Canonical PRC2 function is essential for mammary gland development and affects chromatin compaction in mammary organoids

Distinct transcriptional states are maintained through organization of chromatin, resulting from the sum of numerous repressive and active histone modifications, into tightly packaged heterochromatin versus more accessible euchromatin. Polycomb repressive complex 2 (PRC2) is the main mammalian complex responsible for histone 3 lysine 27 trimethylation (H3K27me3) and is integral to chromatin organization. Using in vitro and in vivo studies, we show that deletion of Suz12, a core component of all PRC2 complexes, results in loss of H3K27me3 and H3K27 dimethylation (H3K27me2), completely blocks normal mammary gland development, and profoundly curtails progenitor activity in 3D organoid cultures. Through the application of mammary organoids to bypass the severe phenotype associated with Suz12 loss in vivo, we have explored gene expression and chromatin structure in wild-type and Suz12-deleted basal-derived organoids. Analysis of organoids led to the identification of lineage-specific changes in gene expression and chromatin structure, inferring cell type–specific PRC2-mediated gene silencing of the chromatin state. These expression changes were accompanied by cell cycle arrest but not lineage infidelity. Together, these data indicate that canonical PRC2 function is essential for development of the mammary gland through the repression of alternate transcription programs and maintenance of chromatin states.

Distinct transcriptional states are maintained through organization of chromatin, resulting from the sum of numerous repressive and active histone modifications, into tightly packaged heterochromatin versus more accessible euchromatin. Polycomb repressive complex 2 (PRC2) is the main mammalian complex responsible for histone 3 lysine 27 trimethylation (H3K27me3) and is integral to chromatin organization. Using in vitro and in vivo studies, we show that deletion of Suz12, a core component of all PRC2 complexes, results in loss of H3K27me3 and H3K27 dimethylation (H3K27me2), completely blocks normal mammary gland development, and profoundly curtails progenitor activity in 3D organoid cultures. Through the application of mammary organoids to bypass the severe phenotype associated with Suz12 loss in vivo, we have explored gene expression and chromatin structure in wildtype and Suz12-deleted basal-derived organoids. Analysis of organoids led to the identification of lineage-specific changes in gene expression and chromatin structure, inferring cell type-specific PRC2-mediated gene silencing of the chromatin state. These expression changes were accompanied by cell cycle arrest but not lineage infidelity. Together, these data indicate that canonical PRC2 function is essential for development of the mammary gland through the repression of alternate transcription programs and maintenance of chromatin states. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111

