Background
Studies of premalignant breast lesions provide insights into the mechanisms rendering the breast epithelium susceptible to oncogenic transformation as well as identify interventions that could prevent breast cancer. Atypical hyperplasias (AH) develop in the terminal duct lobular units of the breast and are further subdivided into either atypical ductal hyperplasia (ADH) or lobular neoplasia (LN). Lobular neoplasia encompasses atypical lobular hyperplasia (ALH) and lobular carcinoma in situ (LCIS). In addition, flat epithelia atypia (FEA) is a subtype of atypical ductal epithelium lacking architectural changes as seen in ADH. The 10-year risk of progression to invasive cancer is estimated to be 7% for all AH [
1] with a cumulative incidence approaching 35% at 30 years. The overall risk of developing breast cancer is increased by ~ 4-fold among women with atypia and is similar for both ductal and lobular lesions [
1,
2]. However, the risk is most prominent among women with higher breast density [
3] suggesting that mechanisms underlying breast density affect the progression of premalignant breast lesions. AH lesions are most often positive for estrogen receptor alpha (ERα+) and approximately 90% of tumors that develop subsequently are ERα-positive. Thus, AH represent a precursor lesion to low-grade ERα-positive tumors. Selective receptor modulators or aromatase inhibitors prevent progression of AH to invasive carcinomas by about 60% [
4‐
6], further supporting an important role for estrogen signaling in malignant progression of AH.
The expression of several genes and proteins have been evaluated in AH and their relationship with risk of progression. CK5/6 and ERα can aid the morphologic interpretation of usual ductal hyperplasia and AH lesions and increase the sensitivity of distinguishing among these lesions [
7,
8]. However, levels of ERα in AH detected by immunohistochemical staining did not predict risk of breast cancer [
9]. The extent of Ki67 immunoreactivity in normal breast tissue has been associated with an increased risk of breast cancer [
10]. Combined measures of proliferation, tumor suppressor activity, and inflammatory signaling within AH, using immunohistochemical scoring for Ki67, p16, and COX-2 respectively, have been used to evaluate breast cancer risk [
11]. Elevated levels of EZH2 were shown to be an early marker for progression of preneoplastic lesions [
12], while other studies identified increases in DNA methylation within promoter elements of tumor suppressor genes such as
APC,
DLEC1,
HOXA1,
RASSF1A, and
SFRP1 [
13,
14]. Progressive methylation of genes in early lesions was reported for
RASSF1A and
RARB2, suggesting the potential value of these targets [
15]. Higher levels of estrogen receptor beta (ERβ) are associated with a decreased risk of progression of AH [
16], suggesting that selective agonists of ERβ may offer potential therapy for chemoprevention. Gene expression profiles have also been used to identify early changes in AH as well as adjacent tumors [
17,
18]. These studies suggest the presence of molecular changes within breast epithelial cells associated with the transition to AH and risk of progression to breast cancer.
In the present study, patients diagnosed with AH and no history of breast cancer (prior to or within the 2-year follow-up after AH diagnosis) were selected. Laser capture microdissection was used to collect both histologically normal benign epithelium (HNB) and AH tissues from each patient. The complete transcriptome was evaluated by microarray, and gene expression patterns were used to define signatures that distinguish AH lesions from HNB tissues. Although ADH and LN tissues have distinct morphologic features, they did not form distinct clusters based on gene expression patterns suggesting that these premalignant lesions share underlying alterations in transcriptional programs. Pathway analyses identified genes encoding ERα, epidermal growth factor receptors (ERB-B), and androgen receptor (AR) as central nodes in the expression profiles. ERB-B2 and WNT signaling pathways were also strongly over-represented among the genes differentially expressed in AH. As methylation and loss of SFRP1 expression had been associated with premalignancy, we determined if it may be responsible for altered expression of a subset of genes altered in AH. Knock down of SFRP1 expression in normal breast epithelial cells (76N-Tert) identified 13 genes within the AH signature that had not previously been connected to SFRP1. SFRP1-regulated genes were also observed in mammary tissue from mice bearing deletion of the Sfrp1 gene. Re-expression of SFRP1 in an ERα-positive breast cancer cell line (MCF7), which has lost expression of the endogenous SFRP1 gene, had the opposite effect providing additional confirmation of an SFRP1-regulated gene network. Antagonism of estrogen-induced responses in progesterone receptor levels was demonstrated by addition of recombinant SFRP1 to human breast explant cultures. These findings demonstrate that SFRP1 expression is diminished in AH resulting in deregulation of a larger program of genes and loss of restraint on ERα signaling which may contribute to development of premalignant breast lesions.
