LMO2 Regulates Epithelial-Mesenchymal Plasticity of Mammary Epithelial Cells.

Introduction
Cellular plasticity has recently been recognized as a key hallmark in tumorigenesis and tumor progression [1, 2]. Epithelial-mesenchymal plasticity (EMP), a form of cellular plasticity, allows cells to exist along a spectrum from epithelial to mesenchymal states, creating hybrid cellular states [1, 3, 4]. This hybrid state is now recognized as important for inducing stem cell-like properties, drug resistance, and metastasis in cancer cells [2, 5–8].
Mammary epithelial cells (MECs) have been used extensively as a model system to study regulators of cellular plasticity [9–11]. The mammary gland is a dynamic organ, with most of the development occurring postnatally during puberty in response to hormones [12]. It is a bi-layered gland of luminal and basal epithelial cells where the luminal lineage gives rise to milk-producing cells and basal/myoepithelial cells have contractile function to enable the release of milk [12]. The CD49fhigh/EpCAMmed-low basal/myoepithelial cells in the mammary gland are enriched for in vivo reconstitution potential [13, 14] and display characteristics of quasi-mesenchymal states (plastic cell state) with expression of both epithelial marker (EpCAM) and mesenchymal marker (smooth muscle actin) [15, 16]. The molecular regulation of the quasi or hybrid state in MECs remains poorly understood, although it is an active area of research due to its potential implications in tumor development and progression.
We have previously demonstrated that the hematopoietic stem cell factor and transcriptional adaptor protein [17], LMO2, is enriched in a minority population of immature tumor cells [8]. Our published results demonstrate that LMO2+ cells are metastatic, predict poor distant recurrence-free survival in patients, and LMO2 is functionally required for metastasis [8]. However, the function of LMO2 in normal mammary epithelial cells is unknown. Here, we explore the role of LMO2 in the normal mammary gland using lineage-tracing, knockout models, organoid assays, and RNA sequencing to find that LMO2 is not essential for mammary gland development but regulates cellular plasticity under regenerative conditions.

Results

Lmo2 + Epithelial Cells are Present in the Luminal and Basal Layer of the Mammary Gland
To determine cell fate and lineage hierarchy of Lmo2-expressing cells in the mammary gland, we generated Lmo2CreERT2/RosamTmG mice (Fig. 1A). To test whether Lmo2CreERT2 marks Lmo2-expressing cells, we pulsed the mice at 8 weeks with a single dose of tamoxifen and sorted unswitched (TdTomato+) and switched (GFP+) mammary epithelial cells (MECs). qPCR analysis of the GFP+ cells 36 h after tamoxifen pulse showed significant enrichment in expression of Lmo2 [8], thus validating the efficacy of the Lmo2CreERT2 mouse model in labeling Lmo2-expressing MECs (Fig. 1B). Flow cytometry analysis of the mammary gland showed that Lmo2+ cells were present both in the luminal and basal lineages of the mammary epithelium (Fig. 1C, Supplementary Fig. 1 A, B). To understand whether Lmo2+ cells contributed to the developing mammary gland we pulsed the mice with tamoxifen at puberty (P28) followed by analysis at 12 weeks. Although we could detect Lmo2-lineage traced cells in the luminal and basal layers, we only observed small clusters of 2–3 cells (Fig. 1D). These data suggest that while Lmo2+ cells are present during the development of the ductal epithelium, they have limited proliferative capacity and thus limited contribution during postnatal development. Similarly, pulsing mice at 8 weeks and analyzing at 16 weeks demonstrated a very small population of persistent Lmo2-lineage traced cells (Fig. 1E).

The mammary gland undergoes profound remodeling stages during pregnancy, which are largely driven by the expansion of specific progenitor populations (e.g., Notch1/Blimp1-expressing luminal adaptive secretory precursors (LASPs) [18, 19], Nfatc1- and Lgr6-expressing basal progenitors [20, 21], PI-MECs-1 [22], PI-MECs-2 [23], and Axin2+ progenitors [24]). To test whether pregnancy alters the relative abundance of Lmo2+ cells, we pulsed female mice at 4 weeks, mated at 12 weeks, and analyzed at pregnancy day 15.5. We found that the Lmo2-lineage traced population expanded during pregnancy and were enriched in alveolar structures, although only 10–20 alveoli per gland were GFP+ with varying contributions of Lmo2-lineage traced cells per alveoli (Fig. 1F). However, no GFP+ cells were detected in the second pregnancy (data not shown).
To further understand the expression of Lmo2 across development, we analyzed previously published single cell RNA-sequencing data across developmental stages [25]. We found that the percentage of Lmo2-expressing cells is highest in HR-low cells (alveolar progenitors) and increases during pregnancy and lactation (Supplementary Fig. 2A-C), suggesting that the Lmo2+ population expands in response to hormones to give rise to milk-producing cells during pregnancy. Thus, Lmo2 likely marks a sub-population of lineage restricted progenitors with limited proliferative potential. Together, these data demonstrate that a minority population of Lmo2-expressing cells are present in both the basal and luminal lineages, and can expand/differentiate during pregnancy but are likely lost post-involution.

