The KMT1A-GATA3-STAT3 Circuit Is a Novel Self-Renewal Signaling of Human Bladder Cancer Stem Cells

Bladder cancer is the most common urological malignancy worldwide after renal carcinoma (1), causing an estimated 430,000 new cases and 150,000 deaths per year (2, 3). Bladder cancer stem cells (BCSCs) have been isolated using defined surface markers, such as CD44, CD47, CD90, cytokeratin 5 (CK5), CK14, CK17, the 67-kDa laminin receptor (67LR), and aberrantly glycosylated integrin α3 (4, 5). These cells were more tumorigenic than bladder cancer non-stem cells (BCNSCs) and exhibited self-renewal and differentiation abilities to form a heterogeneous tumor (5). However, how BCSCs maintain their self-renewal remains elusive.

The alteration of chromatin regulation played a vital role in cancer stem cell reprogramming (6) and bladder cancer tumorigenesis (7). KMT1A (SUV39H1) encodes an evolutionarily conserved histone methyltransferase-catalyzing H3K9me3 modification, which plays an essential role in heterochromatin formation and gene silencing (8). Previous data showed that KMT1A was highly expressed in hematologic malignancies and solid tumors (9, 10). The inhibition of KMT1A induced the differentiation and apoptosis of acute myeloid leukemia cells, implying the potential effect of targeting KMT1A in tumor therapy (11).

The deregulation of transcription factors has been indicated in the self-renewal of cancer stem cell (12) and the tumorigenesis of bladder cancer (13). GATA3 belongs to the GATA family of transcription factors, contains two GATA-type zinc fingers, and is an important regulator of T-cell development (14). Recent studies showed that the loss of GATA3 in bladder cancer promoted cell migration and invasion (15). However, the regulatory mechanisms of GATA3 downregulation in human bladder cancer have not been illustrated. In addition, STAT3 belongs to the signal transducer and activator of transcription (STAT) family and is phosphorylated and activated to form homo- or heterodimers that translocate to the cell nucleus (16). STAT3 plays a key role in many cellular processes such as cell growth, apoptosis, and oncogenesis (17). Previous research indicated that the expression of STAT3 correlated with the metastasis and drug resistance of bladder cancer (18). However, the molecular mechanisms of STAT3 upregulation in bladder cancer remained elusive. Here, we demonstrated the regulatory mechanisms of KMT1A-GATA3-STAT3 circuit in BCSCs for the first time.

Primary human bladder cancer and normal bladder tissues were obtained from The Second Affiliated Hospital of Kunming Medical University College (Kunming, China) with informed consent and approved by the Research Ethics Board at The Second Affiliated Hospital of Kunming Medical University. The fresh sorted BCSCs and BCNSCs were immediately applied for the expression of mRNA and protein in transcriptome microarray assay, quantitative real-time PCR (qRT-PCR), immunofluorescence (IF) staining, flow cytometry sorting, and tumorsphere formation experiments. Besides, the sorted BCSCs were cultured in DMEM/F-12 medium supplemented with 20 ng/mL EGF, 20 ng/mL bFGF, 1% N2, and 2% B27, and BCNSCs were cultured in DMEM/F-12 medium supplemented with 15% FBS for 2 to 3 weeks until the sufficient cell numbers (approximately five to six passages) and then utilized for Western blot (WB), genetic manipulation, generation of xenograft, limiting dilution transplantation, chromatin immunoprecipitation (ChIP), and DNase I digestion experiments. Detailed clinical information for the bladder cancer patients was summarized in Supplementary Table S1 and Supplementary Table S2. Animal work was permitted by the Institutional Animal Care and Use Committee (IACUC) of the Institute of Biophysics (IBP), Chinese Academy of Sciences (CAS) (Beijing, China), and conducted in accordance with its recommendations and ethical regulations. All experimental protocols were approved by the IACUC, IBP, CAS. The mice were maintained under standard conditions according to the institutional guidelines for animal care.

