SET Domain–Containing Protein 4 Epigenetically Controls Breast Cancer Stem Cell Quiescence

As a major global health problem, cancer is one of the leading causes of morbidity worldwide. Because of the heterogeneity of tumor cells, the efficacy of chemoradiotherapy treatment is often suboptimal, as indicated by the high death rate of patients with cancer (1, 2). Current therapies are particularly limited by the emergence of therapy-resistant cancer cells (3–5). Increasing evidence has revealed that a small fraction of cancer cells, termed cancer stem cells (CSC), are responsible for therapeutic resistance, where the quiescence in CSCs is a crucial mechanism for resistance and survival (6–9). Standard therapies mainly target the tumor bulk, but they fail to eradicate the resistant CSCs that may cause relapse in patients after clinical treatments are finished (2, 7, 10).

Cellular quiescence is a reversible and nondividing state, the counterpart to proliferation (11). Previous studies identified that quiescent CSCs were more resistant to chemotherapy and could retain the capacity to proliferate after chemotherapy withdrawal (12–14). Although several molecular players in the regulation of cellular quiescence have been reported, we know remarkably little about the determinants of quiescence and the mechanisms of the transition between active and quiescent states in CSCs (7, 10, 15). Epigenetic studies have shown that heterochromatin is involved in maintaining the reversibility of cellular quiescence, in which the methylation of histones contributes to heterochromatin formation (16, 17). Heterochromatin exists in two varieties, constitutive heterochromatin (cHC) and facultative heterochromatin (fHC), that silence gene expression by virtue of their highly condensed structure (18, 19). The two varieties are distinguished by their distinct epigenetic signatures, fHC involving high levels of trimethylation of lysine 9 of histone 3 (H3K9me3) and lysine 27 of histone 3 (H3K27me3), and cHC exhibiting high levels of trimethylation of lysine 20 of histone 4 (H4K20me3). In contrast, euchromatin displays a high level of acetylation of lysine 9 of histone 3 (H3K9ac; refs. 16, 18, 20, 21).

The family of SET domain–containing proteins (SETD), histone lysine methyltransferases, has been reported to play a role in the regulation of chromatin structure, gene expression by catalyzing the methylation of histone proteins and cell proliferation in several cell lines (22–25). To study cellular quiescence regulation in CSCs, we used Artemia, the brine shrimp, as a model system. This primitive crustacean undergoes cellular quiescence for prolonged periods during embryonic diapause, a state of obligate dormancy to cope with environmental stresses (26). Previously, we reported that SETD4 regulates cellular quiescence by catalyzing H4K20me3 during Artemia diapause entry (27).

Here, we show that SETD4 facilitates heterochromatin formation via H4K20me3 catalysis in BCSCs that are located at certain promoter regions and that it regulates the expression of a set of genes in the quiescent regulation of BCSCs. Indeed, the quiescent BCSCs play critical functions in resistance to chemoradiotherapy and in relapse and correlate with malignancy in clinical patients. We demonstrate an evolutionarily conserved mechanism of CSC quiescence and establish a new cellular narrative for tumorigenesis and relapse.

NOD/SCID female mice were purchased from the Shanghai Laboratory Animal Center (SLAC) of China. They were housed under a 12-hour light/dark cycle (light between 06:00 and 18:00) in a temperature-controlled room (22 ± 1°C) with free access to water and food. All mice were maintained with the approval of the Animal Ethics Committee of the Zhejiang University (Hangzhou, China) and in accordance with the university's animal experiment guidelines.

HEK293T, MKN45, MCF7, T47D, and HCC1937 were purchased from the Tumor Cell Bank of Chinese Academy of Sciences (Shanghai, China). All cells were obtained directly from cell bank and passaged in the laboratory for fewer than 6 months after receipt. All cells were authenticated using short tandem repeat profiling. Mycoplasma detection was performed using a Mycoplasma Detection Set (TaKaRa; 6601) for all the cell lines. They were cultured according to the vendor's instructions at 37°C in a humidified atmosphere with 5% CO₂.

The following antibodies were used: H3K4me3 (Millipore; 2207275), H3K9me3 (Abcam; ab1773), H3K27me3 (Abcam; ab6174), H3K36me3 (Abcam; ab9050), H3K79me3 (Abcam; ab2621), H4K20me1 (Santa Cruz Biotechnology; sc-134221), H4K20me2 (GeneTeX; GTX630545), H4K20me3 for WB (Cell Signaling Technology; 5737s), H4K20me3 for IF (Abcam; ab9053), GAPDH (Cell Signaling Technology; 2118), H3K9ac (Abcam; ab10812), HP1-α (Santa Cruz Biotechnology; sc-130446), SUV4-20h2 (Santa Cruz Biotechnology; sc-366867), H3 (Abcam; ab1791), H4 (Abcam;ab10158), H3S10ph (Cell Signaling Technology; 53348), RbS807/811ph (Cell Signaling Technology; 9308s), Ki67 (Abcam; ab16667), SETD4-(R) (Sigma-Aldrich; HPA035405), SETD4 (Sigma-Aldrich; HPA024073), SETD4-(m) (Santa Cruz Biotechnology; sc-514060), CD44-FITC (eBioscience; 11-0441-81), CD44 (Cell Signaling Technology; 3570), CD24-PE (eBioscience; 12-0242), CD24 (eBioscience; 14-0242-81), CD24-647 (BioLegend; 311110), CD133 (HuaAn-Biotec; EM1701-28), ALDH-1 (Abcam; ab52492), Sox2 (Abcam; ab97959), Oct4 (Abcam; ab18976), Nanog (Abcam; ab109250), PCNA (Abcam; ab29), LC3B (Sigma-Aldrich; L7543), Active-caspase-3 (Abcam; ab32042), and H2AS139ph (γ-H2A; Novus; NB100-384).