Methods
Patient specimens
This is a retrospective study using formalin-fixed and paraffin-embedded (FFPE) archival tissue blocks. A search of pathology electronic files (CoPath) included patients with isolated AH lesions (atypical ductal hyperplasia, flat epithelial atypia, atypical lobular hyperpalsai, classic type lobular carcinoma in-situ) on core biopsy with subsequent excisional biopsy, isolated AH lesions on primary excisional biopsies, and reduction mammoplasties. Exclusion criteria included patients with a prior history of breast cancer or breast cancer within 2 years of initial AH diagnosis or insufficient AH lesion on subsequent excision. Original diagnoses were supported by at least two pathologists. A subspecialized breast pathologist (GMC) reviewed all cases retrieved for the study for concordance of original diagnosis. Cases that, upon review by GMC, did not meet histopathologic criteria for AH (ductal or lobular) were excluded. Characteristics of patients and diagnoses are in Table
1. Patient 14 was found to have a diagnosis of severe ADH bordering on ductal carcinoma in situ (low grade) upon review of original slide material. Institutional Review Board approval was obtained from Baystate Health, Springfield, MA (protocol number 182463).
Table 1
Patient characteristics and array identifiers
1 | 43 | FEA | Left | JJ013 | JJ014 |
2 | 46 | LCIS | Right | JJ001 | JJ002 |
3 | 47 | ADH | Right | JJ005 | JJ006 |
4 | 58 | LCIS | Left | JJ015 | JJ016 |
5 | 40 | ADH, FEA | Right | JJ003 | JJ004 |
6 | 63 | ADH | Left | JJ007 | JJ008 |
7 | 52 | LCIS | Right | JJ009 | JJ010 |
8 | 62 | ADH, FEA | Left | JJ012 | JJ011 |
9 | 46 | LCIS | Left | JJ017 | JJ018 |
10 | 40 | ADH | Right | JJ019 | NA@ |
11 | 49 | ADH, ALH | Left | DJJ021 | DJJ022 |
12 | 53 | ALH, FEA | Right | DJJ023 | DJJ024 |
13 | 53 | LCIS | Left | DJJ025 | DJJ026 |
14* | 58 | ADH | Right | DJJ027 | DJJ028 |
15 | 60 | LCIS, ALH | Left | DJJ029 | DJJ030 |
16-Block1** | 52 | ADH, FEA | Right | DJJ031 | DJJ032 |
16-Block2** | 52 | ADH, FEA | Right | DJJ033 | NA*** |
18 | 47 | ALH | Left | DJJ035 | DJJ036 |
19 | 55 | ALH | Right | DJJ037 | DJJ038 |
20 | 70 | ADH severe | Right | DJJ039 | DJJ040 |
21 | 44 | FEA | Right | DJJ041 | DJJ042 |
22 | 50 | ADH/FEA | Left | DJJ043 | DJJ044 |
23 | 54 | ADH/DCIS | Left | DJJ045 | DJJ046 |
Microscopic evaluation
Atypical hyperplasias (AH) arise in the terminal duct lobular units of the breast and are divided into ductal and lobular subtypes based on cytomorphologic characteristics. Ductal lesions included in the study are atypical ductal hyperplasia (ADH) and flat epithelial atypia (FEA); lobular lesions included atypical lobular hyperplasia (ALH) and/or classic lobular carcinoma in situ (LCIS), representing a spectrum and also known as lobular neoplasia. Subjects with either ductal or lobular atypia were included in the analysis to assess differences in transcriptional features. Areas of AH and benign ducts/lobules were marked for microdissection by the breast pathologist. RNA of sufficient quantity and quality was obtained from 21 AH lesions (from 20 patients). The tissues included 8 lobular lesions (ALH and/or LCIS), 11 ductal lesions (ADH and/or FEA), and 2 that were a mixture of ductal and lobular lesions. All subjects were female; the mean age was 51.9 years (SD = 7.9 years).
