Abstract

   Although immune checkpoint blockade (ICB) therapies have been approved
   for bladder cancer (BLCA), only a minority of patients respond to these
   therapies, and there is an urgent need to explore combined therapies.
   Systematic multi‐omics analysis identified S100A5 as a novel
   immunosuppressive target for BLCA. The expression of S100A5 in
   malignant cells inhibited CD8^+ T cell recruitment by decreasing
   pro‐inflammatory chemokine secretion. Furthermore, S100A5 attenuated
   effector T cell killing of cancer cells by inhibiting CD8^+ T cell
   proliferation and cytotoxicity. In addition, S100A5 acted as an
   oncogene, thereby promoting tumor proliferation and invasion. Targeting
   S100A5 synergized with the efficacy of anti‐PD‐1 treatment by enhancing
   infiltration and cytotoxicity of CD8^+ T cells in vivo. Clinically,
   there was a spatially exclusive relationship between S100A5^+ tumor
   cells and CD8^+ T cells in tissue microarrays. Moreover, S100A5
   negatively correlated with immunotherapy efficacy in our real‐world and
   several public immunotherapy cohorts. In summary, S100A5 shapes a
   non‐inflamed tumor microenvironment in BLCA by inhibiting the secretion
   of pro‐inflammatory chemokines and the recruitment and cytotoxicity of
   CD8^+ T cells. Targeting S100A5 converts cold tumors into hot tumors,
   thus enhancing the efficacy of ICB therapy in BLCA.

   Keywords: bladder carcinoma, effector immune cells, immunotherapy, S100
   family, tumor microenvironment
     __________________________________________________________________

   Systematic multi‐omics analysis identifies S100A5 as a novel
   immunosuppressive target for bladder cancer, which shapes a
   non‐inflamed tumor microenvironment in bladder cancer by inhibiting the
   secretion of pro‐inflammatory chemokines and the recruitment and
   cytotoxicity of CD8^+ T cells. Targeting S100A5 converts cold tumors
   into hot tumors, thus enhancing the efficacy of immunotherapy in
   bladder cancer.

   graphic file with name ADVS-10-2300110-g006.jpg

1. Introduction

   Bladder carcinoma (BLCA) is one of the most prevalent carcinomas
   worldwide, with 573 278 new cases and 212 536 new deaths reported in
   2020.^[ [56]^1 ^] Despite radical cystectomy and platinum‐based
   chemotherapy, advanced BLCA, including locally advanced and metastatic
   BLCA, is commonly regarded as an incurable disease with extremely poor
   prognosis.^[ [57]^2 , [58]^3 ^] Owing to their high tumor mutation
   burden (TMB), immune checkpoint inhibitors (ICIs) have gained
   increasing attention for the treatment of advanced BLCA.^[ [59]^4 ,
   [60]^5 ^] Since 2016, five ICIs, including atezolizumab, durvalumab,
   avelumab, pembrolizumab, and nivolumab have been approved by the US
   Food and Drug Administration (FDA) for the treatment of advanced
   BLCA.^[ [61]^2 ^] However, only a minority of patients respond to ICIs
   treatment, suggesting an urgent need to identify novel biomarkers that
   can not only accurately predict ICIs response but also have the
   potential to be promising immunotherapy targets.^[ [62]^5 , [63]^6 ^]

   Numerous factors affect ICI efficiency. Immune profiles of the tumor
   microenvironment (TME) are vital elements that determine how a response
   occurs.^[ [64]^7 ^] The TME is composed of tumor cells, fibroblasts,
   vascular endothelial cells, immune cells, extracellular matrix, and
   extracellular soluble molecules.^[ [65]^8 ^] Based on the presence or
   absence of T cells in the tumor parenchyma, the TME can generally be
   divided into two profiles: non‐inflamed tumors (without T cell
   infiltration into the tumor parenchyma, including immune‐desert and
   immune‐excluded phenotypes) and inflamed tumors (with T cell
   infiltration into the tumor parenchyma, including immune‐inflamed
   phenotype).^[ [66]^9 ^] Theoretically, ICIs, including anti‐programmed
   death 1 (PD‐1) and anti‐programmed death‐ligand 1 (PD‐L1), exhibit no
   therapeutic effects without T cell immunity. Consistent with this,
   inflamed tumors have higher response rates to immunotherapy, including
   ICIs.^[ [67]^10 ^] In addition, the combination of both anti‐PD‐1 and
   cytotoxic T lymphocyte antigen 4 (CTLA‐4) has shown much higher
   effectiveness because CTLA‐4 increases the production of tumor‐specific
   T cells by moving the checkpoint for T cell proliferation and
   priming.^[ [68]^11 ^] Inspired by this, we are committed to identifying
   a biomarker that can not only predict the infiltration of
   tumor‐infiltrating immune cells (TIICs) for diagnosis, but also develop
   new combined anti‐PD‐1/PD‐L1 treatment strategies for higher
   therapeutic efficacy.

   Members of the S100 protein family play vital roles in tumor invasion
   and immune evasion by acting as Ca^2+ mediators and extracellular
   factors.^[ [69]^12 ^] In breast cancer, tumor cells secrete S100A7 to
   recruit tumor‐associated macrophages (TAMs).^[ [70]^13 ^] In addition,
   S100A8 and S100A9 can recruit myeloid‐derived suppressor cells (MDSCs)
   to maintain an immunosuppressive state in the TME of breast cancer.^[
   [71]^12 ^] In melanoma, S100A4 can increase the release of inflammatory
   cytokines and promote tumor immune responses.^[ [72]^14 ^] S100A5 has
   been reported to play a vital role in the recurrence of the World
   Health Organization (WHO) grade I meningiomas.^[ [73]^15 ^] In
   addition, S100A5 was significantly upregulated in the BLCA.^[ [74]^16
   ^] However, their comprehensive roles in TME and immunotherapy are
   unclear, especially in BLCA. Therefore, in this study, we focused on
   S100A5 by comprehensively analyzing the expression patterns and
   immunological roles of multiple S100 family proteins in BLCA.
   Pan‐cancer analyses revealed that the immunosuppressive role of S100A5
   in the TME was most evident in BLCA. Based on these findings, we
   validated the oncogenic and immunosuppressive roles of S100A5 in BCLA
   both in vitro and in vivo. Moreover, the value of predicting
   immunotherapy response was revealed in our real‐world and multiple
   immunotherapy cohorts.

2. Results

2.1. Focusing S100A5 in BLCA through Comprehensive Analyses

   S100 family proteins including S100A1‐16, S100B, and S100P were
   summarized based on the review of Bresnick et al.^[ [75]^12 ^]
   According to the principles of normalization cancer immunotherapy,^[
   [76]^49 ^] molecular targets should meet two conditions for potential
   immunotherapy: specific expression in carcinoma cells and inhibition of
   TIICs infiltration. Therefore, we first analyzed the expression
   patterns of these proteins in the TCGA‐BLCA and Xiangya BLCA cohorts
   and found that S100A5, S100A7, S100A11, S100A14‐16, and S100B were
   significantly higher in carcinoma tissues than normal tissues in both
   cohorts (Figure [77]S1A,B, Supporting Information). Then, 28 immune
   cells were identified using the single‐sample gene set enrichment
   analysis (ssGSEA) algorithm and systematically correlated with S100
   proteins. As shown in Figure [78]S1C,D (Supporting Information), only
   S100A5 and S100A6 were significantly negatively correlated with TIICs
   infiltration in both cohorts. Combined with the expression patterns, we
   focused on S100A5 in this study.