MCF7 or HCC1937 cells were digested by accutase (Gibco; A1110501) and solid tumors from MCF7-CDXs, HCC1937-CDXs or clinical patients were cut up into small pieces and then digested with ultrapure collagenase III (LS004180) in DMEM at 37°C for 3 to 4 hours. Single cells were filtered through a 45-mm nylon mesh and then resuspended in 100 μL (per 10⁶ cells) HBSS containing 2% FBS for FACS. Antibodies of CD24 and CD44 were added and incubated for 20 minutes on ice. Flow cytometry was performed using a FACS Vantage (BD). Cells were routinely sorted twice and reanalyzed for purity of 90%. The population of CD44^high/CD24^low was used as the BCSCs in this study. For PKH26 staining, BCSCs were labeled with PKH26 (Sigma-Aldrich; PKH26GL-1KT) dye according to the manufacturer's instructions. The labeled cells were cultured under tumorsphere formation conditions for 2 weeks; then, the tumorspheres were dissociated by accutase and subjected to FACS.

Cells were plated at a density of 4,000 to 8,000 cells per well in the 6-well ultralow attachment plates and cultured in tumorsphere formation conditions [DMEM/F12 (Corning; 10-092-cv) supplemented with 10% serum replacement (SR; Thermo Fisher Scientific; 10828028), 20 ng/mL EGF, 5 ng/mL heparin sodium (MedChemExpress; 9041-08-1), 20 ng/mL bFGF (PeproTech; 96-100-18B-500)] at 37°C in a humidified 5% CO₂ incubator.

FACS-sorted qBCSCs or chemoradioresistance qBCSCs were cultured in tumorsphere formation conditions (described above) plus 50 ng/mL exosomes that has been isolated from the cultured medium of MCF7 and HCC1937 cell lines. The exosome-depleted FBS was used in the culture of the MCF7 and HCC1937 cell lines. The total exosome was isolated from the cultured medium using Cell Medium Exosome Isolation Kit (Life Technologies; 4478359) according to the manufacturer's instructions and after 20 hours of culture, the time just before one cell begins dividing into two. These cells were used as the A-qBCSCs in this study.

For mammary fat pad orthotopic xenograft experiments, 6 to 8 weeks of age (at the time of injections) female NOD/SCID mice were used. Cells were resuspended in HBSS/Matrigel (Corning; 354234) and injected into the lower right and left mammary pads of each mouse. Animals were euthanized when the tumors were approximately 0.5 to 1.2 cm in the largest diameter, to avoid tumor necrosis. The tumor width (w) and length (l) was recorded with a caliper and tumor size was calculated using the formula (l × w²/2).

Section samples were fixed with 4% paraformaldehyde and embedded in OCT (Sakura; 4583). Cell samples were fixed with 4% paraformaldehyde. The samples were incubated with the appropriate antibodies. Tumor tissue sections were stained with hematoxylin and eosin (Beyotime, c0105) according to the manufacturer's instructions. Briefly, samples were stained in hematoxylin staining solution for 10 minutes and stained in eosin staining solution for 1 minute then dehydrated and sealed with a neutral gum and detected on the microscope.

Total proteins were extracted by RIPA lysis buffer (Beyotime, P0013B) containing protease inhibitor cocktail (MedChemExpress; HY-K0010). Each protein sample (25 μg) was subjected to SDS-PAGE and then transferred to a nitrocellulose membrane for Western blot analysis using Bio-Rad system, according to the manufacturer's instructions.

qRT-PCR reactions were performed on the Bio-Rad MiniOpticon system using SYBR Premix Ex Taq (TaKaRa; RR420A). Gene-specific primers were used (Supplementary Table S1). The relative amounts of mRNAs were analyzed using the comparative CT method, as described previously (28).

On the basis of the sequence of the human SETD4 gene (NM_017438.4) in GenBank, the pLent-EF1a-SETD4-P2A-GFP-CMV-Puro overexpression plasmid (Vigene biosciences; LT88002) was synthesized and transfected with a viral packaging plasmid of 10 μg of plasmid containing the vector of SETD4, 10 μg of pMD2.G (Addgene; 12259) and 15 μg of psPAX2 (Addgene; 12260) by Lipo 3000 (Invitrogen; L3000015) into HEK293T cells overnight, and the viral supernatant was collected 48 hours later. The viral supernatant was filtered through a 0.45-μm filter, and the freshly sorted BCSCs were used for infection in the presence of 5 μg/mL Polybrene (Sigma-Aldrich; 107689-10G). RNAi was performed to knockdown the overexpressed SETD4 gene at posttransfection day 3, and images were obtained after further incubation with Scramble or SETD4 siRNA (Santa Cruz Biotechnology; sc-91446) for 4 days, respectively. The SETD4 and control siRNAs (100 pmol/L per 10,000 cells) were transfected into overexpressed SETD4 BCSCs using the siRNA transfection system (Santa Cruz Biotechnology; sc-45064), according to the manufacturer's instructions.

