Stress Signaling from Human Mammary Epithelial Cells Contributes to Phenotypes of Mammographic Density

Histologic examination demonstrates that tumor stroma is morphologically distinct from disease-free stroma. It is characterized by increases in extracellular matrix (ECM), fibroblasts, immune and endothelial cells, cytokines and growth factors levels (1), and by fewer and smaller adipocytes (2, 3). Collectively, these changes are termed desmoplasia.

Seminal studies demonstrated that the stroma contributes to tumor initiation, progression, and outcome. Tumor stromal cells (fibroblasts, adipocytes endothelial, and immune cells) promote epithelial cell proliferation, mutagenesis, angiogenesis, and migration and impair apoptosis and immunosurveillance (1, 4). Irradiation of mouse mammary gland stroma drives tumorigenesis of nonirradiated epithelial cells (5) and involuting mammary stroma facilitates progression of premalignant cells in xenografts (6, 7). Gene expression profiles of breast tumor–associated stroma are strongly associated with clinical outcome (8) and are remarkably similar to those of stroma surrounding ductal carcinomas in situ (DCIS; ref. 9), suggesting that tumor stroma develops early in tumorigenesis.

Selected desmoplastic features are also observed in nonmalignant breast tissues of women with high mammographic density (HD; refs. 1–3, 10, and 11). Mammographic density (MD) is determined by the relative amounts of radiolucent material (fat) and radiodense material (epithelial cells, fibroblasts and ECM) within the breast upon imaging. Radiodense areas, referred to as MD, exhibit histologic characteristics of malignant stroma, including low adipocyte content, high ECM, and stromal cell content (1–3, 10, 11). Women with MD > 75% have a 4- to 6-fold increase in invasive breast cancer compared with women with negligible MD (12, 13). It is estimated that almost one third of breast cancers may be attributable to high MD (13).

Previously, we showed that CD36 is necessary and sufficient to coordinately control adipocyte and matrix accumulation, two phenotypes that histologically define MD (3). CD36, a widely expressed transmembrane receptor, modulates cell type- and ligand-specific phenotypes, including adipocyte differentiation, angiogenesis, apoptosis, TGFβ1 activation, cell–ECM interactions, ECM deposition, and immune signaling (3, 14). CD36 expression is negligible in tumor stroma, in contrast to surrounding histologically disease-free tissue, and is inversely correlated with tumor size and grade (3). Strikingly, disease-free HD tissues (MD > 70%–75%) have reduced CD36 levels in multiple stromal components (adipocytes, endothelial cells, macrophages, and fibroblasts) compared with low mammographic density (LD) tissues (25% < MD < 50%), suggesting that CD36 repression is a multicellular coordinated stromal program active in tissues at high risk for tumorigenesis (3).

In addition to desmoplasia and HD, increased ECM is seen in other pathologic conditions associated with a DNA damage response (DDR) and/or telomere malfunction: fibrosis after γ-irradiation (15), dyskeratosis congenita (16), and pulmonary (17) and liver fibrosis (18). Previously, we demonstrated that DNA damaging agents and/or telomere malfunction in mortal, nontumorigenic variant human mammary epithelial cells (vHMEC) induce a DDR and activin A–dependent COX-2 expression (19). Fibroblasts from reduction mammoplasties (RMF) cocultured with DNA-damaged vHMEC induce many genes consistent with a desmoplastic phenotype (e.g., ECM and inflammatory cytokines) and activin A secretion by damaged vHMEC is necessary and sufficient for this induction (20). In vivo, DCIS lesions whose epithelial cells exhibit shorter telomeres, increased activin A and COX-2, are surrounded by activated fibroblasts and increased immune infiltrate (20), suggesting that microenvironmental alterations, mimicking aspects of desmoplasia, occur even in the absence of an invasive tumor.

This manuscript presents in vitro and in vivo data supporting our hypothesis that HD is generated by factors(s) secreted by DNA-damaged epithelial cells (including damage caused by shortened telomeres) that repress CD36 causing induction of desmoplastic-like phenotypes in adjacent fibroblasts.

Human tissues were accrued after informed consent and studied under institutionally approved protocols 10-02471 and 10-03756.