Cell line EJ was obtained from KeyGen BioTECH (KG046). Cell lines SV-HUC-1, T24, 5637, J82, BEAS-2B, A549, MCF 10A, MCF7, RWPE-1, PC-3, FHC, HT29, 293T, and 786-O were obtained from the American Type Culture Collection (CRL-9520, HTB-4, HTB-9, HTB-1, CRL-9609, CCL-185, CRL-10317, HTB-22, CRL-11609, CRL-1435, CRL-1831, HTB-38, CRL-3216, and CRL-1932). Cell lines L02 and Huh7 were provided by Dr. Hong Zhu (the First Affiliated Hospital of Soochow University, Suzhou, China). All cell lines cultured in RPMI 1640 or DMEM medium (Gibco) supplemented with 10% FBS (Gibco), 100 U/mL penicillin (Gibco), and 100 μg/mL streptomycin (Gibco), and incubated at 37°C with 5% CO₂ in a humidified atmosphere. Cell lines were authenticated prior to use and tested for mycoplasma contamination routinely.

Primary human bladder cancer and normal bladder tissues were minced and digested using 20U type IV collagenase (Gibco) and DNase I (Gibco) at 37°C for 4 hours. Then, the cells were filtered by screen cloth to yield single cells and washed twice using PBS. The single cell from primary tissue and bladder cancer cell lines were stained using FITC-conjugated BCMab1 and PE-conjugated anti-CD44 antibody or the same isotype FITC/PE-conjugated antibody for 30 minutes on ice. Labeled cells were analyzed and sorted by flow cytometry (BD FACSAria II).

In the discovery step, a non–muscle-invasive tumor (#1) and a muscle-invasive tumor (#2) samples were selected and total RNA was isolated with an RNA isolation kit (Qiagen) from BCMab1⁺CD44⁺ and BCMab1⁻CD44⁻ cells derived from EJ, bladder cancer samples #1 and #2 according to the manufacturer's instructions. Previous to microarray hybridization, RNA concentration, and purity were determined using a UV2800 ultraviolet spectrophotometer (UNIC). Biotinylated cDNA was prepared according to the standard Affymetrix protocol from 250 ng total RNA by using Ambion WT Expression Kit. cDNA (5.0 μg) was hybridized for 16 hours at 45°C on GeneChip Human Transcriptome Array 2.0. GeneChips were washed and stained in the Affymetrix Fluidics Station 450 and scanned by using Affymetrix GeneChip Command Console that installed in GeneChip Scanner 3000 7G. Data were analyzed by Robust Multichip Analysis (RMA) algorithm using Affymetrix default analysis settings. Values presented were RMA signal intensity. The significantly differentially expressed genes (P < 0.001 with FDR <0.1, 2.0-fold cutoff) for each subset (BCSCs vs. BCNSCs) in the discovery cohort were then extracted and combined to yield 159 differentially expressed genes. Then the 159 differentially expressed genes were clustered by DAVID software (https://david.ncifcrf.gov). Microarray data have been deposited in the NCBI GEO under accession number GSE90903.

Total RNA from cells was extracted using an RNA isolation kit (Qiagen) and subjected to cDNA synthesis using M-MLV Reverse Transcriptase (Promega). The cDNA was then used as the templates for semi-quantitative PCR and qRT-PCR of the candidate genes, running in an ABI 7300 analyzer (Applied Biosystems). Primer sequences are listed in Supplementary Table S3. SYBR Green (Qiagen) was used as the fluorescent probe. Relative expression levels of the target genes were compared with a housekeeping gene, GAPDH. The fold change of differentially expressed genes in BCSCs compared with BCNSCs was calculated with the method of 2^−ΔΔCt (19).

Bladder cancer cell lines and primary bladder cancer samples were lysed using RIPA buffer (50 mmol/L Tris-HCl [pH 7.4], 150 mmol/L NaCl, 0.5% sodium deoxycholate, 0.1% SDS, 5 mmol/L EDTA, 2 mmol/L PMSF, and 1% Nonidet P-40) (20)] for 2 hours. Proteins were separated using polyacrylamide gel electrophoresis and transferred to a nitrocellulose membrane (Millipore). The membranes were blocked using skim milk, probed by primary antibodies and horseradish peroxidase–conjugated secondary antibodies and developing using Pierce ECL Western Blotting Substrate (Thermo Scientific).

Cells (5 × 10³) of bladder cancer cell lines and primary bladder cancer cells were seeded in an ultra-low attachment surface 6-well plate (Corning). Cells were maintained in KnockOut DMEM/F-12 medium supplemented with 20 ng/mL EGF, 20 ng/mL bFGF, 1% N2, and 2% B27. The number of tumorspheres was counted after 2 weeks of cultivation. The whole assay was repeated 4 times.

