A genetic mosaic mouse model illuminates the pre-malignant progression of basal-like breast cancer

Breast cancer is the most frequently diagnosed cancer type and the second leading cause of cancer death in those assigned female at birth (Siegel et al., 2022). Human breast cancer is a heterogeneous disease classified into six molecular subtypes with distinct prognosis: luminal A, luminal B, HER2-enriched, normal-like, claudin-low and basal-like (Perou et al., 2000; Prat et al., 2010; Tobin et al., 2015). Basal-like breast cancer (BLBC) accounts for 15-20% of breast cancer cases and is the most aggressive subtype, with earlier onset, increased chance of metastasis and absence of hormonal-therapy targets (Fulford et al., 2007; Millikan et al., 2008; Tobin et al., 2015; Turner et al., 2004). BLBCs show a high prevalence of p53 (also known as TP53 in human) mutations (∼80%) and deficiency in homology-directed DNA repair (∼50%); the latter is often caused by germline mutations in BRCA1/2, somatic epigenetic inactivation of BRCA1/2, or the loss of other essential genes for homology-directed DNA repair pathway (collectively termed as ‘BRCAness’) (Lord and Ashworth, 2016; McCabe et al., 2006; Network, 2012; Tian et al., 2019). Early detection and prevention of BLBC can fundamentally improve patient care, particularly for individuals with germline BRCA1 mutations who are at a high risk of BLBC. Therefore, it is imperative to gain a deep understanding of how BLBC initiates and progresses during the pre-malignant stages.

Genetically engineered mouse models (GEMMs) present an invaluable pre-clinical platform for studying human cancers. Conditional knockout of Brca1 and p53 (also known as Trp53 in mouse) in mouse mammary epithelial cells led to mammary tumors resembling human BLBC (Hollern et al., 2019; Liu et al., 2007; Molyneux et al., 2010; Xu et al., 1999). These models are useful for studying BLBC at the malignant stage, but are quite limited in examining cancer initiation and pre-malignant progression. First, conditional knockout models generate numerous rather than rare mutant cells at the cancer initiation stage; thus, they do not accurately mimic human cancer initiation from sporadic mutant cells (Liu et al., 2011; Muzumdar et al., 2007), which may impact pre-malignant development. Second, even if rare mutant cells can be generated, unequivocally pinpointing subtle aberrant behaviors of BRCA1 mutant cells at the pre-malignant stage remains challenging.

To overcome these limitations, we used a mouse genetic system known as mosaic analysis with double markers (MADM) developed by our laboratory (Fig. 1A). MADM consists of a pair of chimeric GFP- and RFP-coding sequences (separated by a loxP-containing intron) knocked into homologous chromosomes. Each knock-in cassette is syntenic with either the wild-type or mutant allele of one or more tumor suppressor genes on the same chromosome. From a non-labeled heterozygous animal, Cre/loxP-mediated inter-chromosomal mitotic recombination followed by X segregation of chromosomes generates a homozygous mutant cell labeled with GFP and its sibling wild-type cell labeled with RFP. Mutant cells are rare (0.1-1% or even lower) owing to the low frequency of inter-chromosomal recombination (Muzumdar et al., 2007; Zong et al., 2005), thereby approximating sporadic cancer initiation. The permanent GFP labeling of rare mutant cells enables spatially resolved investigation of their aberrant behavior at any time point during tumorigenesis (Fig. 1A) (Liu et al., 2011; Yao et al., 2020; Zong et al., 2005). Furthermore, along with a GFP⁺ mutant cell, MADM simultaneously generates a sibling RFP⁺ wild-type cell, which serves as an internal reference that enables the detection of subtle abnormalities of mutant cells. Here, we applied MADM to establish a mouse genetic mosaic model for BLBC, in which cancer initiates from sparse Brca1, p53-deficient cells in mammary glands. Our analyses of GFP⁺ mutant cells up to early malignancy provide important insights into cancer initiation and pre-malignant progression, and could enable future studies on the early detection and prevention of cancer.

To establish a MADM-based mouse model of breast cancer, we prepared two stock mouse lines through a multi-generational breeding scheme (Fig. S1). For one stock line, we bred the Brca1 and p53 mutant alleles (Jacks et al., 1994; Xu et al., 1999) onto the MADM-TG allele (Hippenmeyer et al., 2010). For the other stock line, we introduced the MMTV-Cre transgene (Wagner et al., 2001) into the MADM-GT line (Hippenmeyer et al., 2010) to target mammary epithelial cells. Finally, we inter-crossed the two stock lines to generate MADM-Brca1-p53; MMTV-Cre mice (Fig. S1B; hereafter referred to as MADM-mutant mice), in which sparse GFP⁺Brca1, p53-null cells are predisposed to becoming cancerous (Fig. 1A).