LMO2 Regulates in Vivo Reconstitution Potential and Organoid Formation of MECs
We next sought to determine whether LMO2 is required for normal mammary gland development. Germline Lmo2 deletion in mice causes embryonic lethality due to failure in fetal hematopoiesis [26]. Hence, to understand if LMO2 is necessary for the formation of the ductal epithelium, we crossed Lmo2fl/fl mice [27] with Keratin14Cre mice [28] (hence forth referred to as “Krt14Cre”) to specifically delete Lmo2 in the epithelial compartment. Krt14Cre/Lmo2fl/fl mice developed a normal epithelial tree with a mild developmental delay at 5 weeks that was subsequently rescued by week 8 (Supplementary Fig. 3A-C). Flow cytometry analysis of the mammary gland showed no difference in the ratio of basal to luminal cells in Krt14Cre/Lmo2+/+ and Krt14Cre/Lmo2fl/fl in 8–10-week-old mice (Fig. 2A, B).

To test whether Krt14Cre leads to efficient deletion of Lmo2, we performed genotyping and real-time quantitative PCR (qPCR) analysis. While we could detect the deletion product (Supplementary Fig. 3D), qPCR analysis on sorted MECs showed that the Lmo2 transcript is expressed in MECs sorted from Krt14Cre/Lmo2fl/fl mice, albeit at low levels (Supplementary Fig. 3E). This could be due to an inherent inefficiency of Cre that allows cells to escape Krt14Cre mediated deletion and prevents us from fully assessing the impact of Lmo2 deletion in the mammary gland.
Lineage tracing measures a cell’s fate under normal homeostasis while reconstitution assays measure the potential of the cell under austere regenerative conditions [29, 30]. The ability of CD49fhigh/EpCAMmed−low basal cells to regenerate the mammary gland upon transplantation [13, 14] reflects an intrinsic plasticity of the cells. To test whether Lmo2 loss alters reconstitution potential, we used a GFP-tagged lentiviral-Cre-mediated Lmo2 deletion in MECs and performed transplantation assays. We found that acute loss of Lmo2 significantly reduced in vivo reconstitution potential, leading to both fewer and less filled mammary fat pads, suggesting that LMO2 contributes to regeneration of the mammary gland in vivo (Fig. 2C).
To confirm our findings using organoid models, we also generated shRNA against Lmo2 (Supplementary Fig. 3 F, see Methods for shRNA sequences) and transduced sorted luminal and basal cells with either a control vector or shRNA against Lmo2. Consistent with our in vivo transplantation data, we find that Lmo2 knockdown significantly reduced organoid forming capacity of both luminal (Fig. 2D) and basal cells (Fig. 2E). Taken together, these data implicate LMO2 as a positive regulator of in vivo regeneration and clonogenic potential in mouse mammary epithelial cells.

LMO2 Regulates Epithelial-Mesenchymal Cell States in MCF10A Cells
To investigate the function of LMO2 in human mammary epithelial cells, we used the MCF10A cell line as our model, which consists of a heterogeneous population of cells and has been extensively used to study breast tumor initiation and cellular plasticity [31–33]. LMO2 knockdown (Supplementary Fig. 4 A) significantly reduced MCF10A colony formation in 2D (Fig. 3A). Similarly, LMO2 knockdown significantly impaired acini formation in 3D culture (Fig. 3B), suggesting a role for LMO2 in supporting regenerative growth under clonogenic conditions in human mammary epithelial cells.

To test whether the decrease in colony and acini formation with LMO2 knockdown is due to changes in proliferative potential, we performed a WST-1 proliferation assay. LMO2 knockdown in MCF10A cells had no effect on proliferation (Supplementary Fig. 4B). To understand whether LMO2 loss induces apoptosis or decreases cell viability, we performed flow cytometry analysis on MCF10A cells at 4-, 7- and 11-days post-transduction with either a control vector or LMO2 knockdown. We found no significant differences in apoptosis by Annexin V staining or cell viability at all the different time points examined (Supplementary Fig. 4C). Thus, our data suggest that LMO2 promotes clonogenicity through mechanisms independent of proliferation or survival pathways.
Recent studies have shown that cells expressing both epithelial and mesenchymal markers (hybrid, E/M) have higher regenerative and clonogenic capacity than their differentiated counterparts [5, 25, 34–37]. To test whether LMO2 knockdown impacts the hybrid cell state, we performed flow cytometry analysis with CD49f and EpCAM, markers previously used to identify mesenchymal/basal and epithelial cells respectively [16]. We found that MCF10A cells primarily consist of two populations: a double positive/hybrid population (EpCAM+/CD49f+) and a basal/mesenchymal population (EpCAM-/CD49f+) (Fig. 3C top panel, left). Interestingly, LMO2 knockdown significantly reduced the percentage of EpCAM+/CD49f+ and increased the percentage of EpCAM-/CD49f+ cells (Fig. 3C top panel, D, E), suggesting differentiation towards a basal/mesenchymal state. Similarly, we found that LMO2 knockdown reduced the percentage of E-cadherin+ cells (Fig. 3C middle panel, F). To further understand whether LMO2 regulates the intermediate epithelial-mesenchymal state, we utilized the epithelial marker CD104 and mesenchymal marker CD44, which have been previously demonstrated to mark hybrid E/M cells in breast cancer [38]. We found that LMO2 knockdown reduced the proportion of cells in the hybrid state as indicated by a decrease in CD44+/CD104+ cells (Fig. 3C bottom panel, G). Collectively, our data suggests that LMO2 regulates a hybrid cell state in MCF10A cells and LMO2 loss promotes differentiation towards the basal/mesenchymal state.
To determine whether ectopic expression of LMO2 alters the epithelial and mesenchymal populations, we transduced MCF10A cells with ZsGreen and LMO2 overexpression (Supplementary Fig. 5 A). Ectopic expression of LMO2 resulted in no difference in colony formation (Supplementary Fig. 5B) and acini formation (Supplementary Fig. 5 C). Flow cytometry analysis showed increased percentage of EpCAM+/CD49f + cells and a decrease in the basal/mesenchymal (EpCAM-/CD49f+) population (Supplementary Fig. 5D, E). However, we did not see any difference in the mesenchymal or hybrid population defined by CD44 and CD104 markers (Supplementary Fig. 5 F, G), likely due to heterogeneity in marker expression. Overall, our data suggests that LMO2 promotes an epithelial cell state and loss of LMO2 leads to cells becoming mesenchymal.

