shared presence of tissue-resident and exhausted CD8⁺ T cells in tumor and urine, yet absence in blood, indicates urine captures T cells that once infiltrated tumor that are not otherwise detectable in circulation.

We found no significant differences in frequencies of CD8⁺ T cell subsets between tumor and urine across patients (Fig. S4 B). However, there were significantly proportionally more CD8T_{NAÏVE_1} and CD8T_{EM_REST} cells in blood and an absence of exhausted and tissue-resident clusters in blood across patients.

Correspondingly, urine and tumor CD8⁺ T cells more highly expressed the checkpoint markers PDCD1, CTLA4, HAVCR2, LAG3, and TIGIT than blood (Fig. 3 D). Conversely, circulating CD8⁺ T cells more highly expressed TCF7. The checkpoint molecules were most highly expressed by exhausted CD8⁺ T cells and TCF7 by the CD8⁺ T_NAIVE and T_REST-EM clusters (Fig. S3 E). Overall, tumor and urine exclusively contained exhausted and tissue-resident CD8⁺ T cells, whereas blood was enriched in naïve and resting effector memory CD8⁺ T cells.

To delve into tissue-specific differences in CD8⁺ T cells, we performed DEG analysis. Urine and tumor CD8⁺ T cells had striking transcriptional similarity with only one DEG (P value <0.05; log₂FC > |1.0|), GNLY, upregulated in tumor (Fig. 3 E). Conversely, there were 20 DEGs up in tumor versus PBMC, many belonging to the heat shock protein (HSP) family. RGS1 was again overexpressed by tumor versus blood CD8⁺ T cells (Fig. 3 F). GSEA revealed these DEGs were related to response to heat and oxidative stress characteristic of a metabolically unfavorable TME (Fig. 3 F). We also found 40 DEGs upregulated in urine versus blood CD8⁺ T cells. A near majority of these urine versus PBMC DEGs overlapped with tumor versus PBMC DEGs, including HSP genes and RGS1 (Fig. 3 H and Fig. S4 C). Accordingly, several overlapping urine versus PBMC pathways pertained to the stress response as found in our tumor versus PBMC analysis (Fig. 3 I and Fig. S3 G).

We repeated our analysis of functional signatures in CD8⁺ T cells using gene sets in Table S2. We found circulating CD8⁺ T cells demonstrated higher cytolytic effector scores, and conversely, urine and tumor CD8⁺ T cells had higher anergy, activation, and tumor metabolic disruption scores (Fig. S3 H). We concluded urine CD8⁺ T cells were transcriptionally similar to those from tumor with metabolic signatures suggestive of passage through the TME.

To assess the myeloid compartment across tissue, we integrated monocyte and macrophage clusters, encompassing 11,661 cells. Reclustering revealed three CD14⁺ monocyte clusters expressing FCN1 (log₂FC = 0.3, 0.8, 1.1); a monocyte cluster (IL1B_Mono) upregulating IL1B (log₂FC = 1.9) and the monocyte-associated marker VCAN (log₂FC = 0.6); a CD16⁺ monocyte cluster expressing FCGR3A (log₂FC = 2.2); and three macrophage clusters upregulating macrophage markers (CD163, CD68, and MRC1). Among these macrophage clusters, one upregulated SPP1 (log₂FC = 3.1), another highly expressed C1QC (log₂FC = 3.1) and CXCL9 (log₂FC = 2.0), and the third upregulated IL1B (log₂FC = 2.1) (Fig. 4 A and Fig. S5 A).

We observed a striking paucity of macrophages in blood compared with tumor and urine (Fig. 4, B and C). It is well-established that there are few macrophages found circulating in blood, and indeed the macrophage subsets we identified were enriched in tumor and urine. When we examined for macrophage-related genes such as MRC1, CD163, and CD68, tumor and urine had higher expression, whereas blood had higher expression of monocyte-related genes: FCN1, VCAN, and LYZ (Fig. 4 D). The tumor-associated macrophage (TAM) markers SPP1, IL1B, C1QC, and CXCL9 were also higher in tumor and urine.

Interestingly, recent reports in different cancers have linked some of these TAM subsets we identified in BC to clinical prognosis. In particular, SPP1 versus C1QC and SPP1 versus CXCL9 dichotomies have been reported to identify human TAM polarity in studies of colon and head and neck cancer patients (Zhang et al., 2020; Bill et al., 2023). In these contexts, SPP1⁺ TAMs tracked with poorer clinical outcomes, associated with hypoxic pathways, and were implicated in tumor-promoting angiogenesis, whereas C1QC⁺ and CXCL9⁺ TAMs were enriched in complement activation and antigen presentation genes and associated with immune-activating pathways and positive clinical outcomes. In our data, we observed a similar GSEA for these two TAM clusters (Fig. S5, B and C). In pancreatic cancer, a population of IL1B⁺ TAMs, similar to what we describe in BC, was found to contribute to proinflammatory reprogramming of cancer cells and disease progression (Caronni et al., 2023). When we investigated the frequency of these myeloid clusters across our cohort, there were no significant differences between urine and tumor, reiterating the reflection of TAMs in the urine (Fig. S4 D).