Paraffin-embedded serial tissue sections, from 13 LD (25% < MD < 50%) and 14 HD (MD > 70%) disease-free women (Supplementary Table S1), were assessed for γH2AX levels by immunohistochemistry (3) and for telomere content by telomere-specific FISH (20). For immunohistochemistry, antigen retrieval was performed in citrate buffer (pH 6.0) for 30 minutes at 93°C before incubation with a γH2AX antibody (1:800; Millipore, #05-636) for 60 minutes at room temperature. For FISH, telomere-specific (Cy3-labeled) and centromere-specific (FITC-labeled) peptide nucleic acid probes were hybridized to tissue sections and the telomere to centromere intensity ratio calculated.

Human mammary epithelial cells (HMEC) and vHMEC, which have silenced p16 (21), isolated from 12 biopsies of known MD (LD-HMEC/vHMEC or HD-HMEC/vHMEC; Supplementary Table S2) and four RMF (Supplementary Table S3), and human mammary fibroblasts (HMF), isolated from 14 biopsies of known MD (LD-HMF or HD-HMF; Supplementary Table S2) and 10 RMF (Supplementary Table S3), were propagated in MEGM and RPMI-1640 + 10% FBS, respectively (3, 4, 22, 23).

LD-vHMEC and HD-vHMEC exposed to 50 μmol/L etoposide or vehicle control (DMSO) for 24 hours were assessed for γH2AX intensity and foci number, cell viability, apoptosis, and proliferation 0, 1, 3, 6, 12, and 24 hours after drug removal. Immunofluorescence detection of γH2AX was performed using the γH2AX antibody above (1:500; ref. 20). Cell viability (fraction of live/dead cells) and apoptosis (caspase 3/7 activity) were measured using an ApoTox-Glo Triplex Assay (Promega) and proliferation (BrdUrd incorporation) using the Cell Proliferation Assay Kit (Cell Signaling Technology). LD-vHMEC and HD-vHMEC exposed to 20 μmol/L etoposide or vehicle for 3 hours were assessed for long-term survival in a colony-formation assay 9 days after drug removal, and the surviving fractions (plating efficiency of cells exposed to etoposide/plating efficiency of cells exposed to vehicle) calculated (24).

Telomere content was assessed with genomic DNA isolated from HMEC, vHMEC, and HMF using quantitative polymerase chain reaction (qPCR) as described (25). Telomeric DNA was expressed as the ratio of telomere cycle threshold (Ct) to the Human β globin gene Ct. Each sample was run in triplicate wells on each plate and averaged. Each target was analyzed in triplicate plates.

qPCR was performed on a CFX-96 thermocycler using SsoFast Master Mix (BioRad) and TaqMan primer probe sets for each gene (Applied Biosystems) and the data analyzed using the standard curve method. Expression of β-d-glucuronidase (GUSB) was used to normalize for variances in input cDNA. For all experiments, each sample was run in triplicate wells on each plate and averaged, and each gene was analyzed in triplicate plates.

LD-vHMEC and HD-vHMEC were plated in MEGM and conditioned their media for 48 hours. RMF were plated in RPMI-1640 + 10% FBS and grown for 24 hours in RPMI-1640 + 1% FBS before the media were replaced by a 1:1 mix of RPMI-1640 + 2% FBS and conditioned media (or control unconditioned media). RNA was isolated from RMF 48 hours after treatment with conditioned media.

vHMEC were plated in MEGM and conditioned their media for 26 hours. Activin A and TGFβ1 protein levels were measured in the conditioned media using Duo-Set ELISA Kits #DY338 and DY420, respectively (R&D Systems).

RMF, plated in RPMI-1640 + 10% FBS, were grown for 24 hours in RPMI-1640 without serum before the addition of: activin A (Sigma-Aldrich); prostaglandin E2 (Cayman Chemicals); COX-2 inhibitor, NS398 (Cayman Chemicals); protein kinase A (PKA) inhibitor, H89 (Sigma-Aldrich); phosphatidylinositol 3-kinase (PI3K) inhibitor, LY294002 (Sigma-Aldrich); TGFβ receptor 1 (TGFβR1) inhibitors, LY364947 and SB431542 (Sigma-Aldrich); p38 mitogen-activated protein kinase (p38 MAPK) inhibitor SB203580 (Sigma-Aldrich); and MAPK kinase (MAPKK) inhibitor, UO126 (Sigma-Aldrich).

RMF were plated in 0.4-μm pore Transwell dishes (Costar) and cocultured with vHMEC expressing either control luciferase short hairpin (sh-luciferase), activin A short hairpin (sh-activin A), vector control, TRF2, or hTERT (19, 20). RNA was isolated from RMF 48 hours after initiation of coculture.