The rarity of GFP⁺ mutants in MADM not only closely mimics human cancer initiation but also enables clonal analysis of pre-malignant expansion (Greaves and Maley, 2012; Knudson, 1971; Muzumdar et al., 2007). We assessed the abundance of GFP⁺ mutant cells in mammary glands from MADM-mutant mice at 3 months of age, an age shortly after the peak in MMTV-Cre expression (Buono et al., 2006; Wagner et al., 1997). Using whole-mount fluorescence imaging of mammary glands, we found a high abundance of heterozygous cells (GFP and RFP double positive, which thus, appear yellow) generated from MADM recombination events in G1 or post-mitotic cells (G0) (Fig. S2A). These heterozygous cells are clearly present in the entire mammary ductal system and confirm the in vivo recombination by MMTV-Cre. Among many yellow heterozygous cells, we observed sparse and scattered GFP⁺ mutant cells along mammary ducts (Fig. 1B). For a higher-resolution view, we sectioned mammary glands and performed confocal imaging to visualize the abundance of GFP⁺ mutant cells. We found that GFP⁺ mutant cells were often singular (Fig. 1C) and accounted for ∼2% of all mammary epithelial cells, while RFP⁺ wild-type sibling cells accounted for ∼0.6% (Fig. 1D). This difference in percentage could be caused by a growth advantage of mutant cells, or a survival disadvantage of wild-type cells, or a mix of both. Notably, almost all MADM-labeled cells were positive for cytokeratin 8 (CK8; also known as KRT8), (a marker of mammary luminal cells) but negative for CK14 (a marker of mammary basal cells) (Fig. S2B-F), suggesting that they arose from the luminal layer in which the reported cell of origin for BLBC resides (Lim et al., 2009; Molyneux et al., 2010; Shehata et al., 2012). We are aware that the conventional Cre reporter showed that MMTV-Cre is expressed in both luminal and basal cells (Wagner et al., 2001). One possible explanation for this discrepancy is that MADM-labeling relies on inter-chromosomal recombination, which requires a much higher concentration of Cre than intra-chromosomal recombination for conventional floxed alleles (Zong et al., 2005). Therefore, although both luminal and basal cells express MMTV-Cre, the level could be much lower in basal cells than in luminal cells (Rabieifar, 2019), leading to luminal-specific recombination in MADM.

To assess the progression of initiated mutant cells, we collected mammary glands from MADM-mutant mice at different time points (ten mice at each age) and evaluated the overall expansion of GFP⁺ cells through whole-mount imaging. At 3 months, the expansion of GFP⁺ mutant cells was barely noticeable, but, from 6 months onward, GFP⁺ foci became visible, gradually expanded and eventually formed GFP⁺ tumors (Fig. 1E). By examining a cohort of 15 mice with an endpoint of 12 months of age, we found that 13 (87%) developed GFP⁺ tumors with a median latency of 11 months (Fig. 1F). Despite the rarity of cancer-initiating mutant cells, the MADM-mutant mice showed a tumor latency only slightly longer than the ∼9 months latency of a conditional knockout mouse model that initiates cancer with numerous mutant cells (Xu et al., 1999). This suggests that the abundance of cancer-initiating cells is not a rate-limiting factor for the kinetics of breast cancers driven by Brca1 and p53 deficiency. Among 43 GFP⁺ tumors from 33 mice, most mice had one to two tumors but rarely three or more tumors (Fig. S2G). Finally, we did not observe obvious tropism within the five pairs of mammary glands, except for a slightly higher incidence in the largest fourth pair and a relatively lower incidence in the second pair (Fig. S2H).

