Functional delineation of the luminal epithelial microenvironment in breast using cell-based screening in combinatorial microenvironments

Abstract

Microenvironment signals are potent determinants of cell fate and arbiters of tissue homeostasis, however understanding how different microenvironment factors coordinately regulate cellular phenotype has been experimentally challenging. Here we used a high-throughput microenvironment microarray comprised of 2640 unique pairwise signals to identify factors that support proliferation and maintenance of primary human mammary luminal epithelial cells. Multiple microenvironment factors that modulated luminal cell number were identified, including: HGF, NRG1, BMP2, CXCL1, TGFB1, FGF2, PDGFB, RANKL, WNT3A, SPP1, HA, VTN, and OMD. All of these factors were previously shown to modulate luminal cell numbers in painstaking mouse genetics experiments, or were shown to have a role in breast cancer, demonstrating the relevance and power of our high-dimensional approach to dissect key microenvironmental signals. RNA-sequencing of primary epithelial and stromal cell lineages identified the cell types that express these signals and the cognate receptors in vivo. Cell-based functional studies confirmed which effects from microenvironment factors were reproducible and robust to individual variation. Hepatocyte growth factor (HGF) was the factor most robust to individual variation and drove expansion of luminal cells via cKit+ progenitor cells, which expressed abundant MET receptor. Luminal cells from women who are genetically high risk for breast cancer had significantly more MET receptor and may explain the characteristic expansion of the luminal lineage in those women. In ensemble, our approach provides proof of principle that microenvironment signals that control specific cellular states can be dissected with high-dimensional cell-based approaches.

INTRODUCTION

Tissues can be viewed as ecosystems of interacting cells that are simultaneously producing cues and responding to physical and biochemical cues in their immediate surrounding. The dynamic and reciprocal relationship between cells and the microenvironment establishes and maintains cell states[1]. Defined as the sum of cell-cell, cell-extracellular matrix (ECM), cell-soluble factor interactions--in addition to geometric and physical states--the microenvironment is a complex milieu[2–4]. Omics strategies have been used to dissect and explain the complexities of genomes, proteomes, and metabolomes. While proteomics has been used to characterize the content of microenvironments[5,6], there are few methods for efficiently linking cellular functions and phenotypes with combinations of microenvironment components. The microenvironment microarray (MEMA) platform was designed to interrogate the functional impact of diverse microenvironmental signals, including insoluble ECM and soluble growth factors, on cell phenotype, and provides image-based read-outs at the single-cell level [7–9].

The mammary gland consists of a bilayered branching epithelia enmeshed in a basement membrane and surrounded by stroma. The stroma consists of adipose, endothelial, mesenchymal, and immune cells. Evidence suggests that breast cancers arise from the innermost luminal layer of the epithelia, either from the luminal cells themselves or from luminal biased progenitor cells [10–12]. In addition, striking increases in the proportion of luminal cells and progenitors is a consequence of aging that is associated with increased cancer risk [13–15]. The potential links between luminal cells and breast cancer merits a better understanding of their regulation. Primary normal human mammary epithelial cells (HMEC) in low stress media are an excellent in vitro model of luminal epithelial cells because they can be supported for about eight passages, thereby enabling empirical examination [16]. Moreover, HMEC at early passage retain in vivo lineage-specific transcriptional and proteomic profiles[14,17,18]. Despite some successes in growing luminal cells for a short time in culture, there remains a general lack of understanding of the factors in vivo that regulate luminal cell homeostasis. Identifying these factors will provide insights into biological mechanisms of cell state control and could also serve as factors that enable more robust in vitro expts.

Utilizing the MEMA platform together with primary human mammary epithelial cells, we identified combinations of microenvironment proteins that supported expansion of keratin 19 (KRT19) expressing luminal epithelial cells. The ECM and growth factor proteins that comprised the microarrays represent signals found in diverse tissues, providing an opportunity for broad assessment of factors that may support growth and maintenance of luminal epithelial cells. The microenvironment protein combinations we identified as supportive of luminal cell growth are known to be present in breast tissue, and in some cases, have known effect on luminal cells. Transcriptome profiling of different cell types from breast stroma and epithelia demonstrated that the luminal-supportive factors are produced by adipose, fibroblasts, and myoepithelial cells. We focus on hepatocyte growth factor (HGF), which is produced in adipostromal cells, and is known as a luminal cell mitogen[19,20]. We show that HGF maintains luminal epithelia by working through a luminal biased progenitor intermediate. These studies shed light on the mechanisms and molecular signals that may be involved in early breast cancers, as HGF is increased in women at high risk for breast cancer through increased receptor tyrosine-protein kinase Met (MET) expression in luminal cells. Overall, our studies demonstrate the utility of the MEMA platform to functionally define cellular microenvironments.