RNA-Sequencing Identifies Molecular Pathways Regulated by LMO2
As LMO2 is a transcriptional adaptor protein [26], we next sought to identify transcriptional programs regulated by LMO2 that maintain epithelial and mesenchymal cell states in MCF10A cells. Hence, we performed RNA-sequencing analysis on control and LMO2 knockdown cells at 6- and 21-days post-transduction (Fig. 4A). LMO2 knockdown resulted in a significant increase in the expression of genes associated with mesenchymal differentiation such as FN1, MCAM, and FBN1 [39, 40] and down regulation of genes involved in maintaining stem/progenitor cells such as ALDH1A3 [41], LIF [42], and CSF-1 [43] (Fig. 4B). MCAM has been demonstrated as a negative regulator of clonogenicity and regenerative capacity of mammary epithelial cells, where MCAM loss promotes recruitment of macrophages and expansion of MECs through non-canonical WNT signaling [44]. To validate RNA-sequencing analysis, we performed flow cytometry for MCAM and found that LMO2 knockdown increases the percentage of MCAM+ MCF10A cells (Fig. 4D, E), although it did not reach statistical significance for the second shRNA construct. Similarly, we found a trending but non-significant decrease in MCAM in the LMO2 overexpression cells as compared to control (Supplementary Fig. 5H).

Moreover, we saw an increase in expression of genes involved in antigen presentation (HLA-A, HLA-B, and HLA-F) and interferon response (CXCL10, RSAD2) (Fig. 4B, Supplementary Table 1). Gene set enrichment analysis (GSEA) [45] using the Hallmark gene sets showed an upregulation of pathways associated with interferon-α and interferon-γ signaling (FDR q-value < 0.25) (Fig. 4C, Supplementary Table 1), potentially reflecting transcriptional programs that are necessary for hematopoietic stem cell maintenance [46–48]. Although there was an upregulation of specific mesenchymal markers such as MCAM, FN1, FBN1 and LUM, epithelial to mesenchymal transition (EMT) was not significantly enriched in the GSEA at this time point (6 days post transduction) (FDR q-value = 0.28) (Supplementary Table 1).
Previous studies have shown that EMT occurs over time [49–51]. To further characterize this phenotypic shift, we performed RNA-sequencing analysis at 21 days post-transduction with the first shRNA construct. Consistent with our flow cytometry analysis, we found a significant upregulation of mesenchymal genes such as ZEB1, FN1, LUM, CDH2 and POSTN [4] (Fig. 4F). Moreover, the only significantly enriched Hallmark gene set in LMO2 knockdown was epithelial to mesenchymal transition (Fig. 4G, Supplementary Table 1). Thus, our data demonstrates that LMO2 knockdown induces a mesenchymal cell state and promotes upregulation of MCAM, a negative regulator of MEC regenerative capacity.
We next considered whether LMO2 expression is altered as a MEC transitions from an epithelial to a mesenchymal cell state. To test whether LMO2 expression is dependent on the epithelial or mesenchymal state of the cell we treated MCF10A cells with TGFβ1, a known inducer of EMT [52, 53]. After 96 h of treatment, as expected MCF10A became mesenchymal (EpCAM-/CD49f+) and there was a reduction of the double positive (EpCAM+/CD49f+) (Fig. 5A, B). Moreover, we found an increase in the mesenchymal population (CD44+/CD104-), decrease in the hybrid population (CD44+/CD104+) (Fig. 5C, D) and increase in MCAM expression (Fig. 5E). Interestingly, treatment of the MCF10A cells with TGFβ1 for 96 h led to a significant reduction in LMO2 expression while upregulating mesenchymal markers, such as MCAM (Fig. 5F). These data are consistent with our findings where loss of LMO2 results in cells becoming more mesenchymal (Fig. 3).