To compare urine versus tumor monocytes and macrophages, we performed DEG analysis, revealing few gene differences (log₂|FC| > 1 and P < 0.05) (Fig. 4 E). However, we again saw more DEGs in our tumor versus blood analysis. Genes upregulated in tumor myeloid cells were related to anti-inflammatory IL-10 as well as IL-4 and IL-13 signaling, relating to Th2 skewing away from a more directly tumoricidal Th1 response. We again saw tumor myeloid cells upregulated genes in hypoxia, oxidative stress, and autophagy pathways, indicating exposure to TME metabolic perturbations like with T cells (Fig. 4, F and G; and Table S2). When we evaluated urine versus blood DEGs, we saw similar pathways upregulated in urine monocytes and macrophages (Fig. 4, H and I; and Fig. S4 E). Overall, urine contained multiple TAM populations, both immunostimulatory and protumorigenic, and lacking in blood.

We also examined neutrophils, which are known to have both immunosuppressive and cytotoxic roles in cancer (Hedrick and Malanchi, 2022). We integrated 1,571 neutrophils across tissue and clustering resolved subsets recently described in a pan-cancer neutrophil report (Wu et al., 2024) including clusters highly expressing S100A12, VEGFA, CXCL8, CXCR2, and NFKBIZ (Fig. 4 J). As we performed a Ficoll-based density separation of blood to obtain PBMC that removes most polymorphonuclear neutrophils, we expected a de-enrichment of neutrophils in blood. When we compared across tissues, urine captured the breadth of neutrophil profiles seen in tumor (Fig. 4, K and L). Conversely, PBMC lacked cells in the proinflammatory CXCL8⁺ and chemotactic CXCR2⁺ clusters, though, this could be due to processing. Furthermore, as neutrophils have high RNAse content, they are notoriously difficult to capture with scRNAseq, and we likely lost additional neutrophils across tissue due to this (Wigerblad et al., 2022). Thus, though we could not draw many conclusions based on aforementioned caveats, we found BC urine reflected the diversity of intratumoral neutrophil profiles, matching our findings in other compartments.

The 4,130 NK cells in our data clustered into 10 subsets, the majority of which we were able to align with signatures from a recent pan-cancer NK cell atlas (Tang et al., 2023) (Table S2). Each cluster varied in expression of NCAM1/CD56 and FCGR3A/CD16, markers used to distinguish CD56^brightCD16^lo and CD56^dimCD16^hi NK cell types. There were two CD56^brightCD16^lo subclusters (c3_IL7R, c4_CREM) while the rest were CD56^dimCD16^hi. Using their signatures, we distinguished subsets they identified including: c0_DNAJB1, c1_CX3CR1, c3_IL7R, c4_CREM; c7_KLRC2, c8_IL32, and c9_CCL3 (Fig. 5, A and B). We also identified c2_PTGDS based on overexpression of PTGDS encoding prostaglandin D2 synthase (log₂FC = 1.1), c5_MYO1F based on overexpression of MYO1F (log₂FC = 0.7), a nuclear-cytoplasmic shuttle protein that may be involved in transporting NK cytotoxic granules, and c6_SYNE2 based on overexpression of the nuclear protein SNYE2 that binds to cytoplasmic F-actin may play a role in granule transport and cytotoxicity (log₂FC = 0.8) (Fig. 5, A and B).

Interestingly, we observed three clusters exclusive to tumor and urine: c4_CREM, c7_KLRC2, and c9_CCL3, not distinguishable in blood (Fig. 5, C and D; and Fig. S4 F). Two clusters expressed RGS1, a pan-cancer NK tissue residency marker: c4_CREM (log₂FC = 1.3) and c9_CCL3 (log₂FC = 2.0) (Fig. 5 E). The third cluster, c7_KLRC2, appeared to be an adaptive NK cluster due to its high expression of an adaptive NK cell score encompassing KLRC2/NKG2C, B3GAT1/CD57, LILRB1, and FCGR3A/CD16 (Costa-García et al., 2019; Schlums et al., 2015) (Fig. 5 F). The presence of adaptive NK cells could be due to past viral infection by a common virus like cytomegalovirus or stimulation with proinflammatory cytokines. We also found expression of tissue-residency markers CD69, CXCR6, ITGAE/CD103, ITGA1/CD49a, and RGS1 (Peng et al., 2013; Sojka et al., 2014; Freud et al., 2017; Dogra et al., 2020; Tang et al., 2023) higher in tumor and urine than blood NK cells (Fig. 5 G). These data show the existence of tissue-resident and adaptive NK cells in urine, likely originating from tumor and absent in blood.

Conversely, c5_MYO1F, c6_SYNE2, and c8_IL32 clusters were enriched in PBMC and detectable in tumor but not urine. As MYO1F and SYNE2 may be involved in transport of cytotoxic granules and IL32 has been reported to enhance NK cell cytotoxicity (Park et al., 2012), these cell profiles could indicate greater cytotoxicity capacity and explain why they are enriched in blood (Fig. S4 F). We found no significant differences in NK subset frequency between tumor and urine (Fig. S4 F). In summary, urine NK cells appeared to well represent tumor NK cell composition.

