Unique Transcriptomic Changes Underlie Hormonal Interactions During Mammary Histomorphogenesis in Female Pigs

Abstract

Successful lactation and the risk for developing breast cancer depend on growth and differentiation of the mammary gland (MG) epithelium that is regulated by ovarian steroids (17β-estradiol [E] and progesterone [P]) and pituitary-derived prolactin (PRL). Given that the MG of pigs share histomorphogenic features present in the normal human breast, we sought to define the transcriptional responses within the MG of pigs following exposure to all combinations of these hormones. Hormone-ablated female pigs were administered combinations of E, medroxyprogesterone 17-acetate (source of P), and either haloperidol (to induce PRL) or 2-bromo-α-ergocryptine. We subsequently monitored phenotypic changes in the MG including mitosis, receptors for E and P (ESR1 and PGR), level of phosphorylated STAT5 (pSTAT5), and the frequency of terminal ductal lobular unit (TDLU) subtypes; these changes were then associated with all transcriptomic changes. Estrogen altered the expression of approximately 20% of all genes that were mostly associated with mitosis, whereas PRL stimulated elements of fatty acid metabolism and an inflammatory response. Several outcomes, including increased pSTAT5, highlighted the ability of E to enhance PRL action. Regression of transcriptomic changes against several MG phenotypes revealed 1669 genes correlated with proliferation, among which 29 were E inducible. Additional gene expression signatures were associated with TDLU formation and the frequency of ESR1 or PGR. These data provide a link between the hormone-regulated genome and phenome of the MG in a species having a complex histoarchitecture like that in the human breast, and highlight an underexplored synergy between the actions of E and PRL during MG development.

Successful breastfeeding, the production of milk for dairy and animal production, as well as the survival of all mammals, requires the coordinated growth and functional differentiation of the mammary gland (MG) epithelium before the onset of lactation (1). To this end, the majority of MG development occurs during the formation of lobuloalveoli or the maturation of more complex terminal ductal lobular units (TDLUs) (2), followed by their functional differentiation before the onset of milk secretion. These changes and the factors controlling them also affect breast cancer risk, as highlighted by the fact that hormone-induced terminal differentiation of the mammary epithelium during a full-term pregnancy can reduce a woman’s risk for developing breast cancer (3).

Underlying all these processes are the growth-promoting and/or differentiative effects of ovarian 17β estradiol (E) and progesterone (P), and growth hormone and prolactin (PRL) from the anterior pituitary (2). Independently, each of these hormones can stimulate proliferation in normal mammary epithelial cells (MECs) and breast cancer cells (4-6), where E promotes ductal elongation during puberty (2, 7, 8) while P stimulates branching of the mammary ducts as well as alveolar development during pregnancy (9). A primary function of PRL is during terminal differentiation and lobuloalveolar development (10). That said, most of our understanding about how hormones act on the MG stems from studies in rodents that undergo a relatively simple histomorphogenesis of their mammary parenchyma (11), which is also surrounded by an adipose-rich stroma with limited connective tissue (12).

Importantly, none of these hormones act on the MG in isolation during puberty, pregnancy or lactation, as highlighted in the classic study in mice showing that E, P, and either PRL or growth hormone are essential for the MG to undergo full development (13). Indeed, cooperativity between E and P stimulates tertiary branching and alveolar development across a range of species (2) through E-induced P receptor (PGR) expression (14) and epithelial expression of E receptor α (ESR1) (15). This cooperative effect of E and P also occurs in the breasts of postmenopausal women following combined hormone replacement therapy (16). In a similar way, cooperativity between P and PRL can stimulate epithelial proliferation in the absence of E (17). Beyond these examples, experiments investigating how one or more hormones acts on the MG are generally performed using single-gene knockout mice (18) or pituitary-intact ovariectomized (OVX) mice/rats (19), thereby ignoring potentially significant and biologically relevant hormone interactions. In the same way, previous studies of gene expression changes following hormone stimulation have almost always examined the effects of just 1 or 2 hormones or growth factors (14, 20). For example, recombination of ESR1^–/– or amphiregulin knockout (AREG^–/–) cells with wild-type mammary cells elucidated a vital role for AREG in E-stimulated proliferation in the peripubertal female mice (21). Other tissue recombination experiments (eg, RANKL^–/– and PGR^–/– mice) in adult mice used microarrays to reveal a vital role for P-induced receptor activator of nuclear factor κB-ligand (RANKL [TNFSF11]; 22) in P-stimulated proliferation (19, 23). However, the regulation of genes and signaling pathways such as these is not quite so cut and dried. For example, tissue recombination experiments using PRLR^–/– epithelium in pregnant PRLR^+/+ mice along with microarray analyses revealed that the PRLR is required for expression both of Areg and Tnfsf11 (24), while P also regulates both the expression of Areg in rodents (22) and signals MG proliferation and terminal end bud (TEB) formation via its protein (25).

The human breast has a unique histomorphogenesis that differs appreciably from that in the mouse MG (2, 8), consistent with its endocrinological regulation during the menstrual vs the estrous cycle (26). Furthermore, despite the many ongoing efforts to define the endocrine regulation of MG function in mice, the role of individual hormones during normal human breast development remains unclear, as does the response to their combinations. Indeed, there are numerous considerations as to why the extrapolation of data from mice cannot inform a detailed understanding of human breast development. While some have used mice bearing xenografted breast tissue to address some of these considerations (27), additional challenges arise such as the fact that mouse PRL does not activate the human PRLR (28).

One approach for understanding the mechanisms underlying human breast development is to analyze MG development, its endocrine regulation, and the transcriptional responses in a species that has genetics, physiology, and MG biology similar to that of humans. In this way, the MG of pigs is histomorphologically similar to that in the human breast, including identical development and characterization of the TDLU (8, 29) and a similar adipose and connective-tissue stroma (11). In addition, the pig genome is 3 times more similar to the human genome than is that of mice (30), and pigs have a reproductive endocrinology more akin to that in humans (31, 32).

Given the potential utility of the pig MG for modeling human breast development, we hypothesized there would be distinct transcriptomic signatures to define specific histomorphological changes in the MG in response to interactions between different primary hormone signals. To that end, we report herein our alignment of a range of phenotypic outcomes in the MG against its full transcriptomic profile during the response to all combinations of E, P, and PRL.

Materials and Methods

Animals

Tissues used for this study were from female pigs that were manipulated to become hormone-insufficient before repletion with exogenous hormones, as described previously by Horigan et al (8). Briefly, peripubertal female Yucatan miniature pigs (n = 32) were surgically OVX, then were treated daily with 2-bromo-α-ergocryptine methanesulfonate salt (Bromo) to induce hypoprolactinemia (0.1 mg/kg/d, intramuscularly; Sigma-Aldrich; dissolved in 100% ethanol and diluted in 0.9% w/v saline) for a further 8 days. Beginning on day 9, pigs (n = 4/group) were subsequently treated for 5 days with all possible combinations of 17β-estradiol (0.1 mg/kg/d; Sigma-Aldrich), medroxyprogesterone 17-acetate (as a source of P; 0.25 mg/kg/d; Sigma-Aldrich), and Bromo and/or haloperidol (to induce hyperprolactinemia; 1.5 mg/kg/d; Sigma-Aldrich). An additional group of females (n = 4) underwent sham surgery and were then treated with saline for the subsequent 13 days to serve as controls for normal gene expression under the influence of endogenous hormones. Twenty-four hours before euthanasia, pigs were injected with 5-bromo-2-deoxyuridine (BrDU; 5 mg/kg intraperitoneally) to label proliferating cells. At necropsy, tissue samples for RNA extraction were collected from all glands on the left side of the mammary chain and snap-frozen in liquid nitrogen. Samples for immunohistochemistry and whole-mount histological analyses were harvested from all glands on the right side of the MG and fixed in 4% (wt/vol) paraformaldehyde. Pigs were housed, treated, and euthanized in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals, as approved by the University of Vermont Institutional Animal Care and Use Committee.

Microarrays

Total RNA was extracted from the fourth MG (counting cranial to caudal) of each OVX, hormone-treated, or sham-operated control pig (n = 36) using TriReagent. Following DNase treatment (33), and RNA integrity was analyzed on an Agilent 2100 Bioanalyzer. The RNA (1 µg) was reverse-transcribed (RT) to double-stranded complementary DNA (cDNA) using Superscript II (Thermo Fisher Scientific) and oligo-dT priming as described previously (34). Biotinylated-cRNA was then synthesized from double-stranded cDNA by in vitro transcription (Enzo Life Sciences BioArray) and hybridized to Affymetrix GeneChip Porcine Genome Arrays for 16 hours at 45 °C following the manufacturer’s instructions. GeneChips were incubated with streptavidin-conjugated phycoerythrin, followed by a biotin-coupled polyclonal antistreptavidin antibody, then streptavidin-phycoerythrin. Data were collected with the Affymetrix 7G GeneChip scanner, robust multichip average (RMA) normalized, filtered, and log₂-transformed. The porcine GeneChip array contained 23 935 probe sets. Of these, 3143 probe sets were excluded from further analysis given they were undetectable in at least 2 samples within at least 1 treatment group, leaving 20 792 probe sets for statistical analyses. Data were submitted to the NCBI in the Gene Expression Omnibus (GSE106212) https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE106212. Affymetrix porcine gene annotations release 36 (April 13, 2016) was used to identify probe sets and for gene annotation. Sus scrofa gene annotations were also updated manually using the National Center for Biotechnology Information Gene database and GenBank. Mus musculus gene annotations for the data from Kouros-Mehr and Werb (35) and Berryhill et al (36) were updated using the Mouse Genome Informatics Web Site (37).

Cell Culture

MCF-7 human breast cancer cells (RRID: CVCL_0031) were cultured in 5% CO₂ with Dulbecco’s modified Eagle’s medium containing 10% fetal bovine serum (FBS; Gemini Bio-Products) and penicillin/streptomycin. On day 0, subconfluent cells were changed to Dulbecco’s modified Eagle’s medium containing 10% charcoal dextran-treated FBS (Omega Scientific) and penicillin/streptomycin. On day 2, cells were treated with either E (1 µM, dissolved in ethanol) in 10% charcoal-treated serum, 10% FBS, or 0.1% ethanol in 10% charcoal-treated serum (control). On day 4, cells were harvested for either total RNA, or to determine cell number, which was measured by methylene blue assay against a standard curve of known cell numbers (38).