Introduction
A central question in biology is how different cell types maintain distinct cell fates despite containing the same genetic material. Organization of DNA into open or closed chromatin states by posttranslational modifications (PTMs) of histones has emerged as a critical mechanism underpinning cell diversity and reflecting lineage-specific gene expression, developmental programs, or disease processes [1]. The highly conserved Polycomb repressive complex 2 (PRC2), which catalyzes trimethylation of histone 3 on lysine 27 (H3K27me3), is associated with global gene repression and suppression of alternative differentiation programs. The canonical PRC2 complex is composed of the intimately associated core proteins histone methyltransferase Enhancer of Zeste homolog 2 (Ezh2) or an alternative related subunit Ezh1 [2], embryonic ectoderm development (Eed), Suppressor of Zeste 12 protein homolog (Suz12), and histonebinding protein accessory proteins [3]. Upon recruitment of PRC2 to chromatin, Ezh2/Ezh1 deposits the H3K27me3 mark associated with chromatin compaction [3]. PRC2 is required for deposition of H3K27me3 and for maintenance of this PTM upon cell division [4]. Additionally, the requirement of PRC2 activity for H3K27 mono-and dimethylation (H3K27me1 and H3K27me2) remains unclear [5]. Early studies showed that both Suz12 and Eed are nonredundant and essential for a functional PRC2 complex [6]. Deletion of Suz12 or Eed resulted in elevated expression of Hox genes in Drosophila and mammalian cells and marked increases in gene networks that control developmental lineages [7,8] because of loss of PRC2 integrity and H3K27me3-mediated repression [9]. In contrast to Eed and Suz12, Ezh2 function can be compensated, partially [10] or completely [11], by Ezh1. Nevertheless, like mice lacking Suz12 or Eed, Ezh2-deficient mice are not viable and die during early implantation stages [6].
Members of PRC2-in particular, Ezh2-are often found dysregulated in human cancers. In breast cancer, Ezh2 overexpression is associated with aggressive breast cancers and poor prognosis and inversely correlated with H3K27me3 expression [12]. It remains unclear whether Ezh2 overexpression is a consequence or cause of breast oncogenesis [13]. High levels of Ezh2 may not be sufficient to induce mammary tumors in mice [14], suggesting additional driver mutations are required. To further understand the basis of this dysregulation in cancer, it is imperative to determine the normal functions of PRC2 and Ezh2 in maintaining gene expression programs in the mammary gland.
The mammary gland in both humans and mice is a bilayered structure composed of two cellular lineages: an inner luminal layer and an outer myoepithelial layer that contacts the basement membrane [15]. There is increasing evidence for a differentiation hierarchy composed of stem cells, committed progenitors, and mature epithelial cells [15]. In the steady state, mouse mammary stem cell (MaSC)/basal, luminal progenitor, and mature luminal cell subsets display distinct patterns of H3K27me3 [16]. The MaSC/basal subset demonstrates the lowest levels of H3K27me3. Higher levels of H3K27me3 correlate with reduced gene expression and increase upon cell specialization [16]. These data support a model whereby mammary epithelial cell (MEC) differentiation requires narrowing of transcriptional programs and suppression of alternate cell fates. Accordingly, genetic ablation or knock-down of Ezh2 in the mammary gland resulted in a developmental delay [16,17] but did not entirely prevent mammary gland development. This is likely due to residual H3K27 methylation, presumably established through compensatory methyltransferase activity of Ezh1. It is therefore probable that studies to date have underestimated the importance of PRC2 in directing mammary cell fate.
To reexamine the contribution of the canonical PRC2 complex to mammary gland development, we deleted Suz12 in vivo and in mammary organoids [18]. Here, we show a nonredundant function for Suz12 in mammary progenitor cells due to loss of PRC2 function. Similar findings were made upon deletion of Eed. Through the application of assay for transposaseaccessible chromatin using sequencing (ATAC-seq) to probe chromatin accessibility in Suz12deficient organoid cultures, we have identified regions of PRC2-dependent chromatin compaction and consequent changes in gene expression. Interestingly, the chromatin state in basal-derived organoids was reliant on PRC2 function, and Suz12 deletion led to gene derepression, thus highlighting a crucial role for PRC2 in the maintenance of chromatin states within the mammary epithelial hierarchy.