Analysis of RNA integrity
The integrity of RNA in tissue sections was assessed by amplifying 5′ fragment (nucleotides 1355–1472) and 3′ fragment (nucleotides 1650–1717) of the β-actin gene (Additional file
2: Table S5) by quantitative RT-PCR (RT-qPCR). An 8-μm section from each tissue block was placed on a glass microscope slide in RNase-free conditions, deparaffinized in 3 changes of xylene, and allowed to air-dry under vacuum in a desiccator for 30 min. The tissue sample was scraped from the slide using a razor blade directly into 150 μl digestion buffer containing 10 μl proteinase K (miRNeasy FFPE Kit, Qiagen, Germantown, MD) and incubated at 55 °C for 2 h. The samples were subsequently incubated at 80 °C for 15 min and transferred to ice for 3 min. The samples were centrifuged at 13,000×
g for 20 min, and the supernatant was transferred to new tubes. The RNA was harvested following DNase digestion using the miRNAeasy FFPE kit as described in the manufacturer’s instructions (Qiagen). The cDNA was prepared using 100 ng total RNA, oligo dT primers, and the Transcriptor first strand cDNA synthesis kit according to the manufacturer’s instructions (Roche, Indianapolis, IN). Amplification of 5′ and 3′ β-Actin targets were performed using the KAPA SYBR Fast Master Mix (Thermo Fisher, Waltham, MA) containing 200 nM forward primer, 200 nM reverse primer, and 5 μl cDNA. The conditions for mRNA amplification were performed as follows: 40 cycles each of 1 cycle at 95 °C for 2 min, 1 cycle at 95 °C for 15 s, and 1 cycle at 60 °C for 30 s; 1 cycle at 95 °C for 15 s, 1 cycle at 60 °C for 15 s, 20 min ramp, and 1 cycle at 95 °C for 15 s. The
Ct value of the 3′ β-Actin target was subtracted from the
Ct value of the 5′ β-Actin target to determine the amplification ratio. Specimens with ratios < 5 were used for microdissection and transcriptome-wide profiling.
Microdissection and RNA isolation
The H&E stained sections of the AH samples were used to estimate the total area for microdissection. A minimum area of 10 × 106 μm2 was required to ensure a minimum of 50 ng total RNA. Consecutive tissue sections (8-μm thick) were cut using RNase-free conditions and mounted on membrane slides (MMI, Rockledge, FL). The first and every 4th section were H&E stained for microscopic evaluation to confirm that AH tissue was present in unstained microdissected areas. The AH lesion and benign glandular tissue were marked by a breast pathologist (GMC) for microdissection. The benign glandular areas were selected to be at least 1 cm away from the AH lesion in the same tissue block or a different block. The tissues on membrane slides were deparaffinized in 3 changes of xylenes and allowed to air-dry under a vacuum in a desiccator for 30 min prior to laser capture microdissection. The unstained sections were oriented for microdissection aided by landmarks defined on the H&E stained slides. Areas to be microdissected were circumscribed using MMI Cell Tools software (Version Celltools-4.4 #261, Rockledge FL). Microdissected AH and HNB tissues from each patient were collected separately onto caps (MMI Inc., Rockledge FL). The microdissected tissue was collected in 150 μl digestion buffer containing 10 μl proteinase K (miRNeasey FFPE Kit, Qiagen), was kept overnight at 55 °C, and was stored at − 80 °C until further processing. Total RNA was isolated from microdissected tissue using the miRNeasy FFPE Kit (Qiagen) according to the manufacturer’s instructions and quantified using a NanoDrop™1000 (Thermo Fisher Scientific).