Human BLBCs are characterized by a high proliferation index, and lack of estrogen receptor (ER; also known as ESR1), progesterone receptor (PR; also known as PGR) and HER2 (also known as ERBB2) overexpression (Palacios et al., 2005; Perou et al., 2000; Rakha et al., 2008). We assessed six MADM tumors for proliferation and hormone receptor expression by immunohistochemistry and found them highly positive for Ki67 (also known as MKI67; ∼70% of cells) and mostly negative for ER, PR and HER2 (Fig. 2A; Fig. S3A), matching the histopathological features of human BLBCs. To further examine whether the MADM tumors resemble human BLBCs at the molecular level, we performed RNA sequencing of 12 MADM tumors and extracted a panel of 50 genes (PAM50) used to stratify breast tumor subtypes (Perou, 2011). We co-clustered PAM50 signatures of our MADM tumors with the profiles of 1104 human breast tumors from The Cancer Genome Atlas (TCGA) dataset annotated for five breast cancer subtypes. To mitigate overall differences in gene abundance across species, we identified a unique set of mouse-to-human orthologs across all TCGA and MADM tumors and normalized each sample to obtain relative expression values for each gene (see Materials and Methods for details). We found that MADM tumors clustered with the human basal-like subtype but not others (Fig. 2B). Another hallmark of human BRCA1-mutated BLBC is the high frequency of copy number variations (CNVs) for genomic loci containing oncogenes or tumor suppressors (Annunziato et al., 2019; Weigman et al., 2012). To determine whether MADM tumors also share this hallmark, we conducted whole-exome sequencing on six MADM tumors and paired normal somatic tissues. We found recurrent amplification of multiple chromosomal segments harboring oncogenes – such as Met, Myc and Fgfr1 – along with recurrent deletion of the tumor suppressor gene Rb1 (Fig. 2C), closely corresponding to human BLBCs (Fig. S3B) (Network, 2012). Collectively, the histopathological, transcriptomic and genomic analyses of MADM tumors demonstrate that our MADM-mutant mice represent an authentic model for human BLBC.

We next investigated the progressive phenotypic alterations of MADM mutant cells throughout pre-malignancy with a cohort of mice at the intermediate ages between cancer initiation and tumor formation. We performed tissue clearing with the clear, unobstructed brain/body imaging cocktails and computational analysis (CUBIC) method (Susaki et al., 2015) and then conducted whole-mount, 3D imaging using light-sheet microscopy (Fig. S4A) to look for morphological abnormalities by comparing mutant ducts with heterozygous ducts (Fig. 3A). At 3 months of age, we observed short stretches of mutant cells that occupied a continuous region without causing noticeable alterations in ductal morphology. At 6 months of age, some GFP⁺ mutant cells extended side branches resulting in a slightly more complex morphology than that of heterozygous ducts. This hyper-proliferation of mutant cells preceding prominent changes in tissue organization is consistent with observations in BRCA1-mutant carriers (Martins et al., 2012; McKian et al., 2009). Upon further expansion at 8 months, some mutant branches developed extensive epithelial buds reminiscent of alveologenesis during early pregnancy (Richert et al., 2000) even though these are virgin MADM-mutant mice. The alveologenesis was pervasive in late-stage mammary glands and exhibited prominently distinct histology when compared with the internal control heterozygous ducts (Fig. S4B,C). We further quantified the number of buds (alveoli) per 100-µm primary ducts and found that mutant ducts had over tenfold more alveoli than the controls (Fig. S4D,E). Binning the data into four levels based on the number of alveoli per 100 µm primary ducts – level 0 (<5), level I (5-20), level II (20-50) and level III (>50) (Fig. S4F) – we found that high-level alveologenesis (II-III) exclusively occurred within the GFP⁺ mutant regions (Fig. S4G), suggesting that this ‘hyper-alveolarization’ is a characteristic feature of pre-malignant lesions. In 10-month-old mice, we occasionally observed tiny GFP⁺ spherical masses (<1 mm in diameter) that had lost ductal morphology but were not yet palpable, which were termed as ‘micro-invasion’ hereafter (Fig. 3A).

To further determine whether the hyper-alveolarized ducts and the micro-invasions reflect a sequential progression of mammary cancer in MADM-mutant mice, we evaluated whether there is a temporal sequence in the occurrence of these structures. Because multiple expansion levels of mutant cells often co-exist in the same gland (Fig. S4B), we plotted the most-advanced GFP⁺ lesion observed in the mammary gland of each MADM-mutant animal at a series of ages, and observed a progressive emergence of focally expanded GFP⁺ cells, hyper-alveolarized GFP⁺ ducts and micro-invasions, culminating in the formation of GFP⁺ tumors at ∼1 year of age (Fig. 3B). Histological analysis further supported a temporal sequence of these pre-malignant structures (Fig. 3C). Although mammary epithelial cells from control wild-type mice appeared normal across all ages (Fig. S5A), mutant cells in hyper-alveolarized ducts present at 8 months displayed abnormal nucleomegaly (∼1.5× larger than wild-type nuclei), small but conspicuous nucleoli and loss of the basally oriented nuclear polarity (Fig. S5B). The micro-invasions contained residual hyper-alveolarized lobular units along with more advanced microinvasive carcinoma (measuring less than 1 mm) composed of individual cells and larger cords that elicited a stromal response (Fig. S5C). Finally, the frank tumors were dominated by infiltrating cells with a desmoplastic stromal response; no remnants of alveoli were visible (Fig. 3C). The progressive nature from hyper-alveolarizations to micro-invasions to full-blown tumors was further supported by Ki67 staining, revealing a gradual increase in cell proliferation (Fig. 3D,E). Thus, the MADM-mutant model revealed that cancer-initiating cells progress through a visually identifiable sequence of ductal morphology alterations before the emergence of mammary tumors.