To probe the functional state of these NK cells, we examined their expression of cytotoxic granules, perforin, and granzyme B, and several activating and inhibitory receptors (Shimasaki et al., 2020). Blood NK cells had higher levels of PRF1 and GZMB and activating receptors, CD160, NCR3, NCR1, KLRK1, and FCGR3A/CD16, suggesting higher cytotoxic and antibody-dependent cellular cytotoxicity capacity. Conversely, urine and tumor NK cells had decreased expression of cytotoxicity genes and higher expression of inhibitory receptors: KIR2DL1/CD158a, KIR2DL3/CD158b, KIR3DL1/CD148e, KIR3DL2/CD158k, LAG3, PDCD1/PD-1, CD96, and HAVCR2/Tim-3 (Fig. 5 H).

DEG analysis of urine and tumor NK cells revealed few differences (log₂|FC| > 1, P < 0.05) (Fig. 5 I). However, there were many differences comparing tumor and urine NK cells to blood. Tumor NK cells had higher levels of RGS1 than blood (Fig. 5 J), matching T cells and reports from other cancers (Tang et al., 2023). We again found tumor versus blood NK cells enriched in heat and oxidative stress and unfolded protein response pathways (Fig. 5 K). Similar genes and pathways were upregulated in urine versus blood NK cells (Fig. 5, L and M; and Fig. S4 G). Upon applying functional gene signatures, circulating NK cells demonstrated higher cytolytic effector scores, and tumor and urine NK cells had higher hypoxia, glucose deprivation, and anergy scores, suggesting shared metabolic disruption in tumor and urine NK cells (Fig. S4 H). Overall, urine and tumor NK cells had several similarities in activating/inhibitory receptors and metabolic status.

We subsequently integrated B and plasma cells across tissues. Reclustering the 4,944 cells resolved six distinct populations (Fig. 6 A). We detected two naïve B cell clusters expressing IGHD (log₂FC = 0.5–0.9) and IGHM (log₂FC = 0.5–0.8), an IgG⁺ plasma cell cluster expressing IGHG1 (log₂FC = 1.0), and a memory IgA⁺ B cell cluster expressing the memory marker CD27 and IGHA1 (log₂FC = 0.5, 0.8) (Fig. 6 B and Fig. S4 I).

We discovered two clusters shared by urine and tumor but lacking or absent in blood. These included memory IgG⁺ B cells expressing CD27 and IGHG1 (log₂FC = 1.1, 1.0) and a GC-like B cell cluster expressing CD83 (log₂FC = 1.3), associated with light-zone GC B cells, and the chemokine receptor CXCR4 (log₂FC = 1.7), essential for B cell migration to the dark zone of the GC (Allen et al., 2004; Holmes et al., 2020) (Fig. 6, B–D; and Fig. S4 I). This GC-like cluster also most highly expressed BCL6, a canonical GC B cell marker (Cattoretti et al., 1995); however, another well-described GC marker, AICDA/AID (Crouch et al., 2007), was expressed in few B cells in our data overall, so we did not incorporate it in our annotation (Fig. S4 I). In the tumor, GC-like B cells are associated with TLSs, organized structures resembling secondary lymphoid organs that have been associated with adaptive immune responses and clinical benefit (Sautès-Fridman et al., 2019), suggesting prognostic utility for identifying these cells in urine. Globally, there were no significant differences in B cell composition between tumor and urine (Fig. S4 J).

Upon DEG analysis, urine B cells had fewer transcriptional differences to tumor compared to our blood versus tumor analysis (Fig. 6, E and F). The few differences observed were likely due to a slight proportional increase in memory IgG⁺ B cells in urine over tumor (Fig. 6 C and Fig. S4 J). Comparing tumor versus blood B cells, tumor B cells were enriched in pathways related to response to heat and oxidative stress and autophagy (Fig. 6 G). Similar pathways were enriched in our urine versus blood B cell comparison (Fig. 6, H and I; and Fig. S4 K). In conclusion, there was an enrichment of GC-like and memory IgG⁺ B cells in tumor and urine and more transcriptional similarity between urine and tumor B cells than with blood.

Curiously, across the immune subsets, we evaluated several pathways upregulated in urine and tumor as opposed to circulating blood cells were conserved across cell types. This is likely indicative of tumor and urine immune cells’ exposure and response to a metabolically harsh and demanding tumor milieu. This supports the capacity of UDCs to capture tumor-induced signatures better than circulating immune cells.

Finally, we compared DCs across tissue. We integrated pDCs and cDCs and clustered these 982 cells, resolving cDC1s, pDCs, cDC2s, mature regulatory (mreg) DCs, and monocyte-derived DCs (Fig. 6, J–L). We applied published annotations in Table S2 to identify cDC1s, cDC2s, and mregDCs (Maier et al., 2020), as well as CLEC9A and XCR1 to confirm cDC1 cluster annotation (Fig. S4 L). We determined pDCs by expression of TCF4 and LILRA4 and monocyte-derived DCs by FCN1 (Fig. 6 M). The mregDC signature is a gene program enriched in immunoregulatory and maturation genes expressed by cDCs after uptake of tumor antigens (Maier et al., 2020). We saw a slight enrichment of mregDCs in tumor and urine compared with blood, and conversely, an enrichment of pDCs in blood as compared with tumor and urine. Though there were too few cells conserved across tissues to perform robust DEG analysis, we found no significant differences in DC composition between tumor and urine (Fig. S4 M). Overall, our data revealed urine contains a diversity of DCs seen intratumorally.