Cells were fixed by 4% paraformaldehyde for 20 minutes and penetrated by 0.5% TritonX-100 for 30 minutes. After blocking using 10% FBS, primary antibodies were added and incubated overnight at 4°C. After washing 3 times with PBS, fluorescence-conjugated secondary antibodies were added for observation by confocal microscopy.

For generation of xenografts, 2 × 10⁶ bladder cancer cell lines and primary bladder cancer cells were injected subcutaneously into NOD/SCID mice (n = 4–6). Seven days later, the volume of xenografts was measured twice per week (V = (π/6) (a × b × c)). The mice were euthanized after 8 weeks. In the serial transplantation experiments, the first formed tumor was picked up and cut into pieces under sterile condition and further digested using collagenase for 2 hours. The cell suspension was filtered, stained, and isolated to yield BCMab1⁺CD44⁺ cells for serial passage (21).

Limiting dilution transplantation assays were performed as previously described (22). Briefly, 10, 100, 1,000, and 10,000 cells of each group were mixed with Matrigel (BD Biosciences) and subcutaneously implanted into NOD/SCID mice (n = 5–10). The xenograft volume was serially measured (V = (π/6) (a × b × c)). The mice were euthanized after 4 months to calculate the rate of tumor formation and the frequency of tumorigenic cells (http://bioinf.wehi.edu.au/software/elda/).

ChIP was performed as previously described (23). Briefly, 10 million cancer cells were cross-linked with 1% formaldehyde and resuspended in lysis buffer. The cell lysate was sonicated on ice resulting in an average DNA fragment length of 500 bp. After centrifugation, immunoprecipitation was performed in ChIP dilution buffer overnight in the presence of IgG, KMT1A, H3K9me3, and GATA3 antibodies with agitation. A protein A agarose/Salmon Sperm DNA (Merck Millipore) slurry was added and incubated for 2 to 4 hours at 4°C with agitation. The antibody–agarose complex was centrifuged and washed 5 times, and the immunoprecipitated fraction was eluted. The cross-linking was reversed by incubation at 65°C for 4 hours. The DNA was recovered by phenol/chloroform extraction and precipitated, and the abundance of specific sequence was measured by qRT-PCR using the corresponding primer sequences (Supplementary Table S3).

Cell nuclei were isolated and lysed for DNase I digestion assay as described (24). After digestion at 37°C for 5 minutes, total DNA was extracted to perform qRT-PCR assays using promoter-specific primers (Supplementary Table S3).

The STAT3 (Gene ID: 6774) promoter-targeting site (–1710∼–1530) for sgRNA design was 5′-ACTGCCTCCCTGATAACATAGGG-3′, and the primers were ordered from Sangon Biotech Company. The construction of the vectors and positive cells screening was performed as described previously (25). Briefly, genomic DNA was phenol-chloroform extracted from G418-resistant cell colonies. The cells were identified using PCR of 50 ng genomic DNA, 10 pmol of each primer (forward 5′-CTCGTACCACCTCATATCCA-3′ and reverse 5′-ATATGGCCTCTCCTATCTGC-3′). All the positive clones were confirmed by Sanger sequencing.

The shRNAs against KMT1A, STAT3, and GATA3, as well as scrambled shRNA were designed, synthesized, and cloned into the pSicoR vector. The resultant lentiviral vectors containing the KMT1A, STAT3, GATA3, and scrambled shRNA were named shKMT1A, shSTAT3, shGATA3, and shCtrl lentivirus. Sorted BCSCs were cultured in DMEM/F-12 medium supplemented with 20 ng/mL EGF, 20 ng/mL bFGF, 1% N2, and 2% B27 for maintaining an undifferentiated state. One week later, BCSCs were transfected by the shKMT1A, shSTAT3, shGATA3, or shCtrl lentivirus supernatants to obtain cell lines stably expressing the KMT1A, STAT3, GATA3, or Ctrl shRNA by the method of sorting of GFP⁺ cells. ShRNA sequences are listed in Supplementary Table S4.

The Student t test was used to compare the mean values of two groups, whereas a linear correlation analysis was used to determine relationship between the expression levels of two genes. In the gene expression and survival analysis, the average of gene expression was first calculated. Bladder cancer samples expressing higher levels of KMT1A, GATA3, or STAT3 than the average were defined as the high group and the remaining samples as the low group. The overall survival of each group was calculated by a Kaplan–Meier analysis, and the difference between those two groups was examined using the log rank test. A value of P less than 0.05 (∗, P < 0.05; ∗∗, P < 0.01; and ∗∗∗, P < 0.001) was regarded to indicate a significant difference.