The mammary epithelium is composed of two anatomically and functionally distinct compartments – the alveolar regions that produce milk during lactation and the ductal regions that drain milk to the nipple (Fig. 4A) (Visvader, 2009; Visvader and Stingl, 2014). It was unclear whether the hyper-alveolarized structures consisting of mutant cells originated from alveolar or ductal mutants. Because MADM generates one RFP⁺ wild-type sibling cell alongside each original GFP⁺ mutant cell (Fig. 1A), we investigated this question using the ‘twin-spot’ analysis that directly compares mutant cells with their wild-type sibling cells in a clone-by-clone fashion within ductal and alveolar compartments, respectively (Espinosa and Luo, 2008; Muzumdar et al., 2007; Terry et al., 2020).

For the twin-spot analysis, we selected mice at 4 months of age because the focal expansion of GFP⁺ mutant cells was evident at this age, while hyper-alveologenesis was not yet present (Fig. 3B). As a baseline, we first analyzed 40 twin spots in four MADM-wild-type mice that lacked Brca1 and p53 mutant alleles (Fig. S1B), and confirmed that GFP⁺/RFP⁺ cells did not expand prominently and remained similar in number in both ductal and alveolar regions (Fig. 4B,D; data for each individual mouse plotted in Fig. S6A). In contrast, when we performed the twin-spot analysis in four MADM-mutant mice, we readily observed prominent expansion of GFP⁺ mutant clones over RFP⁺ wild-type clones in the ductal region (Fig. 4C,E; data for each individual mouse plotted in Fig. S6B). Surprisingly, the mutant clones in alveoli, even though harboring the same Brca1 and p53 mutations as those in ducts, did not expand and showed no difference in clonal size from the neighboring RFP⁺ wild-type clones (Fig. 4C,E). To conduct the twin-spot analysis more rigorously, we compared the sizes of the GFP⁺ and RFP⁺ sibling clones in a strictly pairwise manner, and found that only GFP⁺ mutant clones within the ductal regions exhibited significantly larger sizes than their RFP⁺ sibling clones (Fig. 4F), indicating that the initial clonal expansion of Brca1, p53-null cells occurs exclusively in the ductal region. This finding is particularly interesting because, although multiple studies with human tissues and conventional mouse models have noted the outgrowth of alveolar-like buds during the pre-malignant development of breast cancer and interpreted it as aberrant alveolar cell expansion (Bach et al., 2021; McKian et al., 2009; Poole et al., 2006; Tao et al., 2017), our results showed that such outgrowth most likely originates from luminal progenitors in the ducts that are either intrinsically fated for the alveolar lineage (i.e. alveolar luminal progenitors within the ducts) or acquire the aberrant alveolar fate due to p53/BRCA1 loss.

BLBCs express basal-cell markers [CK5/14, α-SMA (also known as ACTA2), P63, etc.] yet arise from luminal progenitor cells (Lim et al., 2009; Liu et al., 2007; Molyneux et al., 2010; Shehata et al., 2012). This luminal-to-basal transition is thought to be critical for cancer progression, as it promotes stemness and invasiveness of BRCA1- or p53-mutant mammary epithelial cells in vitro (Bai et al., 2022; Kim et al., 2011; Liu et al., 2008; Mizuno et al., 2010). However, in vivo mapping of luminal-to-basal transitions during tumorigenesis is lacking. To determine whether luminal-to-basal transition occurs in the two representative pre-malignant stages of MADM-mutant mice (hyper-alveolarized ducts and micro-invasions), we leveraged the single-cell resolution of MADM to carefully assess the mutant cells at each stage. Normal mammary luminal cells present a cuboidal shape, whereas basal cells show an elongated spindle shape (Rios et al., 2014). Within hyper-alveolarized ducts in ∼8-month-old MADM-mutant mice, mutant cells mostly exhibited a cuboidal shape with relatively homogeneous cell size (Fig. 5A, top row). In contrast, mutant cells in micro-invasions in ∼10-month-old MADM-mutant mice showed heterogeneous morphologies, with a fraction adopting an elongated spindle shape reminiscent of basal cells (Fig. 5A, bottom row). When we quantified the cell size and circularity of mutant cells in hyper-alveolarized ducts and micro-invasions from four MADM-mutant mice, we found that the mutant cells in micro-invasions were significantly larger in size and lower in circularity than those in hyper-alveolarized ducts (Fig. 5B,C), implying a transition of cell state between these two stages.