We were next interested in whether longitudinal sampling of UDCs could capture temporal changes to the local microenvironment, so we evaluated UDCs from a BC patient receiving intravesical BCG therapy. For this regimen, patients receive initial debulking surgery and then are administered weekly BCG in a saline solution via catheter for 6 weeks. Upon instillation, patients avoid voiding for one to two hours for BCG to remain in contact with their bladder walls.

We followed a patient and obtained voided urine before and 1 hour after treatment, “pre-BCG” and “post-BCG” urine, respectively, each week (Fig. 7 A). Post-BCG urine contained the initial injected BCG along with urine generated over the hour, containing UDCs recruited by BCG. We observed the live UDC count trend upward over the 6-week course (Fig. 7 B). Across all weeks, there was a significant increase in live UDCs in post- versus pre-BCG urine, suggesting recruitment of cells by BCG (Fig. 7 C and Table 2).

We performed scRNAseq on UDCs from each urine sample to capture their phenotype. Unfortunately, contrasting to the aforementioned BC urine, we noted considerable dropout of cells differing from our initial loading. Unlike in earlier samples where patients had an intact tumor, this patient underwent debulking surgery, causing reduced UDC quantities, which likely contributed to the lower number of UDCs sequenced (Table 2). Despite the dropout, we were able to integrate 1,423 UDCs from 9 of the 12 samples with sufficient cell numbers (Fig. 7 D).

When we examined the patient’s UDCs, we discovered immature neutrophils, epithelial cells, monocytes, two T cell, umbrella cells, macrophages, and mature neutrophils (Fig. 7 D and Fig. S5 A). We saw fluctuation in cell composition across the 6 weeks and between the pre- versus post-BCG samples overall (Fig. 7 E). Specifically, we noted an increase in immature neutrophils between pre- versus post-BCG urine at each week, with a prominent increase on Week 5 (Fig. 7 F).

To evaluate BCG-induced changes to UDCs, we performed DEG analysis on all pre- versus post-BCG UDC samples. We did not have enough UDCs to perform this analysis within cell subsets, so we performed a global analysis to derive insight into molecular programs induced by BCG. 1 hour of BCG treatment upregulated proinflammatory genes including cytokines, IL1B, and CXCL8 (Fig. 7 G). GSEA revealed BCG-induced several proinflammatory pathways including IL-6 and IL-8 signaling, macrophage classical and alternative activation, TREM1 signaling, and neutrophil extracellular trap signaling (Fig. 7 H).

We then looked specifically at Week 1, as the pre-BCG sample was never exposed to BCG. For Week 1, we similarly observed the initial BCG treatment–induced several proinflammatory genes (Fig. 7 I). Pathway enrichment revealed an increase in genes related to macrophage activation, leukocyte extravasation, TREM1 signaling, and more (Fig. 7 J). Comparing Week 1 pre-BCG urine to Week 6 post-BCG urine, representing the earliest and latest timepoints, we found the overall treatment course induced a granulocyte and macrophage-dominated proinflammatory response (Fig. S5, B and C).

As we also obtained tumor and urine from BC Patient 7 before and after they received gemcitabine and cisplatin, we investigated whether their UDCs could capture chemotherapy-induced changes to the tumor. This patient had a pathological response with no evidence of disease at their second resection. Upon examining the urothelial cell compartment, we saw a chemotherapy-induced decrease in epithelial markers KRT13 and KRT19 as well as PSCA, commonly overexpressed in BC (Amara et al., 2001), and ELF3, an epithelial transcription factor often mutated in BC (Guneri-Sozeri and Erkek-Ozhan, 2022) (Fig. S5 D). Interestingly, these four genes were also downregulated in urine-derived urothelial cells in response to chemotherapy (Fig. S5 E). Our analysis of longitudinal samples from patients receiving BCG and chemotherapy underscores the potential to monitor pharmacodynamic changes to UDCs that may reflect changes to the TME.

To determine if UDC populations in BC patients were present in HDs, we analyzed publicly deposited scRNAseq of pooled urine from 12 healthy adults (GSE165396; Wang et al., 2021). The sequencing depth of these cells is in Table S1. After removing dead cells expressing 15% or more of mitochondrial genes, 841 UDCs remained. To compare HD and BC UDCs, we integrated them. Additional UDCs from BC patients allowed us to resolve urothelial cells, CD4⁺ T cells, neutrophils, monocytes, CD8⁺ T cells, macrophages, NK cells, endothelial cells, B cells, myofibroblasts, neuronal cells, prostate epithelial cells, fibroblasts, mast cells, plasma cells, and pDCs, from highest to lowest frequency in HD urine that has not previously been reported (Fig. 8 A).