Suz12 or Eed deletion results in marked delay of mammary ductal outgrowth and perinatal lethality
Mammary gland development proceeds through distinct phases that include puberty and cycles of pregnancy, lactation, and involution. Suz12 expression, like Ezh2 [16,17], was detected at all stages of mammary gland development (S1A Fig) but was particularly high during puberty (4-6 week old mice). To examine the role of Suz12 in the mammary gland, we crossed mice bearing floxed Suz12 alleles with Cre transgenic mice that express Cre under control of the mouse mammary tumor virus promoter (MMTVcre). Genotyping of offspring revealed that two-thirds of MMTVcre T/+ Suz12 fl/f mice did not survive to weaning (S1 Data), and examination of newborn pups revealed abnormal lung development (S1B Fig), likely due to activity of MMTVcre in this tissue [19], and consistent with the reported role for PRC2/ Ezh2 in lung [20].
MMTVcre T/+ Suz12 f/f mice that survived beyond birth appeared normal and did not differ from wild-type (Wt) mice with respect to bodyweight (S1C Fig). Examination of whole mounts and histological sections of mammary glands from MMTVcre T/+ Suz12 f/f mice during puberty revealed a heterogeneous phenotype. Some glands were indistinguishable from Wt or heterozygous littermates, while others comprised a small ductal tree, characteristic of the rudimentary ductal tree found in newborn mice (n = 4) (Fig 1A and S1D Fig). These small but otherwise normal mammary glands indicate that gene deletion leads to severe impairment of ductal growth during puberty [21,22]. Notably, Suz12 mRNA expression in Representative images of whole mounts, HE stained sections (A), and immunohistochemical staining for Suz12, Ezh2, and H3K27me3 (B) in mammary glands from 6 week old MMTVcre T/+ Suz12 f/f and control mice. Arrows in whole mounts indicate the leading edge of the mammary epithelium. Isotype-control stained sections are shown in the inset. Scale bars: 4 mm (whole mounts, HE) and 50 μm (immunohistochemistry MMTVcre T/+ Suz12 f/f and MMTVcre T/+ Suz12 f/+ glands was found to be comparable (S1E Fig), demonstrating that these ductal structures were derived from epithelial cells in which Suz12 had not been deleted. Moreover, protein expression of Ezh2, Suz12, and H3K27me3 was retained in sections from 5-6 week old MMTVcre T/+ Suz12 f/f mammary glands, indicative of strong selection against Suz12-deleted cells and retention of a functional PRC2 complex in the mammary epithelium ( Fig 1B). Cell fate was also retained as assessed by immunostaining for estrogen receptor (ER), progesterone receptor (PR), and forkhead box A1 (Foxa1) (S1F Fig).
We next determined whether deletion of the second core subunit of canonical PRC2 member Eed produced a similar phenotype to deletion of Suz12. As seen with Suz12 deletion, conditional deletion of Eed with MMTVcre resulted in perinatal lethality of mice (S1 Data), and surviving mice appeared normal, except for a pronounced absence or delay in growth of the ductal tree (S2A Fig). Similar to Suz12 deletion, Eed mRNA expression was found to be indistinguishable between mammary epithelium from MMTVcre T/+ Eed f/f and Wt mice (S2B Fig). In addition, Eed, Ezh2, and H3K27me3 proteins were detected by immunofluorescence (IF) and immunohistochemistry (IHC) at comparable levels to those in Wt mammary glands (S2C and S2D Fig). Taken together, these results suggest that canonical PRC2 complexes are critical for the proliferation and/or survival of MECs, and cells deleted for PRC2 function cannot contribute to the developing mammary gland.

Suz12 is required for the integrity of PRC2 and histone H3K27me2/me3 in MECs
To investigate which MEC subsets were affected by Suz12 loss, we acutely deleted Suz12 using the inducible and conditional Rosa26-creERT2 (R26creERT2) mouse model. R26creERT2 KI/+ Suz12 f/f mice injected with tamoxifen die within 2 weeks because of hematopoietic failure (personal communication, S. Lee to E. Michalak), thus precluding the use of this model for in vivo studies. Basal/MaSC-enriched (basal, CD29 hi CD24 + ), committed luminal progenitor (CD29 lo CD24 + CD14 + ), and mature luminal (CD29 lo CD24 + CD14 − ) cell subsets were sorted from R26creERT2 KI/+ Suz12 f/f and control mice and subjected to in vitro colonyforming assays to determine the effect of Suz12 deletion on the activity of mammary progenitor cells following induction by 4-hydroxytamoxifen (4OHT). Consistent with expression of Suz12 in all mammary epithelial subsets (Fig 2A), we observed fewer colonies in 2D colony forming assays on irradiated feeder cells ( Fig 2B) upon induction of Suz12 deletion. In contrast to the hematopoietic system, in which loss of one allele of Suz12 [14] or Eed [23] enhances the activity of stem cells, deletion of one Suz12 allele did not affect the clonogenic activity of progenitor cells (Fig 2B). Unsorted MECs efficiently deleted Suz12 and yielded sufficient protein for western blot analysis ( Fig 2C). As expected, Ezh2 protein was completely lost upon Suz12 deletion, while Ezh2 mRNA levels were not reduced (S3A Fig), suggesting rapid Ezh2 protein degradation leads to the absence of functional PRC2 upon Suz12 deletion. This is supported by the loss of H3K27me2 and H3K27me3 that accompanies Suz12 deletion ( Fig 2C). Conversely, expression of the active H3K27me1 mark [4] did not change. No change in cleaved poly ADP ribose polymerase (PARP) was detected, suggesting that Suz12-deleted MECs do not undergo appreciable apoptosis. However, we observed increased expression of cyclin-dependent kinase p16 and p19 alternate reading frame (p19Arf), consistent with a known role for Ezh2/PRC2 in regulating cdkn2a gene expression (S3B Fig). These data suggest that Suz12 exerts an essential function in the mammary gland through maintenance of functional tri-as well as dimethylation of target loci.