cDNA synthesis, amplification, and labeling
The Ovation® FFPE WTA System (NuGEN, San Carlos, CA) was used to prepare amplified cDNA from FFPE-derived total RNA because amplification is initiated at the 3′ end as well as randomly throughout the whole transcriptome in the sample which makes this system ideal for amplification of RNA obtained from FFPE samples. Fifty nanograms of RNA was used to prepare the cDNA according to the manufacturer’s instructions. The cDNA was then purified using columns from the QIAquick PCR Purification kit (Qiagen). Buffer PB from the purification kit was added to the cDNA reaction, loaded on to the column, and centrifuged for 1 min at 13,000×g. The flow through was discarded, and 80% ethanol was added to the column and centrifuged for 1 min at 13,000×g. The 80% ethanol wash step was repeated, and the purified cDNA was eluted with nuclease-free water. An aliquot containing 5 μg of cDNA was fragmented and labeled using the Encore® Biotin Module (NuGEN) according to the manufacturer’s instructions. The biotin-labeled cDNA was hybridized to Affymetrix 1.0 ST microarrays by Genome Explorations (Memphis, TN).
Analysis of microarray data
The data were normalized using the Single-Channel Array Normalization (SCAN) and Universal exPression Codes (UPC) methods from the BioConductor R package “SCAN.UPC” [
19]. This package produces standardized expression measures that are used to estimate whether a given gene or probe is active in a specific sample [
20,
21]. ComBAT, an empirical Bayesian framework, was used to adjust data for batch effects [
22]. The normalized data are available from the NCBI Gene Expression Omnibus Repository [
23] series record GSE118432. Limma [
24] was used to identify differentially expressed genes in a paired-sample model, with HNB and AH samples paired by patient. AH arrays JJ019 and DJJ033 were excluded from the analysis because paired HNB data were not available for these patients. A total of 1039 probesets were selected with adjusted
p < 0.05. Two methods were employed to identify gene signatures distinguishing AH and normal benign tissue. Agglomerative clustering was performed using AGNES [
25] to visualize gene expression patterns. Prediction Analysis of Microarrays (PAM) was used as an alternative approach to define a minimal gene expression signature [
26].
Network analysis
The differentially expressed probesets were mapped to 812 unique genes and used for network analysis [
27]. In cases where there were more than one probeset for a gene, the data were averaged. Protein interaction networks were constructed using the STRING database available within the network analyst tool (
http://www.networkanalyst.ca/). Over-representation of KEGG pathways was determined, and pathways were visualized using Cytoscape [
28].
Analysis of genes dependent on SFRP1
The 76N-Tert cell line was derived from normal breast epithelial cells [
29] and expresses endogenous
SFRP1. Generation and cultivation of engineered human cell lines (TERT-pSUPER, TERT-siSFRP1, MCF7-pCDNA, MCF7-SFRP1) has been described previously [
30‐
32]. Total RNA was extracted from cell lines using an acid-phenol extraction procedure [
33], according to the manufacturer’s instructions (Trizol, Invitrogen, Carlsbad, CA). Relative levels of mRNA were determined by using the 1-Step Brilliant® SYBRIII® Green RT-qPCR Master Mix Kit (Stratagene) containing 200 nM forward primer, 200 nM reverse primer, and 100 ng total RNA. The conditions for cDNA synthesis and target mRNA amplification were performed as follows: 1 cycle of 50 °C for 30 min, 1 cycle of 95 °C for 10 min, and 35 cycles each of 95 °C for 30 s, 55 °C for 1 min, and 72 °C for 30 s. Expression of each gene was normalized to levels of β-actin mRNA. The PCR primer sequences used are described in Additional file
2: Table S5.
Animals
The study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Baystate Medical Center Institutional Animal Care and Use Committee (Permit Number: 132681). Ten-week-old female C57BL/6-
Sfrp1+/+ mice (
n = 6) and C57BL/6-
Sfrp1−/− mice (
n = 6) were individually housed in plastic cages with food and water provided continuously and maintained on a 12:12 light cycle. The
Sfrp1 knockout allele has been described previously [
34,
35]. Mammary tissue was collected from mice, flash-frozen, and stored at − 80 °C until processed for RNA isolation and used to quantify relative levels of transcripts by RT-qPCR using primers described in Additional file
2: Table S5.