   We then performed pan‐cancer analyses of S100A5 in 33 types of
   carcinomas. In most carcinomas, S100A5 was expressed at significantly
   higher levels in tumor tissues than in normal tissues, including
   bladder cancer tissues (Figure [79]S2A,B, Supporting Information). In
   addition, we found that S100A5 was significantly higher in the tumor
   tissues in our TMA (Figure [80]S2C, Supporting Information). S100A5 was
   also expressed in several normal tissues (Figure [81]S2D, Supporting
   Information). Next, we explored the immunological role of S100A5 in
   multiple carcinomas. As shown in Figure [82]S3A (Supporting
   Information), the negative correlation between S100A5 and
   immunomodulators was most evident in BLCA. We selected four important
   immune checkpoints, including PD‐L1, CTLA‐4, PD‐1, and lymphocyte
   activation gene‐3 (LAG‐3), and found that S100A5 had the most exclusive
   association with these immune checkpoints in BLCA (Figure [83]S3B–E,
   Supporting Information). Furthermore, S100A5 was significantly
   negatively correlated with most TIICs in the BLCA (Figure [84]S3F,
   Supporting Information). However, there was no significant negative
   correlation between S100A5 and TIICs in other carcinomas, such as lung
   adenocarcinoma (LUAD), sarcoma (SARC), breast invasive carcinoma
   (BRCA), and colon adenocarcinoma (COAD) (Figure [85]S3F, Supporting
   Information). In summary, we found that only S100A5 shaped a
   non‐inflamed TME specifically in BLCA, by correlating multiple S100
   proteins with multiple cancer types.

2.2. S100A5 Shapes a Non‐Inflamed TME in BLCA

   The cancer‐immune cycle includes the following steps: first, the death
   of cancer cells releases cancer antigens; then, antigen‐presenting
   cells (APCs), such as dendritic cells (DCs), capture and present
   antigens to T cells, resulting in the activation of effector T cells.
   Effector T cells traffic to the tumor sites and infiltrate the tumor
   bed. Finally, effector T cells recognize and kill tumor cells and
   release more antigens.^[ [86]^50 , [87]^51 ^] Based on the bulk RNA‐seq
   data of TCGA‐BLCA, we downloaded the levels of these steps from the
   tracking tumor immunophenotype (TIP)
   ([88]http://biocc.hrbmu.edu.cn/TIP/). As shown in Figure [89]1A;
   Table [90]S1 (Supporting Information), most of these steps were
   significantly negatively correlated with S100A5, indicating that S100A5
   may inhibit the cancer‐immune cycle and shape a non‐inflamed TME in
   BLCA. Furthermore, we directly evaluated the infiltration of 28 immune
   cell types using the ssGSEA algorithm (Figure [91]1B; Table [92]S1,
   Supporting Information) and found significant negative correlations
   between S100A5 expression and most immune cells, including activated
   CD8^+ T cells, type 1 T helper (Th1) cells, natural killer (NK) cells,
   activated CD4^+ T cells, activated DCs, and NK T cells. To eliminate
   the influence of this algorithm, six other algorithms (CIBERSORT, EPIC,
   mMCP‐counter, quanTIseq, TIMER, and xCell) were used to calculate
   immune cell infiltration. As expected, S100A5 expression was
   significantly and negatively correlated with most cytotoxic T
   lymphocytes (CTLs) and NK cells, regardless of the algorithm
   (Table [93]S1, Supporting Information). Importantly, the negative
   correlation between S100A5, the cancer‐immune cycle, and TIICs was
   validated in the Xiangya cohort (Figure [94]1C; Table [95]S2,
   Supporting Information), indicating a robust predictive value of S100A5
   for TIICs in BLCA. In addition to TIICs, S100A5 expression negatively
   correlated with the effector genes of CD8^+ T cells, DCs, macrophages,
   NK cells, and Th1 cells (Figure [96]1D; Table [97]S3, Supporting
   Information). Next, we calculated the pan‐cancer T cell‐inflamed score
   (TIS) and found that the S100A5 was significantly negatively correlated
   (R = −0.29, p < 0.001; Figure [98]1E). Furthermore, we summarized the
   TIS genes and found that S100A5 was significantly negatively correlated
   with almost all TIS genes (Figure [99]1F, bottom left; Table [100]S3,
   Supporting Information). Finally, we found that S100A5 expression was
   significantly negatively correlated with most ICI genes, such as PD‐1,
   PD‐L1, LAG‐3, and CTLA‐4, which were reported to be lower in the
   non‐inflamed TME (Figure [101]1F, upper right; Table [102]S3,
   Supporting Information).^[ [103]^52 ^] Numerous public databases were
   used to validate the results. As showed in Figure [104]2A, S100A5 was
   significantly negatively correlated with immune cell infiltration in
   [105]GSE87304, [106]GSE48276, and [107]GSE48075. Similar results were
   found for [108]GSE120736, [109]GSE31684, [110]GSE32894, [111]GSE69795,
   and E‐MTAB‐1803 (Figure [112]S4–S8, Supporting Information). In
   summary, using bulk RNA‐seq data, we found that S100A5 shaped a
   non‐inflamed TME in BLCA.

Figure 1.

   Figure 1
   [113]Open in a new tab

   Correlations between S100A5 and tumor microenvironment (TME)
   characteristics in TCGA‐BLCA and Xiangya cohort. A) Correlation between
   S100A5 and cancer‐immunity cycles in TCGA‐BLCA cohort. Different colors
   represent different cycles; Positive correlation is marked in red,
   while negative correlation is marked in blue. B) Correlation between
   S100A5 and tumor‐infiltrating immune cells (TIICs) using ssGSEA
   algorithm in TCGA‐BLCA cohort. C) Validation the role of S100A5 with
   cancer‐immunity cycles and TIICs using ssGSEA algorithm in Xiangya
   cohort. Yellow represents higher infiltration cells, while blue
   represents lower infiltration cells. D) Different expression patterns
   of effector genes of CD8^+ T cells, dendritic cells (DCs), macrophage
   cells, natural killer (NK) cells and type 1 T helper (Th1) cells
   between high and low S100A5 groups; Red represents higher expressed
   genes, while blue represents lower expressed genes. E) Correlation
   between S100A5 and pan‐cancer T cell‐inflamed score (TIS). F)
   Correlation between S100A5 and TIS related genes (bottom left) and
   immune checkpoint inhibitor (ICI) genes (upper right); Positive
   correlation was marked in red, while negative correlation was marked in
   blue. ns, not statistically significant. *p < 0.05; **p < 0.01; ***p <
   0.001.

Figure 2.

   Figure 2
   [114]Open in a new tab

   S100A5 specifically expressed on tumor cells and inhibited
   pro‐inflammatory cytokine and chemokines secretion. A) Correlation
   between S100A5 and tumor‐infiltrating immune cells (TIICs) using the
   ssGSEA algorithm in the [115]GSE87304, [116]GSE48276, and [117]GSE48075
   cohorts. B) Venn Diagram showing common differentially expressed genes
   (DEGs) between the high S100A5, stromal score and immune score groups.
   C) KEGG enrichment results for common DEGs. D) tSNE plot of all single
   cells and S100A5 expression patterns in Xiangya scRNA‐seq. E,F) GSEA
   shows the enrichment of cytokine and chemokine secretion related
   pathways E) and T cell infiltration‐related pathways F) between
   different S100A5 expression groups in malignant epithelial cells in the
   Xiangya scRNA‐seq cohort. NES: normalized enrichment score. G) S100A5
   expression levels in the TME ([118]GSE130001) and blood cells
   ([119]GSE145281). H) tSNE plot of all single cells and S100A5
   expression patterns in the PRJNA662018 scRNA cohort. iCAF, inflammatory
   cancer‐associated fibroblasts; mCAF, myo‐cancer‐associated fibroblasts.
   I,J) GSEA shows GO I) and KEGG J) enrichment of cytokine and chemokine
   secretion‐related pathways between different S100A5 expression groups
   in malignant epithelial cells in the PRJNA662018 scRNA cohort. K)
   Volcano plot showing the major DEGs between the knockdown and
   overexpression S100A5 groups in T24 cell. ns, not statistically
   significant. *p < 0.05; **p < 0.01; ***p < 0.001.