SETD4⁺ qBCSCs asymmetric division was determined by the incorporation of EdU (Ribobio; C00054). The SETD4⁺ qBCSCs (n = 20) were incubated in tumorsphere formation medium containing 50 μmol/L EdU for 40 and 60 hours. All samples were fixed with 4% paraformaldehyde for immunofluorescence analysis of SETD4 or Ki67 by detection with second antibody conjugated with Alexa Fluor 594, and EdU was incubated with Alexa 488–conjugated Apollo staining reaction solution (Ribobio; C10310-3) for 30 minutes.

Construction of GST-SETD4 and SETD4 mutants: The open reading frame of GST-SETD4 and four GST-SETD4 mutants were cloned using specific primers (Supplementary Table S1). The in vitro histone methylation transferase reactions were modified versions of protocols described previously (29) and were performed in 50 μL of methylase activity buffer containing 10 μg of core histones as substrates, 10 mmol/L AdoMet (Sangon; A6555-5g) as a methyl donor, and GST, GST-SETD4, dose-enhanced GST-SETD4 and four mutant types of SETD4 as catalyzers. After incubation for 60 minutes at 30°C, the reaction products were examined by Western blotting or mass spectrometry analyses.

Samples were fixed with 2.5% glutaraldehyde overnight and dehydrated with ethanol, then infiltrated with the mixture of absolute acetone and Spurr resin at room temperature. After infiltration, samples were sectioned in ultratome (Leica EM UC7) and then observed using a transmission electron microscope (TEM; JEM-1230, JEOL Inc., Japan) at 80 kV. Images of the cells were captured using a digital camera.

Cells were seeded into ultralow attachment 6-well plates at a density of 8,000 cells per well. The cells were incubated with 100 nmol/L Taxol (Sangon; A601183-0100) and 1 mmol/L 5-FU (Sangon; A100597-0001) for 10 days and/or exposed to X-ray with 30 Gy and cultured for the further 4 days. The medium was changed every 2 days. After treatments, trypan blue (Beyotime; C0011) analysis was performed according to the manufacturer's instructions. The surviving cells were harvested using a dead cell removal kit (Miltenyi Biotec; 130-090-101).

Cellular suspensions were loaded on a single-cell instrument (10x Genomics) to generate single-cell GEMs. Sequencing libraries were loaded on an Illumina Hiseq PE150 a 150 bp paired-end module. The Cell Ranger Single Cell Software Suite 1.3 was used to perform sample demultiplexing, barcode processing and UMI counting. All barcodes, the total UMI counts of which exceed m/10, were considered as cells. For visualizing data in 2-d space, Cell Ranger passes the PCA-reduced data into t-stochastic neighbor embedding (t-SNE). The graph-based clustering algorithm consists of building a sparse nearest-neighbor graph, followed by Louvain modularity optimization (30). The value of k, the number of nearest neighbors, is set to scale logarithmically with the number of cells.

The RNAs of bulk cells were extracted using TRIzol, reverse-transcribed, and included in the cDNA library. The library of bulk cells was sequenced on an Illumina Hiseq X Ten platform with a 150 bp paired-end module. Cuffdiff (v2.2.1) was used to calculate FPKMs for the coding genes of each sample. Genes with corrected P values less than 0.05 and the absolute value of |log₂ (fold change)| >2 were assigned as significantly differentially expressed. The significantly differentially expressed genes were selected (also reported in the previous literature related to stem cell quiescence regulations) and showed in the heatmap. GO enrichment analysis of differentially expressed genes was implemented with Perl module (GO::TermFinder). R functions (q value) were used to test for statistical enrichment of differentially expressed genes among the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways. Gene set enrichment analysis (GSEA) was performed according to the instructions provided on the GSEA website (http://software.broadinstitute.org/gsea/).

Chromatin immunoprecipitation sequencing (ChIP-seq) was performed using Anti-H4K20me3 according to the manufacturer's instructions of Millipore EZ ChIP Kit (Millipore; 17-371). The DNA were constructed for the library and sequenced on an Illumina Hiseq 2500 platform. Peaks were called for aligned reads using MACS2. Differentiated enriched peaks were analyzed using differential peaks and notated using notate peaks.pl in the homer software. Heatmaps were created to present the differentiated enriched peaks, according to the peak enrichment value. Illustrative read coverage graphs of H4K20me3 patterns across candidate genes were analyzed by the Integrative Genomics Viewer.