Northern Blot

Total RNA (5 µg) was electrophoresed on a 1% agarose/formaldehyde gel before transfer to Hybond-XL (Amersham Biosciences) and cross-linking for 2 hours at 80 °C. Prehybridized blots (0.5-M sodium phosphate pH 7.2, 10-mM EDTA pH 8.0, 7% sodium dodecyl sulfate; 65 °C for 2 h) were then hybridized overnight at 65 °C with either pig CSN2 (25 ng, 441-817 bp of Genbank Accession No. X54974) or RN18S (25 ng, 1416-1526 bp of Genbank Accession No. AY265350) cDNA labeled with α-³²P dCTP (6000 Ci/mmol, 50 µCi, Amersham Biosciences) by random priming. Blots were washed to a stringency of 0.1X SSC, 0.1% sodium dodecyl sulfate at 65 °C, and scans of the resultant autoradiographs were quantified using National Institutes of Health Image (https://imagej.nih.gov/nih-image/).

Reverse Transcription and Quantitative Polymerase Chain Reaction

Total RNA was extracted from MCF-7 cells using Qiazol following the manufacturer’s instructions and DNAse-treated as described earlier. Integrity of all RNA was checked using formaldehyde-agarose gels and RT as previously described (33). Primer and probe sequences are presented in Table 1. Expression of human CLSPN, SKA2, CKAP2, and NBL1 was determined using PrimeTime Std quantitative polymerase chain reaction (qPCR) Assays (Integrated DNA Technologies). qPCR was performed using either Fast SYBR Green Master Mix (Applied Biosystems), with 0.2-µmol L^–1 primers, or TaqMan Gene Expression Master Mix (Applied Biosystems), 0.9-µmol L^–1 primers, 0.2 µmol L^–1 hydrolysis probe on a 7500 Fast Real-Time PCR System (Applied Biosystems) using default cycling conditions and the Tm in Table 1 for the combined annealing and extension steps. All reactions were performed in duplicate. Each PCR run included a no-template control containing all reagents except cDNA, and an RT negative-control reaction that was performed in the absence of Moloney murine leukemia virus RT. Melting curve analyses confirmed a single amplicon. To quantify gene expression, standard curves were prepared using 5 to 6 5-fold serial dilutions of a sample known to have high levels of expression for the particular messenger RNA (mRNA). All standard curves had linear regression coefficients of determination of at least 99%. The mRNA and RN18S or RNA18SN1 levels in each sample were interpolated from the Ct values by using the standard curves. Data were expressed as the ratio between the gene’s mRNA and RN18S or RNA18SN1 expression levels, yielding a normalized relative expression level for each mRNA.

Table 1.

Polymerase chain reaction (PCR) primers and probes used for quantitative PCR

Accession No.	Sus scrofa gene	Primers (F, R)	Tma, °C	Eb,%
XM_003131674	SKA2	F CAAACGGACCTGGAGCTATTACC R GACATGTGCGCTTTTAATTGCT	59	89
XM_003130956	CKAP2	F AGCGCAGACATACCATCGTAAA R TCCACTCACTCAGACGAGCTTTT	59	103
XM_021095924.1	CLSPN	F AGCGAGACTCAGCGCCTTA R GGGTTTACGTTTGTAGAATTCATGA	59	104
XM_021095465.1	NBL1 f	F TTGGGAAGCTCTTTCTGAATATCTG R GGCTCAGCCAGGGTCCTT	59	86
	ATP6V1C2 c	F GGTAGTCCCTCGGTCAACCA R CGAAACAGAGTCACGGTGAAGA	59	97
	TNXB f ,c	F GGTCAGAGTACCAGGTCACTGT R ATGGTGGTGATGGTCTTGGA	58	81
	JUN f ,c	F TCCAGTAACGGGCACATCAC R TGCTCGTCAGTCACGTTCTTG	58	102
	SLC35B2 c	F AGGCGGAAGAACTACCTAGAGACA R CCTTAGGCTCATTGCCAAACA	59	99
	CDO1 c	F GGAAAACCAGTGTGCCTACATCA R GAAGGCTCACAGCAGGTTCTG	58	102
	IGFBP3 c	F CCAGGAAACGGCAGTGAGTCC R TCCATGCTGTAGCAGTGCACG	60	80
	RN18S f ,d	F ACGGCTACCACATCCAAGGA R CCAATTACAGGGCCTCGAAA	60	86
Accession No.	Homo sapiens gene	Primers (F, R) and probes (P) (5′-3′)	Tm a , °C	E b , %
	RNA18SN1 f ,d	F ACGGCTACCACATCCAAGGA R CCAATTACAGGGCCTCGAAA	60	86
NM_182744	NBL1 e	F TGTAGCTGAAGCACTGTCCTA R CAAGAACATCACCCAGATCGT P 6-FAM/AGGCCAAGT/ZEN/CCATCCAGAACAGG/IowaBlack FQ	60	99
NM_001190481	CLSPN e	F CTCAGACAGTTTCCCTGAACTC R ATTCCATCCTATCAGCCTTGC P 6-FAM/ACAGGCCGT/ZEN/GGGACCAGTTTT/IowaBlack FQ	60	96
NM_182620	SKA2 e	F CCGCAGTTTTCTCTTCTTTAGTC R TGAGCAGAAAGAGAGTAAGAGC P 6-FAM/CAAACAGAC/ZEN/CTGGAGCTGTCACCA/IowaBlack FQ	60	95
NM_018204.5	CKAP2 e	F TTTGCCAGCTTTCCACTCA R TTGACCAGCGAAGACATACTG P 6-FAM/CGAGCTTTT/ZEN/CTCTCTTCCGAGGTTTCT/IowaBlack FQ	60	95

Abbreviations: F, forward; R, reverse.

^a Annealing and extension temperature.

^b Primer pair efficiency (E).

^c Published in (39)

^d Published in (40).

^e PrimeTime Std quantitative PCR Assay (Integrated DNA Technologies).

^f Both primers are located within one exon.

Immunohistochemistry

Immunohistochemistry for ESR1 and PGR was performed on one fore- and one hind- MG sample to account for any variation due to MG position. Immunohistochemistry for phosphorylated STAT5 (pSTAT5) was performed on one MG from a randomly selected location. Tissue was fixed overnight in 4% paraformaldehyde, dehydrated, and embedded in paraffin before sectioning at 4 μm. Rehydrated sections were pretreated with Triton-X (0.3% in phosphate-buffered saline [PBS]) and hydrogen peroxide (3%). Antigen retrieval was performed by steaming sections in citrate buffer (Target Retrieval Solution, DakoCytomation). Slides were blocked for endogenous biotin (A/B Block, GeneMed) and nonspecific binding using 10% horse serum in PBS. Rabbit polyclonal anti-pSTAT5 (RRID:AB_2315225; 1:50; Cell Signaling Technology), rabbit polyclonal anti-PGR antibody (RRID:AB_2164331; 1:50; Santa Cruz Biotechnology Inc), or mouse monoclonal anti-ESR1 antibody (RRID:AB_61341; 1:50; Ab-11, Lab Vision Corp) were incubated overnight at 4 °C in 10% horse serum. Sections were then rinsed in PBS-Tween20 before incubation with either a biotinylated antimouse immunoglobulin G secondary antibody (1:1000, Jackson ImmunoResearch Laboratories Inc) or a biotinylated F(ab′)2 fragment donkey antirabbit immunoglobulin G (H + L) secondary antibody (5 μg/mL; Jackson ImmunoResearch Laboratories Inc). Horseradish peroxidase-conjugated streptavidin (1:350, Jackson ImmunoResearch Laboratories Inc) was then applied before immunoreactivity was detected as a blue (Vector SG, Vector Laboratories Inc) or maroon (Vector Red, Vector Laboratories Inc) precipitate. Immunohistochemistry for BrDU using a mouse monoclonal biotinylated anti-BrDU antibody (RRID:AB_2532919; 1:50; Zymed Laboratories Inc) as previously described (8). Sections were counterstained with nuclear fast red or hematoxylin. The proportion of MECs that were positive for ESR1 or PGR was quantified by counting at least 200 MECs within each of 5 ducts and 5 TDLUs in each MG. The proportion of MECs positive for pSTAT5 was quantified by counting at least 200 MECs within each of 3 ducts and 3 TDLUs from each animal.

Phenotyping of Mammary Gland Histomorphogenesis

Mammary tissue from fore and hind MGs were fixed in 4% paraformaldehyde, sectioned into whole mounts, and stained with carmine alum, as described previously (8). Parenchymal structures were visualized under a 4× objective and classified according to their histomorphological features as described in the human breast, namely as terminal buds (TBs) or TDLU-1, -2, or -3, and their relative abundance quantified by manual counting using ImagePro (Media Cybernetics Inc), as described previously (8).

Statistics

Microarray calculations were performed using the R Programming Language and Environment for Statistical Computing (41) along with Bioconductor (42) packages. Quality was assessed using the “simpleaffy” package (43) of Wilson and Miller. Normalized probe set statistics were calculated using the RMA methodology of Bolstad and colleagues (44, 45) based on the “affy” package provided by the authors.

The top percentile of probe sets with respect to variance were analyzed using principal component analysis, probe set clustering, and sample clustering. Principal components were based on the RMA expression statistics, centered, and scaled by probe set. Twenty-four clusters were defined using the k-means algorithm (46) based on the Kendall tau distance. Unsupervised average-linked hierarchical clustering, using Euclidian distance as the metric, was performed on the RMA-normalized data using the “Heatplus” package provided by Ploner (47), to visualize probe set/sample relationships in the form of a centered and scaled heat map and dendrogram.

Linear modeling was performed using the Bioconductor limma package described by Smyth (48). Hypotheses were associated with a coefficient (D), the log posterior odds (B) statistic of Smyth (49), and a P value based on an F test.

To determine hormonal effects on gene expression, microarray expression data (n = 32 OVX hormone-treated pigs, where the sham-operated controls were excluded from analysis of variance [ANOVA] analyses) were analyzed using 3-way ANOVA (3 factors; E, P, and PRL at 2 levels), followed by a Benjamini-Hochberg (B-H) false discovery rate (FDR) procedure to control the FDR at 0.01. The overlap between these hormonally regulated gene lists was visualized in Venn diagrams (50).