2.3. Validating the Role of S100A5 in TME in Terms of Molecular Subtypes

   Bladder cancer can be divided into molecular subtypes based on
   transcriptome profiling, which can precisely classify patients
   according to their prognosis and therapeutic options.^[ [120]^35 ,
   [121]^40 , [122]^42 ^] We found that the low S100A5 group patients were
   more likely to be the basal subtype, while those belonging to the high
   S1005A group could be the luminal subtype based on seven molecular
   classification systems in both TCGA‐BLCA and Xiangya cohorts
   (Figure [123]S9A and Table [124]S4, Supporting Information). In
   addition, the area under the ROC curves (AUCs) were more than 0.85 in
   the TCGA‐BLCA cohort (Figure [125]S9B and Table [126]S4, Supporting
   Information) and reached 0.90 in the Xiangya cohort (Figure [127]S9C
   and Table [128]S4, Supporting Information), indicating a robust high
   predictive value of S100A5 for molecular subtypes. Urothelial, Ta, and
   luminal differentiation pathways were enriched in the high S100A5
   group, whereas basal differentiation pathways were enriched in the low
   S100A5 group (Figure [129]S9A and Table [130]S4, Supporting
   Information). It is generally considered that the basal subtypes (low
   S100A5 group) possess more cytotoxic lymphocytes and NK cell
   infiltration and could be more sensitive to immunotherapies.^[ [131]^35
   ^] In addition, the interferon response and immune differentiation
   pathways were enriched in the low S100A5 group (Figure [132]S9A and
   Table [133]S4, Supporting Information). We successfully validated these
   results using [134]GSE48075 and [135]GSE48276 cohorts
   (Figure [136]S9D–F, Supporting Information). In summary, we confirmed
   that S100A5 expression was negatively correlated with immune cell
   infiltration and response to immunotherapies in terms of BLCA molecular
   subtypes.

2.4. Combining Bulk RNA‐seq and scRNA‐seq Revealed that S100A5 could Inhibit
Pro‐Inflammatory Cytokine and Chemokines Secretion

   Dividing the TCGA‐BLCA cohort into two groups based on the median
   expression of S100A5, we filtered 1803 differentially expressed genes
   (DEGs) between the high and low S100A5 groups (|log2FC| >1 and
   adj.P.value <0.05) (Table [137]S5, Supporting Information).
   Surprisingly, there was only one common gene between the low S100A5,
   stromal score, and immune score groups (Figure [138]S10A and
   Table [139]S5, Supporting Information). Moreover, there were no common
   genes among the high S100A5, stromal score, and immune score groups
   (Figure [140]2B; Table [141]S5, Supporting Information). These results
   confirmed the exclusive role of S100A5 with stromal and immune scores
   in terms of DEGs. We identified 967 common genes (Figure [142]S10B and
   Table [143]S5, Supporting Information) and performed Gene Ontology (GO)
   and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses using these
   DEGs. The majority of KEGG pathways were enriched in cytokine‐ or
   chemokine‐related pathways (Figure [144]2C; Table [145]S6, Supporting
   Information). Also, GO pathways were enriched in immune‐related
   pathways, such as T cell activation, response to interferon‐γ, immune
   receptor activity, cytokine and chemokine activity (Figure [146]S10C–E
   and Table [147]S6, Supporting Information).

   These analyses were conducted using bulk RNA‐seq, the fundamental
   principle of which is that all genes are expressed equally in every
   cell. It is difficult to investigate TME heterogeneity at the
   single‐cell level using bulk RNA‐seq. So, we collected three muscle
   invasive bladder cancer (MIBC) samples from our hospital and performed
   scRNA‐seq to overcome this limitation. A total of 19 852 cells were
   grouped into six major clusters after quality control and integration
   (Figure [148]2D, left). Clusters were annotated based on the
   well‐established marker genes as previous reported: epithelial cells
   (EPCAM), myeloid cells (LYZ), T cells (CD3D), fibroblasts (COL1A1), B
   cells (CD79A and CD19), and endothelial cells (PECAM1 and VWF)^[
   [149]^27 , [150]^28 ^] (Figure [151]S11A–H, Supporting Information).
   The UMIs and genes detected in the epithelial and immune cells are
   shown in Figure [152]S11I,J (Supporting Information). As shown in
   Figure [153]S12 (Supporting Information), CNVs accumulated mainly in
   EPCAM^+ epithelial cells, confirming the malignant origin of the marked
   epithelial cells. According to an article reported by Peng et al.,^[
   [154]^32 ^] we calculated the CNV scores of every epithelial cell and
   defined the epithelial cells with a CNV score of more than 0.02 as
   malignant epithelial cells, for further analysis. Interestingly, S100A5
   was specifically expressed in bladder epithelial cells (Figure [155]2D,
   right). We focused on malignant epithelial cells and divided them into
   high S100A5 and low S100A5 expression groups. As shown in
   Figure [156]2E, GO pathways related to cytokine and chemokine secretion
   were downregulated in the high S100A5 expression group (Table [157]S7,
   Supporting Information). Furthermore, the high S100A5 expression group
   showed significantly downregulated T cell proliferation and activation
   pathways (Figure [158]2F; Table [159]S7, Supporting Information).
   S100A5 expression was the highest in epithelial cells, whereas no
   expression was observed in immune cells (such as CD4^+ T, CD8^+ T and
   NK cells) using another scRNA‐seq database containing two bladder
   cancer specimens (Figure [160]2G, [161]GSE130001). We also analyzed
   S100A5 expression in peripheral blood cells (including CD4^+ T, CD8^+
   T, NK cells, and et al.) and found that S100A5 was barely expressed in
   these cells (Figure [162]2G, [163]GSE145281).

   These results were further validated using public scRNA‐seq databases.
   Chen et al. reported the first and largest scRNA‐seq map of BLCA
   (PRJNA662018)^[ [164]^27 ^] and we clustered 59 045 cells from tumor
   tissues into eight major clusters after quality control and integration
   using their cohort (Figure [165]2H, left; Figure [166]S13, Supporting
   Information). As expected, S100A5 was specifically expressed in the
   epithelial cells (Figure [167]2H, right). We then separated the
   malignant epithelial cells following the same role in Xiangya scRNA and
   found that malignant epithelial cells with high S100A5 expression
   significantly inhibited chemokine‐ and cytokine‐related pathways in
   both GO (Figure [168]2I; Table [169]S8, Supporting Information) and
   KEGG (Figure [170]2J and Table [171]S8, Supporting Information)
   analyses. In another scRNA‐seq cohort ([172]GSE135337), we clustered
   36 534 cells into five major clusters (Figure [173]S14 and [174]S15A,
   Supporting Information). S100A5 was specifically expressed on
   epithelial cells (Figure [175]S15B, Supporting Information) and
   significantly negatively correlated with cytokine, chemokine and T cell
   infiltration related pathways (Figure [176]S15C–E and Table [177]S8,
   Supporting Information). Similar results were obtained for
   [178]GSE145137 (Figure [179]S15F–I and Table [180]S8, Supporting
   Information).

2.5. S100A5 Inhibited CD8^+ T Cells Recruitment through Decreasing
Pro‐Inflammatory Chemokines Secretion

   These results are mainly based on large and systematic bioinformatics
   analyses. We then validated these results in vitro and in vivo. S100A5
   short hairpin RNA (sh‐S100A5) and S100A5‐cDNA (oe‐S100A5) and their
   negative controls (sh‐NC and oe‐vector respectively) were successfully
   transfected on human bladder cancer cell lines (T24 and 5637)
   (Figure [181]S16A–D, Supporting Information). RNA‐Seq was performed on
   three parallel sh‐S100A5 and oe‐S100A5 cell lines. As expected, S100A5
   was significantly downregulated in the sh‐S100A5 group compared to the
   oe‐S100A5 group (Figure [182]2K, T24; Figure [183]S16E, 5637,
   Supporting Information). In contrast, several important chemokines,
   including CCL2, CCL5, CXCL10, and CXCL11 were significantly upregulated
   when S100A5 was downregulated (Figure [184]2K; Table [185]S9,
   Supporting Information). Among these chemokines, CXCL10 and CXCL11 were
   reported to play vital roles in the process of recruiting immune
   cells.^[ [186]^53 , [187]^54 ^] Moreover, pathways related to
   cytokine/chemokine secretion/interaction (Figure [188]S16F,G and
   Table [189]S10, Supporting Information) and immune cell infiltration
   (Figure [190]S16H and Table [191]S10, Supporting Information) were
   significantly upregulated in the sh‐S100A5 cell line compared to those
   in the oe‐S100A5 cell line. Similar results were also found in the 5637
   cell line (Figure [192]S17A–C and Table [193]S9 and [194]S10,
   Supporting Information).