This study was started under the agreement of the patients, and informed consent was obtained according to the World Medical Association Declaration of Helsinki. And the Institutional Review Board of the hospital approved the study. All human studies were reviewed and approved by the Research Ethics Board at The Second Affiliated Hospital of Kunming Medical University (Kunming, China). Animal work was permitted by the IACUC, IBP, CAS, and conducted in accordance with its recommendations and ethical regulations. All experimental protocols were approved by the IACUC, IBP, CAS.

Our previous studies indicated that the monoclonal antibody BCMab1 recognized aberrant glycosylated integrin α3 and could be used in the isolation of human BCSCs combined with CD44 (4). To identify differentially expressed genes in human BCSCs, we performed a transcriptome microarray analysis of human BCSCs (BCMab1⁺CD44⁺) and BCNSCs (BCMab1⁻CD44⁻) isolated from non–muscle-invasive tumor (#1), muscle-invasive tumor (#2) samples, and the bladder cancer cell line EJ in a discovery step (Supplementary Fig. S1 and Supplementary Table S1). Microarray data were analyzed by RMA algorithm using Affymetrix default analysis settings. Values presented were RMA signal intensity. The significantly differentially expressed genes (P < 0.001 with FDR < 0.1, 2.0-fold cutoff) for each subset (BCSCs vs. BCNSCs) were extracted and combined to yield 159 differentially expressed genes (Fig. 1A and B; Supplementary Tables S5 and S6). BCSCs highly expressed the stemness-related genes including SOX2, GLI1, CD44, and STAT3, which revealed a stem cell gene signature (26) differed from that of BCNSCs (Fig. 1A). In qRT-PCR analysis, the elevated expression levels of SOX2 (351%), GLI1 (333%), CD44 (230%), and STAT3 (220%) were validated in BCSCs than those in BCNSCs (Fig. 1C). Clustered by DAVID Tools (27), the 56 upregulated genes were mainly enriched in the histone modification, chromosome organization, and transcriptional regulation signaling pathways (Fig. 1D), whereas the 103 downregulated genes were mainly concentrated in the centromeric heterochromatin, mRNA-3'-UTR binding, and translation regulator activity pathways (Supplementary Fig. S2A). The differentially expressed genes were confirmed by qRT-PCR in BCSCs compared with BCNSCs (Fig. 1E; Supplementary Fig. S2B). Because epigenetic alterations have been implicated in bladder cancer tumorigenesis (7) and BCSCs exhibited enrichment in the histone methylation signaling pathway, we focused on the elevated genes (KMT1A and NSD1) responsible for histone methylation.

In the validation step, methyltransferase KMT1A was highly expressed in bladder cancer samples (451%) compared with peri-tumor tissues (Fig. 1F; Supplementary Fig. S3A). Furthermore, the expression of KMT1A was higher in bladder cancer samples than that in peri-tumor tissues, but the expression of NSD1 did not show a significant difference between bladder cancer and peri-tumor tissues (Supplementary Fig. S3B and S3C; Supplementary Tables S6 and S7).

Previous studies indicated that bladder cancer could be divided into basal and luminal subtypes according to gene expression (28). Whether KMT1A was differently expressed in basal and luminal subtypes remained unknown. The basal and luminal biomarkers (28) were used for a hierarchical clustering analysis according to the data of GSE48075 and GSE48276 (29). The expression of KMT1A was higher in the basal tumors compared with that in luminal ones in both fresh and formalin-fixed and paraffin-embedded bladder cancer samples (Supplementary Fig. S4). Therefore, KMT1A could be a basal marker of bladder cancer.

Moreover, KMT1A was highly expressed not only in bladder cancer cell lines, but also in liver cancer, prostate cancer, and colorectal cancer cell lines (Supplementary Fig. S5A and S5B). According to the data from The Cancer Genome Atlas (TCGA), KMT1A was also highly expressed in esophageal carcinoma, stomach and esophageal carcinoma, stomach adenocarcinoma, lung squamous cell carcinoma, head and neck squamous cell carcinoma, bladder urothelial carcinoma, and liver hepatocellular carcinoma (Supplementary Fig. S5C). However, the expression of NSD1 did not show a significant difference between bladder cancer and normal tissues and was only highly expressed in LIHC (Supplementary Fig. S5D).