Human breast explant cultures
The tissue was aseptically minced and placed on Surgifoam gelatin sponges (Ferrosan, Sueborg, Denmark) in 60-mm tissue culture dishes containing phenol-red free DMEM/F12 (Gibco) 2% charcoal stripped serum, insulin, and gentamycin treated with vehicle (100% EtOH), 10 nM 17β-estradiol (E2; Sigma), or 10 nM E2 with 1 μg/mL rSFRP. Explant cultures were maintained for 24 h in 5% CO2 air and subsequently formalin-fixed and paraffin-embedded.
Progesterone receptor staining
Immunohistochemistry (IHC) was performed on a DakoCytomation autostainer using the Envision HRP Detection system (Dako, Carpinteria, CA). Mammary tissue blocks were sectioned at 4 μm, deparaffinized in xylene, rehydrated in graded ethanols, and rinsed in Tris-phosphate-buffered saline (TBS). Heat-induced antigen retrieval was performed in a microwave at 98 °C in 0.01 M citrate buffer. After cooling for 20 min, sections were rinsed in TBS and incubated with rabbit polyclonal anti-PR 1:500, (Cell Signaling; #8757) for 30 min at room temperature. Immunoreactivity was visualized by incubation with diaminobenzidine for 5 min. Tissue sections were counterstained with hematoxylin, dehydrated through graded ethanols and xylene, and cover-slipped. Images were captured with an Olympus BX41 light microscope using (SPOT™Imaging Solutions, Detroit, MI). PR staining of epithelial cells was quantified using ImageJ.
Statistical analyses
The mean expression of genes in parental cells (TERT-pSUPER, MCF7-SFRP1) versus SFRP1 knockdown/overexpressing cells (TERT-siSFRP1, MCF7-SFRP1) and Sfrp1+/+ versus Sfrp1−/−mammary gland tissues were compared using paired t tests.
Discussion
The pathways disrupted in AH offer insights into the early molecular changes that are associated with increased risk of malignancy. Prior analyses of AH have most often utilized regions of AH that co-exist with carcinomas [
17,
18]. This approach was necessary because tissues from percutaneous core biopsies and excisional biopsies with AH in women without cancer are often exhausted for histopathological diagnosis. The recovery and quality of RNA from formalin-fixed, paraffin-embedded (FFPE) tissues is poor and limits the methods for genome-wide transcriptional profiling to 3′ ends of mRNA for prior studies [
37]. However, AH adjacent to tumors may already harbor alterations similar to the tumor cells [
38] which can confound results. To overcome these limitations, we optimized methods for RNA isolation and amplification to allow reproducible analysis of transcriptional profiles using microdissected AH lesions from women without a prior history of breast cancer.
Using these methods, we confirmed increases in mRNA levels of
ESR1 and decreases in
KRT5 as were previously reported using immunohistochemical detection of these proteins [
7], as well as differences in
CDH1 (encoding E-cadherin) in ductal and lobular AH [
36]. Paired analysis of AH and HNB tissues within individuals identified a 99-gene signature that discriminated 90% of the AH and 81% of the HNB (Fig.
3). HNB from 4 individuals clustered with the “AH class” which may reflect false-positives. Alternatively, it may also reflect a limitation of using tissue that is adjacent to AH lesions which can harbor genetic alterations present in the AH [
39,
40]. Field effects have been reported at margins of 2 cm [
41]. This is especially likely for patient 15 for whom the HNB clustered next to the AH suggesting similarity at the molecular level. Conversely, AH from patients 1 and 4 were in the HNB class which may indicate false negatives, but may also indicate lesions that express the morphologic features of AH yet have a molecular signature more similar to HNB. The lesion from patient 1 was an FEA which is consistent with a low potential for malignant progression. These inherent uncertainties in diagnosis lead to an underestimate of the transcriptional changes in AH. Conversely, the differences in expression detected using the 99-gene signature reflect a robust set of biomarkers that can aid concordance in the diagnosis of AH [
42,
43].