RESULTS

High-throughput screen to identify microenvironmental proteins that support the growth of luminal epithelial cells

We used a high-throughput MEMA screen to examine the impact of 2640 combinations of 55 soluble ligands and 48 insoluble ECM proteins on cellular phenotype [7,8] (Fig 1A and B). Soluble ligands and ECMs include growth factors, cytokines, and those that have implicated developmental, heparin-binding and mitogen functions (Table 1). In the MEMA assay, insoluble proteins are spotted onto arrays to form ~400 uM growth pads upon which cells can grow; addition of soluble ligands yields pairwise combinations of microenvironment factors. Each treatment condition was assessed with a median of 14 replicates, yielding over 44,000 individual cell culture spots. Primary human mammary epithelia cells (strain 240L) were seeded onto plates, allowed to attach to individual growth pads, treated with ligand for 72H (Fig 1C), then imaged to assess for expression of cytokeratins KRT5 (basal marker), KRT19 (luminal marker) and incorporation of EdU (cell proliferation) (Fig 1D). All images underwent quantitative image analysis to identify morphological and phenotypic changes associated with microenvironment perturbation. We used unsupervised hierarchical clustering to identify microenvironment conditions that induced similar cellular phenotypes. Microenvironment perturbations induced cellular phenotypes associated with differentiation state, nuclear morphology, cell cycle state and cell count (Fig 1E). We identified the ligands that most strongly modulate the proportion of KRT19+ cells by immunofluorescent (IF) imaging (Fig 2A). BMP2 and CXCL1 were among the ligands that significantly reduced luminal cell proportions while NRG1, HGF, and TNFS11 lead to increased luminal cell proportions (Fig 2B).

Figure 1.

MEMA cellular arrays use immunofluorescence imaging to quantify phenotypic responses to over 2600 microenvironment perturbations A) Identical MEMAs are printed into 8 well plates with ECM proteins randomly located around a grid of COL1 spots (shown as small black triangles). B) 40,842 of the 44,707 microenvironment pads passed quality control. C) Cells are incubated for 3 days in 1 of 55 ligands or plain media then fixed and stained for morphology and biomarker features. D) 4 channel immunofluorescent images of each spot are quantified to generate over 200 data and metadata values per cell E) Hierarchical clustering of image features in rows and microenvironments in columns shows clustering of phenotypes.

Table 1.

Roles in mammary gland and breast cancer for factors identified by MEMA to modulate luminal cell numbers

ME	Known functions in mammary gland	Known functions in breast cancer	Expressed	References
NRG1	Activate luminal cell mammopoiesis and lactation	Levels are shown to be elevated in some breast cancers, however gene is frequently silenced by methylation in breast cancer, might have also tumor suppressor gene	Myoepithelial cells	[47–49]
BMP2	Mammary progenitor niche component	Dysregulated BMP2 signaling drives mammary epithelial cell malignant transformation	Myoepithelial cell, Stroma	[50]
CXCL1	Role in immune cell attraction during mouse mammary tissue involution	Stromal CXCL1 expression predicted poor prognosis	Epithelial cell, Stroma	[50,51]
TGFB	Critically important to mammary morphogenesis and secretory functions through regulation of epithelial cell proliferation, apoptosis and ECM. Inhibits the synthesis of milk casein proteins.	Drives mammary epithelial cell malignant transformation	Stroma during development and pregnancy	[52,53]
FGF2	Regulate ECM production and enhance fibroblast-induced mammary epithelium branching	Drives mammary epithelial cell malignant transformation	Stroma	[54,55]
PDGFB	Increase epithelial cell proliferation, while levels are decreased during involution. Vary depending on hormonal status	Drives tumor vasculature, epithelial mesenchymal transition and invasion	Epithelial cells and stroma	[56]
TNFSF11/RANKL	Induce expansion of mammary progenitor cells during the normal oestrous cycle and during pregnancy	Drives the onset of hormone-induced mammary cancer		[57]
WNT3A	In vitro inhibit mammary progenitor cell differentiation. and support progenitor cell functions	Drives breast cancer cells proliferation and invasion	Not expressed in mammary gland, expressed in breast cancer	[58]
SPP1	Drives mammary gland morphogenesis and lactation	Promote migration, invasion. Prognostic for poor outcome.	Myoepithelial cells	[59–63]
Hyaluronan	Modulate growth factor induced mammary gland branching, furthermore hyaluronan size seems to be determinant for this modulation activity	Promote cancer cell survival, migration, invasion. Prognostic for poor outcome.	Myoepithelial cells and stroma	[64,65]
VTN	Drives mammary gland morphogenesis and lactation	Increased expression in breast cancer, support tumor growth	Stroma, leak from vessels	[66]
OMD	unknown	Increased expression in older patients with breast cancer	Adipocytes	[67]

Figure 2.