Suz12 deletion diminishes mammary organoid proliferation
To develop a system more amenable to molecular studies, we employed a 3D mammary organoid system, in which organoids grown from single cells in defined medium recapitulate many features of mammary tissue architecture and function [18]. Single basal or luminal progenitor cells sorted from R26creERT2 KI/+ Suz12 f/f mice were plated, and deletion of Suz12 was induced by addition of 4OHT on day 1. Suz12 loss resulted in diminished numbers and smaller basaland luminal progenitor-derived organoids over a 2 week culture period. Notably, PCR analysis of the resulting organoids revealed that selection against cre-mediated excision of both floxed alleles of Suz12 had occurred (S3C Fig). To circumvent this issue, small organoids were allowed to form before induction of deletion. Indeed, addition of 4OHT on day 4 resulted in smaller cystic-like organoids (Fig 3A and 3B There were no notable differences in cleaved caspase 3 (CC3) in 2 week old Suz12-deleted organoids, but rather, a striking decrease in proliferative (Ki67 + ) cells was evident ( Fig 3D). Consistent with the idea that cells lacking Suz12 are nonproliferative, repassaging resulted in the expansion of rare cells that had escaped Suz12 deletion (S3F and S3G Fig). Together, these data suggest that loss of Suz12 reduces the fitness and proliferation of mammary organoids.

Suz12 loss results in changes to chromatin accessibility and gene transcription
To determine potential target genes that are dependent on Suz12 and therefore canonical PRC2 function, we performed RNA sequencing (RNA-seq) analysis of Suz12 Wt or Suz12deleted basal-and luminal-derived mammary organoids grown from R26creERT2 KI/+ Suz12 f/f mice. Suz12 was significantly down-regulated in both populations ( To explore whether the changes in gene expression accompanying loss of Suz12 reflect alterations in chromatin compaction associated with loss of repressive H3K27me2/me3 marks, we performed global mapping of chromatin accessibility using ATAC-seq. Differential accessibility analysis identified 2,767 windows, of which 2,415 (87.3%) were more accessible upon Suz12 deletion (Fig 4B and S3 Data). Further, 1,377 of these windows overlapped with the transcription start site or gene body of 766 independent genes in Suz12-deleted basal organoids (referred to as differentially accessible [DA] genes from hereon). Nevertheless, the frequency of reads mapping to nucleosome-free regions versus mononucleosomes was not significantly altered. We observed only minor changes to insert length, suggesting no gross changes in nucleosome occupancy (S5A and S5B Fig), and a comparison of GC enrichment indicated a minor increased frequency of highly GC-rich regions, usually found at regulatory elements within promoters (Fig 4C). Moreover, model-based analysis of ChIP-seq (MACS) peaks analysis showed a small number of unique peaks upon Suz12 deletion ( Fig 4D). Next, we compared the DA windows identified by ATAC-seq with gene expression from their associated gene to determine if changes in chromatin compaction elicited by Suz12 deletion were sufficient to induce changes in gene transcription. A strong correlation was evident between open chromatin and transcriptional up-regulation in Suz12-deleted basal-derived organoids ( Fig 4E).
However, analysis of Wt organoids did not reveal a correlation between genes assigned from ATAC-seq analysis and changes in their expression. Almost one-third (28%) of genes that became more accessible upon Suz12 deletion were significantly up-regulated by RNA-seq ( Fig 4F). These 218 common genes included Cdkn2a (S6A Fig and S4 Data), a known target of Comparison of the Suz12-deleted gene expression signatures derived for both basal and luminal organoids showed that they were most similar to claudin-low and normal-like breast tumors (S7G Fig), similar to that observed for the basal/MaSC cell signature [25]. Likewise, the gene expression profile of Ezh2-deleted MECs was most concordant with claudin-low breast tumors [16]. The striking similarity between gene expression signatures of organoids lacking key PRC2 complex genes and claudin-low breast tumors-which exhibit a metaplastic, clinically aggressive phenotype in patients-suggests that PRC2 may play a role in tumor phenotype and behavior.