To map treatment differences, the log₂-transformed data from 20 792 probe sets was averaged across 4 animals per treatment. Pairwise Pearson correlations were computed in R for all whole transcriptomes (averaged across treatment groups) using all differentially expressed probe sets; this correlation matrix was then visualized as a heat map after hierarchical clustering. Expression profiles of differentially expressed genes were clustered in R using a soft clustering algorithm in the Mfuzz package, which allows genes to be assigned to more than one cluster and quantifies the degree to which the cluster represents the expression profile for each gene (51). After inspecting overlap plots for 4, 6, 8, and 10 clusters, we established that 6 clusters were optimal, for which trajectories were then generated.

Previously published phenotypic data for the MGs of these pigs (%BrDU⁺ MEC, incidence of TDLU-1, -2, -3, and TB (8), expression of porcine PRL receptor long form (PRLR-LF; 33), as well as new data for %ESR1⁺ and %PGR⁺ MECs (this publication), were regressed against the expression levels for 20 792 probe sets from the 32 OVX hormone-treated pigs using the R statistical package, followed by a B-H FDR procedure to control the FDR at 0.01. We also regressed %BrDU⁺ MEC data against 20 792 probe sets for just the OVX E-treated animals (n = 16). The gene expression profile for selected genes that reached greatest significance was validated by RT-qPCR.

RT-qPCR (pig MG and MCF7 cells) and immunohistochemistry data were analyzed for the main effects of E, P, and PRL and their interactions by 3-way ANOVA using PROC GLM (SAS version 9.3, SAS Institute Inc). Data were pretransformed to achieve normality and/or homogeneity of variance. Treatment and either date (pig study) or experiment number (MCF7 cells) were fixed effects. For immunohistochemistry data, MG position and epithelial structure (duct or TDLU) were also considered fixed effects. Multiple means comparisons were made by post hoc Tukey-Kramer (pig MG) or Dunnett (MCF7) test, while controlling for multiple testing. The t test was used to analyze methylene blue assay results. The correlation between the frequency of ESR1- and PGR-positive cells in the same MG was calculated by linear regression analysis in Prism8 (GraphPad Software). Values were considered statistically significant at P less than .05.

Functional enrichment analyses were performed on gene lists of interest using the Database for Annotation, Visualization and Integrated Discovery (DAVID) version 6.8 (david.ncifcrf.gov) (52, 53). Functional categories that were included from the default options were KEGG pathways, all 3 default GOTERM for Gene Ontology (GO), and the 2 UP categories. The background list for Homo sapiens was used because our preliminary analyses found that the results were limited when genes were input as Sus scrofa. For each gene list of interest, annotations with a B-H FDR procedure multiple testing correction P value less than or equal to .05 were deemed significant.

Pathway enrichment analyses were conducted using Ingenuity Pathway Analysis (IPA; Qiagen). An IPA Core analysis was performed on each list of genes that was significantly regulated by the main effects of the 3 hormones or their interactions. An IPA comparison analysis was then run on these 7 (3 main effects, 4 interactions) core analyses. Disease-specific pathways were disabled and only pathways with a B-H multiple testing correction P value less than .05 (–log[P value] > 1.3) are reported, unless otherwise stated. In addition, an IPA upstream analysis was conducted on these same 7 lists of hormone-regulated genes to identify transcriptional regulators. The P value and z score both were used to identify statistically significant regulators. The IPA defines z scores of greater than 2 or less than –2 as indicating statistically significant activation/inactivation states, respectively.

Results

We previously reported certain hormone-induced phenotypic changes within the MG of hormone-deficient pigs following replacement therapy with combinations of exogenous E, P, and PRL (8). To identify genes regulated by combinations of these hormones relative to those regulated by each hormone alone, we analyzed RNA isolated from the MGs of 32 hormone-treated and 4 sham control pigs using porcine microarrays. We detected 20 792 probe sets, which represented 11 506 annotated Sus scrofa genes. Comparing the individual hormones, the greatest number of genes that changed their expression was following treatment with E (1760 genes) and PRL (347 genes) and their combination (Table 2). From the entire transcriptome for the 36 samples across all treatment groups, we generated a heat map for the top percentile of individual probe sets with respect to variance (Fig. 1) and a heat map of whole transcriptome distances (Fig. 2) to identify which hormone treatments evoked the most-pronounced transcriptional changes. The transcriptomic profiles of all MGs exposed to E were distinct from those without (see Fig. 1). We identified hormonally-regulated probe sets using ANOVA, then used these to construct a heat map of whole transcriptome distances between treatment groups (see Fig. 2). This correlation matrix revealed that treatment with E + PRL and E + P + PRL generated similar transcriptional outcomes (96% correlation) that were distinct from those for all other hormone treatments, implying a major response that was driven by the combination of E + PRL. The responses to E alone and E + P were also similar (98% correlation), as was the response to PRL and P + PRL (99% correlation; see Fig. 2). Consistent with data in Fig. 1, all treatment groups devoid of E were similar to each other (96%-99% correlation; see Fig. 2). The proportions of genes that were regulated by more than one hormone, as well as the interactions between hormones, are shown in Fig. 3.

Table 2.

Transcriptomic analysis of mammary tissues from pigs treated with all combinations of 17β-estradiol, medroxyprogesterone 17-acetate, and haloperidol to induce hyperprolactinemia

No. of genes	E	P	PRL	E * P	E * PRL	P * PRL	E * P * PRL
Statistically significant at FDRa≤.01b	1760	23	347	43	212	5	5

Values are the numbers of genes affected by one of the hormones (E, P, or PRL), or their interactions (denoted by an asterisk) as determined by 3-way analysis of variance. Total number of probe sets in the analysis was 20 792, representing 11 506 annotated genes

Abbreviations: E, 17β-estradiol; P, medroxyprogesterone 17-acetate; PRL, prolactin.

^a Benjamini-Hochberg false discovery rate.

^b α ≤ .01; P ≤ 2.19 × 10^–4.

Figure 1.

Heat map/dendrogram of the top 242 differentially expressed probe sets in the porcine mammary gland (MG). Pigs (n = 32) were ovariectomized (O) and treated with 2-bromo-α-ergocryptine (Bromo) to inhibit prolactin (L) release for 9 days followed by 5 days’ treatment with combinations of 17β-estradiol (E), medroxyprogesterone 17-acetate (P), and/or Bromo and/or haloperidol to induce hyperprolactinemia (L). Pigs not treated with haloperidol continued to receive Bromo. Four control pigs were not ovariectomized (underwent a sham operation) and were injected with saline (A----, B----, C----, D----). The MG RNA was extracted and analyzed using Affymetrix Porcine GeneChip microarrays. The actions of E (either from hormone replacement or due to the presence of ovaries) were the major determinant of MG gene expression. The x-axis dendrogram represents the similarities/differences in gene expression profile between different animals. The y-axis dendrogram represents the similarities/differences between probe set expression levels across all animals. A, B, C, D = 4 individual pigs per treatment group.

Figure 2.

Heat map showing the Pearson correlation for whole transcriptomes between different hormone treatment groups. Pigs were ovariectomized and treated with all combinations of 17β-estradiol (E), medroxyprogesterone 17-acetate (P), and/or bromo-α-ergocryptine (Br) and/or haloperidol to induce hyperprolactinemia (PRL), per Fig. 1. Probe set expression per treatment group was averaged across biological replicates. Annotated probe sets identified as differentially expressed in the mammary gland in response to any main hormone effect or a hormone interaction (listed in Supplementary Tables 1-7) (54)) were included in the analysis. Samples were clustered using hierarchical clustering with the dendogram showing the distance between treatment groups.

Figure 3.

Distribution of fold-change expression of probe sets hormonally regulated in the mammary gland. Pigs were ovariectomized and treated with all combinations of 17β-estradiol (E), medroxyprogesterone 17-acetate (P) and/or bromo-α-ergocryptine (BROMO), and/or haloperidol to induce hyperprolactinemia (PRL), per Fig. 1. The fold change in expression of probe sets regulated by a main effect of A, E; B, P; or C, PRL was calculated by either comparing E-, P-, or PRL-treated pigs (n = 4) to BROMO-treated pigs, or comparing all pigs receiving E, P, or PRL (n = 16) to all pigs not receiving E, P, or PRL (n = 16). Dotted lines represent 1.5-fold induction or suppression. D to G, Venn diagrams illustrating the overlap for individual hormone-regulated genes (per Table 2) across different combinations of E, P, and PRL. For clarity and ease of interpretation, each Venn diagram presents a maximum of 5 sets of genes. D, Genes unique to, or overlapping between, E regulated, PRL regulated, P regulated, E * P regulated, and E * PRL regulated. E, Genes unique to, or overlapping between, E regulated, PRL regulated, P regulated, E * P regulated, and P * PRL regulated. F, Genes unique to, or overlapping between, E regulated, PRL regulated, E * P regulated, E * PRL regulated, and P * PRL regulated. G, Genes unique to, or overlapping between, E regulated, P regulated, E * P regulated, and E * P * PRL regulated. E * P * PRL is only included in this Venn diagram because the genes regulated by this interaction include one gene that was also regulated by E * P and one gene that was also regulated by E.

Genomic Responses to Individual Actions of 17β-Estradiol, Progesterone, and Prolactin

Given the range of MG development in response to the different hormone combinations, we next examined the expression of all annotated keratin genes to ensure that treatment effects on gene expression did not reflect pronounced changes in the epithelial/stromal composition of the MG. As shown in Supplementary Fig. 1 (54), there was no treatment effect on the expression of KRT5, KRT8, KRT10, KRT15, KRT17, KRT18, or KRT19.

Among the 1760 genes regulated by E, 44% of probe sets were upregulated 1.5-fold or greater, and 31% were downregulated 1.5-fold or greater, by the main effect of E (Supplementary Table 1A; Fig. 3A) (54). Among these, the canonical E-induced genes PGR (55) and AREG (21) were both markedly induced in the E-treated animals compared to those treated with only Bromo (see Supplementary Table 1A) (54). Using DAVID, functional analysis of E-regulated genes that were upregulated by at least 2-fold revealed 6 of the 7 clusters to be mitosis related, whereas the fifth top was “adenosine 5′-triphosphate (ATP) binding/kinase” and other significant clusters included “extracellular matrix” (ECM) and “collagen” (see Supplementary Tables 1E and 1F) (54). Functional analysis of E-regulated genes that were downregulated by at least 2-fold identified the top functional term to be “signal peptide,” with 2 other top clusters being “lipid degradation” and “gluconeogenesis.” An IPA upstream analysis of E-regulated genes indicated the most influential upstream activators (P < .001; z score > 2) were ERBB2, E2F1, β-estradiol, TBX2, ESR1, EP400, RABL6, CCND1, FOXM1, VEGF, PTGER2, the group E2f and MYC, while the most influential upstream inhibitors (P < .001; z-score < –2) were TP53, CDKN1A, let-7, FOXO3, TCF3, RB1, and the group Rb (Fig. 4). The most statistically significant pathways identified by IPA as being regulated by a main effect of E were all activated by E (B-H P < .001; z score ≥ 2) and included mitotic roles of polo-like kinase, cyclins and cell cycle regulation, actin nucleation by ARP-WASP complex, estrogen-mediated S-phase entry, and signaling by Rho family GTPases (B-H P < .001; z score ≥ 2).