Hits are chosen from the ends of the normalized luminal proportion distributions A) Ligands are arranged based on the proportion of luminal cells growing on the control (collagen type I) spots. Ligand hits are chosen from the upper end of the normalized values. B Images from the spots in the tails of the distribution show the range of luminal cells. C) Percentage of luminal cells grown in collagen type I coated wells with ligands picked from the MEMA screen. D) Comparing the MEMA screen to the validation data shows good correlation at the extremes.

Microenvironment mediated luminal phenotypes that were robust to individual variation

We next sought to replicate key findings from the high-throughput MEMA screen using standard cell culture methods. Well plates were coated with collagen type 1 (COL1) and HMEC (Strain 240L, same used for the MEMA screen) were cultured in these plates until 90% culture confluency (5–6 days) in the presence of the most significant luminal cell regulatory ligands identified in the MEMA screen: BMP2, CXCL1, TGFbeta, FGF2, PDGFbeta, Wnt3a, TNSF11, HGF, and NRG1. These studies confirmed that the shift in the proportion of KRT19+ cells in 240L cultures were in general agreement with MEMA results (Pearsons=0.978, p<0.001; Fig 2C, 2D). We used flow cytometry as an orthogonal approach to assess the shift in luminal populations by quantifying CD227⁺/CD271⁻ cells (CD227, sialomucin 1, is a luminal marker; CD271, nerve growth factor receptor is a myoepithelial marker). BMP2 decreased and HGF increased luminal proportions, confirming the KRT19 results from the MEMA screen (Fig 3A). To test if the BMP2 and HGF effects generalized to other strains, we tested three HMEC strains from different individuals, including the strain tested in the MEMA screen as a comparator (strain IDs: 240L, 124 and 172). BMP2 repressed growth of KRT19+ cells, whereas HGF promoted their growth in all three HMEC strains: 240L, 124 and 172, however the overall effect sizes varied between strain (Fig 3B). In summary, MEMA results with 240L cells using orthogonal assays revealed that the top hits were reproducible, but repeating the validation assays with two additional HMEC strains (strains 124 and 172) showed some inter-individual variation. HGF and BMP2 elicited similar effects in multiple cell strains and were therefore the focus of deeper functional studies.

Figure 3.

Validation of MEMA results. (A) 240L HMEC were cultured in COL1 coated plastic culture dishes with soluble ligands identified in MEMA experiments. After 5–7 days, luminal epithelial cell population proportion was analyzed from subconfluent cultures using two orthogonal methods and markers; immunofluorescence (IF)-KRT19 and FACS -CD227. The luminal epithelial cell population was calculated as a proportion from population size in COL1 only. The figure shows correlation curve between these two methods & markers. Significant (*** p<0.001, Dunnet post hoc test) upregulation of luminal epithelial cell population (both methods) was found when cells were treated with HGF. Each ligand culture had 4–6 replicates. (B) Three primary HMEC strains; 124, 172 and 240L were cultured similarly as in 3A and whisker plots show KRT19+ luminal epithelial cell population proportion relative to COL1 only, in BMP2 and HGF treated cultures. Plots shows data distribution from 4–8 replicates. (C) Bar graph showing proportions of KRT19+ cells in MEMA to identify pairwise combinations of HGF with other ECM and matricellular proteins, in order to identify factors that modulate the luminal expansion effect of HGF. (D) 240L HMEC were cultured in culture plastic coated with the indicated ECM molecules from 3C and HGF was added to the culture medium. Whisker plots shows proportion of KRT19+ cells relative to culture plastic alone from 2 replicate experiments. (E) Cell counts in HGF plus VTN, COL1, LMWHA and SPP1 all show good concordance between the MEMA and coated well assays. Error bars show standard deviation.

The mammary gland contains a complex basement membrane comprised of diverse ECM proteins. We therefore analyzed the MEMA data to identify ECMs that modulated HGF-induced luminal proportions. The MEMA experiments identified several ECMs that synergize with HGF to further amplify the KRT19+ cells: osteomodulin (OMD), low molecular weight hyaluronan (LMWHA), and osteopontin (SPP1); whereas HGF combined with vitronectin (VTN) did not show increase of KRT19+ cells (Fig 3C). We assessed these effects using a multi-well validation method, which revealed that VTN inhibited HGF-induced KRT19+ cell expression while low molecular weight hyaluronan and SPP1 enhanced the HGF effect (Fig 3D). Comparison of the median cell counts in these conditions shows good concordance between the MEMA and coated well assays (Fig 3E). Taken together, orthogonal methodologies validated the high-throughput MEMA screen to identify microenvironment proteins that modulate growth of luminal epithelial cells and demonstrated that the synergy between HGF and the matrix factors OMD, LMWHA, and SPP1 was both replicable and robust.