In addition, the expression of KMT1A was significantly higher in BCSCs compared with either BCNSCs, normal bladder stem cells (NBSCs), or normal bladder non-stem cells (NBNSCs; Supplementary Fig. S6A), in which NBSCs and NBNSCs were isolated by the markers of CD44 and Pan-CK (25). Moreover, KMT1A was highly expressed in BCSCs (489%) and tumorsphere cells (537%) compared with those in BCNSCs and non-sphere tumor cells, respectively (Fig. 1G; Supplementary Fig. S6B). Notably, the expression of KMT1A positively correlated with that of CD44 in bladder cancer cell lines and bladder cancer samples (Supplementary Fig. S6C). Furthermore, the expression of KMT1A and H3K9me3 modification was upregulated in BCSCs compared with that in BCNSCs (Fig. 1H and I), and KMT1A was mainly located in the nuclei of BCSCs (Fig. 1I). Taken together, these results indicated that KMT1A was highly expressed in human BCSCs.

To determine the function of KMT1A in human BCSCs, the expression of KMT1A and H3K9me3 modification was significantly decreased in BCSCs using a shRNA against KMT1A (shKMT1A) (Fig. 2A and B; Supplementary Fig. S7A and S7B). Notably, depletion of KMT1A remarkably decreased the number of tumorspheres by 82% and CD44-positive cells by 84% of BCSCs compared with shCtrl BCSCs (Fig. 2C and D; Supplementary Fig. S7C and S7D). Moreover, depletion of KMT1A significantly inhibited the formation of BCSC xenograft tumors and secondary (2nd), tertiary (3rd), and quaternary (4th) xenografts (Fig. 2E and F). More importantly, shKMT1A BCSCs displayed significantly reduced tumor initiation and a 95% decrease in tumorigenic cells compared with shCtrl BCSCs in the limiting dilution xenograft assay (Fig. 2G). Patients expressing higher levels of KMT1A had a shorter mean survival time than patients expressing lower levels of KMT1A in Bae's cohort (GSE13507) and Hoglund's cohort (GSE37815) (Supplementary Fig. S7E; Supplementary Tables S7 and S8). Therefore, our findings demonstrated that KMT1A positively regulated self-renewal and tumorigenicity of human BCSCs.

To further investigate the underlying mechanisms of KMT1A in the regulation of BCSCs, the expression levels of multiple stemness-related genes were analyzed in shCtrl and shKMT1A BCSCs. When KMT1A was depleted, the expression levels of STAT3 and CD44 decreased by 80% compared with those in shCtrl BCSCs, and the expression levels of other stemness-related genes remained unchanged (Fig. 3A and B; Supplementary Fig. S8A and S8B). Consistently, depletion of KMT1A significantly reduced the tyrosine phosphorylation at the 705 residue of STAT3 (pY-STAT3) in BCSCs (Fig. 3B; Supplementary Fig. S8B). We also found that the expression of STAT3 was significantly higher in the KMT1A^high bladder cancer samples than that in KMT1A^low ones in McConkey's cohort (GSE48276) (Supplementary Fig. S8C). In addition, the expression of STAT3 positively correlated with that of KMT1A in bladder cancer samples (Supplementary Fig. S8D). Moreover, STAT3 was highly expressed (236%) in BCSCs and tumorsphere cells (493%) compared with those in BCNSCs and non-sphere tumor cells, respectively (Fig. 3C; Supplementary Fig. S8E). Consistently, both the expression levels of STAT3 and pY-STAT3 were higher in BCSCs than those in BCNSCs (Fig. 3D). Furthermore, activated STAT3 colocalized with KMT1A in the nuclei of BCSCs (Fig. 3E). These results suggested that STAT3 was highly expressed and activated in human BCSCs, which correlated with the expression of KMT1A.