   To determine the key downstream cytokines/chemokines of S100A5, we
   applied ProcartaPlex multiple immunoassays to detect multiple cytokines
   and chemokines in cell culture supernatants from sh‐S100A5 and
   oe‐S100A5 cell lines. As shown in Figure [195]3A (T24), the secretions
   of CCL2, CCL3, CCL4, CCL5, and CXCL10 were obviously increased when
   knocking down S100A5. Moreover, the secretion protein levels of these
   chemokines and another vital chemokines (CXCL9)^[ [196]^53 ^] were
   validated using ELISA. We found that all these chemokines were secreted
   at significantly higher levels in the sh‐S100A5 cell line
   (Figure [197]3B, T24). Furthermore, RT‐PCR results revealed that the
   mRNA expression levels of these chemokines were significantly higher
   when S100A5 was knocked down (Figure [198]3C, T24). In contrast, when
   S100A5 was overexpressed, the secreted protein levels of these
   chemokines were significantly decreased according to ProcartaPlex
   multiple immunoassays (Figure [199]S17D,E, T24, Supporting Information)
   and ELISA (Figure [200]S17F, T24, Supporting Information). mRNA
   expression was also significantly decreased, as shown by RT‐PCR
   (Figure [201]S17G, T24, Supporting Information). Similar results were
   observed in 5637 cells (Figure [202]S17H–K, Supporting Information). We
   conducted a CD8^+ T cell migration assay to determine whether S100A5
   inhibits CD8^+ T cell infiltration by downregulating these chemokines
   (Figure [203]3D). We found that the knockdown and overexpression of
   S100A5 significantly increased and then decreased (Figure [204]3E, T24)
   CD8^+ T cell migration, respectively, compared to their negative
   controls. Similar results were also found in the 5637 cells
   (Figure [205]3F, 5637). In summary, we found that S100A5 inhibited
   CD8^+ T cell recruitment by decreasing chemokines secretion.

Figure 3.

   Figure 3
   [206]Open in a new tab

   S100A5 inhibited CD8^+ T cell recruitment and cancer cell killing
   process by regulating CD8^+ T cell proliferation and cytotoxicity. A)
   Heatmap of multiple cytokines and chemokines detected by ProcartaPlex
   multiple immunoassays between S100A5 knockdown and the negative control
   groups in T24 cell culture supernatants. Red indicates higher secretion
   and blue indicates lower secretion. B,C) CCL2, CCL3, CCL4, CCL5, CXCL9,
   and CXCL10 levels detected using ELISA B) and RT‐PCR C), respectively,
   between S100A5 knockdown and negative control groups in the T24 cell
   line. D) Diagram of the CD8^+ T cell migration assay. E,F) Relative
   migration of CD8^+ T cells among S100A5 knockdown, S100A5
   overexpression, and their negative controls in T24 cell line culture
   supernatants E) and 5637 cell line culture supernatants F). G)
   Representative images and histogram plots of the T cell‐mediated tumor
   cell‐killing assay between the S100A5 knockdown and negative control
   groups in the T24 cell line. The OD values were normalized to the mean
   value in the sh‐NC group without T cell co‐culture. H) Representative
   images and histogram plots of the T cell‐mediated tumor cell‐killing
   assay between S100A5 overexpression and negative control groups in the
   T24 cell line. The OD values were normalized to the mean value in the
   oe‐vector group without T cell co‐culture. I,J) Flow cytometry analysis
   shows different Ki‐67 I) and perforin J) expression on effector T cells
   after co‐cultured with S100A5 knockdown and overexpression of T24 cell
   lines and their negative control cell lines. ns, not statistically
   significant. *p < 0.05; **p < 0.01; ***p < 0.001.

2.6. S100A5 Attenuated Effector T Cells Killing Cancer Cells through
Inhibiting CD8^+ T Cells Proliferation and Cytotoxicity

   In addition to recruitment, we explored whether S100A5 expression could
   inhibit T cell function. First, we subdivided T cells into CD4^+ and
   CD8^+ T cells based on classical markers (including CD3D, CD3E, CD3G,
   CD4, CD8A, and CD8B) in Xiangya scRNA‐seq (Figure [207]S18A–C,
   Supporting Information). Long et al. divided CD8^+ T cells into
   progenitor exhausted T cells and terminally exhausted T cells, and
   revealed that patients with more progenitor exhausted T cells had a
   higher tumor‐killing efficiency and anti‐PD1 response rate.^[ [208]^55
   , [209]^56 ^] We found that T cell cluster 1 had significantly lower
   progenitor exhaustion scores and higher exhaustion scores than cluster
   3 (Figure [210]S18D, Supporting Information); therefore, we defined
   cluster 1 as composed of terminally exhausted CD8^+ T cells and cluster
   3 as composed of progenitor exhausted CD8^+ T cells (Figure [211]S18E,
   Supporting Information). Then “CellChat” package was used to infer
   interactions and communication probability between high/low S100A5
   expression epithelial cells and T cell subgroups (Figure [212]S18F,
   Supporting Information). Surprisingly, epithelial cells with high
   S100A5 expression had a larger number of interactions and interaction
   weights with T cell subgroups than epithelial cells with low S100A5
   expression (Figure [213]S18G, Supporting Information), indicating that
   epithelial cells with high S100A5 expression had more communication
   with T cells and might inhibit their functions. Similar results were
   obtained for PRJNA662018 scRNA‐seq (Figure [214]S19, Supporting
   Information).

   We then performed a T cell‐mediated tumor cell‐killing assay to
   validate the results described above. In both the T24 and 5637 cell
   lines, S100A5 knockdown significantly enhanced the ability of T
   lymphocyte (activated from human peripheral blood mononuclear cells,
   PBMCs) to kill cancer cells (Figure [215]3G, T24; Figure [216]S20A,
   5637, Supporting Information), whereas S100A5 overexpression
   significantly inhibited this ability (Figure [217]3H, T24;
   Figure [218]S20B, 5637, Supporting Information). Mechanistically, we
   found that the key proliferation marker Ki‐67 (Figure [219]3I) and
   cytotoxicity marker perforin (Figure [220]3J) were significantly
   upregulated in CD8^+ T cells when lymphocytes were co‐cultured with the
   sh‐S100A5 cell line, but downregulated when lymphocytes were
   co‐cultured with the oe‐S100A5 cell line. When CD8^+ T cells were
   directly cultured with recombinant S100A5, the proliferation marker
   Ki‐67 (Figure [221]S20C, Supporting Information) and the cytotoxicity
   marker perforin (Figure [222]S20D, Supporting Information) were
   significantly downregulated compared with the control groups. In
   addition, the exhausted marker PD‐1 was significantly upregulated when
   cultured with recombinant S100A5 (Figure [223]S20E, Supporting
   Information).