Assay for transposase-accessible chromatin with high-throughput sequencing (ATAC-seq) was performed as reported previously (31). Briefly, nuclei were extracted from BCSCs^SETD4 and BCSCs^GFP, and the nuclei pellet was resuspended in the Tn5 transposase reaction mix. The transposition reaction was incubated at 37°C for 30 minutes. Libraries were purified using AMPure beads and then sequenced on an Illumina Hiseq X Ten platform. The data analysis methods were described in the ChIP-seq sections above.

For quantification, at least three experiments were analyzed using ImageJ software. All statistical analyses of the data were performed using means ± SD. For statistical comparison, we performed a one-tailed Student t test. The value of P < 0.05 was considered significant [P > 0.05 considered not significant (NS)] and the exact P value is stated in the figures.

All deep sequencing data that support the findings of this study have been deposited in the Gene Expression Omnibus (GEO). The accession number for the RNA-seq data of BCSCs, qBCSCs, and A-qBCSCs is GSE123810. The RNA-seq data of BCSCs^SETD4 and BCSCs^GFP have been deposited under the accession code GSE123842. The single cell RNA-seq data of BCSCs from solid tumors taken from mice and BCSCs from MCF7 cell lines are available under accession codes GSE124888 and GSE124887, respectively. ChIP-seq data have been deposited under accession code GSE123842. ATAC-seq data are available under accession codes GSE131586.

On the basis of previous studies (32, 33), we obtained BCSCs by isolating a population of CD44^high/CD24^low cells using FACS from luminal type (MCF7) and basal type (HCC1937) human breast cancer cell lines (left FACS plots in Fig. 1A), which expressed high levels of pluripotency markers, such as ALDH-1, Sox2, Oct4, Nanog and displayed abilities of tumorsphere formation and tumorigenesis in NOD/SCID mice (Supplementary Fig. S1A–S1D). PKH26 has been used for marking nondividing or quiescent cells in previous reports (10, 12, 14) and in the current study. A few cells with PKH26 label retention (PKH26⁺) were identified in the BCSC formed tumorspheres (Fig. 1A), indicating that they were in a nondividing or quiescent state and named as qBCSCs. The PKH26⁺ qBCSCs isolated from the tumorspheres (middle FACS plots in Fig. 1A) could be activated in a tumorsphere formation medium, named as activated qBCSCs (A-qBCSCs), and it was these that subsequently formed the tumorspheres (Fig. 1A). However, we did not observe tumorsphere formation or tumorigenesis in the PKH26⁻ population dissociated from the tumorspheres. We found that both the FACS-sorted qBCSCs and the qBCSCs in tumorspheres had very low expression levels of the proliferation markers Ki67 and PCNA (Fig. 1B–D), and very low phosphorylation levels of H3S10 (H3S10ph) and RbS807/S811 (RbS807/S811ph), in contrast to BCSCs and A-qBCSCs (Fig. 1E). Furthermore, FACS-sorted qBCSCs, BCSCs, and A-qBCSCs all exhibited CD44^high/CD24^low, ALDH-1^high, and similar levels of the pluripotency markers (Supplementary Fig. S1E–S1G). Importantly, A-qBCSCs were more capable of tumorsphere formation and tumorigenesis than BCSCs (Fig. 1F and G). However, no tumor formation was observed following injection of the same numbers of qBCSCs from the tumorspheres. In this study, we found that qBCSCs could be activated under conditions of tumorsphere formation medium and the addition of exosomes from the culture medium of MCF7 and HCC1937 accelerated the activation of qBCSCs, especially for tumorigenesis of A-qBCSCs in mice. These results indicate that PKH26⁺ cells in tumorspheres are qBCSCs and have a high tumorigenesis potential upon activation.

To characterize the molecular signatures of qBCSCs, RNA-seq of BCSCs, qBCSCs, and A-qBCSCs from the MCF7 cell lines were performed. The differentially expressed genes in qBCSCs are showed in the heatmap (Fig. 2A; Supplementary Fig. S2A). GSEA showed downregulated expression of genes involved in cell activation, proliferation, and signaling pathways of Wnt, TGFβ, Notch, and JAK-STAT3, and upregulated expression of genes in BMP, p53, BMI1, hedgehog, Brac1, and HES1 in qBCSC pathways, as compared with A-qBCSCs (Fig. 2B; Supplementary Fig. S2B). This pattern has previously been reported as a signature of quiescent stem cells (7, 8, 34–40). The differential expression of related genes in qBCSCs was also validated by qRT-PCR (Supplementary Fig. S2C). Gene Ontology (GO) analysis showed that the expression of specific genes up- or downregulated in qBCSCs were significantly enriched for GO terms linked to the regulation of chromatin stability, proliferation, differentiation, metabolism, and related signaling pathways (Supplementary Fig. S2D).