Immunocytochemistry was performed in RMF using primary antibodies against CD36 (1:20, #9154; Santa Cruz Biotechnology), fibronectin (1:100, #610077; BD Transduction), and α-smooth muscle actin (αSMA; 1:50, #M0851; Dako; ref. 3).

RMF, plated in RPMI-1640 + 10% FBS, were grown for 24 hours in RPMI-1640 without serum before exposure to 80 ng/mL activin A for 48 hours. The media were then replaced with RPMI-1640 + 10% FBS, ±80 ng/mL activin A and ±10 μmol/L 15-deoxy-D12,14-prostaglandin J2 (PJ2; Cayman Chemical) to induce adipocyte differentiation. Lipid accumulation was assayed after 1 week by Oil Red O staining (3).

Image acquisition, statistical analysis, and image analysis are described in Supplementary Methods.

We hypothesized that HD epithelial cells have an elevated DDR and secrete factor(s) that repress CD36 expression and reprogram adjacent fibroblasts. To test this hypothesis, levels of γH2AX, a DNA damage marker, were assessed by immunohistochemistry (Fig. 1A, left) in biopsies obtained from disease-free women with measured MD (Supplementary Table S1). HD tissues had higher γH2AX levels (1.9-fold, P < 0.0001) than LD tissues (Fig. 1A, right and Supplementary Fig. S1A). To determine whether the increased DDR observed in HD tissues was associated with telomere malfunction, telomere DNA content was assessed by FISH (Fig. 1B, left). Telomere DNA content was reduced (1.2-fold, P < 0.0001) in HD epithelial cells compared with LD epithelial cells (Fig. 1B, right and Supplementary Fig. S1B). CD36 levels, measured by immunohistochemistry in this cohort, were lower (4.5-fold, P < 0.0001) in HD tissues than LD tissues (3). These in vivo data are consistent with our hypothesis that factors secreted by DNA-damaged HD epithelial cells repress CD36 in adjacent fibroblasts.

To further assess the differences between LD and HD tissues in vitro, we isolated and propagated 3 cell types (3, 4, 22, 23) from biopsies of disease-free women with measured MD (Supplementary Table S2): (i) human mammary epithelial cells (HMEC) with an intact p16/Rb pathway; (ii) vHMEC with a compromised p16/Rb pathway; and (iii) human mammary fibroblasts (HMF). We primarily used vHMEC for these experiments because: (i) HMEC have limited proliferative capability (∼10 population doublings) compared with vHMEC (∼30–65 population doublings); (ii) HMEC, which have intact cell-cycle checkpoints (p16/Rb), undergo growth arrest upon telomere malfunction, whereas vHMEC, which lack p16, still proliferate (19, 22); and (iii) vHMEC preexist in vivo in disease-free breast tissue (21).

To determine whether HD-vHMEC had a heightened basal DDR, as observed in vivo, basal γH2AX levels were measured. γH2AX levels were slightly higher (1.1-fold, P = 0.05) in HD-vHMEC than LD-vHMEC (Fig. 2A, left and Supplementary Fig. S2A). Cell propagation in culture leads to selective expansion of cells capable of surviving and proliferating. Therefore, cells with the most pronounced DDR and the highest levels of γH2AX (“jackpots”) are likely lost from the population, potentially explaining why the difference in γH2AX levels between LD and HD cells is greater in vivo than in vitro. One might predict that “jackpots” would exist transiently in culture and in greater numbers in HD-vHMEC. To assess this possibility, cells with the 0.5% highest γH2AX levels (“jackpots”) were counted. “Jackpots” were more frequent (4.8-fold, P < 0.0001) in HD-vHMEC, as predicted (Fig. 2A, right).

Because the induction of a DDR typically leads to growth arrest or apoptosis, we asked whether HD-vHMEC showed reduced proliferation and/or increased apoptosis when compared with LD-vHMEC. HD-vHMEC showed decreased proliferation (1.5-fold, P = 0.01) when assessed by BrdUrd incorporation (Fig. 2B) but no significant difference in apoptosis when assayed for caspase-3/-7 activity (Supplementary Fig. S3B).