To clarify whether the morphological change represents a luminal-to-basal transition, we assessed the expression of luminal cell marker (E-cadherin; also known as CDH1) and two basal cell markers (α-SMA, CK5) in mutant cells by immunofluorescent staining. In hyper-alveolarized ducts, GFP⁺ mutant cells maintained the expression of E-cadherin and were negative for basal markers (Fig. 5D, top and bottom panels, top rows). In micro-invasions, by contrast, some mutant cells co-expressed E-cadherin and basal cell markers (Fig. 5D, top and bottom panels, bottom rows), suggesting an incomplete transition from the luminal to basal cell state. We quantified this observation by analyzing a total of ∼700 cells in mammary tissues from four mice with hyper-alveolarized ducts or micro-invasions. In hyper-alveolarized ducts, E-cadherin-positive mutant cells were almost exclusively negative for α-SMA and CK5, but, in micro-invasions, ∼50% of mutant cells expressed both basal cell markers and E-cadherin (Fig. 5E). The incompleteness of the luminal-to-basal transition in micro-invasions was further supported by the prevalent expression of another luminal marker CK8/18 in GFP⁺ mutant cells, comparable to that within hyper-alveolarized mutant ducts (Fig. S7A). Only sporadic ER⁺ cells were found at both stages (Fig. S7B), which is consistent with the current understanding that ER⁻ luminal progenitor cells serve as the cancer cell of origin of BLBC with BRCA1 mutations (Lim et al., 2009). Taken together, the cellular heterogeneity and the incomplete luminal-to-basal transition in micro-invasions revealed by our study likely present an excellent therapeutic opportunity and warrant further comprehensive molecular and cellular analysis.

In this study, we established a mouse genetic mosaic model for BLBC that shares histopathological, transcriptomic and genomic similarities with human BLBCs. Taking advantage of the spatial resolution provided by MADM, we identified multiple morphologically distinct pre-malignant stages, including focal expansion of mutant cells, hyper-alveolarization of mutant ducts and micro-invasion. Surprisingly, the clonal analysis revealed that hyper-alveolarized mutant structures originate from ductal rather than alveolar cells. Further progression from hyper-alveolarized structures to micro-invasions resulted in loss of ductal organization and an incomplete luminal-to-basal transition, manifested by enlarged cell size, elongated cell shape, elevated cell proliferation and gain of basal marker expression without losing luminal marker expression. Overall, our MADM-based mouse model provides a useful tool for studying the pre-malignancy BLBC, which should empower pre-clinical research on early detection and cancer prevention.

MADM-based cancer models offer several unique advantages. First, MADM creates a small number of homozygous mutant cells within a heterozygous animal, closing mimicking the sporadic loss of heterozygosity (Lohmussaar et al., 2020) of tumor suppressor genes in cancer. For germline BRCA1 mutation carriers, cancer is often initiated by the sporadic loss of the wild-type BRCA1 allele (Cornelis et al., 1995; Maxwell et al., 2017; Nones et al., 2019). Second, the generation of mutant cells by MADM is coupled with permanent GFP labeling through a mitotic recombination event, allowing mutant cells to be unequivocally identified and tracked throughout the entire process of tumorigenesis (Muzumdar et al., 2007; Zong et al., 2005). Because both GFP and tdTomato are expressed at the high enough level to be visible under a fluorescence stereomicroscope within fresh, unstained mammary glands (Fig. 1E), MADM enables the gross evaluation of pre-malignant phenotypes and targeted sample collection prior to various downstream analyses that require unfixed tissues. Third, MADM generates sibling wild-type cells that are labeled with RFP, providing an internal control for the GFP⁺ mutant cells (Beattie et al., 2017; Terry et al., 2020). Without RFP⁺ wild-type siblings as a reference (Fig. 4), it would be difficult to distinguish the clonal expansion of GFP⁺ mutant ductal cells from stochastic neutral drift (Lopez-Garcia et al., 2010; Snippert et al., 2010). Although our study is focused on breast cancer modeling, it should be noted that, owing to the modular nature of the MADM system and the availability of a genome-wide library of MADM mice (Contreras et al., 2021), one could establish many other cancer models to study their initiation and pre-malignant progression upon the sporadic LOH of relevant tumor suppressor genes.