Evidence of an epithelial-stromal circuit

We next sought to determine which cell types within the mammary gland—if any—produced the microenvironment proteins that were identified on MEMA, as well as their known or putative receptors. We performed whole transcriptome RNA sequencing on adipose, stromal fibroblasts, luminal epithelial (LEP) and myoepithelial (MEP) cells from 21 women who underwent breast reduction or prophylactic mastectomy surgeries. Transcriptome analyses showed, for example, HGF is produced exclusively by adipocytes and fibroblasts (Fig 4A), whereas its receptor MET is expressed by myoepithelial and luminal epithelial cells—suggesting a direct interaction between these distinct cell populations (Fig 4B). The ECM factors that modulate HGF effects are produced exclusively by non-luminal cells: OMD is produced entirely by adipocytes, SPP1 is produced by myoepithelial cells, and all the hyaluronan synthases (HAS1, HAS2, and HAS3) are produced either in the stroma or by the myoepithelial cells (Fig 4A). The mRNAs corresponding to putative receptors for these factors were all expressed in luminal epithelial cells (Fig 4B). RNA-sequencing of the different cell types found in breast combined with the cell-based functional MEMA screening created a model of the luminal epithelial cell microenvironment in breast tissue (Fig 4C). These studies identified specific molecular factors that may mediate cell-cell communication in the mammary gland to modulate epithelial cell populations.

Figure 4.

RNA-sequencing of cell types from breast identify the cellular sources of microenvironment factors and the cell types with the cognate receptors. (A) Relative mRNA expression (TMM values) of microenvironment factors in adipocytes (yellow), fibroblast (blue), myoepithelial cell (red) and luminal cell (green), shown in pie plots. (B) Bar graphs showing RNA expression (TMM-log2 values) of the receptors for the microenvironment factors, delineated by cell type. Results are mean±SD from 23 epithelial, 20 fibroblast and 25 adipocyte tissues. (C) Summary figure of mRNA expression results depicting a model of the luminal epithelial cell microenvironment. Factors that were shown to modulate luminal cell numbers in MEMA are produced in myoepithelial cells, fibroblasts, and fat.

MET expression is correlated with HGF-driven increases in luminal cell proliferation and amplification of cKit+ epithelial cells

Because RNA sequencing analysis showed the presence of MET receptor in both myoepithelial and luminal cells, we performed an HGF dose response study to determine the effect of growth in both lineages. The optimal concentration of HGF was deemed to be 40nM because it was the lowest concentration that begot maximal effect on cell number (Fig 5A). HGF was more mitogenic to FACS-enriched luminal cells (CD227+/CD271−) compared to FACS-enriched myoepithelial cells (CD227−/CD271+) or compared to luminal cells co-cultured with myoepithelial cell (Fig 5A). Myoepithelial cells (CD227−/CD271+) treated with 40 nM HGF showed a 5% upregulation of cell number compared to non-treated control wells, while luminal cells (CD227+/CD271−) showed a 114% increase. These results indicate that despite the expression of the MET receptor in both epithelial cell lineages, the mitogenic effect of HGF is more potent in the luminal cell population.

Figure 5.

HGF effect on luminal epithelial cell proliferation and differentiation. (A) Line plots show the effect of HGF concentration on proliferation of the epithelial lineages. 240L HMEC were FACS sorted using CD227+ (luminal, LEP) and CD271+ (myoepithelial, MEP) markers. Sorted cells, LEP or MEP, or unsorted LEP+MEP were cultured in 96-well plates and treated with increasing concentration of HGF (0–800nM, added every other day) and relative cell number was analyzed after three days using Cell titer Glo assay. Results are mean±SD from 5 replicate experiments. (B) Bar graphs showing cell numbers of 240L HMEC that were FACS sorted using progenitor marker cKit or luminal epithelial cell (LEP) marker CD227, and subsequently treated with HGF (40ng/ml). After 5 days, cultures were fixed, and cell number/well counted under the microscope. Results show mean from 2 replicate experiments. (C)Three strains of HMEC were treated with soluble ligands identified in MEMA experiments and the cKit+ progenitor population was analyzed from subconfluent cultures using FACS. Bar graphs show fold changes compared to COL1 only cultures. With HMEC strain 240L experiment was repeated four times and error bars show mean±SD. With HMEC strains 124 and 172 experiment was repeated once. There was significant (** p=0.00124, Dunnet post hoc test) upregulation of the ckit+ population when 240L cells were treated with HGF. (D) MET receptor mRNA expression in FACS sorted lineages of LEP (CD227+/CD271−), MEP (CD227−/CD271+), cKit+/Axl−, or cKit+/Axl+ cells 240L and 124 HMEC. Results are mean of 4 samples and significance *p<0.001 by Tukey’s multiple comparison test.