To determine whether HD epithelial cells had reduced telomere content, as observed in vivo, telomere DNA was measured by qPCR in LD and HD vHMEC, HMEC, and HMF (all at comparable population doublings: see Supplementary Fig. S2B and Supplementary Methods). Telomere DNA was reduced 1.4-fold in both HD-vHMEC and HD-HMEC compared with LD-vHMEC and LD-HMEC (P = 0.007 and 0.03, respectively; Fig. 2C, left and middle, respectively) and 1.3-fold in HD-HMF compared with LD-HMF (P = 0.0009; Fig. 2C, right), demonstrating that this phenotype is not restricted to the epithelial compartment. These in vitro data demonstrate that HD-vHMEC have increased basal γH2AX levels, reduced proliferation, and reduced telomere content compared with LD-vHMEC, recapitulating our observations in vivo and indicating that LD and HD epithelial cells are intrinsically distinct.

To assess whether HD-vHMEC have a differential response to exogenous DNA damage, LD-vHMEC and HD-vHMEC were exposed to etoposide for 24 hours to induce double-strand DNA breaks, and γH2AX intensity quantitated (Fig. 3A). LD-vHMEC and HD-vHMEC exhibited increased γH2AX intensity after etoposide exposure (2.0-fold, P < 0.0001; 2.6-fold, P < 0.0001; respectively). However, γH2AX intensity was higher (1.4-fold, P < 0.0001) in HD-vHMEC.

To assess the ability of LD-vHMEC and HD-vHMEC to recover from etoposide-induced DNA damage, γH2AX levels were measured 0, 1, 3, 6, 12, and 24 hours after drug removal (Fig. 3B). HD-vHMEC consistently exhibited significantly higher γH2AX intensity than LD-vHMEC. Importantly, although γH2AX intensity in LD-vHMEC returned to basal levels by 12 hours, γH2AX intensity in HD-vHMEC remained elevated for 24 hours, suggesting that DNA repair is less efficient in HD-vHMEC.

Although γH2AX intensity is frequently used as a readout of DDR, and correlates well with the amount of DNA damage (26) and γH2AX foci (27), we nonetheless counted the number of γH2AX foci in a subset of LD-vHMEC and HD-vHMEC (Supplementary Fig. S3A). Consistent with γH2AX intensity measurements, both LD-vHMEC and HD-vHMEC exhibited increased γH2AX foci number (6.3-fold, P < 0.0001; 9.8-fold, P < 0.0001; respectively) after etoposide exposure. Furthermore, although not statistically significant, γH2AX foci number was higher in HD-vHMEC 0 and 24 hours after etoposide removal (1.1- and 1.4-fold, respectively).

Cell viability (Fig. 3C) and apoptosis (Supplementary Fig. S3B) were assessed 0, 1, 3, 6, 12, and 24 hours after etoposide removal. In spite of their increased and sustained DDR, HD-vHMEC exhibited increased cell viability compared with LD-vHMEC for the entire time course. Both LD-vHMEC and HD-vHMEC showed increased levels of apoptosis upon etoposide exposure. However, HD-vHMEC underwent less apoptosis than LD-vHMEC (although not statistically significant) for the entire time course, consistent with their increased viability. We also assessed the long-term survival of these cells after exposure to etoposide using a colony formation assay (Fig. 3D), and found that HD-vHMEC formed more colonies than LD-vHMEC 9 days after etoposide removal (2.2-fold, P = 0.006).

These in vitro data demonstrate that HD-vHMEC have elevated and persistent γH2AX levels, increased viability/survival and decreased apoptosis following exogenous DNA damage, compared with LD-vHMEC. Therefore, LD-vHMEC and HD-vHMEC have intrinsic differences in both their basal and induced DDR.

Having demonstrated that LD-vHMEC and HD-vHMEC are intrinsically different, we tested our hypothesis that secreted factor(s) from HD epithelial cells repress CD36 expression in adjacent stromal cells. CD36 mRNA levels were assessed in RMF (Supplementary Table S3) after exposure to control unconditioned media or conditioned media from LD-vHMEC or HD-vHMEC (Fig. 4A, left). Conditioned media from both LD-vHMEC and HD-vHMEC repressed CD36 in RMF compared with unconditioned media (1.6-fold, P < 0.0001; 2.2-fold, P < 0.0001, respectively). However, consistent with our hypothesis, conditioned media from HD-vHMEC repressed CD36 in RMF more than conditioned media from LD-vHMEC (1.4-fold, P = 0.0005).