In our previous report, we found that SETD4 was expressed abundantly in the quiescent cells of Artemia diapause embryos (27). Similarly, we also found high expression levels of SETD4 in qBCSCs (PKH26⁺) as compared with PKH26⁻ cells in tumorspheres (Fig. 2C). To further confirm the specificity of SETD4 expression in qBCSCs, we analyzed SETD4 expression in various cells, including FACS-sorted BCSCs, qBCSCs, A-qBCSCs, and four cancer cell lines (T47D, MKN45, MCF7, HCC1937). The results showed that SETD4 was abundantly expressed only in qBCSCs, while its expression was very low in BCSCs and A-qBCSCs (Fig. 2D and E) and remained undetected in any of the four cancer cell lines or in the arrest state as triggered by starvation treatment (Supplementary Fig. S3A).

Because SETD4 was detected at a high level in qBCSCs, but at a low level in BCSCs, and to explore the function of SETD4 in the regulation of BCSC quiescence, GFP-fused SETD4 (GFP-SETD4) was overexpressed in BCSCs (BCSCs^SETD4; Supplementary Fig. S3B). We found that the capability of tumorsphere formation and the expression levels of Ki67, H3S10ph, and RbS807/S811ph in BCSCs were all inhibited by the overexpression of GFP-SETD4, but not inhibited in BCSCs that were overexpressing GFP (BCSCs^GFP; Fig. 2F and G; Supplementary Fig. S3C). To validate the function of overexpressed SETD4 in the regulation of BCSC quiescence, RNAi was performed to knock down the overexpressed SETD4 gene. The BCSCs^SETD4 treated with scramble siRNA were maintained in a quiescent state and could not divide to form tumorspheres. In contrast, those treated with siRNA of SETD4 produced tumorspheres in a similar manner to BCSCs^GFP (Fig. 2H). In addition, we found very low levels of H3S10ph and RbS807/S811ph and distinct lack of any Ki67 signal in BCSCs^SETD4 in contrast to the BCSCs^GFP (Supplementary Fig. S3D and S3E). These results indicate that SETD4 is required for maintenance of BCSCs quiescence.

We previously established that SETD4 catalyzes H4K20me3 during diapause formation and regulates cell quiescence in Artemia (27). Here, we found a specific enrichment of H4K20me3 in qBCSCs (PKH26⁺) in tumorsphere and in FACS-sorted qBCSCs, but not in BCSCs or A-qBCSCs (Fig. 3A–C). An in vitro histone methylation transferase assay showed that the level of H4K20me3 was enhanced upon the supplementation of GST-SETD4 (Supplementary Fig. S4A–S4C). We repeated the assay using four SETD4 mutations, none of the four mutants of SETD4 showed any methyltransferase activity on H4K20me3 (Supplementary Fig. S4D). Moreover, we observed that H4K20me3 was increased in BCSCs^SETD4 but not in BCSCs^GFP (Fig. 3D; Supplementary Fig. S4E), and this effect was eliminated by knockdown of the overexpressed GFP-SETD4 (Fig. 3E; Supplementary Fig. S4F). Likewise, SETD4 overexpression also induced the increase of H4K20me3 in MCF7 and HCC1937 cell lines (Supplementary Fig. S4G). These results indicate that SETD4 functions in the specific catalysis of H4K20me3 in the qBCSCs.

Analysis of TEM revealed a striking increase of condensed heterochromatin in the nuclei of qBCSCs (Fig. 3F) and BCSCs^SETD4 (Fig. 3G), in contrast to that in BCSCs, A-qBCSCs and BCSCs^GFP. In addition, we observed increased H4K20me3 (a marker for cHC) and low levels of H3K9ac (a marker for euchromatin) in qBCSCs (Fig. 3C; Supplementary Fig. S5A) and BCSCs^SETD4 (Supplementary Fig. S5B–S5D), in contrast to those in BCSCs, A-qBCSCs, and BCSCs^GFP. Here, we found that HP1-α, which plays an essential role in heterochromatin formation (41), was also enriched in qBCSCs and BCSCs^SETD4 (Fig. 3C; Supplementary Fig. S5A, S5B and S5E). However, we did not observe any significant differences in levels of the fHC markers of H3K9me3 and H3K27me3 between BCSCs, qBCSCs, and A-qBCSCs or between BCSCs^SETD4 and BCSCs^GFP (Supplementary Fig. S5A, S5B, and S5F–S5I). These results indicate that qBCSCs and BCSCs^SETD4 contain more cHC and less euchromatin than BCSCs, A-qBCSCs, and BCSCs^GFP. Thus, we conclude that SETD4 controls BCSCs quiescence by cHC formation via H4K20me3 catalysis.

In epigenetic regulation, local gene expression is influenced through modifications of chromatin that recruit transcription factors that can either activate or repress gene transcription (42). To explore epigenetic regulation by H4K20me3 in BCSCs quiescence, we performed ChIP-seq in BCSCs^SETD4 and BCSCs^GFP from the MCF7 cell line. We found that in BCSCs^SETD4, H4K20me3 was distributed on all 23 chromosomes (Supplementary Fig. S6A). Interestingly, quiescent BCSCs^SETD4 showed marked enrichment of H4K20me3 modifications relative to BCSCs^GFP, a result compatible with widespread repression of gene expression. The overall pattern of H4K20me3 modifications for the unique signature genes of BCSCs^SETD4 (vs. BCSCs^GFP) is shown as a heatmap (Supplementary Fig. S6B).