Human mammary epithelia expressing the receptor tyrosine kinase cKit are enriched for luminal-biased progenitor activity[13,21], thus we examined the impact of HGF on the expansion of cKit+ HMEC. HGF caused a 2 to 4-fold expansion of cKit+ cells and a corresponding 3-fold increase in luminal cells (Fig 5B). The other luminal-expanding factors identified on the MEMA did not have as much effect on the cKit+ cells as compared to HGF as determined by FACS (Fig 5C). MET expression levels in myoepithelial, luminal and cKit+ progenitor cells were then compared. cKit+/Axl− cells are suggested to present luminal biased progenitors and cKit+/Axl+ cells express bipotent progenitor phenotype. Our results show that cKit+/Axl+ progenitor cells had 3-fold greater expression of MET as compared to more differentiated cells (Fig 5D). These data suggest that HGF exerts its mitogenic effect on luminal cells via expansion of the luminal-biased progenitor cells.

Luminal epithelial cells or luminal-biased progenitors are thought to be the cell of origin for most breast cancers, and we have demonstrated that women who are germline carriers of high-risk breast cancer mutations have an expanded luminal epithelia relative to age-matched average-risk women[15]. We examined HGF and MET expression in average-risk HMEC, which were obtained from reduction mammoplasty, and compared that to HMEC from high-risk women undergoing prophylactic mastectomy because they were germline carriers of BRCA1, BRCA2, ATM, or PALB2 mutations. HGF expression levels did not change with risk status in adipose, fibroblasts, luminal, or myoepithelial cells (Fig 6A). MET levels were unchanged with risk in adipose, fibroblasts, and myoepithelial cells, whereas luminal cells from high-risk strains showed significantly elevated MET expression relative to average-risk strains (Fig 6B).

Figure 6.

HGF signaling in high-risk mammary tissue. HGF (A) and MET receptor (B) mRNA expression in normal and high breast cancer risk mammary tissue cells; adipocytes (ADI), fibroblast (FIBRO), luminal epithelial cell (LEP) and myoepithelial cell (MEP). Results are TMM mean±SD from 23 epithelial, 20 fibroblast and 25 adipocytes replicative tissue. Significance between normal and high-risk group ** p=0.00119, Dunnet post hoc test.

DISCUSSION

Tissue microenvironments are potent determinants of cell fate and lineage specificity. Phenotype-microenvironment association screens are a tractable way to generate detailed functional understandings of how tissue microenvironments work to maintain lineage-specificity. We used the MEMA platform to identify combinations of tissue microenvironment components that direct mammary progenitor cell differentiation[8], that maintain chemo-resistant cancer stem cell phenotypes [22], and that modulate sensitivity to pathway-targeted cancer drugs [23,24]. Because a majority of breast cancers are luminal subtype, and likely are derived from cells that reside in the luminal microenvironment, we sought to identify microenvironment components that supported normal mammary luminal epithelial cell expansion. Indeed, maintenance of normal luminal epithelial cells in culture has been a persistent barrier to the ability to study them in vitro. The MEMA we used contained 2640 combinations of ECM and growth factors that are enriched across diverse tissue types, including breast, brain, and bone. Thus, it is noteworthy that many of the components that regulated luminal cell expansion were already known to impact mammary morphogenesis or breast cancer. Thus while some of these particular findings are not novel in the strictest sense, we show that the MEMA approach was an efficient way to identify microenvironmental regulators of a cell state, which could be useful for identifying potential in vivo reguators or in vitro regulators the elicit specific behaviors from cell systems (Table 1).

We used quantification of KRT19 expressing cells as an assay for expansion or contraction of the luminal epithelial cell population. Expression of KRT19 in luminal subtype cancers is likely a vestige of the lineage of origin, which is thought to be mature luminal epithelial cells. Data from the luminal-like breast cancer cell line MCF7 suggested that KRT19 might be involved in proliferation through association with cyclin Ds[25]. Conversely, KRT19 silencing was also shown to reduce proliferation of luminal-like BT474 and SKBR3 breast cancer cells due to decreased AKT signaling in an EGR1 dependent manner [26]. KRT19 is a well-defined biomarker of luminal cells in a normal context, as we are using it, though its mechanistic role is not yet understood.

During the course of validation, we deeply examined the role of HGF because we found it to be highly reproducible between the MEMA and other experimental platforms, as well as robust in that the luminal cell expansion effect was detectable in primary HMEC strains from three different individuals. HGF is known to play roles in breast tumor progression[27,28], and in normal mammary gland morphogenesis in mice, as well as a luminal cell mitogen in cultured cells [20]. The major lineages, luminal and myoepithelial, express MET, the receptor for HGF. Using FACS-enriched lineages we observed that the mitogenic effect of HGF is limited to luminal cells, inspite of the fact that myoepithelial cells express more MET than luminal cells. That luminal cell proliferation was relatively decreased when cocultured together with myoepithelial cells suggests that myoepithelial cells provide additional regulatory signals. For example, BMP2 is mainly expressed by myoepithelial cells, and was predicted to suppress luminal cell proliferation based on the MEMA results. This interpretation is consistent with a luminal microenvironment in which positive and negative signals are integrated to balance and maintain the luminal population.