Figure 4.

Heat map comparisons of z scores for 88 upstream regulators of gene expression. Pigs were ovariectomized and treated with 17β-estradiol (E), medroxyprogesterone 17-acetate (P), and either 2-bromo-α-ergocryptine or haloperidol to induce hyperprolactinemia (PRL) as per Fig. 1. Ingenuity Pathway Analysis was used to identify upstream regulators of genes expressed in the mammary gland and statistically significantly regulated by a main effect of either A, E, PRL, or the interaction between E * PRL; or B, E, P, or the interaction between E * P.

Only 23 genes had expression that was regulated by P (Table 2; Supplementary Table 2) (54); 83% of P-regulated probe sets were upregulated and 12.5% were downregulated 1.5-fold or greater by a main effect of P (see Supplementary Table 2A; Fig. 3B) (54). Among these, the most altered was ATF7IP2 (activating transcription factor 7 interacting protein 2), which was upregulated 2.95-fold by P. Interestingly, GSTA1 (glutathione S-transferase α 1) was upregulated 1.7-fold by P alone, but downregulated 25-fold by the addition of P to E (Supplementary Table 2C) (54). The opposite was true for NRCAM (neuronal cell adhesion molecule; see Supplementary Table 2) (54), which was downregulated 2.1-fold by P alone vs Bromo (see Supplementary Table 2A) (54), but was upregulated 6.4-fold by the addition of P to E (see Supplementary Table 2B) (54). Other genes known to be P responsive, such as soluble RANKL (TNFSF11) and calcitonin (CALCA), were not represented on the S scrofa microarray (22,56).

A total of 347 genes were regulated by PRL (see Table 2); 42% of probe sets were upregulated and 24% of probe sets were downregulated 1.5-fold or greater by a main effect of PRL (Supplementary Table 3A; Fig. 3C) (54). Unsurprisingly, 4 of the top 5 genes that were upregulated by a main effect of PRL were milk protein genes (Supplementary Table 3B) (54), which in the case of CSN2 was confirmed by Northern blot (See Supplementary Fig. 2) (54). Using DAVID for functional analysis of PRL-regulated genes that were upregulated or downregulated at least 2-fold, the significant key word with the most enrichment was “milk protein” (B-H P = .001), while the most significant GO term was “secreted” (B-H P  <.001; Supplementary Table 3D) (54). The 2 most enriched clusters of terms were based on “secreted” and “antimicrobial” (Supplementary Table 3E) (54). There were no statistically significant annotations for the genes downregulated at least 2-fold by PRL. Among the 347 PRL-regulated genes, 69 were also regulated by E (Fig. 3D). An IPA upstream analysis of these PRL-regulated genes identified that the most influential upstream activators (P < .001; z score > 2) were the cytokines tumor necrosis factor, OSM, interleukin (IL)-1B and PRL, along with PGR and FOXO1 (Fig. 4A). No upstream inhibitors of PRL-regulated genes were identified. The most statistically significant PRL-regulated pathways identified by IPA were increased fatty acid metabolism (P < .001; z score > 2) and decreased incorporation of both lipid and fatty acid (P < .001; z score < –2), as well as increases for pathways for organism mass, inflammatory response, activation of antigen presenting cells, engulfment by macrophages, cell cycle progression of lymphocytes, and transport of carboxylic acid (P < .001; z score > 2).

Synergistic Responses to Combinations of 17β-Estradiol, Progesterone, and Prolactin

Consistent with our previously-reported histological findings (8), the greatest interactive response by the MG to any combination of E, P, and PRL (statistically significant, and vs any change in response to each hormone individually) was when pigs were treated with E + PRL (212 genes changed; see Table 2, Supplementary Tables 4-7; Figs. 3D and 3F) (54). Among these, 76 genes were also regulated by E alone (Figs. 3D and 3F), 50 were regulated by PRL alone (Figs. 3D and 3F), while 18 were all regulated by E, PRL, and E * PRL (Figs. 3D and 3F). By contrast, statistical testing for the interaction between E and P only identified a synergistic change in the expression of 43 genes relative to the responses to E or P alone (see Table 2; Supplementary Table 4; Figs. 3E and 3G) (54). As shown in Figs. 3E and 3G, 16 of these 43 genes were also regulated by E alone, 11 genes were regulated by P alone, and only 7 genes were regulated by E, P, and E * P. The combination of PRL and P only affected the expression of 5 additional genes relative to each hormone alone (see Table 2; Supplementary Table 6; Figs. 3E and 3F) (54), while the interaction between E, P, and PRL differentially affected the expression of an additional 5 genes beyond the response to all other combinations of these hormones (see Table 2; Supplementary Table 7; Fig. 3G) (54). These data establish that E and PRL have the greatest synergistic effect on differential gene expression in the porcine MG. Functional analysis of E * PRL-regulated genes using DAVID identified the statistically significant terms, “cellular response to estradiol stimulus,” “Ubl conjugation,” and “mitochondrion” (B-H P < .05; Supplementary Table 5B) (54). An IPA analysis of the E * PRL-regulated genes found "creatine-phosphate biosynthesis" was upregulated, while "inhibition of matrix metalloproteases" was downregulated (B-H P < .05). The most statistically significant upstream activators (z score > 2) of these synergistically regulated genes were MYC (P < .001), E2F1 (P < .001), E2F2 (P < .001), and CLDN7 (P < .001), while the upstream inhibitors (P < .001; z score < 2) were TGFB1, CDKN2A, TWIST1, and SMARCB1 (Fig. 4A). A comparison between the upstream regulators of E-regulated, PRL-regulated, and E * PRL-regulated genes (see Fig. 4A) revealed a large number of regulators acting in the same direction for all 3 gene lists (eg, activators including MYC, RABL6, E2F1, CSF2, gentamicin, TNF, mibolerone, β-estradiol/estrogen, FOXO1, EGF, MYCN, AR, and HOXA10; and inhibitors including CDNK2A, mir-21, calcitriol, let-7, fulvestrant, KDM5B, and TWIST1. By comparison, there was no consensus between the upstream regulators of E-regulated, P-regulated, or E * P-regulated genes (Fig. 4B). The expression of genes found to have a statistically significant synergistic response to E + PRL from the arrays (Supplementary Fig. 3; Supplementary Table 5) (54) was also evaluated by qRT-PCR (Fig. 5). In the case of the Jun proto-oncogene (JUN), cysteine dioxygenase type 1 (CDO1), tenascin (TNXB), and insulin-like growth factor binding protein 3 (IGFBP3), the level of expression was confirmed as being reduced synergistically (P < .05) by E + PRL relative to the response recorded for either E or PRL alone (see Fig. 5A-5D). Conversely, expression of solute carrier family 35, member B2 (SLC35B2) was confirmed as being synergistically increased by E + PRL (P < .05; Fig. 5E). The qPCR data for ATP6V1C2 confirmed that it was PRL regulated (P < .001; Fig. 5F) but did not confirm a statistically significant interaction between E and PRL.

Figure 5.

Quantified expression for genes responsive to an interaction between the effects of 17β-estradiol (E) and prolactin (PRL). Pigs were ovariectomized and treated with hormones per Fig. 1. Reverse transcription–quantitative polymerase chain reaction (RT-qPCR) was used to analyze RNA from the mammary glands of pigs treated with E and either 2-bromo-α-ergocryptine (Bromo) or haloperidol to induce hyperprolactinemia (PRL). Panels are A, JUN; B, CDO1; C, TNXB; D, IGFBP3; E, SLC35B2; and F, ATP6V1C2. Data are means ± SEM (n = 4 pigs). ^a,b,cDifferent letters indicate statistically significant differences in gene expression between treatments (P < .05).

Clustering Analysis

To better understand the patterns of transcriptional responses to the various hormones, the microarray data were soft-clustered in mFuzz to generate 6 clusters (Fig. 6; Supplementary Table 8) (54). Cluster 1 genes were upregulated by either E alone, or E + P. An IPA analysis revealed the top upstream endogenous regulators of cluster 1 included estradiol, GSK3 inhibitor, and COLQ (all P < .001), while the top molecular and cellular function represented was “cell death and survival” (P < .001). Cluster 2 genes were upregulated by PRL alone, and were inhibited by E. An IPA upstream analysis of cluster 2 identified the top upstream regulator as STAT3 (P < .001), the top molecular and cellular function represented was “cell-to-cell signaling and interaction” (P < .001), and the top canonical pathway was “dendritic cell maturation” (B-H P < .05). Cluster 3 genes were upregulated by E + PRL, which was enriched for genes having the top molecular and cellular function being “cell cycle/mitosis” (P < .001), the top canonical pathways being “cell cycle control of chromosomal replication” and “E-mediated S-phase entry” (B-H P ≤ .001), and the top upstream regulators being proteins involved in cell cycle regulation (all P < .001), namely CDKN1A (57), E2F2 (58), RB1 (59), let-7 miRNA (60), and E2F1 (61). Cluster 4 genes were upregulated by PRL and E + PRL and included SLC35B2, as described earlier (Fig. 5E). The top upstream endogenous regulators of cluster 4 were ERN1 and estradiol (P < .001), where the top molecular and cellular function represented was the same as cluster 1, namely “cell death and survival” (P < .001). Cluster 5 genes were upregulated by E + PRL, but were inhibited by either E or PRL alone. The top upstream regulator of cluster 5 was the nuclear transcription factor HNF4A (P < .001), the top canonical pathway was “sirtuin signaling” (B-H P < .005), while the top molecular and cellular function was “carbohydrate metabolism” (P < .001). Cluster 6 genes were downregulated by E + PRL and included JUN, TNXB, CDO1, and IGFBP3 as described earlier (see Fig. 5). The top upstream endogenous regulators of cluster 6 were PPARG, tumor necrosis factor, and estradiol (P < .001), and the top canonical pathways were "complement system", "E-dependent breast cancer signaling", and "acute-phase response signaling" (B-H P < .05). The top molecular and cellular function represented by cluster 6 was “lipid metabolism” (P < .001). These clusters of similar expression profiles contained subsets of genes that we found by other analyses to be regulated by the main effects of the 3 hormones, or their interactions (eg, genes statistically significantly regulated by E * PRL were present in clusters 3, 5, and 6 (constituting ~ 20% of all probe sets in each cluster), illustrating that the response for a given gene across all hormone treatments was not identical, for example, for all genes regulated by a significant interaction between E * PRL.