Discussion
Using in vivo and in vitro studies, we show that deletion of a core component (either Suz12 or Eed) of the canonical PRC2 complexes leads to a complete block in mammary gland development in vivo and markedly curtails progenitor cell activity in vitro. Strikingly, Suz12-or Eeddeleted cells were strongly selected against and could not contribute to the developing mammary gland. This resulted in very small glands consisting of Wt cells. These findings are consistent with previous reports for nonredundant genes that are required for stem cell function [26]. Assays using primary MECs and organoids suggest that this is due in part to de-repression of cdkn2a, resulting in increased p16 and p19 expression and reduced proliferation owing to cell cycle arrest. Thus, previous studies [16,17] have underestimated the importance of PRC2 in governing mammary progenitor activity and differentiation, owing to some functional redundancy between Ezh1 and Ezh2. Moreover, these results are consistent with PRC2 promoting mammary epithelial expansion, rather than inhibiting it [27]. By exploiting mammary organoids to bypass the severe developmental phenotype associated with loss of Suz12, we examined changes in chromatin structure in the presence and absence of PRC2 and correlated these with gene expression changes.
PRC2 complexes are reported to colocalize with H3K27me3 on the promoters of around 10% of all genes [5]. Upon Suz12 deletion, we observed comparable changes in gene expression (7%-11%) in both basal-and luminal-derived organoids. There was significant overlap between the up-regulated genes in basal-and luminal-derived organoids, and the magnitude of change was highly correlated. Assessment of chromatin accessibility with ATAC-seq indicated that the chromatin of basal-derived organoids is more open upon Suz12 loss. Recent findings in human breast epithelial subsets [28] predict that bivalent promoters and primed enhancers would be the most affected by loss of H3K27me3 in Suz12-deleted basal-derived organoids. Since we did not observe a significant increase in GC content of DA reads identified by ATAC-seq, the mechanism is most likely to involve enhancers. Coupling RNA-seq with ATAC-seq analysis in basal-derived organoids revealed that deletion of Suz12 led to more than one quarter of genes becoming significantly transcriptionally active coincident with more accessible chromatin. We have previously shown that the MaSC/basal subset demonstrates the lowest levels of H3K27me3 across transcriptional start sites (TSSs) [16] while also producing slightly less RNA overall [29]. This observation supports the notion that this gene subset marked by H3K27me3 is transcriptionally regulated through H3K27me3-mediated repression accompanied by chromatin compaction and is dependent on canonical PRC2 function. Our results indicate that open chromatin is a good predictor of gene activation in basal-derived organoids. In the luminal compartment, there is an increased level of H3K27me3 [16], while these cells are more transcriptionally active overall [29]. In luminal-derived organoids, loss of PRC2 resulted in increased chromatin accessibility and an increase in reads mapping to regions with high GC content. In contrast to basal-derived organoids, the low correlation with gene expression in luminal-derived organoids suggests that additional mechanisms beyond chromatin accessibility serve to ensure proper gene repression. This may be due to DNA methylation or the combination of additional activating and repressive histone marks that maintain gene repression in this more committed cell type, as observed for human luminal cells [28].
Our studies indicate that PRC2 is responsible for both H3K27me3 and H3K27me2 marks in MECs, as evidenced by their loss in Suz12-deleted cells. Notably, we show here that the absence of these repressive marks through deletion of a single protein results in large changes in gene expression, consistent with the notion that chromatin states are the sum of a limited repertoire of PTM combinations on histone tails. While these data do not exclude noncanonical functions in the presence of a functional PRC2 complex [30], they suggest that Suz12 is a limiting factor for expression of Ezh2 protein and thus a functional PRC2 complex in the mammary epithelium.
In the context of breast cancer, our data indicate that Suz12-deleted organoids, like Ezh2null basal cells [16], displayed similarities with signatures of claudin-low breast cancers, which are thought to arise from MaSCs. While Ezh2 up-regulation was initially associated with aggressive breast cancers, several studies now indicate that Ezh2 overexpression may be a consequence rather than a cause of breast cancer [13]. Indeed, deletion of Ezh2 accelerated tumors in a mouse model of Brca1-deleted breast cancer [31] and in a breast cancer model of Notch activation [13]. Additionally, the status of the cdkn2a locus of an individual tumor might predict its response to Ezh2 loss. The observation that Suz12 deficiency in normal mammary cells leads to de-repression of cdkn2a, paralleling that seen with loss of Ezh2 [32], would predict that loss of PRC2 function in breast cancer would result in decreased tumor proliferation via up-regulation of p16 Ink4a and p14Arf expression. However, if this locus is perturbed, as is seen in many breast cancers [33], then loss of Ezh2 might have quite different consequences. Further work will be required to determine if loss rather than overexpression of PRC2 contributes to tumor progression in other cellular contexts.
In summary, our findings establish an essential role for PRC2 in the maintenance of mammary progenitor function and point to a critical function for PRC2 in maintaining chromatin states to ensure appropriate gene expression in MECs.