HGF/MET overexpression is a negative prognostic factor for breast cancer patients, and it is suggested that MET dysregulation plays a role in development of acquired drug resistance[29]. The role of HGF/MET in breast cancer initiation and susceptibility is not known. Of the factors we examined, HGF had the greatest effect on cKit+ cells. RNA-sequencing of adipose, fibroblasts, luminal, and myoepithelial both from average-risk and age-group matched high-risk breast tissues revealed a naturally occurring variation in MET expression; luminal cells from high-risk epithelia expressed more as compared to luminal cells from normal-risk. In addition, all HGF was produced from stromal fat and fibroblasts. We used to our advantage the natural variation in MET expression to demonstrate that the mitogenic effect of HGF was correlated with the level of MET receptor present in luminal cells, suggesting a regulatory relationship. Others have implicated cKit+ cells as breast cancer cells of origin for basal-subtype breast cancers[30], which is the most prominent in subtype in high-risk women with germline BRCA1 mutations[31]. It is tempting to speculate that high MET expression is related to the characteristic expansion of luminal cells in high-risk epithelia[15], and that HGF plays a role in amplifying the putative cancer cells of origin in breast tissue. Additionally, these findings could offer insights into MET-targeted breast cancer therapy, hinting at the potential benefits of MET-targeted therapy for high-risk women with germline BRCA1 mutations, given their elevated expression of the MET receptor.

The number of potential interactions in a tissue microenvironment that are deterministic of cell fate and function is not infinite, but there is significant complexity. Transgenic and gene silencing approaches have led to detailed understandings of individual microenvironment components[32–34], however few studies have examined how multiple components interact with and counterbalance one another. The MEMA approach simplifies microenvironments into hundreds or thousands of overlapping pairwise combinations of factors, and by using linear modeling approaches, the components with reproducible and dominant activities can be identified. We propose that this cell-based functional approach is powerful for establishing functional relationships between microenvironments and specific cellular states and is an efficient methodology for understanding the microenvironment basis of tissue specificity.

METHODS

Cell culture-

Human mammary epithelial cells were cultured in M87A media supplemented with cholera toxin at 0.5 ng/ml (Sigma-Aldrich) and oxytocin at 0.1nM (Bachem, Switzerland)[35]. Cells were isolated from reduction mammoplasty (RM), contralateral (C), prophylactic mastectomies (PM), peripheral to DCIS (P), strains are listed in Table 2. Epithelial cells were maintained as previously described[35,36]. All discarded breast tissues that were used to establish epithelial cell strains, fibroblasts, or adipose were obtained with institutional review board approval at City of Hope (Duarte, CA) or Lawrence Berkeley National Laboratory (Berkeley, CA). Women were eligible for the present study if they were undergoing a breast reduction or prophylactic mastectomy. Women were defined as HR if they had a germline mutation that increased their risk of breast cancer including BRCA1, BRCA2 and PALB2; women were defined as AR if they had a ≤12% lifetime risk of breast cancer by not possessing any predisposing germline mutation.

Table 2.

Human mammary epithelial strains and associated adipostromal cells used in this study. strain ID, age of individual during surgery, Tissue source (reduction mammoplasty (RM), Prophylactic mastectomy (PM), contralateral (C) or peripheral to DCIS (P)), Cell type (epithelial (E), Fibroblast (F) and adipocyte (A)), breast cancer risk status and risk gene, which experiments/figures has each strain used.