Figure 6.

Clustering of similar gene expression profiles into 6 clusters using R and soft clustering. Pigs were ovariectomized and hormone treated with 17β-estradiol (E), medroxyprogesterone 17-acetate (P); and 2-bromo-α-ergocryptine (Bromo) or haloperidol to induce hyperprolactinemia (PRL), as per Fig. 1. The main hormonal coordinators of gene expression changes for each cluster are indicated above each graph. ctrl, control.

Synergistic Activation of the Mammary Epithelium by 17β-Estradiol, and Prolactin

In considering the overall effects of the various hormone treatments, E and PRL collectively regulated the expression of 2341 out of a total of 11 506 genes (20.3% of the total). We previously reported that the combination of E + PRL also stimulated maximal proliferation of the MG epithelium, alongside morphological development of the parenchyma from TDLU-1 to TDLU-3 (8). At the same time, there is widespread recognition that JAK2/STAT5 signaling, which is stimulated by PRL, directs proliferation and/or differentiation of the mammary epithelium (62) and can be modulated by E (63). Consistent with these roles, PRL induced the expression of both STAT5A and STAT5B, which also correlated positively with the frequency of TDLU3 structures (Supplementary Tables 3 and 15) (54). Given the combinatorial effects of E and PRL, we also assessed whether STAT5 signaling was modified in response to their combination. Indeed, there was a pronounced synergistic induction of pSTAT5 in the MG epithelium of pigs treated with E +PRL relative to the effect of either E or PRL alone (P < .05; Fig. 7).

Figure 7.

Immunohistochemistry for phosphorylated STAT5 (pSTAT5). All pigs were ovariectomized and treated with hormones as per Fig. 1. Mammary tissue was from pigs treated with or without 17β-estradiol (E), and either 2-bromo-α-ergocryptine (Bromo), or haloperidol to induce hyperprolactinemia (PRL). A, Representative pSTAT5 immunohistochemistry images from the 4 treatment groups. B, Quantification of the frequency of pSTAT5-positive nuclei. Data are means ± SEM (n = 4 pigs). ^a,b,cDifferent letters indicate statistically significant differences (P < .05).

Expression of Genes Associated with Epithelial Proliferation

Our hormone ablation and replacement approach generated a wide range of proliferative responses in the MG ranging from 0.5% to 25% of MEC labeled with BrDU within ducts and TDLU (8). We used this quantitative outcome to identify genes associated with hormone-induced proliferation by regressing the expression levels for 20 792 probe sets against %BrDU⁺ MEC for each animal. This approach identified 1669 genes (1899 probe sets) having expression levels that correlated (840 positive, 829 negative) with %BrDU⁺ (FDR < 0.01; Supplementary Table 9A and 9B; Table 3) (54). Among these, the expression of 1197 genes (72%) was statistically significantly regulated by E. Given the limited proliferation in nonestrogenized females, we repeated the regression analysis for the animals treated with E (n = 16). This approach refined the list to 71 genes that were positively or negatively correlated with %BrDU⁺ (Supplementary Table 9C) (54). By comparing the 2 lists (1899 vs 71 probe sets), we identified 38 genes common to both (Supplementary Table 9D (54)). The most significant annotation for these 38 genes is “cell cycle” (B-H P < .001) and the most enriched is “chromosome outer kinetochore” (B-H P < .01; Supplementary Table 9E) (54), where the only 2 clusters containing significant terms were both related to mitosis (Supplementary Table 9F) (54). 29 of these 38 proliferation genes are also regulated by E (Table 4). The most significant annotation for these 29 genes was the UniProtKB key word “cell cycle” (B-H P < .001; Supplementary Table 10A) (54), based on the known function of 16 of these genes (Table 4), including DDIT3, which was negatively correlated with proliferation and includes, among its many annotations, “cell cycle arrest” (see Table 4). The top 2 clusters for the functional annotations for these 29 genes are both based on “cell cycle” (Supplementary Table 10B) (54), confirming the validity of the “reverse genomics” approach. Among these genes, 13 have not previously been implicated in either cell proliferation or E-regulated MG growth. We selected 4 genes for further validation by qRT-PCR based on their range of expression across samples (Supplementary Fig. 4) (54) and on (manual) annotation of the Affymetrix probe sets. Expression of NBL1 was inhibited by E and was negatively correlated with E-induced cell proliferation (Fig. 8A), whereas CKAP2, CLSPN, and SKA2 (Fig. 8B-8D) were induced by E and were positively correlated with E-induced cell proliferation (see Fig. 8; main effect of E, P < .05). The expression of 3 genes (NBL1, CKAP2, and CLSPN; see Figs. 8A-C) was affected by PRL (P < .02), and the combination of E and PRL (P < .02) synergistically affected the expression of CKAP2 and CLSPN (positively; P < .003) and NBL1 mRNA levels (negatively; P < .02).

Table 3.

Regression analysis of phenotypic data for the mammary glands from hormone-treated pigs relative to microarray expression data from 20 792 probe sets

No. of genes	%BrDU+c	%ESR1+d	%PGR+e	%TDLU1f	%TDLU2g	%TDLU3h	No. of TBsi
Statistically significant at FDRa < 0.01b	1669	730	342	1190	557	532	1206

Abbreviations: BrDU, 5-bromo-2-deoxyuridine; ESR1, estrogen receptor-α; MG, mammary gland; PGR, progesterone receptor; TB, terminal bud; TDLU-1, terminal ductal lobular unit-1; TDLU-2, terminal ductal lobular unit-2; TDLU-3, terminal ductal lobular unit-3.

^a False discovery rate.

^b α ≤ .01: P < 4.464 × 10^–4.

^c Percentage of BrDU-positive epithelial cells in the MG (8).

^d Percentage of ESR1-positive epithelial cells in the MG (Fig. 10).

^e Percentage of PGR-positive epithelial cells in the MG (Fig. 11).

^f Percentage of TDLU-1 structures in the MG (8).

^g Percentage of TDLU-2 structures in the MG (8).

^h Percentage of TDLU-3 structures in the MG (8).

ⁱ Incidence of TBs in the MG (8).

Table 4.

17β-estradiol (E)-regulated genes involved in E-induced proliferation

	E-treated animals			All animals			E-regulated
Gene	BrDU slope	BrDU P	BrDU R2	BrDU slope	BrDU P	BrDU R2	P	Gene name
ANLN	16.34	1.17E-04	0.67	5.72	3.61E-08	0.64	3.28E-09	Anillin actin binding protein
BUB1	11.36	2.99E-04	0.62	6.27	7.78E-09	0.68	1.26E-09	BUB1 mitotic checkpoint serine/threonine kinase
CD99L2	–18.75	8.19E-05	0.68	–23.00	1.09E-06	0.55	1.95E-04	CD99 molecule-like 2
CDK1	8.57	1.36E-04	0.66	3.81	1.07E-08	0.67	1.42E-09	Cyclin dependent kinase 1
CENPF	8.48	3.73E-04	0.61	4.57	1.59E-08	0.66	1.11E-08	Centromere protein F
CKAP2	10.26	3.35E-05	0.72	5.34	3.74E-09	0.69	5.45E-09	Cytoskeleton-associated protein 2
CLSPN	11.76	8.38E-05	0.68	8.00	8.04E-09	0.68	1.86E-08	Claspin
DDIT3	–20.44	2.87E-04	0.62	–17.80	2.09E-06	0.53	3.47E-05	DNA damage-inducible transcript 3
DLGAP5	7.43	3.44E-04	0.61	4.63	1.50E-08	0.66	1.70E-08	DLG-associated protein 5
ENDOU	–8.21	4.41E-04	0.60	–7.67	1.84E-08	0.66	1.67E-06	Endonuclease, poly(U) specific
ENO2	–12.43	2.55E-04	0.63	–7.98	3.84E-07	0.58	6.37E-06	Enolase 2
GAS6	–16.59	1.24E-04	0.66	–9.28	2.41E-07	0.59	1.18E-06	Growth arrest specific 6
HEBP1	–18.34	1.15E-04	0.67	–21.45	4.93E-08	0.63	2.73E-05	Heme binding protein
HMMR	7.77	2.46E-04	0.63	3.75	3.86E-06	0.51	7.51E-06	Hyaluronan-mediated motility receptor
KIF23	9.62	2.59E-04	0.63	4.39	8.56E-08	0.62	1.22E-09	Kinesin family member 23
KNL1	10.20	2.02E-04	0.64	4.56	6.43E-08	0.63	2.83E-08	Kinetochore scaffold 1
NBL1	–9.70	2.34E-05	0.73	–7.78	2.36E-07	0.60	5.99E-05	Neuroblastoma 1, DAN family BMP antagonist
NCAPG	9.15	3.56E-04	0.61	4.53	6.96E-08	0.63	1.45E-08	Non-SMC condensin I complex subunit G
NUF2	8.89	2.33E-04	0.63	5.78	1.12E-07	0.61	1.31E-07	NUF2, NDC80 kinetochore complex component
PDGFRL	–14.77	4.04E-04	0.60	–12.43	7.41E-09	0.68	2.87E-07	Platelet-derived growth factor receptor like
RORA	–10.60	2.27E-04	0.63	–6.36	1.12E-05	0.48	1.64E-04	RAR related orphan receptor A
SKA1	7.68	2.54E-04	0.63	4.35	7.94E-08	0.62	6.34E-08	Spindle and kinetochore associated complex subunit 1
SKA2	19.92	7.63E-06	0.77	9.56	1.18E-07	0.61	4.50E-07	Spindle and kinetochore associated complex subunit 2
SMC2	11.40	1.16E-04	0.67	10.41	2.87E-07	0.59	1.12E-04	Structural maintenance of chromosomes 2
SMOC2	–10.76	2.31E-04	0.63	–9.32	7.16E-07	0.56	7.53E-05	SPARC-related modular calcium-binding protein 2
TPPP	–8.59	2.71E-04	0.62	–6.92	6.49E-08	0.63	2.21E-06	Tubulin polymerization promoting protein
VRK1	14.90	3.89E-04	0.61	11.73	1.49E-07	0.61	4.90E-06	VRK serine/threonine kinase 1
YPEL3	–15.35	2.04E-04	0.64	–12.57	2.72E-08	0.65	4.38E-07	Yippee-like 3
ZBTB47	–15.72	1.20E-04	0.66	–15.00	5.83E-08	0.63	3.66E-05	Zinc finger and BTB domain containing 47

Gene expression data from all ovariectomized hormone-treated animals (n = 32) was analyzed by analysis of variance to identify the main effect of E (FDR < 0.01), which was then regressed against %BrDU-positive cells from immunohistochemistry data to extract statistically significant linear relationships (FDR < 0.01). Next, gene expression data from only E-treated animals (n = 16) was regressed against %BrDU-positive cells. Genes that were E-regulated and statistically significantly associated with %BrDU-positive cells in both regressions are shown. Underlining indicates genes with an annotation of cell cycle, mitosis, or centrosomes. Slope, linear regression slope are shown for each gene.