Comparing cell composition between HD and BC urine, HD urine mainly contained neutrophils, monocytes, and macrophages and had a relative paucity of other immune (e.g., T cells) and stromal populations we found in BC urine (Fig. 7, A and B). Though present in BC urine, we did not detect mast cells, plasma cells, or endothelial cells in HD urine. While the frequency of HD UDCs was much less than in BC patients and too low to do robust DEG analyses between BC and HD UDCs, it was interesting to resolve these immune subsets in HD urine. This indicated homeostatic shedding of tissue-resident immune cells in urine in a non-cancer context, though at lower quantities and with different compositions and profiles than in BC.

As we had scRNAseq from NMIBC and MIBC patients, we compared UDC composition across stages (Fig. S5 F). Overall, NMI patients had a higher proportion of urothelial cells and a lower proportion of immune cells in their UDCs. However, there were not enough patients to make statistically robust comparisons. Overall, our work demonstrates that in addition to BC patients, HD urine can contain epithelial, immune, and stromal populations that can be profiled via scRNAseq, though at much lower cell magnitudes.

We next investigated clinical applications of our BC UDC scRNAseq atlas. From a literature search for urine assays for BC clinical care, we identified 12 commercially available diagnostic urine tests for BC, six of which were cleared for use by the FDA, that employed approaches including enzyme-linked immunosorbent assay, immunohistochemistry, DNA mutational detection, reverse-transcriptase polymerase chain reaction (RT-qPCR), and more (Table S3). As our data was transcript-based, we focused on molecular tests: the Cx Bladder test, based on IGFBP5, HOXA13, MDK, CDK1, and CXCR2, utilized for diagnosis and follow-up (O’Sullivan et al., 2012), and the Xpert BC Monitor test, based on UPK1B, IGF2, CRH, ANXA10, and ABL1, used for recurrence monitoring (Pichler et al., 2018).

From our atlas, we found Cx Bladder and Xpert BC Monitor test markers, except for CXCR2, were more highly expressed in BC than HD UDCs (Fig. 8 C). When we examined our scRNAseq to identify cells within the TME expressing these genes, we found the Cx Bladder Test markers were expressed by a diversity of cell types: IGFBP5 was expressed by urothelial cells, neuronal cells, and fibroblasts; MDK by urothelial cells, endothelial cells, fibroblasts, and neuronal cells; CDK1 by neuronal cells; and CXCR2 by neutrophils, which may explain why we found this marker also expressed by HD UDCs as we saw HD urine contained a high proportion of neutrophils (Fig. 8, B and D). The Xpert BC Monitor test markers were expressed mainly by urothelial and nerve cells (Fig. 8 E).

We also sought to deconvolute cell origins of markers in the commercially available Oncuria urine test, a multiplex immunoassay for BC containing many immune-related genes. This test has been validated for NMIBC diagnosis with attempts to apply it to predict BCG response (Murakami et al., 2022; Furuya et al., 2020; Hirasawa et al., 2021). This test examines for the biomarkers: SERPINA1, APOE, ANG, CA9, CXCL8, MMP9, MMP10, SERPINE1, SDC1, and VEGFA. The concentration of these analytes is incorporated into an algorithm that generates a risk score for NMIBC. We applied these features to our UDC atlas and found these markers were expressed by neutrophils, monocytes, macrophages, urothelial cells, neuronal cells, plasma cells, and mast cells (Fig. 8 F). Interestingly, assessment of these markers (except CA9 that was not detected) in UDCs from the patient who received BCG revealed an increase across the duration of BCG therapy (Fig. 8 G), matching a BCG-induced increase in granulocytes we noted earlier (Fig. 7 E). Our ability to resolve cell contributions of markers involved in commercial urine clinical assays reveals a putative application of our novel UDC scRNAseq atlas.

While our main interest was UDCs, urine proteins can also be investigated for concordance with the TME. We mapped the proteomic landscape of BC urine using Olink, assessing for 92 chemokine and cytokine analytes in their Immuno-Oncology Panel in urine from 10 HDs, 15 NMIBC, and 10 MIBC patients. We then performed unsupervised Euclidean hierarchical clustering on their protein profiles without inputting clinical stage details and observed that many of the MI patients clustered together based off similarity in their multiplex plasma profiles (Fig. S5 G). This may indicate that more advanced-stage patients’ urine profiles are similar and reveals a tumor-promoting immune microenvironment that is more inflammatory and devoid of competent adaptive immune cell responders. We also saw cytotoxic proteins, IFNγ and granzyme B, significantly decreased in MI patients as compared with NMI patients and HDs. Conversely, proinflammatory cytokines IL-8 and IL-6 were significantly elevated in BC patients compared with HDs and in MIBC versus NMIBC patients (Fig. S5 H). Both IL-8 and IL-6 have been correlated with BC progression (Urquidi et al., 2012; Chen et al., 2013). Overall, our work provides evidence that BC urine can reveal information about the bladder TME that can be exploited for therapeutic benefit and better understanding of the local immune response.

In this study, we performed, to our knowledge, the first unbiased characterization of cell contents of urine from BC patients using scRNAseq. To understand how UDCs may reflect the bladder TME, we performed scRNAseq across patients’ matched urine, tumor, and blood. Our results clearly showed several subsets of immune cells in BC urine that are more transcriptionally similar to the TME than blood. Across major immune lineages, UDCs consistently and robustly transcriptionally resembled tumor immune cells with few differences. UDCs shared tumor signatures relating to metabolic perturbation, immune suppression, and tissue residency, suggesting their passage through the TME. These findings have several research and clinical implications as urine can be sampled non-invasively and repeatedly in scenarios where tumor biopsies may not be feasible, including during therapy.