Induction of a DDR in vHMEC is associated with increased activin A (19). To assess whether the heightened DDR observed in HD-vHMEC was associated with higher activin A, activin A mRNA, and protein levels were measured in LD-vHMEC and HD-vHMEC. HD-vHMEC had higher levels of activin A mRNA (1.4-fold, P = 0.03) and protein (1.8-fold, P = 0.04) than LD-vHMEC (Fig. 4A, middle and right, respectively).

Activin A is a TGFβ family member and TGFβ is known to induce ECM accumulation/fibrosis (28). Therefore, we evaluated TGFβ1 protein levels in LD-vHMEC and HD-vHMEC and in vHMEC with increased telomere malfunction, that is, overexpressing TRF2 (TRF2-vHMEC), and vHMEC with reduced telomere malfunction, that is, overexpressing hTERT (hTERT-vHMEC; ref. 19). TGFβ1 protein levels were not significantly different between LD-vHMEC and HD-vHMEC nor between TRF2-vHMEC and hTERT-vHMEC (Supplementary Fig. S4A and S4B, respectively), demonstrating that TGFβ1 does not contribute to the DDR in vHMEC in our experimental conditions.

We previously showed that activin A, induced by DDR, is necessary and sufficient for COX-2 induction in vHMEC (19). As predicted, COX-2 mRNA levels were higher (1.5-fold, P = 0.09) in HD-vHMEC than LD-vHMEC (Supplementary Fig. S4C). These data further support that the stress response previously identified by us in DNA-damaged vHMEC (19) is heightened in HD-vHMEC compared with LD-vHMEC.

To determine whether activin A was sufficient to repress CD36 in RMF, RMF were exposed to 80 and 320 ng/mL activin A. Exposure to 80 ng/mL activin A repressed CD36 mRNA (2.6-fold, P = 0.01) and protein levels (Fig. 4B, left and right, respectively). A similar repression was observed at 320 ng/mL. Because CD36 expression is primarily controlled by the transcription factor PPARγ (29), we asked whether activin A–dependent CD36 repression could be mediated by PPARγ. RMF exposed to activin A repressed PPARγ mRNA at both doses (1.3-fold, P = 0.007; 1.4-fold, P = 0.015, respectively; Fig. 4B, left). To assess whether activin A secretion by vHMEC was necessary for CD36 repression in RMF, RMF were cocultured with vHMEC expressing short hairpin RNA to either a control (sh-Luciferase-vHMEC) or activin A (sh-activin A-vHMEC). RMF cocultured with sh-activin A-vHMEC had higher levels of CD36 mRNA (2.0-fold, P = 0.007) than RMF cocultured with sh-luciferase-vHMEC (Fig. 4C). Thus, activin A is sufficient to repress CD36 in RMF and activin A secretion by vHMEC is necessary for CD36 repression in RMF.

Activin A induces COX-2 and secretion of its product, prostaglandin E2 (PGE2), in vHMEC (19). To ascertain whether COX-2 expression and PGE2 secretion by vHMEC was necessary and/or sufficient for CD36 repression in RMF, RMF were exposed to activin A, PGE2, or a COX-2 inhibitor (NS398), or activin A and NS398 together and CD36 mRNA levels measured (Supplementary Fig. S4D). CD36 expression was repressed in RMF exposed to activin A, but exposure to PGE2 did not repress CD36. In addition, exposure of RMF to NS398 did not affect activin A–mediated CD36 repression, demonstrating that COX-2 induction, and the subsequent secretion of PGE2, by vHMEC is neither necessary nor sufficient for CD36 repression in RMF.

To evaluate whether telomere malfunction in vHMEC was necessary and/or sufficient for CD36 repression in RMF, RMF were cocultured with vector-vHMEC (control), TRF2-vHMEC with increased telomere malfunction, or hTERT-vHMEC with reduced telomere malfunction (19). CD36 mRNA levels were repressed in RMF cocultured with TRF2-vHMEC (1.5-fold, P = 0.01) but elevated in RMF cocultured with hTERT-vHMEC (1.8-fold, P = 0.03) compared with RMF cocultured with vector-vHMEC (Fig. 4D). These data demonstrate that telomere malfunction in vHMEC is necessary and sufficient for CD36 repression in RMF and expand our previous report about cell-extrinsic consequences of telomere malfunction (20).