Discussion
Cellular plasticity is key to maintaining regenerative capacity in the mammary gland and plays an important role in tumor progression [2, 9–11, 54]. Previous studies have implicated several important transcriptional regulators of EMT as critical regulators of this process, such as Snail, Slug, Twist1 and Zeb1 [55–57]. However, the role of other proteins beyond the core EMT transcription factors are only beginning to be understood [58–60].
Our previous studies have shown that LMO2 marks a population of immature tumor cells and functionally LMO2 is required in the early steps of the metastatic process [8]. The role of LMO2 in the normal mammary gland, identified here, aligns with the known role of LMO2 in promoting metastasis rather than primary tumor growth. Notably, the in vivo transplantation assay recapitulates key features of the metastatic process, as successful engraftment requires activation of programs involved in invasion, survival, and proliferation within a foreign microenvironment. Our data are also consistent with previous studies showing that deletion of Snai2 or Twist1 in mammary epithelial cells does not significantly impact formation of the ductal tree during development but impacts in vivo reconstitution [61–65]. While we find that Lmo2 knockout does not impact normal mammary gland development, we are unable to unequivocally measure the full impact of Lmo2 deletion in vivo given incomplete recombination efficiency. Thus, our study primarily assesses the functional role of Lmo2 in experimental regenerative conditions (transplantation) but not homeostatic regenerative conditions (e.g. ablation of specific lineages, pregnancy). Nevertheless, our lineage-tracing and transcriptomic analyses also suggest that, under homeostatic conditions, Lmo2 is expressed in a subset of alveolar progenitors that expand during pregnancy. This pregnancy-induced expansion has been well-documented in other progenitor populations, such as Lgr6-expressing or Blimp1+ luminal progenitors [18, 21]. However, in contrast to these populations, we do not observe Lmo2-lineage traced cells after the 2nd round of pregnancy. Thus, we hypothesize that Lmo2 marks a subset of alveolar progenitors with limited proliferative potential that likely differentiate into milk producing cells and are lost during involution. Previous studies have shown that basal and luminal cells exhibit plasticity during pregnancy, contributing to milk-producing cells and luminal to basal transition [24, 66]. As we pulsed mice early during mammary gland development, our study does not assess the role of Lmo2 in cellular plasticity during pregnancy. Given that Lmo2 is expressed in both luminal and basal layers, our study also does not explore whether Lmo2 marks a rare bipotent cell in the basal layer that can expand during pregnancy to give rise to the milk-producing cells.
Epithelial-to-mesenchymal plasticity (EMP) and maintenance of a hybrid cell state has been recently shown to be a critical factor for tumorigenicity, chemoresistance and metastatic potential in cancer cells [6, 7, 37, 67–70]. Since tumor progression often co-opts developmental and regenerative processes, elucidating these processes under normal conditions is crucial for identifying how they are hijacked during tumorigenesis. Earlier studies had shown that EMT and acquisition of mesenchymal features lead to increased stem cell capacity and tumorigenicity [65]. More recent studies have shown that the hybrid epithelial/mesenchymal state has higher stem cell capacity, tumorigenicity and metastatic capability than a fully differentiated mesenchymal cell state [5–7, 37, 38]. Our data are consistent with the current hypothesis where we find that loss of LMO2 in mammary epithelial cells promotes a mesenchymal differentiated state that is less regenerative [5, 13, 14, 37]. Conversely, while LMO2 overexpression promotes an epithelial phenotype, this effect is not uniform across all hybrid state markers investigated. This discrepancy underscores the inherent heterogeneity of hybrid populations and the markers used to identify them. Our work shows that LMO2 knockdown reduces regenerative capacity in vivo and organoid/acini formation in vitro but does not decrease proliferation in 2D culture. However, it remains to be determined whether LMO2’s contributions to clonogenicity/regenerative capacity and EMP are mechanistically linked or functionally independent both in vitro and in vivo.

LMO2 expression is downregulated during TGFβ-induced EMT. While future research is needed to fill in the mechanistic relationship between LMO2 and cellular plasticity, our data demonstrate LMO2’s potent influence in maintaining a hybrid cell state in mammary epithelial cells. In further support of this notion, we find that LMO2 loss induces expression of several mesenchymal markers, including MCAM. Interestingly, loss of MCAM has the opposite effect where it increases clonogenicity and regenerative capacity of MECs through the IL4-STAT6 axis and non-canonical WNT signaling [44]. We speculate that LMO2 promotes cellular plasticity in part through suppression of MCAM expression. However, we are unable to rescue the phenotype of LMO2 loss with MCAM knockdown due to inherent differences in proliferation/transduction efficiency of epithelial/mesenchymal populations. Collectively, our study identifies a minority population of LMO2+ mammary epithelial cells in the mouse mammary gland and reveals a previously unknown role for LMO2 in promoting the regenerative capacity and maintenance of epithelial-mesenchymal cell states of mammary epithelial cells.