Abbreviations: %BrDU, percentage of 5-bromo-2-deoxyuridine; E, 17β-estradiol; FDR, false discovery rate.

Figure 8.

Validation by reverse transcription–quantitative polymerase chain reaction (RT-qPCR) for genes having expression that correlated with estrogen (E)-induced cell proliferation in the mammary gland (MG). Pigs were ovariectomized and hormone treated as per Fig. 1 with 17β-estradiol (E), medroxyprogesterone 17-acetate (P), and either 2-bromo-α-ergocryptine (Bromo) or haloperidol to induce hyperprolactinemia (PRL). Graphs on the left show hormone-regulated gene expression in the pig MG as measured by RT-qPCR. Regression plots on the right show the correlation between the proliferation phenotype (percentage of 5-bromo-2-deoxyuridine [%BrDU]-positive cells) and log₂ or log₁₀ expression of the gene as measured by RT-qPCR. A, NBL1; B, CKAP2; C, CLSPN; D, SKA2. Black squares refer to non-E animals. Open circles refer to E-treated animals. Data are means ± SEM (n = 3-4 pigs). ^a,b,c,dDifferent letters indicate statistically significant differences (P < .05).

To further confirm a role for these genes during E-induced cell proliferation, we examined expression of NBL1, CKAP2, CLSPN, and SKA2 in ESR1-positive MCF7 human breast cancer cells treated with E or FBS (Fig. 9). Expression of 3 genes (SKA2, CLSPN, and CKAP2; see Fig. 9B) was induced during E- as well as FBS-induced proliferation. By contrast, expression of the tumor suppressor NBL1, which in the MGs of pigs, was inhibited by E and had decreased expression in the most proliferative samples in vivo, was intriguingly increased in MCF7 cells during E- and FBS-induced proliferation (see Fig. 9B).

Figure 9.

Effect of 17β-estradiol (E) and fetal bovine serum (FBS) on MCF7 cells and their expression of the 4 genes associated with proliferation in the mammary glands (MG) of pigs treated with E. Cells were cultured in 10% charcoal dextran-treated FBS for 48 hours and then treated with either 1-µM E in ethanol or 0.1% ethanol (Con) in charcoal-treated serum, or cultured with FBS for 48 hours. A, Cell numbers were measured using the methylene blue assay. Data are means ± SEM (n = 5 wells). Data are representative of 2 independent experiments. B, MCF7 cells were treated with E or FBS for 48 hours, RNA harvested, and gene expression measured using reverse transcription–quantitative polymerase chain reaction. Data are means ± SEM (n = 6-9) from 3 independent experiments. *P < .05; **P < .001; ***P < .0001 compared to Con.

Expression of Genes Associated With Terminal Bud Formation

Using the same approach, we examined gene expression changes corresponding to E-induced formation of TB, the frequency of which ranged from 5 to 53 TB/field in each MG (8). The expression of 1206 genes correlated with number of TBs per field (FDR < 0.01; Table 3; Supplementary Table 11A and 11B) (54); 1041 of these were also E regulated (PFDR < 0.01; see Supplementary Table 11) (54). Of a total of 165 significant terms for genes positively correlated with number of TBs per field, the top KEGG pathway was “cell cycle” (B-H P < .001; Supplementary Table 11D) (54). The top clusters of functional terms included many related to mitosis in addition to clusters of “extracellular matrix” (ECM), “mRNA splicing,” and “endocytosis/phagocytosis” (Supplementary Table 11E) (54). The most statistically significant term for genes negatively correlated with number of TBs per field was “extracellular space” (B-H P = .003; Supplementary Table 11F) (54), and the top cluster of functional terms is based on “metal binding” (Supplementary Table 11G) (54). Not surprisingly, of the 1041 genes involved in E-induced TB development, 827 (79%) were also correlated with %BrDU⁺ (Supplementary Table 12A and 12B) (54). The top KEGG pathway represented by these E-regulated genes positively correlated with number of TBs and %BrDU was “cell cycle” (B-H P < .001; Supplementary Table 12C) (54), and most of the clusters of terms are related to mitosis, ATP/GTP binding, mRNA processing, and mRNA transport (Supplementary Table 12D) (54). The top GO term represented by E-regulated genes that was negatively correlated with the number of TBs and %BrDU was “extracellular space” (Supplementary Table 12E (54)). The remaining 214 genes (21%) that did not correlate with %BrDU⁺ may be involved with E-regulated TB development, independent of proliferation (Supplementary Table 13A and 13B) (54). Within this gene list, the genes that positively correlated with E-regulated TB development, independent of proliferation, had 3 significant functional annotations, all related to “extracellular matrix” (ECM; B-H P < .02; Supplementary Table 13C) (54). The genes that negatively correlated with E-regulated TB development, independent of proliferation, have the statistically significant terms “phosphoprotein,” “alternative splicing,” and “cytoplasm” (B-H P < .04; Supplementary Table 13D) (54).

Expression of Genes Associated With Terminal Ductal Lobular Unit Development

As in humans, epithelial structures in the MGs of pigs can be described as TDLU, type 1, 2, or 3. We had previously quantified the frequency of these TDLU types within the MG (%TDLU1 7%-87%, %TDLU2 12%-75%, %TDLU3 0%-17%) in response to hormone replacement, where TDLU-3 developed only following exposure both to E and PRL (8). Using the same approach as described earlier, expression of the 20 792 probe sets was regressed against %TDLU1 (Supplementary Table 14) (54), %TDLU2 (Supplementary Table 15) (54), and %TDLU3 (Supplementary Table 16) (54). We identified 433 genes (FDR < 0.01) having expression that correlated both with %TDLU1 and %TDLU2. Notably, none of these correlations were in the same direction, where opposite directions of correlation for TDLU1 and 2 is consistent with our MG phenotype data where increased %TDLU1 was associated with a decrease in %TDLU2, and vice versa. By further parsing the gene list, we identified that expression of 321 genes was E regulated (FDR < .01) and correlated with %BrDU⁺ (FDR < .01), and in the same direction as for %TDLU2 (Supplementary Table 17A and 17B) (54). The expression for most of these genes (n = 306) also correlated with number of TBs per field, in the same direction as %BrDU⁺ and %TDLU2 (see Supplementary Table 17A) (54). Of these 306 genes, expression of 22% (n = 68) was negatively correlated with number of TBs per field, %BrDU⁺, and %TDLU2 (but positively with %TDLU1), and expression of 78% was positively correlated with number of TBs per field, %BrDU⁺, and %TDLU2 (but negatively with %TDLU1).

Only 216 genes (FDR < .01) had expression that correlated with both %TDLU1 (see Supplementary Table 14) (54) and %TDLU3 (see Supplementary Table 16) (54) in opposing directions, and none were correlated with %TDLU1 and %TDLU3 in the same direction. Of these, 185 were also regulated by E (FDR < .01) and correlated with %BrDU⁺ (FDR < .01) (in the same direction as %TDLU3; Supplementary Table 18A and 18B) (54). The most enriched cluster of functional annotations associated with these 185 genes was DNA repair (Supplementary Table 18D) (54), thereby distinguishing the processes involved in development of TDLU-1 to TDLU-2 from those associated with TDLU-3 development, despite the fact that expression of all these 185 genes also correlated with proliferation. The genes associated with development of TDLU-3 may have differentiative functions that are not currently annotated. The expression of around half of these (n = 98) was also correlated with number of TBs per field in the same direction as %BrDU⁺ and %TDLU3 (see Supplementary Table 18A) (54). Of these, 15 genes were associated with all 5 histomorphology outcomes (%TDLU1, 2, and 3; #TB/field; and %BrDU⁺) and were regulated by E (Table 5); 14 were positively correlated with %TDLU2, %TDLU3, number of TBs per field, and %BrDU⁺. Only 1, TSPAN7, was negatively associated with %TDLU2, 3, number of TBs per field, and %BrDU⁺. While the most statistically significant term represented by these 15 genes was “cell cycle” (7 genes; B-H P < .001), 11 of these 15 genes shared the UniProtKB key word “nucleus” (B-H P < .05; Supplementary Table 19A) (54), which was also the theme of the most statistically significant cluster of terms, ahead of a cluster of mitosis-related terms (Supplementary Table 19B) (54).

Table 5.