While our results are promising with exciting applications, we were limited by a small sample size due to the high costs of scRNAseq. Even within our cohort, there was variability in quantity and profile of UDCs across patients. Despite limitations, we are encouraged by the principles and proof of concepts set forth by this study. With expanded access to sequencing, continued reduction in the price of these technologies, and abundance of urine as a patient-derived material, these shortcomings should be reduced in the future.

Our study is the first to demonstrate the sheer breadth of tumor immune populations in BC urine that are underrepresented or undetectable in blood. This suggests clinical or research applications of BC UDCs for TME assessment and monitoring. As BCG and ICB each have patient subsets that are non-responders, identifying urine molecular signatures predictive of resistance in non-responders could translate to new urine tests for immunotherapy monitoring.

There are no standard screening strategies for individuals at high risk for BC, and tests for recurrence have limitations. Existing FDA-approved BC urine tests focus solely on cancer cells. These include urine cytology, which is based on morphologic assessment of UDCs under the microscope and is limited to detection of cancer cells (Planz et al., 2005); UroVysion, which examines for chromosome changes often seen in BC tumors (Varella-Garcia et al., 2004); BTA test, which detects bladder tumor-associated antigen (BTA) (Pode et al., 1999); ImmunoCyt, which focuses on mucin and carcinoembryonic antigen often found on tumor cells (O’Sullivan et al., 2012); and NMP22 BladderChek, which focuses on nuclear matrix protein 22 (NMP22) released by BC cells (Moonen et al., 2005). Other urine tests have emerged or are in development that examine DNA, RNA, or protein exfoliated from tumors. Identification of cells and transcriptional states unique to urine of BC patients versus healthy individuals offers the potential to discover novel urine biomarkers that may improve BC immunosurveillance.

While this is the first scRNAseq study on urine from BC patients, other studies described scRNAseq of urine from individuals with different renal pathologies (e.g., diabetes, focal segmental glomerular sclerosis, and acute kidney injury). The studies suggest the value of sequencing UDCs for other conditions including other GU cancers, such as upper tract UC, renal cell carcinoma, and prostate cancer, and even non-GU cancers. Urine is an easily accessible material for liquid biopsy applications, and scRNAseq can resolve critical cells and transcriptional profiles contributing to disease progression or treatment response. Future higher-powered UDC scRNAseq studies could resolve signatures distinguishing individuals with BC and other pathologies from HDs, which could be leveraged for prognostic tests.

An exciting, relatively unexplored clinical utility of urine involves using it to monitor and predict response to cancer immunotherapy. As UDCs reflect tumor-infiltrated immune cell composition and transcriptional profiles, UDCs could provide insight into cellular and molecular mechanisms of therapy response or resistance. Our assessment of treatment-induced changes to UDCs from a patient undergoing BCG therapy highlights the potential to use repeated urine sampling to capture temporal and treatment-induced changes to the tumor immune landscape. Higher-powered studies may allow for the detection of urine biomarkers of response or resistance to BCG and ICB that could one day be used in the clinic to replace tumor biopsies.

Importantly, urine can provide a critical and accessible window into tumor mechanisms that would otherwise be unattainable. Future immunotherapy clinical trials in BC should consider performing scRNAseq on longitudinal urine to assess for signatures of response or resistance. This could translate into non-invasive urine-based clinical tests for diagnostic, prognostic, and/or clinical decision-making purposes. Overall, our findings lay the groundwork for promising future research and clinical applications for urine in BC and other cancers.

Urine was collected from BC patients by cystoscope or Foley catheter in the operating room before the TURBT or cystectomy procedure, respectively. The freshly collected urine was then spun down at 500 g for 10 min at 4°C. The pellet was washed with ice-cold phosphate-buffered saline (PBS) (14190144; Gibco) and filtered through a 70-μm cell strainer. UDCs were counted for viability using Trypan blue (25-900-CI; Corning).

Tumor specimens were obtained freshly from the operating room and were grossed by a surgical pathologist. The tumor was then transported at room temperature and immersed in RPMI 1640 media (11875-093; Gibco). Once received, tumors were immediately minced and digested using an enzyme cocktail from a Tumor Dissociation Kit (130-095-929; Miltenyi Biotec) and mechanical dissociation with heat using the 37C_h_TDK_2 program on the gentleMACS (130-096-427; Miltenyi Biotec). Cells were sequentially filtered through 100-, 70-, and 40-μm cell filters. Single-cell suspensions were counted for viability using Trypan blue and those with over 70% viability were sent for sequencing immediately thereafter.

Fresh blood samples were obtained from patients before the start of surgery. Peripheral blood was collected into 10-ml ethylenediaminetetraacetic acid (EDTA) tubes and transported. Once received, whole blood was spun at 300 g for 7 min, and plasma was removed and frozen down. PBMCs were isolated by centrifugation over a Ficoll400 cell separation solution with a density of 1.077 g/ml (17-1440-03; GE Healthcare). Red blood cells were removed using ACK lysis buffer (A10492-01; Gibco). The remaining cells were counted for viability using Trypan blue. Cells were immediately loaded for scRNAseq.