Illustrative read coverage graphs of H4K20me3 patterns across candidate genes showed that in BCSCs^SETD4, H4K20me3 was typically enriched at the promoter regions and negatively correlated with the expression of the genes of MYC, WNT1, EEF1A1, IGF1, SMAD4, but was decreased at the promoter region and upregulated the expression of TP53 gene (Fig. 3H). In addition, ATAC-seq was performed. We found lower open chromatin enriched peaks in BCSCs^SETD4 than that in BCSCs^GFP, in which BCSCs^SETD4 showed more widespread repressions of gene expressions relative to BCSCs^GFP (Supplementary Fig. S6A and S6B). Notably, we found weak ATAC-seq signals at the MYC, WNT1, EEF1A1, IGF1, and SMAD4 promoters in BCSCs^SETD4, which likely explains their repressed gene expression status (Fig. 3I); however, the TP53 promoter exhibited stronger ATAC-seq signals in BCSCs^SETD4 than BCSCs^GFP, suggesting the upregulated gene expression status of TP53 in BCSCs^SETD4. The results of ATAC-seq indicated a lower chromatin accessibility and widespread repression of gene expression in BCSCs^SETD4, which were also consistent with the results of our prior H4K20me3 ChIP-seq analysis (Fig. 3H).

We also compared the gene expression profiles of BCSCs^SETD4 and BCSCs^GFP using bulk RNA sequencing analysis (Supplementary Fig. S6C). Consistent with our analysis of GO terms and qRT-PCR results, KEGG pathway analysis revealed that genes correlated with cell activation and proliferation were downregulated in BCSCs^SETD4 (Supplementary Fig. S6D and S6E). Our data show that SETD4-overexpressed BCSCs (BCSCs^SETD4) are similar to qBCSCs in terms of its global gene expression pattern based on the analysis of transcriptome (Fig. 2A and B; Supplementary Fig. S2A–S2D; Supplementary Fig. S6C–S6E). Thus, SETD4 promotes cHC formation in qBCSCs and epigenetically regulates the expression of a set of genes by catalyzing the H4K20me3 located at the promoter regions.

To investigate BCSCs' resistance to chemoradiotherapy, FACS-sorted BCSCs from MCF7 and HCC1937 cell lines were treated with drugs and radiation. Interestingly, a few cells (average 3.28% and 2.92% of MCF7-BCSCs, average 3.64% and 3.38% of HCC1937-BCSCs) were survived after the drug and radiation treatments, respectively, as determined by trypan blue staining (Fig. 4A). Importantly, we found that all tested surviving cells were SETD4 positive and Ki67 negative (Fig. 4A and B), indicating that these surviving BCSCs were in quiescent state or SETD4-defined qBCSCs. Analysis of TEM revealed that they contained more condensed heterochromatin (Fig. 4C). In addition, Western blot analysis revealed that these SETD4-defined qBCSCs had also abundant H4K20me3, HP1-α, and low amounts of H3K9ac, indicating that they contain more cHC and less euchromatin than did the BCSCs before treatments (Supplementary Fig. S7A). To confirm the resistance of SETD4-defined qBCSCs to chemoradiotherapy, we performed drug and radiation treatments on FACS-sorted BCSCs, qBCSCs, and A-qBCSCs. As expected, FACS-sorted qBCSCs had high survival rates (55.56% and 67.02%) after drug and radiation treatments compared with BCSCs (2.82% and 5.52%) and A-qBCSCs (0.42% and 1.03%; Fig. 4D). Moreover, overexpressed SETD4 enabled BCSCs to survive with resistance to both treatments of chemical drugs and radiation, whereas BCSCs^GFP were all sensitive to the treatments and exhibited widespread cell death (Fig. 4E). However, BCSCs^SETD4 treated with siRNA to SETD4 lost chemoradiotherapy resistance and died, but BCSCs^SETD4 treated with scrambled siRNA maintained their quiescent state and retained resistance to the treatments (Supplementary Fig. S7B). These results indicated that SETD4-defined qBCSCs were resistant to chemoradiotherapy.

To reveal the cellular response of SETD4-defined qBCSCs to chemoradiotherapy, we also analyzed the activities of autophagy, apoptosis, and DNA damage after chemoradiotherapy. As shown in Supplementary Fig. S7C–S7E, no signals were detected for LC3B, activated caspase-3, and γ-H2A in SETD4-defined qBCSCs, while high levels were observed in BCSCs and A-qBCSCs after drug and radiation treatment. The results indicate that SETD4-defined qBCSCs are resistant to chemoradiotherapy and display no evident cellular damage beyond it, in contrast to BCSCs and A-qBCSCs.