Materials and Methods

Mice
All mice used for this study were maintained at the UCSC Animal Facility/Vivarium in accordance with the guidelines of the Institutional Animal Care and Use Committee (Protocol: Sikas2311dn). Lmo2CreERT2 and Lmo2fl/fl mice were a gift from Dr. Terence Rabbits, Tg(KRT14-cre)1Amc/J (RRID: IMSR_JAX:004782, Strain 018964) and (ROSA)26Sortm4(ACTB−tdTomato,−EGFP)Luo/J (RRID: IMSR_JAX:007676) were purchased from JAX. Lmo2CreERT2/Rosa26mTmG BL6 female mice were subjected to intraperitoneal injection of tamoxifen(Sigma: T5648-1G). Tamoxifen was dissolved in corn oil at a concentration of 10 mg/ml. Each mouse received a dose of 1.5 mg, as previously described [8]. Mice were injected at 4 or 8 weeks and subsequently analyzed at 36 h, 12 w, or 14 w post-injections. The mammary glands were either stained and imaged or enzymatically digested and subjected to flow cytometry. Krt14Cre mice were crossed to Lmo2fl/fl to generate mammary epithelial specific deletion of Lmo2.

Cell Lines
MCF10A (ATCC: CRL-10317, 36 year white female) cells and HEK293T cells were certified by the vendor to be bacteria-, fungus-, and mycoplasma-free. Cell lines have not been authenticated, were thawed at passage 4 and passaged less than 10 times. MCF10A cells were maintained in Advanced DMEM 1×/F12 media (Fisher Scientific: 12634010) with 5% horse serum (Life Technologies: 16050−122), 1% PSA (Fisher Scientific: ICN1674049), 20 ng/mL human EGF (Peprotech: AF-100-15−100UG), 100 ng/mL of cholera toxin (Sigma Aldrich: C8052-0.5MG), 10 µg/mL of insulin (Neta Scientific: SIAL-I9278-5ML), and 0.5 mg/mL of hydrocortisone (Sigma-Aldrich Company: H0888-1G). For TGFβ1 treatment, 30k cells were seeded in 6-well plates and treated with 10 ng/mL of TGFβ1 every other day (Peprotech: 100-21−10UG). Cells were detached at 75% confluency using 0.25% trypsin (Fisher Scientific: MT25053CI) for 15 min at 37 °C, then neutralized with MCF10A culture media and sub-cultured at a ratio of 1:4. HEK293T cells were maintained in DMEM with 10% fetal bovine serum (Fisher Scientific: MT350110CV) and 1% PSA (Fisher Scientific: ICN1674049). Cells were detached at 75% confluency using 0.25% trypsin (Fisher Scientific: MT25053CI) for 5 min at 37 °C, then neutralized with HEK293T culture media and sub-cultured at a ratio of 1:10. All cells were maintained at 37 °C with an atmosphere of 5% CO2. Cell lines were routinely tested for mycoplasma contamination with Mycoplasma Detection Kit (Applied Biological Materials Inc: G238). Cells were frozen at 70% confluency in 90% FBS (Fisher Scientific: MT350110CV) and 10% DMSO (Sigma Aldrich: D2650-100ML) slowly at −80 °C in a freezing chamber, at 1 × 106 cells per vial, for at least 24 h before transferring into liquid nitrogen for long term storage. Images of cells were taken on a Zeiss Live Microscope capturing both brightfield and fluorescence.

Mammary Gland Digestion
L2-5 and R2-5 mammary glands were harvested, minced, and chemically digested overnight in Advanced DMEM1x/F12 with 1% PSA, gentle collagenase/hyaluronidase (Stem Cell Technologies: 7919), and DNAse I at 37 °C, 5% CO2, and added humidity as previously described [8]. Briefly, after overnight digestion, glands were mechanically digested by pipetting with a serological pipette until no tissue pieces were visible. Digested glands were washed with staining buffer (Hank’s balanced salt solution, 2% bovine calf serum, 1% PSA) and centrifuged (500 × g) at 4 °C for 5 min. Red blood cells were lysed with 5 mL of ACK lysis buffer for 5 min at room temperature, and cells were washed with 15 mL of staining buffer. Cells were treated with 0.25% trypsin with EDTA and gently pipetted continuously for 2–3 min to digest the basement membrane. Cells were then treated with DNAse I (Fisher Scientific: NC9709009) and dispase (VWR: # MSPP-7913) and pipetted continuously for 2–3 min to prevent clumping. The single-cell suspension was then filtered through a 40 μm mesh strainer and pelleted via centrifugation (500 × g, 4 °C, 5 min). Cells were then resuspended in staining buffer and transferred to flow cytometry tubes for staining.

Transplantation Assay
Live lineageneg (CD45-/CD31-/Ter119-, see Supplementary Table 2) cells were sorted from Lmo2+/+ and Lmo2fl/fl mice. Cells were gated as shown in Supplementary Fig. 1A-B. Cells were transduced with Lenti-Cre-GFP (GFP.Cre empty vector was a gift from Tyler Jacks [Addgene plasmid #20781; http://n2t.net/addgene:20781; RRID: Addgene_20781]) overnight at an MOI of 10 and then counted before transplantation. Cells were resuspended in media with 33% Growth Factor Reduced Matrigel (Fisher Scientific: CB40230C) and injected into the cleared mammary fat pad of weaning age C57BL/6 (JAX:000664) mice (21–28 days) as previously described [37]. All transplants were allowed to grow 8 weeks before analysis.