To gain further insights into the biologic relevance of CD36 regulation by activin A, we analyzed the sensitivity and durability of this regulation in RMF in vitro. RMF exposed to physiologic levels (∼1.2 ng/mL) of activin A (30) for 48 hours showed CD36 repression in a dose-dependent manner, repression being observed (1.4-fold, P = 0.05) with as little as 1.25 ng/mL of activin A (Fig. 5A). Exposure of RMF to two doses of activin A for 2, 4, or 8 days repressed CD36 to a similar extent (2.6- to 3.4-fold, P < 0.05) under all conditions, demonstrating that a 2-day exposure to activin A is sufficient for maximum CD36 repression in RMF (Fig. 5B). Importantly, RMF exposed to activin A for 48 hours then propagated in the absence of activin A for five passages exhibited sustained (and even increasing) CD36 mRNA repression (1.3-fold, P = 0.0006 at P5) for several weeks after activin A removal (Fig. 5C and Supplementary Fig. S5). In summary, CD36 expression in RMF is exquisitely sensitive to physiologic levels of activin A and even a brief exposure to activin A can result in prolonged CD36 repression.

To elucidate the mechanism(s) by which activin A represses CD36 in RMF, RMF were exposed to activin A alone or in the presence of various pathway inhibitors or vehicle control (control RMF; Fig. 5D). Activin A, like other TGFβ family members, signals through the TGFβ family type I and type II receptors. We found that RMF exposed to activin A plus TGFβ family type I receptor (TGFβR1) inhibitors (LY364947 and SB431542) exhibited higher levels of CD36 (5.7-fold, P = 0.0001; 5.8-fold, P < 0.0001, respectively) than control RMF, demonstrating that activin A/TGFβR1 signaling is required for activin A–dependent CD36 repression. Previous studies demonstrated that MAPK represses CD36 and that activin A utilizes the MAPK pathway (19, 31). RMF exposed to activin A plus MAPK kinase (MAPKK) inhibitor (UO126) had higher CD36 levels (2.9-fold, P = 0.004) than controls. Thus, the MAPK pathway is necessary for activin A–dependent CD36 repression. In contrast, RMF exposed to activin A plus PI3K inhibitor (LY294002) exhibited lower CD36 levels (1.8-fold, P < 0.0001) than controls, suggesting that the PI3K pathway is required for the induction of CD36 expression. Finally, RMF exposed to activin A plus PKA inhibitor (H89) or activin A plus p38 MAPK inhibitor (SB203580) had similar levels of CD36 to controls, demonstrating that neither of these pathways are involved in regulating CD36 expression in this context.

If a DDR and increased activin A secretion in HD epithelial cells was responsible for the CD36-modulated desmoplastic-like phenotypes (low adipocyte and high ECM content) observed in HD tissues in vivo (3), activin A should modulate these phenotypes in RMF in vitro.

RMF were placed under proliferative or adipocyte differentiation conditions, in the absence or presence of activin A, and Oil Red O staining, an indicator of intracellular fat accumulation, was measured (Fig. 6A; ref. 32). RMF grown in the absence or presence of activin A both accumulated fat under differentiation conditions compared with proliferative conditions (15.7- and 25.4-fold, respectively, P < 0.0001). However, RMF exposed to activin A accumulated significantly less fat under differentiation conditions than RMF grown without activin A (1.2-fold, P < 0.0001).

RMF were exposed to two doses of activin A and protein and mRNA levels for selected ECM genes were assessed. Exposure of RMF to activin A induced fibronectin and α-SMA protein accumulation (Fig. 6B, left) and tenascin C, fibronectin, and collagen 1A1 mRNAs (4.1-fold, P = 0.0004; 1.8-fold, P = 0.025; 1.7-fold, P = 0.012; respectively, for 80 ng/mL; Fig. 6B, right) similarly at both doses. Thus, activin A can decrease fat accumulation and increase matrix accumulation in RMF in vitro, two prominent phenotypes of HD tissue modulated by CD36 expression.

We previously showed, in vitro and in vivo, that CD36 is dramatically repressed in multiple stromal cell types within disease-free tissues from women with HD compared with women with LD (3). To define the mechanisms that account for CD36 repression, we tested the hypothesis that HD results from stress signaling in epithelial cells that induces the secretion of factors that repress CD36 and reprogram adjacent fibroblasts. Using cohorts previously used by us to study CD36-dependent phenotypes (3), we show that HD epithelial cells have more basal DNA damage (γH2AX intensity) than LD epithelial cells. In addition, HD-vHMEC have more γH2AX foci after exogenous DNA damage, and take longer to resolve these foci, than LD-vHMEC. Paradoxically, HD-vHMEC also have increased viability/survival and decreased apoptosis after exogenous DNA damage, a property that could facilitate the escape of a mutated clone, leading to cancer. HD epithelial cells also have slightly shorter telomeres, both in vitro and in vivo, than LD epithelial cells. This relatively small difference is not surprising because telomere length is a tightly regulated phenotype.