We then investigated whether SETD4-defined qBCSCs were able to cause tumor relapse in MCF7 and HCC1937 cell derived xenografts (CDX) after development of resistance to chemotherapy. Tumors were generated in NOD/SCID mice and had reached their minimum size after chemotherapeutic treatment. Tumors had then relapsed after chemotherapy's completion (Fig. 4F and G). Upon the completion of chemotherapy treatment, high percentage of surviving tumor cells (34.82% in MCF7-CDXs and 32.75% in HCC1937-CDXs) were SETD4⁺ in contrast to the situation of prechemotherapy treatment tumors (1.08% in MCF7-CDXs and 1.32% in HCC1937-CDXs) and the relapsed tumors 2 weeks beyond treatment (2.71% in MCF7-CDXs and 2.19% in HCC1937-CDXs; Fig. 4H). Analysis of Western blot also showed significantly high expression level of SETD4 in the solid tumors after chemotherapy (Fig. 4I). Using immunofluorescence analysis, we found that SETD4⁺ cells in the tumors were CD44^high/CD24^low (Fig. 4J; Supplementary Fig. S8A) and had a low level of Ki67 expression (Fig. 4K; Supplementary Fig. S8B). These results suggest that SETD4-defined qBCSCs in tumor are able to cause relapse beyond resistance to chemoradiotherapy. In addition, the chemotherapy-resistant SETD4⁺ qBCSCs in post-chemo treatment tumors had a high level of HP1-α expression and the enrichment of H4K30me3, but a low abundance of H3K9ac, indicating that they contained more cHC, but less euchromatin than that in pre-chemo treatment tumor cells or relapsed tumor cells (Supplementary Fig. S8C–S8E).

Interestingly, we found that these SETD4-defined qBCSCs could survive for more than 2 months in the presence of chemical drugs, and sustained themselves and formed typical tumorspheres after the drugs were removed in vitro (Fig. 5A). To address how SETD4-defined qBCSCs balance self-renewal during tumorigenesis and relapse, two independent experiments were performed (Fig. 5B). In the first experiment, a SETD4-defined qBCSC divided into two cells after approximately 40 hours activation. One cell was SETD4⁺/EdU⁻/Ki67⁻ and the other was SETD4⁻/EdU⁺/Ki67⁺, indicating that the SETD4⁺ qBCSC sustained itself in a quiescent state with two original DNA strands by an asymmetric division, while producing a daughter cell. The daughter cell subsequently divided symmetrically into two SETD4⁻/EdU⁺/Ki67⁺ cells under conditions without EdU, indicating that the daughter cell contains two newly synthesized DNA strands. In the second experiment, EdU was added only after the SETD4⁺ qBCSC had already divided into three cells. This resulted in the generation of one SETD4⁺/EdU⁻/Ki67⁻ cell and four SETD4⁻/EdU⁺/Ki67⁺ cells that could proliferate to produce a population of cells, indicating that the SETD4⁺ qBCSC is always maintained in a quiescent state during the proliferation process. We conclude that SETD4-defined qBCSCs are able to sustain themselves by asymmetric division and concurrently produce a daughter cell, which then proliferates into a population of cancer cells by symmetric divisions.

Recent advances in single-cell gene expression analysis offers an opportunity to greatly improve the identification and classification of different cell types within a heterogeneous cell population (43, 44). We performed single-cell profiling of 3,765 BCSCs from tumors of MCF7-CDXs and identified clusters using t-SNE analysis. BCSCs were distributed in seven clearly delineated clusters, in which all clusters showed characteristic CD44^high/CD24^low expression (Fig. 6A). Importantly, the BCSCs population contained 5.42% SETD4⁺ qBCSCs (204 cells) in the BCSC population (3,765 cells) that had partitioned into a cluster (cluster 7) as a distinct cell type with high cellular component homogeneity. In addition, sequence data from 3,037 single BCSCs FACS-sorted from tumors of MCF7-CDXs showed that, after drug treatment, 93.45% of the surviving BCSCs were in the defined cluster of SETD4⁺ qBCSCs, and 6.55% were SETD4⁻ and spread into other clusters of BCSCs (Fig. 6B).

Differential gene expression analysis identified molecular signatures for each cell type and provided a comprehensive genetic module repertoire for the BCSCs population (Fig. 6C). Our t-SNE analysis showed that in expression of related genes, some become enhanced (such as HES1, TP53, and BMP2) and others suppressed (such as MKI67, TGFBR3, and WNT10A) in the SETD4⁺ qBCSCs cluster (Fig. 6D). This matched the expression known in quiescent cell types (7, 8, 34–40). We also identified new consensus markers, including SETD4, ANGPTL4 (an inhibitor of tumor angiogenesis; ref. 45), and CA9 (a transmembrane protein of carbonic anhydrase; Fig. 6E; ref. 46). Similar results were also observed using t-SNE analysis of single-cell sequences on 3,575 BCSCs and 1,175 drug resistant qBCSCs derived from the MCF7 cell line (Supplementary Fig. S9A–S9E). Taken together, our results show that SETD4-defined qBCSCs present as a small population (approximately 5%) in BCSCs, representing a distinct cell type within heterogeneous BCSCs and play critical functions in resistance to chemoradiotherapy and relapse.