Flow Cytometry Analysis
Cells were stained with antibodies listed in Supplementary Table 2 for 15 min at room temperature, as previously described [37]. See Supplementary Fig. 1 A for flow cytometry gating strategies. MCF10A cells infected with pSicoR, shLMO2-1, shLMO2−2, ZsGreen, and LMO2 OE were resuspended in staining buffer consisting of HBSS (Fisher Scientific: MT21022CV) with 2% BCS (Sigma Aldrich: 12133 C-500ML) and 1% PSA (Fisher Scientific: ICN1674049) and stained with antibodies shown in Supplementary Table 2. Stained cells were then washed with staining buffer, resuspended with DAPI (Life Technologies: D1306) at 1:10,000, filtered, and analyzed on a BD Biosciences FACSAria cell sorter. Data were analyzed using FlowJo software (10.10.0).

Mammary Organoids
Luminal and basal cells were sorted from the mammary gland of a female 3-month-old C57BL/6 (JAX:000664) mice as shown in Supplementary Fig. 1A. 96-well low attachment plate (Fisher Scientific: 7200603) plates were seeded with 50 µL of growth factor reduced Matrigel (Fisher Scientific: CB40230C) mixed with 11,000 L-Wnt3a irradiated cells per well. Luminal or basal cells were transduced with pRSI12-TdTomato, shLmo2-1, or shLmo2-2 and seeded at 2500 per well. Organoid media was replenished every two days. The organoid media consisted of Advanced DMEM 1×/F12 media (Fisher Scientific: 12634010), 1% PSA (Fisher Scientific: ICN1674049), 10% FBS (Fisher Scientific: MT350110CV), 50 ng/mL human EGF (Peprotech: AF-100-15−100UG),100 ng/mL mouse noggin (Peprotech: 10773-428), 250 ng/mL human R-spondin-1 (Peprotech: 120-38−50UG), 100 × N-2 supplement (Life Technologies: 17502048), 50× B-27 supplement (Life Technologies: 17504044), 10 mM HEPES (Fisher Scientific: 12634028), 10 µM Y-27632 (Fisher Scientific: 125410), and 1× GlutaMAX (Life Technologies: 35050061). Organoids were maintained at 37 °C with an atmosphere of 5% CO2. Images of cells were taken on a Zeiss Live Microscope. The images were uploaded into Biodock, and the AI was trained to quantify the number and size of the organoids.

Collagen/Matrigel Acini Formation
The protocol was modified from the Brugge Lab (https://brugge.hms.harvard.edu/protocols). 250 µL of collagen I (Fisher Scientific: 354249) was neutralized by adding 31 µL of sterile 10× PBS and sterile NaOH until collagen I had a pH of 7.5. Growth factor reduced Matrigel (Fisher Scientific: CB40230C) was mixed with neutralized collagen I at 4:1 and 40 µL was seeded in a 96-well low attachment plate. MCF10A cells were infected with lentiviruses containing pSicoR, shLMO2-1, shLMO2-2, ZsGreen, or LMO2 OE and seeded in MCF10A growth media with 2% horse serum (Life Technologies: 16050−122) at 2% growth factor reduce Matrigel and hEGF (5 ng/mL, Peprotech: AF-100-15−100UG) at a density of 50 cells/well. The acini were imaged on a Keyence BZ-9000 microscope or Zeiss Live Microscope. The images were uploaded into Biodock, and the AI was trained to quantify the number and size of the organoids.

shRNA Cloning and Virus Production
The lentivirus plasmid pSicoR (pSicoR was a gift from Tyler Jacks (Addgene plasmid #11579; http://n2t.net/addgene:11579; RRID: Addgene_11579 [71]) was utilized as the backbone for generating lentivirus plasmids containing the human short hairpin RNA insertions targeting LMO2. The sequences for the LMO2 shRNA are 5′-GACGCATTTCGGTTGAGAA-3′ and 5′-GCATCCTGTGACAAGCGGATT-3′. The cDNA for LMO2 overexpression was purchased from Genescript and cloned into pHIV-Zs Green vector (Addgene, #18121). The lentivirus plasmid pRSI12 (Cellecta) was utilized for the backbone for generating lentivirus plasmids containing mouse short hairpin RNA insertions targeting Lmo2. The sequences for mouse Lmo2 are 5’-TAATCTCCTAAGAAATGCCTC-3’ and 5’-AATTGCACAACTCTAGTCCAT-3’. HEK293T cells were seeded at a density of 5 × 106 cells per 10 cm plate the day before transfection. For lentivirus production, two separate cocktails were prepared. The first cocktail contained 15 µg of lentivirus plasmid per plate, 7.5 µg of pCMV-R8.91 per plate, and 5 µg of pCMV-VSV-G per plate, diluted in 1.5 mL of Opti-MEM reduced serum media (Life Technologies: 31985070). The second cocktail consisted of 1.5 mL of Opti-MEM reduced serum media supplemented with 40 µL of Lipofectamine 2000 (Fisher Scientific: 11668019) per plate. After a 5 min incubation period, the two cocktails were mixed and incubated for 25 min before adding to the HEK293T cells. The transfection mixture was added to the HEK293T cells and incubated for 6 h. Following incubation, the media was replaced with fresh growth media. After 48 h, the lentivirus-containing media was collected and precipitated using lentivirus precipitate solution (ALSTEM: VC100). The precipitated lentivirus particles were then resuspended in an appropriate buffer for subsequent use in experiments or storage at −80 °C.