   Interestingly, when cancer cell lines were not co‐cultured with T
   lymphocytes (Figure [224]3G,H, Figure [225]S20A,B, Supporting
   Information), knockdown or overexpression of S100A5 alone inhibited or
   promoted cancer cell proliferation, respectively. We speculated that
   S100A5 may also act as an oncogene and promote BLCA proliferation and
   invasion. MTT cell viability assays revealed that knockdown
   (Figure [226]S21A, Supporting Information) and overexpression
   (Figure [227]S21B, Supporting Information) of S100A5 significantly
   inhibited and promoted the proliferation of bladder cancer cells,
   respectively. In addition, the colony formation assay confirmed that
   the knockdown (Figure [228]S21C, Supporting Information) and
   overexpression (Figure [229]S21D, Supporting Information) of S100A5
   significantly inhibited and promoted the colony formation ability of
   bladder cancer cells, respectively. In addition, a wound healing assay
   revealed that S100A5 promoted bladder cancer cell invasion
   (Figure [230]S21E–G, Supporting Information). In summary, we found that
   S100A5 could act as an oncogene and inhibit cancer cell death by
   inhibiting CD8^+ T cell proliferation and cytotoxicity.

2.7. Targeting S100A5 Enhanced the Efficacy of Anti‐PD‐1 Treatment and CD8^+
T Cells Recruitment

   Considering that S100A5 could inhibit CD8^+ T cell infiltration and
   proliferation in vitro, we further explored whether it could affect the
   efficacy of anti‐PD‐1 treatment preclinically. We constructed a
   subcutaneous bladder cancer model by subcutaneously injecting S100A5 KD
   and negative control MB49 cells (Figure [231]S22A,B, Supporting
   Information). Consistent with the results for the T24 cell line, we
   found that the knockdown of S100A5 inhibited BLCA proliferation and
   invasion using MTT (Figure [232]S22C, Supporting Information), colony
   formation (Figure [233]S22D, Supporting Information), and wound healing
   assays (Figure [234]S22E, Supporting Information) on the MB49 cell
   line. The in vivo experimental procedure and treatment schedule were
   illustrated in Figure [235]4A. As shown in Figure [236]4B,C, S100A5
   knockdown or anti‐PD‐1 treatment alone significantly suppressed the
   tumor burden in vivo. However, S100A5 knockdown plus anti‐PD‐1
   treatment showed the highest efficacy in inhibiting tumor burden,
   indicating that S100A5 downregulation could enhance the efficacy of
   anti‐PD‐1 treatment. In addition, there was no difference in body
   weight between the groups of mice (Figure [237]4D). Moreover, knockdown
   of S100A5 with anti‐PD‐1 treatment showed the greatest ability to
   prolong survival in mice (Figure [238]4E).

Figure 4.

   Figure 4
   [239]Open in a new tab

   Downregulation of S100A5 enhanced the efficacy of anti‐PD‐1 treatment
   and CD8^+ T cells recruitment in vivo. A) In vivo experimental
   procedure and treatment schedule. B) Tumor images among different
   experimental groups after mice sacrifice. C) Histogram plot of tumor
   volumes among different experimental groups after mice sacrifice. D)
   Scatter diagram plot of body weights among different groups during
   experimental procedure. E) Kaplan‐Meier plot of survival percent among
   different experimental groups. F–H) Representative contour plots and
   the proportion of CD8^+ T cells in T cells F), GZMB^+ cells in CD8^+ T
   cells G), and IFN‐γ ^+ cells in CD8^+ T cells H). ns, not statistically
   significant; **p < 0.01; ***p < 0.001.

   Furthermore, the tumor tissues were digested into single‐cell
   suspensions, and flow cytometry analysis was applied to explore the
   composition of immune cells. As shown in Figure [240]S23A–D (Supporting
   Information), the proportions of leukocytes (CD45^+), lymphoid cells
   (CD45^+CD11b^−), and T cells (CD3^+) were generally the same in each
   group. Then we further explored the proportion of CD8^+ T cells and
   their cytotoxicity indicators (including GZMB, IFN‐γ, TNF‐α and
   Perforin) among each group. We found that S100A5 knockdown or anti‐PD‐1
   treatment alone significantly increased the infiltration of CD8^+ T
   cells (Figure [241]4F) and enhanced their cytotoxicity by upregulating
   GZMB (Figure [242]4G), IFN‐γ (Figure [243]4H), TNF‐α (Figure [244]5A)
   and perforin (Figure [245]5B). However, S100A5 knockdown with anti‐PD‐1
   treatment showed the highest efficacy in recruiting CD8^+ T cells and
   upregulating these cytotoxicity indicators (Figure [246]4F–H and
   Figure [247]5A,B). In addition, immunofluorescence (IF) analysis
   confirmed that both S100A5 knockdown and anti‐PD‐1 treatment promoted
   CD8^+ T cell infiltration, whereas the combination of S100A5 knockdown
   and anti‐PD‐1 treatment showed the highest efficiency
   (Figure [248]5C,D).

Figure 5.

   Figure 5
   [249]Open in a new tab

   IHC and TissueFAXS Cytometry panoramic tissue quantification assay
   depicted the spatial exclusive relationship between S100A5 and effector
   T cells in the TME. (A‐B) Representative contour plots and the
   proportion of TNF‐α ^+ cells in CD8^+ T cells A), and perforin^+ cells
   in CD8^+ T cells B). C) Representative immunofluorescence (IF) images
   of CD8 staining among different experimental groups. Scale bars, 40 µm.
   D) Percent of CD8^+ cells in all cells corresponding to IF images. E)
   Representative images of the expression patterns of CD8, PD‐L1 and
   S100A5 in non‐inflamed and inflamed TME using IHC. Scale bars, 100 µm.
   F) CD8 and PD‐L1 IHC scores between high and low S100A5 groups. G,H)
   Representative multi‐color staining of non‐inflamed G) and inflamed H)
   phenotypes of patients with bladder cancer: S100A5 (purple), CK19
   (azure), CD4 (red), CD8 (green), DAPI (blue). I) The histograms of
   different S100A5^+CD4^+, S100A5^+CD8^+, and S100A5^+CK19^+ percent
   cells among the whole TMA. J) The flow‐like cytometer plots show the
   percent of S100A5^+CK19^+, S100A5^+CD4^+, and S100A5^+CD8^+ cells
   respectively. K) The spatial distribution of CD4^+ and CD8^+ T cells
   within the distance gradients of S100A5^+CK19^+ tumor cells (0–25 µm,
   25–50 µm, 50–100 µm, and 100–150 µm). ns, not statistically
   significant. *p < 0.05; **p < 0.01; ***p < 0.001.

   The above in vivo experiment was still based on the S100A5 knockdown
   cell line, and we designed a nanomedicine carrying siS100A5 for the
   direct targeting of S100A5. Liposomes are one of the most widely used
   carriers of siRNAs because of their high loading efficiency, reliable
   drug protection, good biocompatibility, and targeted delivery.^[
   [250]^57 , [251]^58 , [252]^59 ^] We successfully synthesized liposome
   and loaded them with siS100A5. Transmission electron microscopy (TEM)
   showed the morphology of liposome@siS100A5 (Figure [253]S24A,
   Supporting Information). The average DLS of the liposome@siS100A5 was
   205.6 nm (a little bit larger than that of liposome alone) with
   polydispersity index (PDI) of 0.16 (Figure [254]S24B, Supporting
   Information). The surface charge of the liposome was 43.6 mV and turned
   to be 25.8 mV when siS100A5 (−11.6 mV) was loaded onto it
   (Figure [255]S24C, Supporting Information). Standard curve shows the
   relationship between fluorescence intensity and siS100A5 concentration
   (Figure [256]S24D,E, Supporting Information), and with loading
   efficiency of 78%.