To determine the function of STAT3 in human BCSCs, the expression and tyrosine phosphorylation of STAT3 were significantly decreased in BCSCs using an shRNA against STAT3 (Fig. 3F and G; Supplementary Fig. S9A and S9B). Notably, depletion of STAT3 remarkably decreased the number of tumorspheres by 75% and CD44-positive cells by 74% of BCSCs compared with shCtrl BCSCs (Fig. 3H and I; Supplementary Fig. S9C and S9D). Remarkably, depletion of STAT3 severely impaired the formation of BCSC xenograft tumors and secondary (2nd), tertiary (3rd), and quaternary (4th) xenografts (Fig. 3J and K). Importantly, shSTAT3 BCSCs showed dramatically reduced tumor initiation and a 90% decline in tumorigenic cells compared with shCtrl BCSCs (Fig. 3L). We also observed that patients expressing higher levels of STAT3 had a shorter mean survival time than patients expressing lower levels of STAT3 in Dyrskjøt 's cohort (E-MTAB-4321) and Michor's cohort (GSE31684) (Supplementary Fig. S9E; Supplementary Tables S7 and S8). These results indicated that the expression and activation of STAT3 played an indispensable role in the self-renewal maintenance and tumorigenicity of human BCSCs.

KMT1A imprints H3K9me3 on the promoter of target genes, which induces transcriptional repression (8). Hence, there should be a regulatory factor between KMT1A and the upregulation of STAT3 in BCSCs, and we consequently focused on the multiple potential negative regulators of STAT3, such as GATA1, GATA2, GATA3, NFκB, c-Myc, SOCS3, and P53 (17, 30–33). In the ChIP experiments, we found that KMT1A did not remarkably bind to these promoters except for that of GATA3 in BCSCs (Fig. 4A; Supplementary Fig. S10). Furthermore, five overlapping regions in the promoter of GATA3 were identified to be occupied by H3K9me3 modification, and the expression of GATA3 was significant lower in BCSCs than that in BCNSCs (Fig. 4B; Supplementary Fig. S11). The H3K9me3 modification of the -1351∼-1172 region of the GATA3 promoter was decreased by 90% when KMT1A was depleted (Fig. 4C and D). Moreover, depletion of KMT1A remarkably enhanced the chromatin accessibility of the GATA3 locus in the DNase I digestion assay (Fig. 4E). Consistently, the expression of GATA3 was significantly increased by 330% in shKMT1A BCSCs compared with that in shCtrl BCSCs (Fig. 4F and G). In BCNSCs, the overexpression of KMT1A, instead of KMT1A lacking the SET domain (KMT1A-ΔSET), significantly increased the expression levels of STAT3 and pY-STAT3 (>10-fold), but decreased the expression of GATA3 by 95% compared with those in the control vector-transfected BCNSCs (vec) (Fig. 4H, left), which was comparable with BCSCs (Fig. 4H, right).

The expression of GATA3 was lower in CD44^high bladder cancer samples than that in CD44^low bladder cancer samples in McConkey's cohort (GSE48276) and Michor's cohort (GSE31684) (Supplementary Fig. S12A). Furthermore, the expression of GATA3 negatively correlated with that of CD44 in bladder cancer samples (Supplementary Fig. S12B). Finally, patients expressing higher levels of GATA3 had a longer mean survival time than patients expressing lower levels of GATA3 (Supplementary Fig. S12C; Supplementary Tables S7 and S8). These results showed that KMT1A directly bound to the promoter of GATA3, where it augmented H3K9me3 modification and suppressed the transcription of GATA3 depending on its methyltransferase activity.

The mechanisms of GATA3 in the regulation of the expression of STAT3 were also investigated. Interestingly, GATA3 bound to three regions of the STAT3 promoter, and the binding to the -1710∼-1530 region was the most significant (707% enrichment; Fig. 5A). According to previous studies, the binding motif of GATA family transcription factors is 5'-[A/T] GATA [A/G]-3' (34). This region of the STAT3 promoter contains the distinctive binding sequence 5'-TGATAA-3' of GATA3 which ranges from -1654 to -1649 (Supplementary Fig. S13A). Hence, CRISPR/Cas9 was applied to abrogate the -1654∼-1649 segment of the STAT3 promoter. In a heterozygous mutant BCNSCs (Mut), GATA3 only associated with a single allele of STAT3 promoter (Fig. 5B and C). Similarly, in the DNase I digestion assay, the abrogation of the STAT3 promoter remarkably enhanced the chromatin accessibility of the STAT3 locus in Mut BCNSCs compared with WT BCNSCs (Fig. 5D). In addition, the deletion of a single allele of STAT3 promoter resulted in the significant upexpression of p-STAT3 and STAT3, but did not affect the expression of GATA3 (Fig. 5E). In the tumorsphere formation assay, Mut BCNSCs generated more tumorspheres than WT BCNSCs (Fig. 5F).