Strain ID	age	Tissue source	Cell type	Risk	Used in figure
240L	19	RM	E, F	normal	All
6	40	RM	E, F	normal	4,6A,6B
30	49	RM	E, F	normal	4,6A,6B
51	27	RM	F	normal	4, 6A, 6B
117	56	RM	E, F	normal	4,6A,6B
122	66	RM	E, F	normal	4,5E,6A,6B
124	29	RM	E	normal	3B,5D,4,6A,6B
153	60	RM	F	normal	4, 6A, 6B
160	16	RM	F	normal	4, 6A, 6B
169	35	RM	E, F	normal	4,6A,6B
172	28	RM	E	normal	3B,5D
184	21	RM	E	normal	4,6A, 6B
237	66	RM	F	normal	4, 6A, 6B
C001	52	PM	E,F	High-BRCA2	4,6A,6B
C002	50	PM	E, F	High-BRCA1	4,6A,6B
C003	33	PM	E, F	High-BRCA2	4,6A,6B
C004	40	PM	E, F	High-BRCA1	4,6A,6B
C008	53	PM	E, F	High-BRCA1	4,6A,6B
C009	40	PM	E, F, A	High-BRCA1	4,6
C014	52	PM	E, F, A	High-PALB2	4,6A,6B
C017	36	PM	E, F	normal	4,6A,6B
C018	40	C	E, F	normal	4,6A,6B
C019	54	C	E, F, A	normal	4,6A,6B
C020	44	C	E, F, A	High-BRCA2	4,6A,6B
C023	35	PM	E, F, A	High-BRCA1	4,6
C025	45	C	A	normal	4, 6A, 6B
C028	41	C	A	normal	4, 6A, 6B
C036	51	C	A	normal	4, 6A, 6B
C039	55	C	E, F, A	High-PALB2-VUS and APC-VUS	4,6A,6B
C041	44	C	A	High-TP53, BAP1, VUS	4, 6A, 6B
C044	44	C	A	High-BRCA2	4, 6A, 6B
C045	70	P	A	normal	4, 6A, 6B
C046	52	PM	A	High-BRCA2	4, 6A, 6B
C048	43	C	A	normal	4, 6A, 6B
C049	55	C	A	High-PALB2	4, 6A, 6B
C051	33	C	A	normal	4, 6A, 6B
C053	31	C	A	High-BRCA1-BRCA2	4, 6A, 6B
C057	57	C	A	High-BRCA1-BRCA2-VUS	4, 6A, 6B
C063	36	C	E, F, A	High-BRCA1	4,6A,6B
C067	69	C	A	normal	4, 6A, 6B
C071	71	C	A	normal	4, 6A, 6B
C078	24	C	A	High-BRCA1	4, 6A, 6B
C081	54	C	A	High-PALB2	4, 6A, 6B
C085	39	C	A	High-BRCA1	4, 6A, 6B
C086	66	C	A	normal	4, 6A, 6B

MEMA –

Level 2 data from the HMEC240L staining set 4 microenvironment microarrays were downloaded from Synapse at (https://www.synapse.org/#!Synapse:syn2862345/wiki/394513) and processed using R[37,38]. Each feature was rrscaled [39] to gaussianize the distributions, remove outliers and convert to z scores. Each feature was then RUV [40] normalized using spatial residuals as the controls. This process reduces noise in the data by removing technical spatial artifacts that are common to multiple MEMA arrays. Processing and visualization [39,41] code is available on Github at https://github.com/markdane/MEMA_HMEC. Unsupervised hierarchical clustering using the R hclust function using Euclidean distance and the ‘complete’ method (R, stats package, [37]) was used to see how different microenvironments affect cell phenotype figure 1E. Molecular functions used in figure 1E were downloaded from the Human Protein Atlas https://www.proteinatlas.org/.

MEMA results validation assays-

Microenvironment effect to luminal cell population was studied by coating standard multi-well plates (6-well plate for flow cytometry analysis and 24-well plate for Immunohistochemistry analysis) with COL1 (calf skin, Sigma-Aldrich) 100mg/ml with or without VTN (1ug/ml, Sino biological), OMD (1ug/ml, Sino Biological), low molecular weight hyaluronan (1ug/ml, R&D), SPP1 (1ug/ml, R&D). Cell were cultured on coated plates and ligands: BMP2 (100ng/ml), CXCL1 (4ng/ml), FGF2 (100ng/ml), HGF (40ng/ml), NRG1-b (10ng/ml), PDGF-B (50ng/ml), TGFB1 (0.2ng/ml) from Sino Biological and TNFSF11 (7.5ng/ml, Millipore), WNT3A (10ng/ml, R&D), were added to the culture medium. Fresh medium and ligands were changed every other day. Cultures were continued until sub confluent (5–7 days), then processed as described immunohistochemistry and flow cytometry protocols below.

Cell proliferation-

Microenvironment effect to cell proliferation was studied by sorting passage 4 240L HMEC by using CD117/cKit-PE (Biolegend, clone 104D2, 1:50), anti-human CD227-FITC (BD Biosciences, clone HMPV, 1:250) and anti-human CD271-APC (Biolegend, clone ME20.4, 1:100) antibodies and Instrument: Aria II (BD Bioscience). Sorted myoepithelial cells (CD271+) and luminal cells (CD227+) were plated on 96-well plate and treated with increasing concentration of HGF (0–800ng/ml), after three days cell viability/cell number was studied by using cell titer Glo 2.0 assay (Promega). Sorted cKit+ cell and luminal cells were also plated on 24 well plate, equal number of cells on each well, and cells were treated with or without HGF (40ng/ml), fresh culture medium and HGF was changed every day, after 7 days cultures were fixed and stained as described in immunohistochemistry protocol. Cultures were imaged and cell number on each well was calculated by using automated image analysis and cell segmentation.

Immunohistochemistry –

For in vitro immunofluorescence staining, cells were fixed in 4% paraformaldehyde at rt for 10 minutes, blocked with PBS, 5% normal goat serum, 0.1% Triton X-100, and incubated with anti-KRT14 (1:1000, Covance, polyclonal) and anti-KRT19 (1:1000 628502, Biolegend) overnight at 4°C, then visualized with fluorescent secondary antibodies (Invitrogen) incubated with sections for 2 hours at room temperature.