The observed progression trajectory in the MADM-mutant mice provided multiple insights for understanding the early genesis of BLBC. Although hyper-alveolarization was initially thought to reflect the aberrant expansion of mutant alveolar cells (Tao et al., 2017), our study showed that it actually originates from mutant ductal cells either intrinsically fated for or misdifferentiated toward the alveolar fate upon p53 and BRCA1 mutations. Corroborating our finding, the recent single-cell sequencing of mammary glands from Brca1, p53-deficient mice spanning various pre-malignant ages revealed dysregulation of transcription factors driving alveologenesis in luminal progenitor cells, causing aberrant alveolar outgrowth (Bach et al., 2021). Intriguingly, this finding may point to distinct roles of hormonal signaling between pre-malignancy and malignancy of BLBC: the hyper-alveolarization of mutant ducts resembles alveolar outgrowth during early pregnancy (Macias and Hinck, 2012), a process known to be regulated by the progesterone signaling (Brisken et al., 1998; Humphreys et al., 1997; Lydon et al., 1995); and progesterone receptors are overexpressed in Brca1-deficient mammary epithelial cells of human and mouse, and exposure to exogenous progesterone dramatically increases mammary gland volume in Brca1-deficient mice (King et al., 2004; Ma et al., 2006; Poole et al., 2006). Therefore, although the expression of hormonal receptors is low/absent in malignant BLBCs, progesterone signaling could play significant roles at pre-malignancy, which warrants further study and may offer a new avenue of cancer prevention (Nolan et al., 2016; Sigl et al., 2016; Trabert et al., 2020).

At the stage of micro-invasion, MADM-mutant cells undergo a partial luminal-to-basal transition, which could increase the stemness and invasiveness of Brca1- or p53-deficient cells, as reported in the literature (Bai et al., 2022; Kim et al., 2011; Liu et al., 2008; Luond et al., 2021; Mizuno et al., 2010), and was recently implicated as a critical step at the onset of basal-like tumorigenesis (Landragin et al., 2022 preprint). Our time-course analysis of Brca1, p53-deficient cells in vivo showed that the loss of Brca1 and p53 does not immediately induce a luminal-to-basal transition even after the manifestation of hyper-alveolarization. Instead, mutant cells only reach an incomplete luminal-to-basal transition at ∼10 months after mutant cells progress to micro-invasive lesions. Although we cannot rule out cell-intrinsic mechanisms for this transition, our observation implies that the exposure of luminal cells to extrinsic stromal factors due to basement membrane breaching in micro-invasion could be the trigger for the luminal-to-basal transition, which should be investigated more thoroughly in the future.

Although powerful, MADM has certain limitations. First, the tumor latency tends to be long, e.g. it took ∼8 months for MADM-mutant mice to progress into the pre-malignant stages in this study. If desired, additional clinically relevant mutations could be introduced to accelerate cancer development in our model (Annunziato et al., 2019). Compound mutations that are syntenic with Brca1 and p53 can be introduced using the same scheme as shown in Fig. S1B; for mutations that are not syntenic, mutant alleles can be introduced into the MADM model through conventional breeding schemes (Muzumdar et al., 2016; Yao et al., 2020). Second, owing to the involvement of many genetic elements in the MADM model, it tends to have mixed genetic background that precludes allotransplantation experiments. When necessary, one could backcross stock mice into desired pure genetic background to improve the versatility of this model. Third, MADM relies on mitotic recombination to generate mutant cells, and thus cannot be used to mutate post-mitotic cells. Finally, when constitutively expressed Cre transgene is used, e.g. MMTV-Cre for this study, the birth timing of mutant cells is not clear. When desired, one could use temporally controlled Cre lines, such as Tet or CreER system, to precisely control the timing of tumor initiation. Notwithstanding these limitations, the current MADM model for BLBC enables spatially resolved analysis throughout the pre-malignant progression, and can greatly facilitate studies of early detection and cancer prevention.