In addition, GATA3 was knocked down in BCNSCs (Supplementary Fig. S14A). Depletion of GATA3 upregulated the expression levels of STAT3 and pY-STAT3 in BCNSCs compared with those in shCtrl BCNSCs (Supplementary Fig. S14A and S14B). In the tumorsphere formation assay, shGATA3 BCNSCs generated more tumorspheres than shCtrl BCNSCs (Supplementary Fig. S14C), which suggested that the downregulation of GATA3 was sufficient to upregulate STAT3 in BCNSCs. These results indicated that GATA3 negatively regulated the transcription of STAT3 mainly dependent on the presence of STAT3 promoter.

To determine the function of GATA3 in human BCSCs, GATA3 or GATA3 combined with STAT3 were overexpressed in BCSCs. The overexpression of GATA3 in BCSCs (oeGATA3) resulted in a 77% decrease in the expression levels of STAT3 and pY-STAT3 compared with vec BCSCs. Nevertheless, the co-overexpression of STAT3 and GATA3 in BCSCs (oeGATA3/STAT3) restored the expression levels of STAT3 and pY-STAT3 (Fig. 6A and B; Supplementary Fig. S15A and S15B). Moreover, GATA3 efficiently bound to the -1710∼-1530 region of the STAT3 promoter in oeGATA3 BCSCs compared with vec BCSCs (Fig. 6C and D). Furthermore, oeGATA3 BCSCs produced 86% fewer tumorspheres and 68% fewer CD44-positive cells than vec BCSCs (Fig. 6E and F). However, the co-overexpression of STAT3 and GATA3 rescued the ability of tumorsphere formation and the number of CD44-positive cells of BCSCs (Fig. 6E and F; Supplementary Fig. S15C).

Remarkably, oeGATA3 BCSCs formed smaller tumors than vec BCSCs (Fig. 6G). In contrast, the tumors formed by oeGATA3/STAT3 were similar to those formed by vec BCSCs (Fig. 6G). In the serial transplantation experiment, the overexpression of GATA3 severely suppressed secondary (2nd) and tertiary (3rd) serial tumor formation by BCSCs, but the co-overexpression of STAT3 and GATA3 restored serial tumor formation by BCSCs (Fig. 6H). More importantly, oeGATA3 BCSCs exhibited dramatically reduced tumor initiation and a 90% decline in tumorigenic cells compared with vec BCSCs (Fig. 6I). In contrast, oeGATA3/STAT3 BCSCs rescued tumor-initiating ability and the proportion of tumorigenic cells (Fig. 6I).

In addition, GATA3 was overexpressed in STAT3 promoter mutated BCSCs (Supplementary Fig. S16A and S16B). However, the overexpression of GATA3 did not affect the expression levels of p-STAT3 and STAT3 (Supplementary Fig. S16B and S16C). In limiting dilution transplantation and xenograft assays, oeGATA3 did not significantly decrease tumorigenicity and tumor formation of STAT3 promoter mutated BCSCs (Supplementary Fig. S16D and S16E). Taken together, these data indicated that GATA3 negatively regulated the self-renewal and tumorigenicity of human BCSCs by suppressing the transcription of STAT3, and the overexpression and activation of STAT3 rescued the dysfunction of BCSCs resulting from the overexpression of GATA3.

Bladder cancer could be divided into basal and luminal subtypes according to genomic expression profiles (28). Luminal tumors displayed the upregulation of GATA3 and PPARγ target genes, and the enrichment for CDKN1A, ELF3, FGFR3, and TSC1 alterations (29). Basal tumors were enriched with CD44 and CK5 biomarkers, and the mutations of TP53 and RB1 genes (35). In this study, the expression of KMT1A and STAT3 was upregulated, but the expression of GATA3 was downregulated in BCSCs, which displayed a basal characteristic and correlated with less favorable prognosis. Taken together, KMT1A may serve as a novel indicator of basal subtype of bladder cancer.