Crystal Violet Colony Formation Assay
MCF10A cells infected with lentivirus were seeded in a 6-well cell culture-treated plate at 500 cells per well. The cells were incubated at 37 °C for two weeks, with media changes every 3 days. The cells were fixed with 10% neutral-buffered formalin for 15 min, then stained with 0.01% crystal violet for 60 min. The cells were rinsed with 1× PBS (Life Technologies: 14190144), air dried, and the number of colonies counted.

Annexin V Staining
MCF10A cells were transduced with pSicoR, shLMO2-1, or shLMO2-2 and analyzed on days 4, 7, and 11 after transduction. Culture media was collected, and cells were detached using 0.25% trypsin (Fisher Scientific: MT25053CI) for 15 min at 37 °C, then neutralized with MCF10A culture media. Cells were centrifuged at 300 × g for 5 min at 4 °C, the supernatant was discarded, and the cell pellet was resuspended in 1× binding buffer (Invitrogen: BDB556454) at a concentration of 1 mL per 1 × 10⁶ cells. APC-Annexin V antibody (BioLegend: 640920) was added at 5 µL per 100 µL of 1× binding buffer. Cells were incubated with the antibody for 15 min, followed by a wash with 3 mL of 1× binding buffer. Cells were again centrifuged at 300 × g for 5 min at 4 °C. The final cell pellet was resuspended in 1× binding buffer containing DAPI (Life Technologies: D1306) at a 1:10,000 dilution and analyzed by flow cytometry on the BD FACSAria.

Whole Mount
Mammary glands were dissected and placed on Superfrost slides (Fisher Scientific: 48382-200 (PK)) and placed in Carnoy’s Fixative solution overnight. Carnoy’s Fixative consists of 600 mL of 100% ethanol (VWR International LLC: 71001-628), 300 mL of chloroform (VWR International LLC: MK444004), and 100 mL of Glacial Acetic Acid (Fisher Scientific Company: A385000). The following day, the slides were sequentially immersed in 70% ethanol and 50% ethanol for 15 min each. Afterward, they were rinsed in deionized water for 15 min and stained overnight with Carmine Alum. Carmine Alum consists of 1 g Carmine (Sigma Aldrich: C1022-25G), 2.5 g Aluminum Potassium Sulfate (Fisher Scientific Company: S251521) and 500 mL of distilled water. The Carmine Alum solution was boiled for 20 min and filtered through Whatman paper before storage at 4 °C [72]. The next day, the slides were dehydrated sequentially in 70%, 95%, and 100% ethanol for 15 min each. The glands were then cleared for 5 min in xylene, mounted on slides using Permount (Fisher Scientific: SP15100) and allowed to dry overnight before imaging.

WST-1 Assay
MCF10A cells were seeded at a density of 30,000 cells per well in a 12-well culture plate and transduced with either lentiviral constructs pSicoR, shLMO2-1, or shLMO2-2. Following transduction, cells were re-seeded at 1,000 cells per well in a 96-well plate for viability analysis. Cell viability was assessed every other day using the WST-1 reagent (Neta Science: SIAL-5015944001). After a 2 h incubation with WST-1, absorbance was measured at 440 nm using a microplate reader.

Quantitative PCR (qPCR)
Cells for qPCR were collected from sorted MECs or 2D tissue culture and spun down. RNA was extracted using the RNeasy Micro Kit (Qiagen: 74104) and quantified via ultraviolet spectrophotometer (Nanodrop). cDNA was synthesized using iScript cDNA Synthesis Kit (BioRad: 1708891), then underwent real-time polymerase chain reaction with SYBR Green Universal Master Mix (Fisher Scientific: 4309155) for specific target genes. Primers used were LMO2 F:5’-GGCCATCGAAAGGAAGAGCC-3’ and R:5’-GGCCCAGTTTGTAGTAGAGGC-3’, MCAM F:5’- AGCTCCGCGTCTACAAAGC−3’ and R:5’- CTACACAGGTAGCGACCTCC-3’, Lmo2 F:5’-TCGCTCTCTCTCTTTGGCGT-3’ and R:5’-TGGCTTTCAGGAAGTAGCGG-3’, ACTB F:5’-CACCATTGGCAATGAGCGGTTC-3’ and R:5’-AGGTCTTTGCGGATGTCCACGT-3’, GAPDH F:5’-GTCTCCTCTGACTTCAACAGCG-3’ and R:5’-ACCACCCTGTTGCTGTAGCCAA-3’, Actb F:5’-GGCTGGATTCCCCTCCATCG − 3’ and R:5’-CCAGTTGGTAACAATGCCATGT-3’, Gapdh F:5’-TGGCCTTCCGTGTTCCTAC − 3’ and R:5’- GAGTTGCTGTTGAAGTCGCA − 3’. All expression data was normalized to housekeeping controls.