The following mouse lines were crossed to establish the MADM-mutant and MADM-wild-type mice: TG11ML (stock #022976, The Jackson Laboratory) (Henner et al., 2013), GT11ML (stock #022977, The Jackson Laboratory) (Henner et al., 2013), Brca1^flox (strain #01XC8, NCI Mouse Repository) (Xu et al., 1999), p53^KO (stock #002101, The Jackson Laboratory) (Jacks et al., 1994), MMTV-Cre (stock #003553, The Jackson Laboratory) (Wagner et al., 1997). The breeding schemes are shown in Fig. S1. We exclusively used female mice, primarily focusing on the fourth pair of mammary glands (the abdominal pair) for data collection, unless otherwise specified in figure legends. All animal work was performed in the University of Virginia Animal Vivarium. All procedures, including housing and husbandry were approved by the Institutional Animal Care and Use Committee (IACUC) at the University of Virginia, following national guidelines to ensure the humanity of all animal experiments.

For genotyping, the mouse toe was used to extract DNA for PCR. First, 120 µl of 50 mM NaOH was added to each toe and then incubated at 95°C for 20 min in the PCR machine, followed by addition of 30 µl of 1 M Tris-HCl (pH 7.4) and mixing. One microliter of the toe solution was used for the PCR template in a 20 µl PCR reaction. PCR primer sequences were as follows. (1) MADM TG/GT cassettes: primer-1, 5-TGGAGGAGGACAAACTGGTCAC-3; primer-2, 5-TCAATGGGCGGGGGTCGTT-3; primer-3, 5-TTCCCTTTCTGCTTCATCTTGC-3; PCR products, knock-in (KI) band, 230 bp and wild-type band, 350 bp. (2) MMTV-Cre: primer-1, 5-CACCCTGTTACGTATAGCCG-3; primer-2, 5-GAGTCATCCTTAGCGCCGTA-3; PCR product, KI band, 300 bp. (3) p53^KO: primer-1, 5-ACCGCTATCAGGACATAGCGTT-GG-3; primer-2, 5-CACAGCGTGGTGGTACCTTATG-3; primer-3, 5-GGTATACTCAG-AGCCGGCCTG-3; PCR products, KI band, 700 bp and wild-type band, 450 bp. (4) Brca1^flox: primer-1, 5-CTGGGTAGTTTGTAAGCATCC-3, primer-2, 5-TCTTATGCCCTCAGAAAACTC-3; PCR products, flox/flox band, 365 bp and wild-type band, 297 bp.

Mammary glands were harvested and fixed with 4% paraformaldehyde (PFA) at 4°C for 24 h. For immunofluorescence, tissues were then washed with PBS twice, soaked with 30% sucrose at 4°C for 48 h, and embedded in optimal cutting temperature (OCT) compound (SAKURA). The tissues were sectioned at 20 µm thickness with a Thermo Fisher Scientific NX50 Cryostat. For staining, slides were first blocked in 0.3% Triton-X 100 and 5% normal donkey serum in PBS for 20 min, then incubated with primary antibodies (anti-CK8, Abcam, ab182875, 1:200; anti-CK14, BioLegend, 905301,1:400; anti-E-cadherin, BioLegend, 147301, 1:200; anti-α-SMA, Sigma-Aldrich, A5228, 1:500) diluted in blocking buffer at 4°C overnight. Secondary antibody incubation was performed for 1 h at room temperature in PBS. To stain nuclei, slides were incubated in 4′,6-diamidino-2-phenylindole (DAPI) solution (1 µg/ml in PBS) for 5 min before being mounted with 70% glycerol. Fluorescent images were acquired on Zeiss LSM 700/710 confocal microscope. Images were processed with Zen and Fiji. For immunohistochemistry, PFA-fixed tissue was further processed for paraffin embedding and then sectioned at 4 µm thickness. After antigen retrieval, the primary antibody (Ki67, Epitomics, 4203-1, 1:400) was incubated at 4°C overnight. Horseradish peroxidase (HRP)-conjugated secondary antibodies were then used, and 3,3′-diaminobenzidine (Vector Laboratories, SK-4100) was used to develop color.

PFA-fixed mammary glands were cleared for large-scale 3D imaging with the standard CUBIC method (Susaki et al., 2015). Briefly, tissues were immersed in 50% reagent-1 (25 wt% urea, 25 wt% Quadrol, 15 wt% Triton X-100, 35 wt% dH₂O), shaken at 110 rpm at 37°C for 12 h, and then transferred to 100% reagent-1 with DAPI (1 µg/ml) for shaking until mammary glands became transparent. After reagent-1, tissues were washed three times with PBS, 1 h each time, with shaking to remove reagent-1. Tissues were then shaken in 50% reagent-2 for 12 h at 37°C, followed by 100% reagent-2 (25 wt% urea, 50 wt% sucrose, 10 wt% triethanolamine, 15 wt% dH₂O), shaking for 48 h. A Zeiss Z.1 light-sheet microscopy system was used for acquiring images. Tissues were attached to the holder of the light-sheet microscope with super glue.