Within the signature, there were increases in genes associated with the luminal (e.g.,
ESR1,
GATA3,
KRT18) and decreases in genes associated with the basal breast epithelium (e.g.,
KRT5,
TP63,
ACTA2). This may reflect a clonal expansion neoplastic luminal cells in AH which could result in a decrease in the apparent expression of genes that are markers of the basal epithelium. However, there were also significant decreases in genes associated with the normal luminal epithelium. Immunolocalization of c-KIT is reported in normal breast epithelial cells [
44,
45] with loss of c-KIT in low-grade breast cancers. ELF5 has been associated with differentiation of the luminal epithelium, and its expression is decreased significantly in luminal A, luminal B, and HER2 subtypes of breast cancer [
46]. Levels of these genes associated with luminal epithelium were decreased significantly in the AH samples (Fig.
4; Additional file
2: Table S3). In addition, increased levels of immunohistochemical staining for ERα, GATA3, and FOXA1 proteins have been reported in early breast lesions [
8,
37,
47] which is consistent with increased expression of mRNA detected in AH (Figs.
2,
3, and
4, Additional file
2: Table S3). Therefore, the signature is not solely due to the increase in luminal epithelium in AH tissues. The expression analyses of AH may represent more complex cellular phenotypes that are not characteristic of either the basal or luminal cells within normal breast tissues. Transcriptional profiling of mouse mammary and human breast tissues have also revealed an unexpected complexity of cellular identities and lineages [
48‐
50]. Therefore, it is possible that the gene signature in AH may represent an enrichment of a subclass of breast epithelial cells.
Models of breast progression generally support the evolution of invasive ductal and lobular carcinomas along two distinct lineages of lesions [
51]. This is based on the differences in genomic alterations and gene expression profiles in invasive ductal and lobular carcinomas [
52,
53]. A comprehensive analysis identified mutational hallmarks distinguishing invasive lobular carcinomas [
54]. However, we failed to observe a distinction in expression profiles from ductal and lobular AH by hierarchical clustering (Fig.
3). Similar numbers of AH were profiled for both the ductal and lobular histological classes, and expression of E-cadherin (
CDH1) mRNA levels confirmed the classifications. This raises the possibility that alterations in a common set of pathways contribute to atypical hyperplasias in both ductal and lobular epithelial cells. The subsequent alterations observed during progression toward ductal and lobular carcinomas may be defined by vulnerabilities that differ in the ductal and lobular cell types.
The network analyses provide further insights into the spectrum of molecular changes detected in AH and render the breast epithelium at heightened risk of breast cancer. Among genes that are differentially expressed in AH (0-order network), we observed a network involving
ESR1,
ERB-B receptors and
AR/GATA3/FOXA1 (Fig.
5). Prior studies examining transcriptional profiles of benign hyperplasias also identified increased expression of
ESR1 and
ERBB genes [
8,
47]. A study of matched normal, early neoplasia and carcinoma from a cohort of 25 women identified elevated expression of
ERBB2,
FOXA1, and
GATA3 [
37]. This study by Brunner et al. included only 7 cases of early neoplasias without synchronous cancer which likely limits the threshold for detecting changes in AH. Nonetheless, they observed elevated expression of
KDM4B,
XBP1,
AR,
MYB, and
SPDEF in early neoplasias compared to normal [
37] which were also elevated in our profile (Additional file
2: Table S1).
KDM4B and
XBP1 are part of the interaction network with
ESR1 while
SPDEF,
FOXA1,
GATA3, and
MYB are linked to
AR (Fig.
5). Therefore, our data is consistent with and extends the information provided by prior studies. As the samples used in our study were from AH in patients without breast cancer, these genes define alterations in pathways that may contribute to the formation of premalignant breast lesions. The prominent role of estrogen signaling in the network is consistent with the success of anti-estrogen treatments in preventing progression of AH to carcinomas [
55,
56].