   The anti‐tumor efficiency of liposome@siS100A5 was tested in the
   following four groups: (1) liposome + IgG2a isotype, (2)
   liposome@siS100A5 + IgG2a isotype, (3) liposome + anti‐PD‐1, (4)
   liposome@siS100A5 + anti‐PD‐1. We found that the injection of
   liposome@siS100A5 alone significantly suppressed the tumor burden, and
   this effect was synergized when combined with anti‐PD‐1 treatment
   (Figure [257]S25A,B, Supporting Information). There was no difference
   in the body weight between the groups of mice during the treatment
   procedure (Figure [258]S25C, Supporting Information). To test the
   toxicity of liposome@siS100A5 in the blood system, we detected mouse
   whole blood and blood biochemistry indices after intravenous injection
   of liposome@siS100A5 and revealed that all these blood indices were
   within the normal range (Figure [259]S25D, Supporting Information). In
   addition, H&E staining of the heart, liver, spleen, lungs, and kidneys
   revealed that our treatment caused minimal damage to the major organs
   (Figure [260]S25E, Supporting Information). We found that S100A5 was
   mainly expressed in malignant epithelial cells, whereas very low
   expression was observed in immune cells and peripheral blood cells
   using scRNA‐seq analysis. This could explain why targeting S100A5 had
   minimal adverse effects on normal tissues and blood system.^[ [261]^49
   ^] Furthermore, flow cytometry analysis (Figure [262]S25F, Supporting
   Information) revealed that targeting S100A5 could significantly promote
   CD8^+ T cell infiltration, and this effect could be synergized when
   combined with anti‐PD‐1 treatment (Figure [263]S25G, Supporting
   Information). In summary, we revealed that the downregulation of S100A5
   and targeting S100A5 could enhance the efficacy of anti‐PD‐1 treatment
   and promote CD8^+ T cells recruitment in vivo.

2.8. The Spatial Exclusive Relationship between S100A5^+ Malignant Cells and
Effector T Cells

   Tissue microarrays (TMAs) containing 50 BLCA samples receiving no prior
   treatment were prepared. We defined the samples as “inflamed phenotype”
   samples in which CD8^+ T cells were located in the tumor parenchyma and
   defined the samples as “non‐inflamed phenotype” samples in which CD8^+
   T cells were located in the stroma (not in parenchyma) or no CD8^+ T
   cells either in parenchyma or stroma. The representative image in
   Figure [264]5E shows the exclusive role of S100A5 between CD8 and PD‐L1
   at the protein level. Moreover, both the CD8 and PD‐L1 IHC scores were
   significantly higher in the low S100A5 group (Figure [265]5F). The
   TissueFAXS cytometry panoramic tissue quantification assay was further
   applied to reveal the spatially exclusive role of S100A5^+ bladder
   cancer cells and CD4^+ and CD8^+ T cells. Figure [266]5G,H shows
   representative images of multi‐color staining. In non‐inflamed tumors,
   the wide expression of S100A5 in tumor cells inhibited the infiltration
   of CD4^+ and CD8^+ T cells into the tumor region (Figure [267]5G,
   non‐inflamed). In contrast, none of S100A5 expression tumors exhibited
   an inflamed phenotype, with numerous CD4^+ and CD8^+ T cells
   infiltrating the tumor regions (Figure [268]5H, inflamed). The
   co‐expression of S100A5 on tumor cells and the exclusive role of S100A5
   between CD4^+ T and CD8^+ T cells reached statistical significance in
   the whole TMA cohort (Figure [269]5I). In addition to CD4^+ and CD8^+ T
   cells, S100A5 was barely expressed in NK (CD16^+) cells and macrophages
   (CD68^+) (Figure [270]S26A,B, Supporting Information). Moreover, the
   flow‐like cytometry plots show that S100A5 was positively expressed
   mainly in CK19^+ tumor cells (58.84%), but almost not in CD4^+ T cells
   (2.45%) or CD8^+ T cells (4.15%, Figure [271]5J). For further spatial
   analysis, we quantified the number of CD4^+ and CD8^+ T cells within
   the distance gradients of S100A5^+CK19^+ tumor cells (0‐25 µm,
   25–50 µm, 50–100 µm, and 100–150 µm). As expected, the greater the
   distance from S100A5^+CK19^+ cells, the more CD4^+ and CD8^+ T cells
   showed an increasing trend (Figure [272]5K), which confirmed the
   spatially exclusive relationship between S100A5 and effector T cells.

2.9. S100A5 and Immune Checkpoint Blockade (ICB) Response

   As S100A5 shaping a non‐inflamed TME, theoretically, lower S100A5
   expression could be associated with favorable ICB response rates.
   Therefore, we determined whether S100A5 expression could predict ICB
   response. As our previous study reported, we have built Xiangya immune
   cohort containing 51 BLCA patients treated with ICB from our
   hospital.^[ [273]^19 ^] Patients with higher S100A5 expression (Figure
   [274]6A) showed progressive disease after anti‐PD‐1 treatment
   (Figure [275]6B). In contrast, patients with lower S100A5 expression
   (Figure [276]6C) achieved a complete response after anti‐PD‐1 therapy
   (Figure [277]6D). According to the pathological response, these
   patients were categorized into the complete response (CR), partial
   response (PR), stable disease (SD), and progressive disease (PD)
   groups. We further divided CR and PR patients into the response group,
   whereas SD and PD patients were classified into the non‐response group.
   Low S100A5 IHC score group (IHC score less than 6) possessed
   significantly more patients responding to anti‐PD‐1 treatment than high
   S100A5 IHC score group (IHC score no less than 6) among the whole TMA
   cohort (Figure [278]6E). Importantly, the patients with lower S100A5
   IHC scores exhibited significantly higher disease‐free survival (DFS)
   rates than those with higher scores (Figure [279]6F). Moreover,
   multi‐color IF images showed a negative correlation between S100A5
   expression, CD8^+ T cells infiltration, and PD‐L1 expression
   (Figure [280]6G), revealing the mechanism of response or non‐response
   to treatment. In a large immunotherapy cohort of BLCA (IMvigor210), we
   found that S100A5 expression was highest in the desert immune
   phenotype, TC0 (lowest PD‐L1 expression on tumor cells), and IC0
   (lowest PD‐L1 expression on immune cells) (Figure [281]6H–J).
   Furthermore, we found that patients with CR expressed significantly
   lower S100A5 than patients with PD or SD in IMvigor210
   (Figure [282]6K).

Figure 6.

   Figure 6
   [283]Open in a new tab

   Relationship between S100A5 expression and immune checkpoint blockade
   (ICB) response. A) Representative immunohistochemical (IHC) image of
   high S100A5 expression patient. Scale bar, 100 µm. B) Representative CT
   image for patient with progressive disease after anti‐PD‐1 treatment.
   C) Representative IHC image of low S100A5 expression patient. Scale
   bar, 100 µm. D) Representative CT image for patient with complete
   response after anti‐PD‐1 treatment. E) Relative percentage of patients
   with clinical response to immunotherapy between different S100A5
   expression groups in Xiangya immune cohort. Yellow, immunotherapy
   response group; Blue, non‐response group. F) Disease‐free survival
   (DFS) of patients with different S100A5 IHC scores in Xiangya immune
   cohort. G) Representative multi‐color IF images for S100A5 (green), CD8
   (red), PD‐L1 (yellow) and DAPI (blue) in non‐response and response
   group patients. Scale bars, 100 µm. H) Expression of S100A5 on desert,
   excluded, and inflamed immune phenotypes in IMvigor210 cohort. I,J)
   Expression of S100A5 on patients with different PD‐L1 expression on
   tumor cells I) and immune cells J) in IMvigor210 cohort. K) Correlation
   between S100A5 expression values and immunotherapy response in the
   desert phenotype of IMvigor210 cohort. Different color represents
   different response type. CR: complete response; PR: partial response;
   SD: stable disease; PD: progressive disease. ns, not statistically
   significant. *p < 0.05; **p < 0.01; ***p < 0.001.