We next focused on the role of SETD4-defined qBCSCs in tumorigenesis, chemotherapy resistant and relapse in clinical breast cancer patients. We obtained solid tumors that had been removed from patients with breast cancer, dissociated cancerous cells from them, and subjected these cells to treatment with chemotherapy drugs and radiation. We found that all surviving cells were SETD4⁺, CD44^high/CD24^low and Ki67⁻ (Fig. 7A), suggesting that they were SETD4-defined qBCSCs. Furthermore, the high levels of H4K20me3 and HP1-α and low levels of H3K9ac indicated that these SETD4-defined qBCSCs carried higher contents of cHC and lower contents of euchromatin (Supplementary Fig. S10A). Subsequently, these SETD4-defined qBCSCs could be activated in tumorsphere formation medium for 20 hours and then transplanted into NOD/SCID mice. Interestingly, 8 weeks after injection of only 10 of these cells, tumors had all occurred in all such NOD/SCID mice (Fig. 7B). However, we did not observe tumorigenesis when the same numbers of BCSCs from the same tumors of patients with breast cancer were injected. This indicates that SETD4-defined qBCSCs have roles as the originators of tumor and relapse.

Analysis of clinical samples showed that the ratio of SETD4-defined qBCSCs in solid tumors from advanced stage (stage III) patients was more than 3-fold higher than in tumors obtained from early stage (stages I and II) patients (Fig. 7C; Supplementary Fig. S10B). Moreover, the ratio of SETD4-defined qBCSCs was more than 3-fold higher in solid tumors obtained from patients who had received chemotherapy treatment than in tumors from patients who had not received treatment prior to surgery (Fig. 7D; Supplementary Fig. S10C). These results suggest that the presence of SETD4-defined qBCSCs may correlate with malignancy and chemotherapy resistance in clinical breast cancer patients. We next assessed whether SETD4-defined quiescent CSCs were present in other types of cancer. We obtained solid tumors that had been removed from patients with gastric, cervical, ovarian, liver, and lung cancers (Supplementary Fig. S10D), disaggregated the cells, and subjected them to the chemical drug treatments. We found that all chemotherapy-resistant cells tested from each of the five solid tumors were SETD4⁺, Ki67⁻ and high levels of the cancer stem cell marker (Fig. 7E), indicating that SETD4-defined quiescent CSCs were also present in all examined cancer types. Our results are summarized in Fig. 7F.

Taken together, the results from our previous study on diapause cysts of Artemia (27) and our current work on six types of cancer reveal an evolutionarily conserved mechanism of cellular quiescence epigenetically controlled by SETD4. Although Suv4-20h2 has previously been reported as responsible for catalyzing H4K20me3 in mouse and human fibroblasts (47, 48), we did not observe any significant differences in Suv4-20h2 expression in response to H4K20me3 enrichment in qBCSCs (Supplementary Fig. S11A). We also did not observe SETD4 expression in response to the increase in H4K20me3 in the quiescence of mouse embryonic fibroblasts induced by contact-inhibition, in which Suv4-20h2 catalyzes H4K20me3 (Supplementary Fig. S11B). It seems that both Suv4-20h2 and SETD4 are able to catalyze H4K20me3, but they function in different types of cell, although SETD4 has been studied in the regulation of gene expression and cell proliferation in several cell lines (24, 49). We propose that SETD4 is a determinant of quiescence specifically in CSCs, but not in cancer cells. Thus, SETD4 can be applied to define qCSCs occurring within the large heterogeneity of tumor cells and even, more specifically within the wider CSCs population.

In this study, we did not find any tumor occurrence by injection of qBCSCs during the period of 6 months. Our results indicate that the activation of qBCSCs is required for tumor occurrence. On the basis of evidence of a strong correlation between SETD4-defined qBCSCs and malignancy and chemoresistance in patients with breast cancer, it may be possible to use the SETD4 and/or SETD4-defined qBCSCs as important indicators for assessing the grade of malignancy and likelihood of tumor recurrence in a clinical setting. In the current study, we found that the ability of chemoradiotherapy resistance disappeared completely after SETD4-defined qBCSCs were activated, and therefore activating BCSCs may enable their eradication by subsequent treatments with standard chemoradiotherapy. Our findings suggest that SETD4 and/or SETD4-defined qCSCs could be also used as key targets in clinical treatment for a wide range of cancers.

No potential conflicts of interest were disclosed.

Conception and design: S. Ye, W.-J. Yang

Development of methodology: S. Ye, Y.-F. Ding, W.-H. Jia, X.-L. Liu, J.-Y. Feng

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Q. Zhu, S.-L. Cai, Y.-S. Yang, Q.-Y. Lu, X.-T. Huang, Y.-H. Wang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Ye, Y.-F. Ding, J.-S. Yang

Writing, review, and/or revision of the manuscript: S. Ye, W.-J. Yang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S.-N. Jia, G.-P. Ding, J.-J. Zhou, Y.-D. Chen

Study supervision: W.-J. Yang

We thank S. Zhang for the help with the laser microscopy, X. Song for flow cytometry analysis, J. Li for transmission electron microscopy, Y. Xu for mass spectrometer analysis, and X. Xu for mouse husbandry support. We would like to express our sincere gratitude to Mr. C. Wood for critical reading of the manuscript. This work was supported by the National Major Research and Development Project (2016YFA0101201) and the National Natural Science Foundation of China (31730084).

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.