Genes regulated by 17β-estradiol (E; by analysis of variance, false discovery rate [FDR] ≤ 0.01) that correlated with percentage terminal ductal lobular units-1 (%TDLU1), %TDLU2, %TDLU3, number of terminal buds/field, and uptake of 5-bromo-2-deoxyuridine with an FDR of less than or equal to 0.01

Gene names	Slope of %TDLU1	Slope of %TDLU2	Slope of %TDLU3	Slope of No. of TBs	Slope of %BrDU+	E-regulated P	Function
KPNA2	–26.2	16.1	11.8	19.4	10.0	5.8E-08	Nuclear protein transport
AURKA	–18.4	11.4	8.2	12.6	6.9	1.4E-08	Spindle pole stabilization
BIRC5	–20.3	12.5	9.1	13.0	7.4	2.0E-09	Mitosis/Inhibits apoptosis
FANCD2	–20.3	12.7	8.8	14.0	7.6	1.3E-08	DNA repair
GINS1	–15.7	9.9	7.0	11.0	5.9	1.1E-09	DNA replication
LMNB2	–29.7	18.8	12.7	21.8	10.7	3.9E-08	Structure of nucleus
MAGED1	–41.2	26.6	17.6	28.0	14.7	2.9E-09	Programmed cell death
MCM4	–22.6	15.0	9.4	13.9	8.1	6.4E-10	DNA helicase
MCM6	–24.5	16.0	10.2	17.1	9.1	1.5E-10	DNA helicase
MTHFD2	–30.9	20.1	13.3	19.6	10.9	1.9E-07	One carbon metabolism
MYL7	–16.9	10.8	7.4	10.6	6.2	4.9E-09	Focal adhesion
PLK1	–20.3	12.8	8.8	13.6	7.4	9.8E-09	Mitosis
RAD54L	–23.3	14.4	10.5	15.8	8.5	8.5E-10	DNA repair/recombination
SYCE2	–33.8	20.2	15.4	21.0	12.8	2.0E-11	Cell cycle
TSPAN7	24.6	–15.3	–11.4	–16.3	–9.1	1.3E-08	Cell development, growth, activation, and motility

Nonunderlined genes are positively correlated, and underlined genes are negatively correlated, with proliferation and development of TDLU-2, -3, and TB. Slope equals the linear regression slope for each gene.

Abbreviations: BrDU, 5-bromo-2-deoxyuridine; E, 17β-estradiol; FDR, false discovery rate; TDLU1, terminal ductal lobular units-1; TDLU2, terminal ductal lobular units-2; TDLU3, terminal ductal lobular units-3; TBs, terminal buds/field.

We also examined whether any of these E-regulated genes associated with development of TDLU-1 through TDLU-3 were also responsive to the interactive effect of E + PRL. The expression of only 26 genes was correlated with %TDLU1, %TDLU3, and %BrDU⁺ and was regulated both by E * PRL and E. These 26 genes may be involved in the differentiative effects of E + PRL during progression from TDLU-1 to TDLU-3. Of these, 11 were negatively correlated with %BrDU⁺ and TDLU-3 development, whereas 15 were positively correlated (Table 6).

Table 6.

Genes correlated with percentage terminal ductal lobular units-1 (%TDLU1), %TDLU3, and percentage 5-bromo-2-deoxyuridine (false discovery rate [FDR] < 0.01) that were also regulated by 17β-estradiol (E; FDR < 0.01) and E * prolactin (PRL; FDR < 0.01)

Gene symbol	Slope %TDLU1	Slope %TDLU3	Slope %BrDU+	E * PRL-regulated P	Gene function; if function unknown, then gene name
MSH3	–19.0	8.8	7.5	2.1E-04	Postreplicative DNA mismatch repair
NPM3	–19.1	10.2	7.2	4.4E-05	RNA binding/chaperone
DHFR	–21.4	10.3	8.6	7.3E-05	DNA precursor synthesis
RAD51AP1	–20.5	10.4	8.0	1.2E-04	DNA damage response
RCC2	–18.6	10.7	7.3	1.6E-04	Microtubule binding
DDIAS	–26.2	12.0	11.0	2.1E-04	Antiapoptosis
BRCA1	–25.5	12.5	9.3	1.8E-05	Damaged DNA binding
LOC100511884	–20.8	13.3	8.5	1.2E-06	Histone H2A type 1-like
CCDC167	–29.6	16.1	10.6	4.7E-05	Coiled-coil domain-containing protein 167
PDSS1	–28.8	16.6	11.3	5.7E-05	Ubiquinone biosynthesis
SELRC1	–30.5	17.2	11.3	7.8E-05	Assembly of mitochondrial respiratory chain
H2AFZ	–43.6	20.2	17.7	2.3E-05	Transcription regulation
BPNT1	–41.4	21.5	16.1	9.0E-05	Sulfur metabolism
ERP29	–51.4	28.0	20.2	1.2E-04	Protein processing in ER
FANCI	–51.6	29.5	21.2	1.8E-04	DNA repair
RNASE4	44.0	–22.7	–15.8	9.7E-05	Ribonuclease
FHL1	37.6	–22.2	–14.5	1.1E-06	Muscle development
F2RL2	36.6	–21.3	–14.3	3.1E-06	Receptor for activated thrombin
PLXDC2	28.7	–16.3	–10.4	2.7E-06	Tumor angiogenesis
ACAA1	26.9	–14.6	–11.0	1.7E-04	Fatty acid metabolism
SORL1	23.3	–14.3	–9.3	1.4E-04	Uptake of LDL
EBF1	27.6	–13.7	–10.8	6.2E-05	Transcriptional activator
SYT11	22.7	–13.0	–8.6	6.6E-05	Ca(2+)-dependent exocytosis of secretory vesicles
TNXB	21.8	–11.7	–8.3	1.1E-05	Inhibits cell migration
MAMDC2	15.2	–8.1	–5.4	1.2E-05	MAM domain containing 2
ABCA6	15.3	–7.7	–5.7	5.4E-06	Macrophage lipid homeostasis

Nonunderlined genes are positively correlated, and underlined genes are negatively correlated, with proliferation and development of TDLU-3. Slope equals the linear regression slope for each gene.

Abbreviations: BrDU, 5-bromo-2-deoxyuridine; ER, endoplasmic reticulum; LDL, low-density lipoproteins; PRL, prolactin.

Expression of Genes Associated With Abundance of Estrogen Receptor α

The steroid hormone receptors are critical regulators of MG development and are important prognostic indicators in breast cancer. To gain further insight into the relationship between steroid hormone receptor expression and gene expression underlying MEC proliferation and differentiation, we quantified the frequency of immunohistochemically localized ESR1⁺ MECs within the MGs of pigs treated with the various hormone combinations (Fig. 10A). There was no difference in the %ESR1⁺ MECs within ducts and TDLU structures (P = .79), and there was no effect of MG position in the MG apparatus (P = .8). Treatment with E slightly increased the frequency of ESR1⁺ MECs vs that in OVX females treated with Bromo (18% vs 15% positive, P < .05). Furthermore, there were inhibitory main effects of E (P < .001), P (P < .001), and PRL (P < .001) on the abundance of ESR1⁺ MECs. There were also negative interactive responses for %ESR1⁺ in response to E * P (P < .001), E*PRL (P < .001), and P * PRL (P < .001), so that only 0.1% of MECs remained as ESR1⁺ when all 3 hormones were administered (Fig. 10B).

Figure 10.

Effect of hormone treatments on the abundance of E receptor α (ESR1)-positive epithelial cells in the mammary glands (MGs) of pigs. Epithelial cells that were ESR1-positive (ESR1⁺) were localized by immunohistochemistry. Pigs were ovariectomized and hormone treated as per Fig. 1 with 17β-estradiol (E), medroxyprogesterone 17-acetate (P), and either 2-bromo-α-ergocryptine (Bromo) or haloperidol to induce hyperprolactinemia (PRL). Control pigs were sham-ovariectomized and injected with saline as per Fig. 1. A, Percentage ESR1⁺ epithelial cells in control gilts, ovariectomized Bromo-treated gilts, and ovariectomized gilts treated with all combinations of E, P, and either Bromo or PRL. Data are means ± SEM (n = 4 pigs). ^{a,b,c,d,e,f,g}Different letters indicate statistically significant differences (P < .05). B, Representative images of immunohistochemistry for ESR1 in ducts and terminal ductal lobular units (TDLUs) of the MG. Positive cells were detected using Vector SG (blue) and cells were counterstained using nuclear fast red. Brown patches are melanin. Scale bar equals 0.5 mm. C, Correlation between %ESR1⁺ cells and ESR1 messenger RNA abundance measured by microarray analysis. Black squares equal non–E-treated animals. Open circles equal E-treated animals.

Gene expression data for ESR1 mRNA confirmed it was regulated by P (P < .001), PRL (P < .001), and the interaction between E * PRL (P < .001; data not shown). We also confirmed the validity of the anticipated relationship between ESR1 frequency (by immunohistochemistry) and its gene expression (4 different probe sets). The correlation between these 2 measures ranged from R² = 0.64 for the probe set with the highest level of expression (P = < .001; Fig. 10C), to R² = 0.23 (P = .006) for the probe set with the lowest level of expression. The probe set detecting the highest level of expression were sequences located in the 3′ untranslated region of 2 full-length coding ESR1 transcripts (ENSSSCT00000063430.1 and ENSSSCT00000035147.2).

The microarray data for 20 792 probe sets were regressed against the %ESR1⁺ frequency phenotype data from immunohistochemical analysis across 32 animals (excluding the saline-treated, ovary-intact controls). This analysis revealed 355 genes having expression that was positively correlated with %ESR1⁺, and 376 genes having expression that was negatively correlated with %ESR1⁺ (Table 3; Supplementary Table 20A and 20B) (54). Using DAVID, annotation of these gene sets revealed statistically significant enrichment of genes in the KEGG pathway “DNA replication” (B-H P < .001), while there were also 4 significant terms related to lipid metabolism: “fatty-acid metabolism,” “lipid metabolism,” “lipid biosynthesis,” and “lipid-binding” (B-H P < .05; Supplementary Table 20D) (54). While the top 2 clusters of significant terms are related to mitosis, the third top is based on “mitochondrion” (Supplementary Table 20E) (54). We also excluded the 226 genes that we had previously identified as being E regulated, and performed pathway analysis on the remaining 505 genes (Supplementary Table 20C) (54). The top upstream regulator of this set of genes that correlated with %ESR1 was inhibition by ErbB2 (P < .001; z score = –2.126). DAVID annotation revealed the top KEGG pathways represented by %ESR1-correlated genes (excluding those that were E regulated) were “acetylation” and “phosphoprotein” (B-H P < .001; Supplementary Table 20F) (54), while the top 2 clusters of pathways were based on “mitochondrion” and “lipid/phospholipid biosynthesis” (Supplementary Table 20G) (54).