3. Discussion

   S100 family proteins, as a type of small molecular EF‐hand
   calcium‐binding proteins, play vital roles in the development and
   progression of numerous types of carcinomas and show prognostic value
   and are potential novel targets for treatment.^[ [284]^60 ^] The vital
   role of S100 proteins in carcinomas has been widely reported in
   melanoma, breast carcinoma, and lung carcinoma. S100A4 not only
   activates the nuclear factor‐kappa B (NF‐κB) pathway and leads to the
   release of the tumor necrosis factor (TNF)‐α,^[ [285]^61 ^] but also
   shapes an inflamed TME by inducing the secretion of IL‐8 and C—C
   chemokine ligand 2 (CCL2),^[ [286]^14 ^] playing an oncogenic role in
   malignant melanoma. In breast carcinoma, multiple S100 proteins are
   dysregulated, including S100A2, S100A4, S100A6‐9, and S100A11.^[
   [287]^60 ^] Among them, S100A8 and S100A9 can recruit MDSCs to maintain
   an immune suppression state in the TME,^[ [288]^12 ^] indicating the
   key roles of S100 family proteins in modulating the TME. For lung
   carcinoma, S100A4 expression was associated with worse survival
   outcomes in NSCLC and promoted cancer invasion through the NF‐κB/MMP9
   signaling pathway.^[ [289]^62 , [290]^63 ^] Recently, Liu et al.
   revealed that S100A7 can shape an immunosuppressive TME and decrease
   the efficacy of immunotherapy in LUSC.^[ [291]^64 ^] However, the
   expression patterns and functions of S100 proteins are cancer
   specific.^[ [292]^60 ^] Regarding BLCA, only Yao et al. reported that
   S100A2‐3, S100A5, S100A7‐9, S100A14‐16, and S100P were significantly
   higher in BLCA tissues using real‐time PCR.^[ [293]^16 ^] The
   comprehensive roles of S100 proteins in the TME and immunotherapy
   remain unclear, especially in BLCA. In this study, for the first time,
   we comprehensively analyzed the expression patterns and immunological
   roles of multiple S100 family proteins in BLCA and found that S100A5
   might have the most important value in BCLA. Furthermore, pan‐cancer
   analyses through 33 types of carcinomas revealed that the
   immunosuppressive role of S100A5 in the TME was most obvious in BLCA.

   As a high TMB tumor, an increasing number of studies have demonstrated
   the efficacy of ICIs in BLCA. Five ICIs have gained approval from the
   FDA: atezolizumab, durvalumab, avelumab, pembrolizumab, and
   nivolumab.^[ [294]^2 , [295]^4 , [296]^5 ^] However, only a minority of
   patients respond to ICI treatment, and most patients benefit little
   from ICIs and suffer from treatment‐related toxicity, suggesting an
   urgent need to identify biomarkers of response to ICIs.^[ [297]^5 ,
   [298]^6 ^] To date, numerous biomarkers have been found to predict ICI
   response. PD‐L1 expression in tumor cells has been reported to be
   associated with higher response rates; however, patients without PD‐L1
   expression also show an ICI response.^[ [299]^65 ^] Other biomarkers
   include TMB, alteration of DNA damage repair genes, and interferon‐γ
   associated genes; however, all these biomarkers need additional
   validation for clinical application.^[ [300]^65 ^] As an immune
   rheostats, the treatment function of ICIs largely depends on
   pre‐existing anticancer immune responses.^[ [301]^50 ^] Therefore, the
   immunosuppressive state of TME can be a major obstacle to ICI efficacy.
   The TME is composed of tumor cells, fibroblast cells, vascular
   endothelial cells, immune cells, extracellular matrix, and
   extracellular soluble molecules, and plays a vital role in cancer
   development and evasion of the host immune system.^[ [302]^8 ^] Based
   on the presence or absence of T cells in the tumor parenchyma, the TME
   can generally be divided into two profiles: non‐inflamed and inflamed
   tumors.^[ [303]^9 ^] In this study, we used systematic bioinformatics
   analysis (including bulk RNA‐seq and scRNA‐seq) to reveal that S100A5
   is associated with a non‐inflamed TME phenotype in BCLA. Inflamed
   tumors, characterized by CD4^+ and CD8^+ T cell infiltration and
   elevated signatures of immune activation, are sensitive to
   immunotherapy.^[ [304]^7 , [305]^8 ^] We validated that S100A5
   inhibited CD8^+ T cell infiltration, whereas its knockdown
   significantly promoted infiltration and enhanced the cytotoxicity of
   CD8^+ T cells in vitro and in vivo.

   There are two fundamental strategies for cancer immunotherapy: “immune
   enhancement” strategy, which focuses on increasing immune activation,
   and “immune normalization”, which focuses on restoring immune cell
   deficiency in the TME.^[ [306]^49 ^] Immune‐related adverse events
   (irAEs) limit the clinical application of “immune enhancement”
   strategy, while “immune normalization” strategies, such as ICIs show
   objective response rates without obvious irAEs. However, ICIs have
   shown no therapeutic efficacy in some patients, partly because of a
   lack of pre‐existing anticancer immune responses. The combination of
   “immune enhancement” and “immune normalization” strategies could be
   more effective. For example, combined therapy with anti‐PD‐1 and CTLA4
   therapy has shown much higher effectiveness.^[ [307]^11 ^] The key for
   developing “immune enhancement” strategy is to avoid irAEs. Using
   sc‐RNA‐seq data, we found that S100A5 was specifically expressed in
   tumor cells and barely expressed in other cells, both in the TME and
   blood. These results could be beneficial for developing anti‐S100A5
   drugs that are tumor cell‐specific and less toxic. Niclosamide, an
   inhibitor of S100A4, which inhibit colorectal cancer (CRC) progression,
   has entered phase II clinical trials for the treatment of metastatic
   CRC.^[ [308]^66 , [309]^67 ^] Based on our findings, anti‐S100A5
   therapy might have the ability to turn “cold” tumors into “hot” tumors
   and show higher effectiveness in combination with ICIs. Our work is the
   first step towards the development of anti‐S100A5 drugs that can
   reverse the TME without obvious toxicity.

   Unsupervised clustering of transcriptome profiling data revealed that
   BLCA can be divided into molecular subtypes that can be used to predict
   prognosis and therapeutic options, including chemotherapy and
   immunotherapy.^[ [310]^35 ^] Several molecular subtype models of BLCA
   have been established, including CIT,^[ [311]^37 ^] Lund,^[ [312]^38 ^]
   MDA,^[ [313]^39 ^] TCGA,^[ [314]^40 ^] Baylor,^[ [315]^41 ^] and UNC.^[
   [316]^42 ^] However, the differences in the number, relative size, and
   names of the subtypes between these molecular subtype models inhibit
   their clinical application. Thus, Kamoun et al. reported a consensus
   molecular classification by conducting a network analysis of these six
   molecular subtype models.^[ [317]^35 ^] Consensus molecular
   classification has promoted the clinical use of molecular subtypes.
   However, all of these molecular subtypes are built based on
   transcriptome profiling data (RNA‐seq or microarray), which can be
   complex and expensive for clinical applications. We found that S100A5
   expression could accurately predict all seven classification systems
   with high accuracy (AUC > 0.85), and validated this prediction value
   using other public databases and our own RNA‐seq cohort. Our findings
   greatly simplify molecular classification systems and promote their
   clinical applications. Basal subtypes have been reported to possess
   more cytotoxic lymphocytes and NK cell infiltration, and may be more
   sensitive to immunotherapy.^[ [318]^35 ^] This is consistent with our
   previous finding that S100A5 shaped a non‐inflamed phenotype of BLCA in
   terms of the molecular subtypes of BLCA.

   There are some limitations of our study. First, our validation cohorts
   contained RNA‐seq and microarray data, and different cohorts may have
   led to batch effects and bias in our findings. Second, although S100A5
   could mediate tumor immune evasion by regulating chemokine secretion
   and CD8^+ T cell cytotoxicity, further detailed mechanisms are
   required. Third, to turn cold bladder tumors into hot ones, S100A5
   small‐molecule inhibitors need to be developed in future research.

4. Conclusions

   S100A5 shapes a non‐inflamed tumor microenvironment in BLCA by
   inhibiting the secretion of pro‐inflammatory chemokines and the
   recruitment and cytotoxicity of CD8^+ T cells. Targeting S100A5
   converts cold tumors into hot tumors, thus enhancing the efficacy of
   ICB therapy in BLCA.