We previously reported that DNA damage or telomere malfunction in epithelial cells results in increased secretion of activin A, which can act in a paracrine or autocrine fashion to induce its own expression, and the expression of many protumorigenic genes, in adjacent cells (19, 20). Similarly, we find that HD-vHMEC, with increased DNA damage, have higher activin A levels than LD-vHMEC. We describe a novel phenotype associated with this pathway by showing that CD36 expression in RMF is exquisitely sensitive to physiologic levels of activin A (30), and that even a transient exposure to activin A can persistently repress CD36. Consistent with our hypothesis, conditioned media from HD-vHMEC are more potent in repressing CD36 in RMF than conditioned media from LD-vHMEC. Activin A and telomere malfunction in vHMEC are both necessary and sufficient for this repression. Importantly, activin A also modulates CD36-dependent desmoplastic-like phenotypes in vitro.

Our data suggest that CD36 repression may result from epigenetic modification, in addition to repression of its key regulator PPARγ. First, HD-HMFs maintain CD36 repression in culture over several passages in the absence of exogenous activin A or interaction with HD epithelial cells (3). In addition, exposure to trichostatin A, a histone deacetylase inhibitor, increases CD36 expression more extensively in HD-HMF than LD-HMF (unpublished data). Finally, transient exposure to activin A persistently represses CD36 expression in RMF. Hence, even transient DNA damage in epithelial cells, and the subsequent secretion of activin A, could result in prolonged CD36 repression in adjacent fibroblasts.

We demonstrate that the activin A–dependent repression of CD36 in RMF requires both the activin A receptor (TGFβRI) and MAPK pathways, consistent with reports demonstrating that MAPK can regulate PARγ function and repress CD36 (31) and with activin A's ability to repress PPARγ expression in RMF. Although TGFβ1 has been implicated in CD36 repression in macrophages (31), TGFβ1 levels are not elevated in HD-vHMEC, nor in vHMEC with telomere malfunction, suggesting that TGFβ1 does not participate in CD36 repression in this context. Finally, we show that, unlike many other protumorigenic phenotypes previously described by us (19, 20), CD36 repression in RMF is not COX-2 dependent.

We envision a scenario in which (i) elevated basal DNA damage in HD epithelial cells results in increased activin A secretion; (ii) activin A binds to its receptor on adjacent fibroblasts and activates the MAPK pathway; (iii) MAPK pathway activation results in PPARγ phosphorylation and inhibition; (iv) PPARγ inhibition leads to decreased CD36 transcription and subsequently, the induction of the desmoplastic-like phenotypes observed in HD tissues (Fig. 7). In response, and as an extension, these desmoplastic-like fibroblasts can in turn promote motility of neighboring epithelial cells (20) and induce DNA damage of neighboring epithelial cells through the release of reactive oxygen species (ROS; ref. 33). Damaged epithelial cells would further exacerbate protumorigenic fibroblast phenotypes via the induction of activin A–dependent signaling pathways, thus propagating the DNA damage signal and the associated HD phenotypes throughout the tissue. Therefore, a breast with more DNA-damaged epithelial cells would exhibit more mammographically dense areas, leading to overall high MD. Our study highlights the reciprocal interactions between epithelial and stromal cells and their potential to induce carcinogenic processes and shows, for the first time, that there is a differential level of this cellular cross-talk in LD and HD tissues. The model described above suggests that these reciprocal interactions are initiated by a DNA damage event in the epithelial compartment. However, we cannot rule out that the initiating event occurs in the stromal compartment or, alternatively, simultaneously in both epithelial and stromal compartments. Regardless of the source of the initiating event, our data suggest that the alterations in LD and HD epithelial cells and fibroblasts are “intrinsic” and subsequently maintained in cells purified from tissues and propagated in vitro.