Voided urine was collected from the BC patient at the start of their office visit. The patient then received intravesical BCG administration by their clinical provider. 1 hour after administration, we collected voided urine again. At both timepoints, the freshly collected urine was transported on ice to the lab and spun down at 500 g for 10 min at 4°C. The pellet was washed with ice-cold PBS and filtered through a 70-μm cell strainer. To improve viability, we stained these with LIVE/DEAD fixable aqua stain (L34957; Thermo Fisher Scientific) on ice for 20 min in PBS and then washed them. Live cells were sorted using an IMI5L cell sorter (BD) and immediately loaded for sequencing.

Droplet-based scRNAseq was performed on the 10X Genomics Chromium Single Cell 3′ and 5′ GEX V3 platforms, following manufacturer instructions. UDCs, tumor cells, and PBMC were added to PBS with 0.04% bovine serum albumin (BSA) (9048-46-8; Fisher Bioreagents) for loading on the 10X platform. Following library preparation, sequencing was performed on an Illumina NovaSeq S4 flow cell. Matched specimens from the same patient were processed in parallel during library preparation and sequenced on the same flow cell to limit batch effects.

Collection of human liquid and tissue specimens for this study was approved by the Mount Sinai Institutional Review Board under the Tisch Cancer Institute of the Icahn School of Mount Sinai’s GU Cancer Biorepository (IRB#: 10-1180). All patients provided written informed consent before enrollment in the study.

Sequencing data was aligned and quantified by applying the Cell Ranger Single-Cell Software Suite (version 3.0, 10× Genomics) using the GRCh38 human reference genome. This generated a raw unique molecular identifier (UMI) count matrix, which was then made into a Seurat object using the R package Seurat version 4.0.6 (https://github.com/satijalab/seurat/;Butler et al., 2018). For quality control, genes expressed in less than five cells were discarded. Cells expressing more than 2,500 and fewer than 200 unique genes, and cells expressing 15% or more of mitochondrial content were removed. The UMI count matrix was log normalized. With the FindIntegrationAnchors function of Seurat, the top 2,000 variable genes were used to create anchors. The IntegrateData function was used to integrate the data and create a new matrix consisting of 2,000 features. Cell clusters were identified using the Seurat FindClusters function with a resolution of 0.5, unless otherwise specified, and were visualized with two-dimensional Uniform Manifold Approximation and Projection (UMAP) plots. The Seurat FindAllMarkers function was implemented to identify the most highly expressed genes in the clusters for annotating their cell identities. Log₂FC > 0.5 and adj.p.val <0.05 were used as cutoffs.

For differential gene expression, we utilized the Seurat FindAllMarkers function, applying the Wilcox Rank Sum test. Volcano plots were generated using the EnhancedVolcano R package, and cutoffs of log₂FC > |1| and P < 0.05 were used for display and GSEA. GSEA was performed using Ingenuity Pathway Analysis, and the top 5 or 10 pathways were displayed, excluding pathways involving irrelevant diseases or cell types.

Cells were stained with LIVE/DEAD fixable blue stain (L34957; Thermo Fisher Scientific) on ice for 20 min in PBS and washed. Fc receptors were blocked for 10 min on ice in FACS buffer (2% BSA, 1 mM EDTA in PBS). Staining for surface markers was conducted in FACS buffer on ice for 25 min before washing. Antibodies for surface staining used included: anti-human CD3 in PE-CF594 (Clone: UCHT1; Catalog #: 562280; BD), CD4 in PE (VIT4; 130-098-167; Miltenyi), CD8a in FITC (RPS-T8; 301006; Biolegend), CD14 in Alexa Fluor 700 (M5E2; 557923; BD), CD16 in APC-Cy7 (3G8; 302018; Biolegend), CD19 in APC (HIB19; 302212; Biolegend), CD45 in BV510 (HI30; 563204; BD), CD56 in BV605 (HCD56; 318334; Biolegend), CD66b in PerCPCy5.5 (QA17A51; 396914; Biolegend), and CD163 in BV421 (GHI/61; 333612; Biolegend). Staining with isotype controls was performed for gating, and antibodies used included APC mIgG1 (MOPC-21; 400129; Biolegend); BV421 mIgG1, κ chain (MOPC-21; 400158; Biolegend); BV605 mIgG1, κ chain (MOPC-21; 400161; Biolegend); and PerCPCy5.5 mIgG1, κ chain (MOPC-21; 400149; Biolegend). Cells were assessed using an LSR Fortessa cytometer and the FACSDiva v8.0.2 software. Data were analyzed using FlowJo v10.5.3 (BD).