5. Experimental Section

Sources of Datasets and Preprocessing—Xiangya Cohorts

   As reported in the previous studies,^[ [319]^17 , [320]^18 ^] 60 fresh
   bladder cancer samples was collected (only 57 qualified samples) and 13
   paired adjacent normal samples without prior treatment from the Xiangya
   Hospital and performed RNA sequencing (RNA‐seq) of the qualified
   samples. Among the 57 patients, 56 were successfully followed up. This
   cohort was named the Xiangya BLCA cohort ([321]GSE188715). In addition,
   a tissue microarray (TMA) containing 50 BLCA samples was built and 28
   paired adjacent normal samples without prior treatment and another TMA
   containing 51 BLCA samples with immune checkpoint blockade (ICB)
   treatment.^[ [322]^19 ^]

Sources of Datasets and Preprocessing—Public Databases

   For the Cancer Genome Atlas (TCGA) database, the fragments per kilobase
   per million mapped fragments (FPKM) value of RNA‐seq data of 33 types
   of cancer from the UCSC Xena data portal
   ([323]https://xenabrowser.net/) was downloaded.^[ [324]^20 ^] Specific
   to TCGA‐BLCA, the FPKM value was transformed into transcripts per
   kilobase million (TPM). Duplicate patients and those without complete
   clinical information were filtered out. A total of 403 patients with
   detailed clinical information were included in the analysis. Gene
   Expression Omnibus (GEO) ([325]https://www.ncbi.nlm.nih.gov/geo/)
   databases with more than 50 BLCA samples were downloaded using
   “GEOquery” R package, including: [326]GSE87304 ([327]GPL22995),
   [328]GSE48276 ([329]GPL14951), [330]GSE48075 ([331]GPL6947),
   [332]GSE120736 ([333]GPL10558), [334]GSE31684 ([335]GPL570),
   [336]GSE32894 ([337]GPL6947), [338]GSE69795 ([339]GPL14951), and then
   transformed the gene symbol using the corresponding GPL files.
   E‐MTAB‐1803, a public database containing 85 BLCA samples, was
   downloaded from The European Bioinformatics Institute
   ([340]https://www.ebi.ac.uk/). Moreover, the expression matrix was
   downloaded and clinical information of the BLCA immunotherapy cohort
   (IMvigor210 cohort) from
   [341]http://researchpub.Gene.com/imvigor210corebiologies/.^[ [342]^21
   ^]

Sources of Datasets and Preprocessing—Calculating of TME Cell Infiltration

   Tracking tumor immunophenotype (TIP) was a web‐based analytical
   platform for analyzing the level of seven‐step cancer immune cycles
   ([343]http://biocc.hrbmu.edu.cn/TIP/).^[ [344]^22 ^] The TPM data was
   uploaded of the TCGA‐BLCA and Xiangya cohorts onto TIP and downloaded
   the cancer‐immune cycle. The single‐sample gene set enrichment analysis
   (ssGSEA) algorithm was used to calculate the infiltration of 28 immune
   cells into the BLCA TME. The gene set for each immune cell was obtained
   from a study by Charoentong et al.^[ [345]^23 ^] To eliminate the
   influence of this algorithm, six other algorithms were used to
   calculate the infiltration of immune cells: CIBERSORT, EPIC,
   mMCP‐counter, quanTIseq, TIMER, and xCell. Among these algorithms,
   EPIC, quanTIseq, TIMER, and xCell algorithms were performed using
   “deconvolute” function of the “immunedeconv” R package. For CIBERSORT,
   LM22 was downloaded, an annotated gene set, to define the 22 immune
   cell subtypes from the CIBERSORT web portal
   ([346]http://cibersort.stanford.edu/). For mMCP‐counter, the genes and
   probesets files were downloaded from
   [347]https://github.com/ebecht/MCPcounter and used
   “MCPcounter.estimate” function of the “MCPcounter” R package.^[
   [348]^24 ^]

Other Immunological Characteristics of TME

   MHC, receptors, chemokines, and immune stimulators (122
   immunomodulators) were identified based on a study by Charoentong
   et al.^[ [349]^23 ^] Moreover, 22 ICI genes were identified based on
   the study by Auslander et al.^[ [350]^25 ^] Eighteen genes for
   calculating the T cell‐inflamed score (TIS) were identified based on
   the study by Ayers et al.^[ [351]^26 ^] Pan‐cancer TIS and the effector
   genes of CD8^+ T cells, dendritic cells (DCs), macrophages, natural
   killer (NK) cells, and type 1 T helper (Th1) cells were collected as
   previously described.^[ [352]^17 ^]

Analysis Process of Single Cell RNA Sequencing (scRNA‐seq)

   Three samples from muscle invasive bladder cancers (MIBCs) were
   obtained from Xiangya hospital and performed scRNA‐seq in OE Biotech
   Co, Ltd (Shanghai, China), named Xiangya scRNA cohort. The detailed
   preparation of single‐cell suspensions, sequencing of droplet‐based
   single‐cells and raw data processing have been reported in previous
   studies.^[ [353]^18 , [354]^27 ^] After cell ranger processing, the
   count matrixes were converted into Seurat object by “Seurat” R package
   (version 4.1.0). Cells were regarded with unique molecular identifier
   (UMI) numbers less than 1000, gene numbers less than 200,
   log10GenesPerUMI less than 0.70 and mitochondrial‐derived UMI counts
   over 20% as low‐quality cells and filtered out these cells. Then the
   count matrixes were normalized using “NormalizeData” function and
   regressed out the effect of mitochondrial ratio using “SCTransform”
   function. “SelectIntegrationFeatures” function was used to find the top
   3000 variable features for integration and “PrepSCTIntegration”
   function was used to convert the Seurat object to SCT list object for
   integration. Then, “FindIntegrationAnchors” and “IntegrateData”
   functions were used to integrate the three samples based on the top
   3000 variable features selected above. This process regressed the
   potential batch effect and created a new matrix with 3000 variables.
   Then principal component analysis (PCA) was conducted using “RunPCA”
   function and ran t‐distributed stochastic neighbor embedding (tSNE)
   based on “pca” reduction using “RunTSNE” function. Finally, the main
   cells clusters were identified using “FindNeighbors” and “FindClusters”
   function (res = 0.4) and visualized the cells clusters based on tSNE
   plot. Clusters were annotated combined by “SingleR” package and
   well‐established marker genes reported previously. In particular, cells
   annotated as T cells were further clustered into subclusters. All
   parameters of the functions used above were set to default.

   In addition, a public scRNA‐seq dataset (PRJNA662018^[ [355]^27 ^])
   containing eight BLCA and three normal tissue samples was downloaded.
   Another public scRNA‐seq dataset ([356]GSE135337^[ [357]^28 ^])
   containing seven BLCA tissue samples, was downloaded from the
   supplementary materials of the GEO database. The analysis processes
   were similar to those used for Xiangya scRNA cohort. Specifically, out
   cells were filtered with UMI numbers less than 1000, genes numbers less
   than 250, log10GenesPerUMI less than 0.80, and mitochondrial‐derived
   UMI counts over 10%. Moreover, the log2TPM matrix of one primary
   bladder cancer patient (two other matrices were not available in the
   GEO database) was downloaded and cell type information from the
   supplementary file of [358]GSE145137.^[ [359]^29 ^] Because all the
   cluster and cell type information could be downloaded, the above
   scRNA‐seq analysis processes was omitted and directly used “VlnPlot”
   function to visualize the expression of S100A5 in the TME of BLCA. Two
   other scRNA‐seq databases ([360]GSE130001^[ [361]^30 ^] and
   [362]GSE145281^[ [363]^31 ^]) were analyzed using TISCH
   ([364]http://tisch.comp‐genomics.org/home/).

Estimation of Copy Number Variations (CNVs) in Epithelial Cells

   Setting the epithelial cells from three normal tissue samples in
   PRJNA662018 scRNA as references, the “InferCNV” package was used to