Ethics statement
All animal experiments were conducted using mice bred at and maintained in our animal facility according to the Walter and Eliza Hall Institute of Medical Research Animal Ethics Committee guidelines, approval number 2017.002.

Mice
Suz12 f/f [34], Eed f/f [35], Ezh2 f/f [36], MMTV-cre (line A) [37], and R26creERT2 [38] genetargeted mice have been described previously. CD4cre-Eed f/f T lymphocytes [39] were a gift from R. Allan. MMTV-cre mice were maintained as a pure strain on an FVB/N background, R26creERT2 mice were on a C57BL/6 background, and Ezh2, Eed, and Suz12 conditional knockout mice were on either a C57BL/6 or mixed FVB/N and C57BL/6 background. For analysis of postnatal day 1 lungs, adult female mice were subjected to timed pregnancies and injected with 0.2 mg progesterone on days 17 and 18 of pregnancy to delay parturition. At 19.5 days of pregnancy, pups were recovered by caesarian section and monitored for breathing for 1 hour with constant stimulation before collection of lungs into Bouin's solution.

Mammary cell preparation and cell sorting
Primary mouse MEC cultures were isolated from both inguinal and/or thoracic mammary glands and prepared as described [17]. MEC suspensions and flow cytometry were performed as previously described [40]. Antibodies against mouse antigens were purchased from Biolegend and included CD24-PB (#101820), CD31 (#102410), CD45 (#103112), and Ter119 (#116212) conjugated to APC, CD29-FITC (#102206), and CD14-PE (#123309). Cells were sorted on a FACSAria or FACSDiva (BD PharMingen) and manually counted prior to plating on irradiated fibroblast feeders as described [40]. After 7 days, colonies were fixed and stained with Giemsa and manually counted. To induce R26creERT2-mediated deletion, 0.1 μM 4OHT was added to culture medium for 20 hours on day 1 of culture.