To evaluate urine cytokines across disease stages, we used the Olink proteomics Immuno-Oncology panel. This encompasses 92 paired oligonucleotide antibody-labeled probes targeting proteins related to biological processes such as evasion and promotion of tumor immunity, chemotaxis, tissue remodeling, cell killing, and metabolism, among others. Each biomarker is assessed using a matched pair of antibodies coupled to unique and partially complementary oligonucleotides and measured using RT-qPCR. In a 96-well plate, 1 μl of patient plasma was mixed with 3 μl of an Olink incubation mix and incubated overnight at 4°C. The next day, Olink extension reagent mix containing PCR polymerase was added to each well and placed into a thermal cycler for preamplification (1.5 h). In the detection phase, 2.8 μl from each well was mixed with 7.2 μl of a detection mix and placed on a 96-96 Dynamic Array Integrated Fluidic Circuit along with the 92 oligonucleotide pairs. The chip was processed using the Fluidigm BioMark qPCR reader. Samples were run in singlet with blanks and interplate batch controls. More information regarding assay protocols and validations may be obtained from the supplier (https://www.olink.com). Sample data quality control and normalization were done according to supplier guidelines. Data was assessed using the RT-qPCR analysis software using the ΔΔCt method and normalized protein expression manager. Data was normalized using intraplate controls for urine samples, interplate controls, and negative controls, and expressed on a log₂ scale (normalized protein expression; NPX), which is proportional to protein concentrations. One NPX difference is equivalent to the doubling of protein expression. Unbiased hierarchical clustering was performed using the ComplexHeatmap package in R (https://github.com/jokergoo/ComplexHeatmap; Gu et al., 2016).

Statistical analyses and graph generation were performed in R (version 3.6.0). Non-parametric Wilcoxon signed-rank tests were calculated as indicated in each figure by Prism (GraphPad) software Version 9.0 (ns: P > 0.05, *: P < 0.05, **: P < 0.01, ***: P < 0.001, ****: P < 0.0001). No statistical tests were performed to predetermine the sample size.

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Dr. Nina Bhardwaj ([email protected]).

This study did not generate new unique reagents.

Fig. S1 shows the top DEGs for each cluster in the UDC scRNAseq and flow cytometry of immune UDCS. Fig S2 includes heatmaps of genes used for cluster annotation of the scRNAseq of integrated UDC, tumor, and PBMC as well as comparisons of cell composition across tissue. Fig. S3 includes heatmaps of genes used for annotation of CD4⁺ and CD8⁺ T cell subclusters and comparisons of their transcriptomic profiles across tissue. Fig. S4 contains subcluster annotation information for monocyte/macrophage, B cells, and DCs as well as transcriptomic comparisons across tissue for these cells plus NK cells. Fig. S5 shows the top DEGs for UDCs from the BCG-treated patient, comparisons between Week 1 pre-BCG and Week 6 post-BCG UDCs, comparisons of BC Patient 7’s tumor and urine urothelial cells before and after chemotherapy, and an overview of UDC composition and multiple urine-derived proteins across HD, NMIBC, and MIBC. Table S1 includes a summary of the scRNAseq sequencing depth for each sample. Table S2 lists gene signatures used for analysis and annotation. Table S3 contains a summary of commercially available urine-based clinical tests for BC.

scRNAseq data have been deposited in GEO (Accession #: GSE267718) and are publicly available.

We are immensely grateful to the BC patients who volunteered to participate in these studies, the Mount Sinai GU medical oncology and urology providers who screened, enrolled, and cared for these patients, the Mount Sinai Biorepository and Pathology Core for help with tissue accrual, and the Mount Sinai Genomics Core for scRNAseq processing. Thank you to the Mount Sinai Human Immune Monitoring Core for help with our Olink assessment of urinary cytokines. Thank you, Howard Chen, for help with analysis of our Olink results. Thank you to Drs. Jeremiah Faith, Thomas Marron, William Oh, and Sumit Subudhi for their advice on this project while serving as thesis committee members for M.A. Tran. A special thank you to all members of the Bhardwaj Lab for their support and discussions.

This work was supported by the National Institutes of Health (NIH) 1R01CA249175 and Department of Defense TTSA CA181008. M.A. Tran was supported by an NIH pre-doctoral fellowship, NIH F30 CA275269, and a Medical Scientist Training Program Training Grant: T32GM007280. N. Bhardwaj was supported by a Cancer Research Institute/Oliver R. Grace Clinical and Laboratory Integration Program grant (CRI4377).

Author contributions: M.A. Tran: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Software, Validation, Visualization, Writing—original draft, Writing—review & editing, D. Youssef: Data curation, Investigation, S. Shroff: Data curation, Investigation, D. Chowhan: Data curation, Investigation, K.G. Beaumont: Investigation, Methodology, Project administration, Supervision, Writing—review & editing, R. Sebra: Investigation, Methodology, Resources, Writing—review & editing, R. Mehrazin: Resources, Writing—review & editing, P. Wiklund: Investigation, Resources, J.J. Lin: Resources, Writing—review & editing, A. Horowitz: Resources, Writing—review & editing, A.M. Farkas: Conceptualization, Funding acquisition, Methodology, Supervision, Writing—review & editing, M.D. Galsky: Funding acquisition, Investigation, Writing—review & editing, J.P. Sfakianos: Funding acquisition, Resources, Supervision, Writing—review & editing, N. Bhardwaj: Conceptualization, Funding acquisition, Project administration, Resources, Supervision, Writing—review & editing.