Signaling networks in AH may also include mRNAs and proteins where levels are unchanged, but their activities are stimulated by interactions with other proteins causing post-translational modifications (e.g., by phosphorylation) and formation of larger complexes. Therefore, the 1st-order networks were also interrogated using KEGG pathways. This yielded a complex network of 306 genes and again identified over-representation of ERB-B signaling. WNT signaling was also found to be significantly over-represented in AH (Table
3). While SFRP1 can antagonize WNT signaling, it has been shown to also bind other proteins such as RANKL and thrombospondin, as well as affects signaling and responses via ERα, TGFβ, and p53 [
30‐
32]. Furthermore, loss of
SFRP1 is an early event observed in the MCF10A progression series [
57]. Therefore, we tested whether loss of
SFRP1 expression may regulate a portion of the genes within the AH signature. Inhibition of
SFRP1 in immortalized normal breast epithelial cells mirrored the changes in expression of 13 genes that were differentially expressed in AH (Fig.
6b). Re-expression of
SFRP1 in MCF7 cells reversed the changes in expression of 9 of the 13 genes tested (Fig.
6d). The presence of an SFRP-regulated gene network was also conserved in mouse mammary tissues (Fig.
6c). Consistent with these findings, the
Sfrp1−/− mice exhibit precocious side-branching and hyperplasia of ductal/lobulo-alveolar units [
35]. Together, these data demonstrate a role for SFRP1 in driving a portion of the signature found in AH.
Several of the SFRP1-regulated genes are involved in signal transduction. Loss of SFRP1 expression had consistent effects resulting in increased expression of
FOXA1,
ERRB4, and
KDM4B in both TERT-siSFRP1 cells and in mouse mammary tissues. FOXA1 is a pioneer factor which can open up chromatin to allow for access to ERα transcriptional sites [
58‐
60]. KDM4B is a histone demethylase which is upregulated in ERα+ breast tumors and can regulate the expression of both ERα and FOXA1 as well as modulate ERα and p53 signaling [
61‐
64], ERBB4/HER4 is critical for progesterone receptor (PR) expression [
65] and has also been suggested to be responsible for promoting an autocrine proliferation pathway induced by estrogen [
66,
67]. The increased expression of
FoxA1,
Kdm4b, and
Erbb4 a mammary glands of
Sfrp1−/− mice is consistent with the increased proportion of PR expressing cells and proliferation [
32].
Analysis of ERα activity and endogenous ERα targets in our human cell lines also exhibited enhanced ERE reporter activity when SFRP1 was knocked down and repressed reporter activity when SFRP1 was re-expressed [
32]. To confirm previous studies suggesting SFRP1 control of estrogen-induced PR expression, we added rSFRP1 to explants of normal breast tissue and demonstrated tempered induction of PR protein by estrogen (Fig.
7). Deletion of
Sfrp1 in mice also resulted in enhanced estrogen-stimulated responses and occasional hyperplasias, but was not sufficient for the development of spontaneous mammary tumors [
32,
35]. Therefore, loss of
SFRP1 expression appears to be a key driver leading to broader alterations in gene expression and permitting increased signaling through ERα and derangements in the ERB-B and WNT pathways as well.
Although considered a “benign” lesion because progression to invasive cancer is relatively low, AH presage at least 40,000 breast cancer diagnoses annually [
1,
68]. Therefore, intervention at this early stage offers the possibility of prevention of breast cancer. However, to minimize overtreatment, it is important to identify biomarkers discriminating the small subgroup of women with AH for whom risk is sufficiently high to warrant intervention. Molecular profiling of ductal carcinoma in situ aids in identifying women who may omit radiation following breast conserving surgery using a 12-gene panel (7 biomarkers, 5 reference genes) [
69,
70]. None of these 7 biomarkers were detected among the 1039 genes that were differentially expressed in AH compared to histologically normal epithelium (Additional file
2: Table S1). While the 99-gene signature discriminates AH tissues (Fig.
3), the study is not able to access the utility of the signature in assigning risk. Larger cohorts of pure AH with follow-up of > 20 years for validation of predictive signatures are needed to identify women with AH who will benefit from interventions and reduce the potential for overtreatment [
2].
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.