Flow cytometry-

Trypsinated cells were fixed with 1.6% paraformaldehyde at rt for 10 minutes, by following standard flow cytometer sample preparation protocol. Cells were stained by using: anti- human CD117/cKIT-PE (Biolegend, clone 104D2, 1:50), anti-human CD227-FITC (BD Biosciences, clone HMPV, 1:250) and anti-human CD271-APC (Biolegend, clone ME20.4, 1:100) antibodies and analysis was done by using BD Accuri C6 (BD Bioscience) instrument.

RNA Sequencing and analysis-

Passage four HMECs (listed in Table 1) were sorted into myoepithelial and luminal populations using the markers described above. Fibroblasts were differentially trypsinized from the epithelial cells and obtained for RNA extraction. Adipose tissue samples were cut manually and homogenized with trizol (ThermoFisher Scientific, Cat #: 15596018). RNA extraction from the sorted cells was done using Quick-RNA MicropPrep kit with Zymo-Spin IC colunms (Cat#:R1050) and from the trizol-homogenized adipose samples using Direct-zol RNA miniprep kit (Cat#:R2051). Samples were then submitted to City of Hope integrative genomics core for RNA sequencing.

RNA sequencing libraries were prepared using Kapa RNA mRNA HyperPrep kit (KapaBiosystems, Cat#: KR1352) according to the manufacturer’s protocol (Adipose Tissue samples were processed using the Kapa’s mRNA HyperPrep with RiboErase HMR kit, cat#: KK8561). 100 ng of total RNA from each sample was used for polyA RNA enrichment. The enriched mRNA underwent fragmentation and first strand cDNA synthesis. The combined 2nd cDNA synthesis with dUTPand A-tailing reaction generated the resulting ds cDNA with dAMP to the 3’ ends. The barcoded adaptors were ligated to the ds cDNA fragments. A 10-cycle of PCR was performed to produce the final sequencing library. The libraries were validated with the Agilent Bioanalyzer DNA High Sensitivity Kit and quantified with Qubit. Sequencing was performed on Illumina HiSeq 2500 with the single read mode of 51cycle. Real-time analysis (RTA) 2.2.38 software was used to process the image analysis.

RNA-Seq reads were trimmed to remove sequencing adapters using Trimmomatic 39. The processed reads were mapped back to the human genome (hg19-ensembl-release75_v2) using STAR software version 2.5.3a 40. The HTSeq 41 and RSeQC software 42 were applied to generate the count matrices and strand information, respectively, with default parameters. For visualization of mRNA expression levels, mRNA counts were normalized using TMM (trimmed mean of M values) method 43. RNA-seq data have been deposited in the Gene Expression Omnibus database under accession no. GSE182338.

Illuminal BeadChip transcriptome analysis -

Four lineages (luminal, myoepithelial, cKit+/AXL− and cKit+/AXL+) in 240L HMECs were stained with anti-human CD227-FITC (BD Biosciences, clone HMPV), anti-CD10 (Biolegend, clone HI10a), anti-CD117 (Biolegend, clone 104D2) and anti-AXL (hybridoma cells, clone 10C9) by following standard flowcytometry protocol. Cells were separated by using FACSVantage (BD Biosciences). Total RNAs from four lineages are isolated using Quick-RNA MicroPrep, after enrichment by FACS. Sample preparations for Illumina HumanHT-12 v4 Expression BeadChip arrays were performed in UCLA Neuroscience Genomics Core (UNGC). Raw gene expression data from Illumina HumanHT-12 v4 BeadChips were pre-processed with Bioconductor limma package neqc function which performs normexpr background correction using negative control probes, log2 transformation and quantile normalization between arrays. Normalized data set were pre-filtered to remove gene probes with values less than negative control probes. This was done by calculating detection p-values using limma package detection P Values function, and only gene probes with detection p-values < 0.05 for at least 12 samples were retained. Potential batch effects between chips were checked using Principal Component Analysis (PCA).

MicroEnvironment MicroArray (MEMA) platform is a tool to identify combinations of tissue microenvironment components that direct cell functions.

Hepatocyte growth factor (HGF) maintains mammary gland luminal epithelia by working through a luminal-biased progenitor intermediate.

Mammary epithelial luminal cells from women who are genetically high risk for breast cancer had significantly more HGF receptor MET.

ABBREVIATIONS

ECM

extracellular matrix

HGF

Hepatocyte growth factor

MEMA

microenvironment microarray

HMEC

human mammary epithelial cells

KRT

keratin

MET

receptor tyrosine-protein kinase Met

COL1

collagen type 1

OMD

osteomodulin

LMWHA

low molecular weight hyaluronan

SPP1

osteopontin

VTN

vitronectin

LEP

luminal epithelial

MEP

myoepithelial

HAS

the hyaluronan synthase