Organoid culture
Organoids were cultured as described previously [18] in 8 μl BME (Cultrex) drops (70 basal and 60 luminal progenitor cells per drop) on nontreated 24 well plates and covered by Advanced DMEM/F12 supplemented with growth factors excluding Wtn3a or FGF2. Rock inhibitor (Y-27632) was added to culture medium for the first 4 days, and the medium was refreshed every 2-3 days. To induce R26creERT2-mediated deletion, 0.1 μM 4OHT was added to culture medium for 16-20 hours on day 1 or day 4 of culture. Organoids were photographed using best focus projection images on the Nikon TiE System software after 12-14 days in culture. Prior to proliferation measurements, RNA-seq, or ATAC-seq analysis, 12-16 day old organoids were dissociated into single cells using TrypLE express (Thermo Fisher Scientific). Cell suspensions were mixed 1:1 with CellTiter-Glo Luminescent Cell Viability Assay (Promega) prior to detection of luminescence. Genomic DNA was extracted from organoids using DNeasy Blood and Tissue kit (Qiagen) and PCR used to distinguish the Wt, floxed, and recombined (deleted, referred to as "del") Suz12 alleles as described [34].

Histology and whole mounting
Mice were injected with BrdU Cell Labeling Reagent (0.5 mg/10 g body weight, Amersham Biosciences) 1.5 hours prior to collection. For histology, tissues were fixed in 4% paraformaldehyde overnight and embedded in paraffin. Sections (5 μm) were stained with hematoxylineosin (HE). For whole-mount analysis, mammary glands were harvested and fixed in Carnoy's solution (6:3:1 of 100% ethanol, chloroform, and glacial acetic acid) and stained with Carmine alum. The extent of ductal outgrowth was measured on inguinal whole mounts as the distance from the center of the lymph node to the leading edge of the ductal mass.

ATAC-seq analysis
ATAC-seq was performed as described [24] with the following adaptations. Organoid cultures were dissociated, and 50,000 single cells were lysed and nuclei collected at 500 g for 10 minutes in lysis buffer containing 0.1% NP-40. Pelleted nuclei were tagmented with Nextera Tn5 Transposase (TDE1, Illumina FC-121-1030) for 20 minutes at 22˚C and 30 minutes at 37˚C. Transposed DNA was purified using Qiagen MinElute kit (28204) and fragments PCR amplified as described [24]. ATAC libraries were sequenced on a NextSeq using a 150H kit with 75 bp paired-end reads, and total reads were collated from 2 runs to obtain approximately 60 × 10 6 reads per sample. Sequencing runs were pooled for each replicate and sample, and adapters were trimmed with Trim Galore! (http://www.bioinformatics.babraham.ac.uk/ projects/trim_galore/) and mapped to the mm10 mouse genome using bowtie2 [41], allowing for fragments <2,000 bases in length. Mitochondrial and duplicate reads were removed using Picard-Tools (http://broadinstitute.github.io/picard/), and bam files were run through macs2 (doi: 10.1186/gb-2008-9-9-r137) for peak calling using these parameters: "-nomodel-shift-75-extsize 150-qvalue 0.05." For DA analysis, bam files were loaded in SeqMonk (v1.37.1, https://www.bioinformatics.babraham.ac.uk/projects/seqmonk/), and probes were created using a tiling approach with 150 bp windows end to end across the genome. Raw counts were then processed through the differential gene expression pipeline within edgeR [42], and DA regions were called with an exact test and FDR < 0.05. Reads were quantified by reads per million and log2 transformed for visualizing graphically. For each gene, the 5 Kb region upstream of intron 1 was used in combination with the gene body to define the TSSs plus gene body. GC enrichment was quantified on mapped bam files using DeepTools [43] computeGCBias with an effective genome size of 2,150,570,000. These data have been deposited in the Gene Expression Omnibus (accession code GSE116431).