Approximately 50 mg tissue from each mammary tumor was used for RNA extraction. The tissue was homogenized in 500 µl TRlzol, then 100 μl chloroform was added and mixed thoroughly, followed by centrifugation (12,000 rcf) for 15 mins at 4°C. The upper-layer aqueous phase containing the RNA was transferred to a new tube, and 5 μg polyacrylamide was added, followed by an equal volume of 70% ethanol. Afterward, 700 μl of the sample was used as the input for RNA isolation using an RNeasy Mini Kit (Qiagen) according to the manufacturer's protocol. The quality of RNA samples was evaluated with Bioanalyzer (Agilent Technologies), and samples with an RNA integrity number (RIN) >8 were used for library preparation. Libraries were prepared with a TruSeq Stranded mRNA Library Prep kit (Illumina). Libraries were multiplexed at an equimolar ratio, and 1.3 pM of the multiplexed pool was sequenced on a NextSeq 500 instrument with a NextSeq 500/550 Mid/High Output v2.5 kit (Illumina) to obtain 75-bp paired-end reads. From the sequencing reads, adapters were trimmed using fastq-mcf in the EAutils package (version ea-utils.1.1.2-779) with the following options: −q 10 −t 0.01 −k 0 (quality threshold 10, 0.01% occurrence frequency, no nucleotide skew causing cycle removal). Quality checks were performed using FastQC (version 0.11.8) and MultiQC (version 1.7). Data were aligned to the murine transcriptome (GRCm38.84) using HISATv2 (version 2.1.0) with options for paired-end reads. HISAT read counts were converted to transcripts and normalized to transcripts per million (TPM) using StringTie (version 2.1.6).

TCGA breast cancer expression data were obtained from the UCSC genome browser (Ciriello et al., 2015). Human orthologs for mouse genes were obtained from the Ensembl biomart in R using the getAttributes function. For the human and murine datasets, we obtained the intersection of unique orthologs to obtain a set of 14,980 mice to human ortholog genes to evaluate co-expression. To enable cross-species comparisons, we performed a sample-wise column normalization to obtain new TPM estimates that accounted for gross differences in gene abundance between species. Hierarchical clustering of PAM50 genes was performed using ‘pheatmap’ in R using Euclidean distance and ‘ward. D2’ linkage.

Genomic DNA was prepared from tumors developed in MADM-mutant mice and from the tails of the same mice as the control with a DNeasy Blood and Tissue Kit (Qiagen). Whole-exome sequencing at 100× coverage was performed as a contract service with Genewiz. Raw BCL files were converted to fastq files with bcl2fastq v.2.19, and adapter was trimmed with Trimmomatic v.0.38. Trimmed reads were mapped to the mouse reference genome, and somatic variants and CNVs were called using the Dragen Bio-IT Platform (Illumina) in somatic mode and a panel of normals to remove technical artifacts. The filtered VCF was annotated with Ensembl Variant Effect Predictor (VEP) v95 for the Ensembl transcripts overlapping with the filtered variants. CNVs that passed quality control filters from Dragen were visualized using the GenVisR v.1.16.1 package in R.

Statistical analysis was performed with GraphPad Prism. Bar graphs were presented as the mean±s.e.m. unless otherwise annotated in figure legends. The normality of data distribution was checked with qqplot in R. Depending on whether the data followed a normal distribution, paired two-tailed Student's t-test or Mann–Whitney U test was used as indicated in the figure legends. The Chi-square test or Fisher's exact test was used to test frequency distribution as indicated in the figure legends. Statistical significance is denoted by ‘not significant (n.s.)’ (P>0.05), **P<0.01, ***P<0.001 and ****P<0.0001.

We thank Dr Bing Xu and Xiaoyu Zhao for providing feedback on the manuscript, and Dr Qinlei Gu for the artwork on mammary gland distribution in mice. We also thank Dr Pat Pramoonjago at the Biorepository and Tissue Research Facility, Sheri Vanhoose at the Research Histology Core, Dr Stacey Criswell at the Advanced Microscopy Facility and Shelly Verling at the vivarium for their assistance on the project. We thank Dr Ammasi Periasamy at the UVA Keck Center for assistance with the Zeiss Z.1 light-sheet microscopy system. These core facilities are partly supported by UVA Cancer Center Grant P30-CA044579.