Our data support one potential mechanism that contributes to MD (Fig. 7). However, MD is a complex phenotype likely regulated by multiple pathways. Studies have implicated IGF1 in the acquisition of HD (10). Interestingly, IGF1 mRNA levels are higher (2-fold, P = 0.09) in HD-HMF than LD-HMF and activin A induces IGF1 expression (5.7- to 6.5-fold, P = 0.0002) in RMF (unpublished data). In addition, although we ruled out the involvement of TGFβ1 in the DDR pathway, TGFβ1 could be involved in another context because it promotes fibrosis (28), a phenotype exhibited by HD.

MD is a heritable trait that can be modified by environmental factors (34). The same can be said of telomere length; for example, chronic stress is correlated with shortened telomeres (35, 36). We, and others, have demonstrated that loss of telomere DNA can have cell-extrinsic consequences that may facilitate the development of a protumorigenic stroma (20, 37). This may partially explain why loss of telomere DNA is associated with poor clinical outcome for women with breast cancer and increased risk of cancer (38, 39). This study is the first to suggest that telomere malfunction also contributes to HD.

In the breast, sex hormones drive expansion and involution of epithelial cells during menstrual cycling and lactation. Provocatively, expansion of epithelial cells during the luteal phase of the menstrual cycle is accompanied by modest but significant increases in MD (40) and coincides with the highest serum activin A levels (41). Successive expansion and involution may leave epithelial cells particularly vulnerable to DNA damage or telomere malfunction and select for genetically unstable cells able to resist apoptosis, like HD-vHMEC.

One might anticipate that as women age and their telomeres shorten, increased stress signaling would drive an increase in MD. However, MD remains the same or even decreases with age (42). This apparent contradiction could be explained by age-related lobular involution. This age-dependent loss of mammary epithelial cells would translate into decreased signaling to adjacent fibroblasts and a concomitant decrease in production of HD phenotypes. In fact, there is an inverse relationship between age-related involution and both MD and breast cancer risk (43, 44).

The demonstration that decreasing MD reduces breast cancer risk (45) provides tremendous opportunities for cancer prevention. Several drugs in clinical trials or already approved by the Federal Drug Administration modulate potential therapeutic targets identified in this study. Activin A is inhibited by competitive inhibitors of its receptor (e.g., bimagrumab), soluble receptor traps (e.g., dalantercept and sotatercept), and receptor kinase inhibitors (e.g., LY-2157299; ref. 46), whereas CD36 expression is increased by aspirin, dexamethasone, statins, and adalimumab (47–50). The effects of these drugs on MD remain to be investigated.

No potential conflicts of interest were disclosed.

Conception and design: R.A. DeFilippis, C. Fordyce, T.D. Tlsty

Development of methodology: R.A. DeFilippis, C. Fordyce, J. Zhao, T.D. Tlsty

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Zhao, K. Kerlikowske

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R.A. DeFilippis, C. Fordyce, K. Patten, H. Chang, K. Kerlikowske, B. Parvin, T.D. Tlsty

Writing, review, and/or revision of the manuscript: R.A. DeFilippis, C. Fordyce, K. Kerlikowske, T.D. Tlsty

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Fordyce, J. Zhao, G.V. Fontenay, K. Kerlikowske, T.D. Tlsty

Study supervision: R.A. DeFilippis, T.D. Tlsty

Other (actually performed experiments): R.A. DeFilippis, K. Patten, C. Fordyce

The authors thank Dr. Blackburn (UCSF) for thoughtful discussions and advice, Tlsty laboratory members, particularly Dr. Gascard, for editorial assistance and the Nikon Imaging Center (UCSF) for imaging support. The authors also thank Dr. Au (UCSF), Drs. Hornik and Kim (Kaiser Foundation Research Institute, Oakland, CA), Dr. Sukumar (Johns Hopkins Sidney Kimmel Comprehensive Cancer Center, Baltimore, MD), and Ms. Wiles (Cooperative Human Tissue Network, Nashville, TN) for providing breast tissue samples.

This work was supported by NIH/NCI PO1 CA107584 to T.D. Tlsty, K. Kerlikowske, and B. Parvin (under LBNL contract No. DE-AC02-05CH11231), NIH/NCI RO1 CA097214 to T.D. Tlsty, CBCRP grant 14OB-0165 to T.D. Tlsty, NIH/NCI U54 CA143803 to T.D. Tlsty and B. Parvin (UC Riverside) and R01 Research Supplement for Underrepresented Minorities to C.A. Fordyce.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.