Library Construction, Quality Control and Bulk RNA Sequencing
Bulk RNA sequencing was performed by Novogene. MCF10A cells transduced with either control (pSicoR) or shLMO2 knockdown vectors after 6 and 21 days were grown to 85–90% confluency in a 6-well plate. RNA was isolated according to manufacturer’s instructions (Qiagen RNeasy Plus Micro Kit, 74034). Messenger RNA was purified from total RNA using poly-T oligo-attached magnetic beads. After fragmentation, the first strand cDNA was synthesized using random hexamer primers, followed by the second strand cDNA synthesis using either dUTP for directional library or dTTP for non-directional library. For the non-directional library, it was ready after end repair, A-tailing, adapter ligation, size selection, amplification, and purification. For the directional library, it was ready after end repair, A-tailing, adapter ligation, size selection, USER enzyme digestion, amplification, and purification. The library was checked with Qubit and real-time PCR for quantification and bioanalyzer for size distribution detection. Quantified libraries were pooled and sequenced on Illumina platforms, according to effective library concentration and data amount.

Immunoblotting
Whole lysates were generated from MCF10 cells transduced with ZsGreen and LMO2 OE by RIPA buffer (EMD Millipore: 20–188) with protease and phosphatase inhibitor (Thermo Scientific: 1861281). SDS-PAGE gel (BIO-RAD: 4561094) was run at 100 V for 1 h, and transferred to a polyvinylidene difluoride membrane (EMD Millipore: IPVH85R) for 70 V for 90 min. The membrane was block in 5% BSA (RPI: 9048-46−8) Tris-buffered saline (BIO-RAD: 1706435)–Tween 0.1%(Fisher Scientific: BP337-500) (TBST) at room temperature for 1 h and subsequently probed with primary anti-hLMO2 (R&D: afaf2726) at 1:1000 overnight at 4 °C. Incubation with the secondary antibody for 1 h containing goat-HRP at 1:5000 and actin-HRP at 1:5000 (Invitrogen: MA5-15739-HRP). Visualization was enabled by using the reagent Clarity Western ECL Substrate (BIO-RAD: 170–5060) and BIO-RAD ChemiDoc™ MP Imaging System.

Data Quality Control
Raw data (raw reads) of fastq format were processed through fastp software. In this step, clean data (clean reads) were obtained by removing reads containing adapter, reads containing poly-N and low quality reads from raw data. Q20, Q30 and GC content were calculated. All the downstream analyses were based on clean data with high quality.

Reads Mapping to the Reference Genome
The human reference genome (hg38) and gene model annotation files were downloaded from UCSC genome website directly. The index of the reference genome was built using Hisat2 v2.0.5 [73] and paired-end clean 1 reads were aligned to the reference genome using Hisat2 v2.0.5. We selected Hisat2 as the mapping tool for which Hisat2 can generate a database of splice junctions based on the gene model annotation file and thus a better mapping result than other non-splice mapping tools.

Quantification of Gene Expression Level
FeatureCounts [74] v1.5.0-p3 was used to count the reads numbers mapped to each gene. Gene expression was then converted to transcripts per million (TPM) by normalizing for gene length, then sequencing depth.

Differential Expression Analysis and Gene Set Enrichment Analysis
Differential expression analysis of two conditions (three biological replicates per condition) was performed using the DESeq2 R package (1.20.0) [75]. DESeq2 provides statistical routines for determining differential expression in digital gene expression data using a model based on the negative binomial distribution. Differentially expressed genes were identified by comparing control and LMO2 knockdown cells. The resulting P-values were adjusted using Benjamini-Hochberg’s approach for controlling the false discovery rate. Genes with an adjusted P-value ≤ 0.1 found by DESeq2 were assigned as differentially expressed heatmaps. GSEA [45] was performed on all DEGs at day 6 post-transduction and a pre-ranked list of genes at 21 days post-transduction. The FDR q value < 0.25 was considered significant between control and knockdown conditions using the Broad Institute’s software.

Statistical Analysis
All graphs display the average as central values, and error bars indicate ± SD unless otherwise indicated. P values are calculated using paired or unpaired t test, ANOVA, Wilcoxon rank-sum test, and Mann-Whitney U test, as indicated in the figure legends. All P and q values were calculated using Prism (10.2.2), unless otherwise stated. For animal studies, sample size was not predetermined to ensure adequate power to detect a prespecified effect size, no animals were excluded from analyses, experiments were not randomized, and investigators were not blinded to group allocation during experiments. For organoid/acini quantification, data points affected by Matrigel detachment were not quantified.

Supplementary Information