RNA-seq analysis
Total RNA was extracted from basal-or luminal-derived organoids grown from single sorted luminal or basal populations from the mammary glands of R26creERT2/Suz12 f/f female mice using the RNeasy Mini kit (Qiagen). Two biological replicates were prepared of the basalderived organoids and 3 biological replicates of luminal-derived. RNA-seq was carried out on an Illumina Nextseq 500 to produce 20-65 million 80 bp reads per sample. Read pairs were mapped to the mouse genome (mm10) using the subread aligner [44] implemented in the Rsubread software package. Read counts for Entrez Genes were obtained using featureCounts [45] and its inbuilt mm10 annotation. Gene information was downloaded from the NCBI on 1 February 2017. Statistical analysis used the limma [46] and edgeR [42] software packages. Genes with at least 0.5 read counts per million (cpm) in at least 2 samples were considered to be expressed and were kept in the analysis. Immunoglobulin receptor segments, ribosomal genes, predicted and pseudogenes, and obsolete Entrez IDs were filtered out. Trimmed mean of Mvalues (TMM) scale normalization [47] was applied, and read counts were transformed to log2-cpm with a prior count of 3. Linear models were used to test for expression differences between 4OHT treated versus untreated samples from luminal cells and from basal cells. Each organoid sample was treated as a random block, allowing for correlation between repeats [48]. Differential expression was assessed using the Treat method [49], computing empirical Bayes moderated t statistics relative to a fold change threshold of 1.5, and allowing for an abundance trend in the standard errors and for robust estimation of the Bayesian hyperparameters [50]. The Benjamini and Hochberg method was used to control the FDR. These data have been deposited in the Gene Expression Omnibus (accession code GSE116431).

Expression profiles for breast cancer subtypes
Microarray expression profiles of breast tumors were downloaded from Gene Expression Omnibus series GSE18229 [51]. Probe intensities were normexp background corrected with offset 50 [52] and loess normalized using the limma package. Mouse Entrez Gene IDs were mapped to human using HUGO Gene Nomenclature Committee orthology predictions downloaded November 2016. Suz12-deficient expression signatures were computed for each tumor using a previously described method [25]. Briefly, the sum of products of RNA-seq log2-foldchanges with microarray log2-normalized intensities was computed for all genes DE in the Suz12 deficient basal-or luminal-derived organoids.

Quantitative reverse-transcriptase PCR (qRT-PCR) analysis
For qRT-PCR analysis, total RNA from MECs was reverse-transcribed using Superscript III (Invitrogen), and cDNA were analyzed on a LightCycler 480 (Roche). Input cDNA concentrations were normalized to GAPDH. Product accumulation was evaluated using the comparative Ct method (2 −ΔΔCT ). Primer sequences were: Ezh2 For:  [51] for genes associated with Suz12-deficiency in basal-derived organoids (left) and luminalderived organoids (right). The Suz12-deficient expression score is highest in the claudin-low subtype and lowest in the basal and Her2 subtypes (P = 3.8e-10 for basal-derived and i = 6.2e-11 for luminal-derived organoids by Tukey HSD). ATAC-seq, assay for transposase-accessible chromatin using sequencing; DA, differentially accessible; DE, differentially expressed; Her2, human epidermal growth factor receptor 2; KO, knockout; MACS, model-based analysis of ChIP-seq; MD, mean-difference; RNA-seq, RNA sequencing; Suz12, Suppressor of Zeste 12 protein homolog; Wt, wild type. (TIF) S1 Data. Genotyping results of litters from MMTVcre-Suz12 f/f and MMTVcre-Eed f/f matings using the indicated mating pairs. Observed numbers and percentages of males, females, and total offspring of the indicated genotypes at weaning, as well as the expected mendelian frequency (expressed as percentage of total), are shown. (XLSX) S2 Data. Significantly altered genes found by RNA-seq in Suz12-deleted basal-and luminal-derived mammary organoids. Length = gene length in bases (total of all exons). Remaining columns give the t statistic, P value, and FDR, respectively, from the Treat method relative to fold-change threshold of 1.5. AveExpr, average log2 counts per million; FDR, false discovery rate; logFC, log2 fold-change; RNA-seq, RNA sequencing.