KMT1A was reported to repress the myogenic differentiation of alveolar rhabdomyosarcoma cells by interacting with MyoD (36), which is the first report about KMT1A in the regulation of stemness and differentiation. Previous studies indicated that GLI1, STAT3, BMI1, and β-Catenin participated in the self-renewal of BCSCs (5). However, those signaling molecules that have been identified to be upregulated in BCSCs to date were also highly expressed or functional in NBSCs (37). Here, we found that KMT1A was highly expressed in human BCSCs compared with either BCNSCs or normal bladder tissue, which positively regulated the self-renewal of BCSCs and correlated with a poor prognosis in patients with bladder cancer. Therefore, KMT1A is a novel potential biomarker of human BCSCs and could be a promising target for bladder cancer therapy due to its high expression levels in BCSCs.

GATA3 played an important role in the establishment of human T-cell commitment by repressing Notch signaling and stemness-related genes (14). In addition, GATA3 was reported to be a suppressor of breast and bladder cancer (15, 38). Our results showed that the expression of GATA3 was downregulated in human BCSCs and directly correlated with favorable prognosis in patients with bladder cancer. Mechanistically, KMT1A recruited a suppressing complex (such as HP1) to the promoter of GATA3, imprinted H3K9me3 modification on it, and repressed its transcription in human BCSCs (Fig. 6J). Taken together, our data illustrated the mechanisms of GATA3 downregulation in BCSCs for the first time.

The activation of STAT3 is proved to play a crucial role in normal homeostasis mediated by tissue-specific stem cells and cellular transformation (17, 39). The activation of STAT3 in NBSCs led to direct progression to invasive bladder cancer (40). Metformin exerts anticancer effects including cell-cycle arrest, the reduction of cell proliferation, migration, and invasiveness, and the increase in apoptotic cell death by inhibiting the activation of STAT3 pathways in bladder cancer cells (18). However, the underlying mechanisms of STAT3 in driving BCSCs remained unclear. Our results showed that GATA3 directly bound to the promoter of STAT3 and repressed its transcription, and this process may be completed with the help of other cofactors (Fig. 6J). Transcriptional repression of GATA3 mediated by KMT1A upregulated the expression and tyrosine phosphorylation of STAT3. Furthermore, the overexpression and phosphorylation of STAT3 were indispensable to maintain the self-renewal and tumorigenicity of BCSCs (Figs. 3H–L and 6E–I), and the amount of pY-STAT3 positively correlated with the expression of STAT3 in human BCSCs (Figs. 3B, D, and G, 4H, 5E, and 6B). Therefore, both the overexpression and phosphorylation of STAT3 played important roles in the activation of KMT1A-GATA3-STAT3 circuit in human BCSCs. Overall, these results demonstrated the mechanisms underlying STAT3 upregulation in BCSCs for the first time. Whether there are other negative regulators of STAT3 in human BCSCs need to be further investigated (Fig. 6J).

In conclusion, our findings demonstrate that the KMT1A-GATA3-STAT3 circuit, as a novel self-renewal signaling, drives the self-renewal maintenance of human BCSCs. This finding indicates that both the heterochromatin formation mediated by KMT1A and expression of stemness-related genes triggered by activated STAT3 promote the self-renewal and tumorigenicity of human BCSCs, which provides the novel targets for bladder cancer therapy (Fig. 6J).

No potential conflicts of interest were disclosed.

Conception and design: Z. Yang, C. Li

Development of methodology: Z. Yang, C. Li

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Z. Yang, L. He, K. Lin, C. Li

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Z. Yang, C. Li

Writing, review, and/or revision of the manuscript: Z. Yang, Y. Zhang, A. Deng, Y. Liang, C. Li, T. Wen

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Li

Study supervision: C. Li

We would like to thank Professor Yi Shi (Institute of Microbiology, Chinese Academy of Sciences) for his critical reading of the article. We also thank the patients and the urology surgeons Jiansong Wang and Haifeng Wang (Department of Urology, The Second Affiliated Hospital of Kunming Medical College, Kunming, China), Drs. Junfeng Hao, Guizhi Shi, Shu Meng, and Xiang Shi (Institute of Biophysics, Chinese Academy of Sciences) for their technical supports. Professor Jeffrey S. Damrauer's group (Lineberger Comprehensive Cancer Center, University of North Carolina at Chapel Hill) provided the basal and luminal subtype calls on their meta-dataset of bladder tumors.

This work was supported by the National Natural Science Foundation of China (81602644 to Z. Yang, 81672956 to C. Li, 81472413 to C. Li) and the grant from the Ministry of Science and Technology of China (2010ZX09401-403 to T. Wen).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.