Expression of Genes Associated with Abundance of Progesterone Receptor

Last, we quantified the frequency of PGR⁺ MECs within the MGs of OVX pigs treated with E, P, and/or PRL for 5 days (Fig. 11A). The frequency of %PGR⁺ MECs within ducts and TDLU structures was the same (P = .45), and there was no effect of MG position within the mammary apparatus (P = .51). The overall frequency of %PGR⁺ MEC ranged from 0.1% to 21% across the treatment groups, where the %PGR⁺ MECs in E-treated or sham-operated females was up to 80-fold higher than in Bromo-treated controls (Fig. 11A and 11B), reflecting a clear positive main effect of E (P < .001). However, similar to the regulation of ESR1, there was a negative main effect both of P (P < .001) and PRL (P < .001) on the frequency of %PGR⁺ MECs. Furthermore, negative interactive responses to E * P (P < .001) and E * PRL (P < .001) led to further downregulation of the frequency of %PGR⁺ MECs by approximately 50%. There was also a negative interactive response to P * PRL (P < .001). Similar to ESR1, when all 3 hormones were administered together, the frequency of %PGR⁺ MECs was greatly reduced relative to the MGs of animals treated with E alone (P < .001). As expected, the level of PGR mRNA expression detected by microarray (one probe set) was regulated by E (P < .001; see Supplementary Table 1) (54). Also, in keeping with expectations from the correlation analysis for ESR1, the values for %PGR⁺ MECs (by immunohistochemistry) and PGR mRNA levels (by microarray) were highly correlated (R² = 0.59; Fig. 11C). In E-treated females the frequency of %ESR1⁺ and %PGR⁺ MECs was also highly correlated (R² = 0.69; P < .001; data not shown).

Figure 11.

Effect of hormone treatments on the abundance of progesterone receptor (PGR)-positive epithelial cells in the mammary glands (MGs) of pigs. Receptor-positive epithelial cells were localized by immunohistochemistry. Pigs were ovariectomized and treated as per Fig. 1 with 17β-estradiol (E), medroxyprogesterone 17-acetate (P), and either 2-bromo-α-ergocryptine (Bromo) or haloperidol to induce hyperprolactinemia (PRL). Control pigs were sham-ovariectomized and injected with saline as per Fig. 1. A, Percentage PGR-positive (%PGR⁺) epithelial cells in control gilts, ovariectomized Bromo-treated gilts, and ovariectomized gilts treated with E, P, and PRL. Data are means ± SEM (n = 4 pigs). ^a,b,c,dDifferent letters indicate statistically significant differences (P < .05). B, Immunohistochemistry for PGR in ducts and terminal ductal lobular units (TDLUs) in MGs. Positive cells were detected using Vector SG (blue) and cells were counterstained using nuclear fast red. Brown patches are melanin that is endemic to this breed. Scale bar equals 0.5 mm. C, Correlation between %PGR⁺ cells and PGR messenger RNA abundance from the microarray analysis. Black squares equal non–E-treated animals. Open circles equal E-treated animals.

Regressing the microarray data for the 20 792 probe sets against the %PGR⁺ MECs (by immunohistochemistry) identified 273 genes having expression that was positively correlated, and 73 genes having expression that was negatively correlated, with %PGR⁺ (see Table 3; Supplementary Table 21) (54). Annotation of these genes using DAVID identified the most statistically significant term to be “secreted” (B-H P < .001; Supplementary Table 21D) (54), and the top 3 clusters of terms were based on “secreted,” “extracellular matrix” (ECM) and “collagen” (Supplementary Table 21E) (54), which were also statistically significant terms in the annotation of E-regulated genes. Given that E is the main regulator of PGR expression, we also excluded E-regulated genes from this analysis, leaving 92 genes (Supplementary Table 21C) (54). Using IPA, the top upstream regulators of this set of 92 non–E-regulated genes that correlated with %PGR⁺ MECs were E (P < .001) and IGF2BP1 (P < .001). Within this gene list, there were no statistically significant annotations, with the highest being “phosphoprotein” (B-H P = .07).

Correlation of Gene Expression With That for Porcine Prolactin Receptor Long Form

Along with the steroid hormone receptors, the transmembrane PRLR is essential for transduction of the signal initiated by PRL (64), where expression of PRLR-LF mRNA in the pig MG is tissue specific and differentially regulated by hormones (65), as well as being cell-type specific (33). Given that the PRLR gene was not represented by probe sets on our S scrofa microarray, we established which genes were coregulated with the PRLR-LF by regressing PRLR-LF mRNA expression (by RT-qPCR) against expression levels of the 20792 probe sets (33). This approach identified 23 genes that correlated with PRLR-LF mRNA expression (FDR < .01; Supplementary Table 22) (54); among these, 9 genes were negatively, and 15 were positively, correlated with PRLR-LF expression, including STAT5B, LACTB and SLA-1 (Supplementary Fig. 5) (54).

Discussion

Epithelial cells within the normal breast undergo a complex yet poorly understood histomorphogenesis, as directed by an equally complex yet also unclear interplay between hormones including E, P, and PRL, all within a unique stromal microenvironment (66). Notably, many aspects of this development in humans are distinct from those in widely studied rodents and are challenging to study at the mechanistic level. Our findings here, using the MG of pigs as a manipulable model for endocrine ablation and replacement, provide broad insights into the transcriptional and endocrine regulation of mammary epithelial growth and TDLU development, particularly in response to unique hormone synergies.

A primary objective of this study was to define the molecular responses during TDLU progression in response to interacting combinations of E, P, and PRL. Among all genes expressed in the MGs of pigs, 20% were regulated by E and PRL, individually and/or together, pointing to a specific role for E + PRL during TDLU development. This finding aligns with our previous demonstration that E + PRL stimulates the greatest extent of epithelial proliferation and differentiation during progression from TDLU-1 to TDLU-3 in the MGs of pigs (8). By contrast, in mice, the combinations of either E + P (9, 67) or PRL + P (17) synergistically stimulate the MG epithelium to proliferate while E + P stimulates MG proliferation and ductal development across a range of species (2). Questions still remain as to whether there are distinct differences underlying the endocrine regulation of TLDU formation in pigs, humans, and ruminants.

Several lines of evidence support a specific mechanistic response by the pig MG to the combination of E + PRL. First, E induces the expression of PRLR mRNA in the MG epithelium (8). Second, in our present data, the combination of E + PRL suppressed ESR1 expression more so than E alone, potentially reflecting a shifted differentiative state of the epithelium. Third, beyond our transcriptomic data, the downstream target of PRLR signaling, STAT5, underwent a synergistic increase in the extent of its phosphorylation in response to E + PRL. Whether E primes the MG epithelium for the subsequent actions of PRL, or whether PRL facilitates the downstream consequences of ESR1 activation, is unclear. Certainly in the case of breast cancer cell lines expressing both PRLR and ESR1, PRL enhanced E-induced proliferation downstream of the ESR1, and induced a greater level of ESR1 phosphorylation than did E (68). Likewise E and PRL cooperatively induce extracellularly regulated kinase 1/2 and c-Fos phosphorylation in breast cancer cells (69). Clearly the question remains as to the extent to which E + PRL cooperatively regulates the course of human breast development, or for that matter MG development in any nonrodent species such as pigs and nonhuman primates (70) having complex TDLU. Our finding that P alone had minimal effects on either development of the MG (8), or its transcriptome, in hormone-ablated pigs aligns with a range of other findings in OVX females, including for mice (71), heifers (72), nonhuman primates (70), and in OVX-hypophysectomized rats (73). This consistent lack of response to P alone across a range of species is almost certainly due to the loss/absence of E-inducible PGR (55), as was observed here. At the same time, the complete temporal response to P within the MGs of these hormone-ablated females remains unclear given that we sampled after only 5 days of exposure.

By associating different hormone-regulated phenotypic characteristics (TB development, TDLU type, proliferation index, and ESR1 and PGR status) with the corresponding transcriptomic profile, we identified gene expression signatures underlying various aspects of hormone-regulated development. A key outcome from this type of approach was the identification of genes having expression levels that had a strong correlation with the rate of epithelial proliferation as measured by the incorporation of BrDU, which is the gold standard for proliferation status. While the function of these genes in MG development has not been defined in detail, their role(s) related to various aspects of mitosis are in keeping with the observed phenotypes. Indeed, others have convincingly validated that immunolocalization of CKAP2 is a reliable mitotic marker for breast cancer (74). Perhaps more notable, however, is the future potential for using the expression level for these genes as covariate measures of cell proliferation in MG samples where only RNA is available, absent any matching tissue for immunohistochemical analysis of cell proliferation markers such as Ki67 or phosphorylated histone H3. While others have used expression of genes such as MKI67 (Ki67) in this way as part of the PAM50 panel for subtyping of human breast cancers, that specific gene/set can lack reliability as a biomarker for identifying patients with a low-risk outcome (75). Indeed, our data indicate that the expression of MK167 alone did not correlate as well with BrDU positivity (only R² = 0.545 across all animals and R² = 0.166 among E-treated animals) as did the expression of 26 of the 29 proliferation-associated genes documented in Table 4.

The broad range of epithelial proliferation induced by the various hormone combinations was also manifest as distinct and associated phenotypic changes such as the formation of TBs and TDLUs. Our association analysis revealed 1210 genes having expression that correlated with TB frequency. In a similar way, Kouros-Mehr and Werb (35) isolated TEBs from the MGs of peripubertal mice then used microarrays to identify their unique gene expression signature. On reannotating their data set (http://www.informatics.jax.org/batch), we identified 1189 genes that were upregulated or downregulated by at least 1.5-fold (B-H P < .05) in TEBs compared to the stroma, and that were upregulated or downregulated at least 1.5-fold in TEBs compared to ducts. Comparing that list to the 1210 genes identified in this study revealed 165 genes common to both analyses (see Supplementary Table 11C) (54), including overrepresentation of 29 significant pathways related to cell cycle (26% of genes) and the nucleus (50% of genes), and other significant UniProtKB key words and GO terms including “phosphoprotein” (73% of genes), “acetylation” (42% of genes), and “Ubl conjugation” (25%), “isopeptide bond” (17%), and “protein binding” (67% of genes) (B-H P < .05; see Supplementary Table 11G) (54). This type of phenome-genome approach can also help explain the mechanisms of hormone action on the MG, particularly by taking into account changes in tissue composition and development. As a case in point, Ppara was among the 165 TB/TEB-associated genes that Deroo et al (76) previously concluded was negatively regulated by E given its downregulation in the MG of OVX mice following 4 weeks treatment with E. Our data indicate that PPARA is not directly regulated by E, but rather its expression declines in association with increasing TB number, in keeping with the findings of Kouros-Mehr and Werb (35), who showed it was downregulated in mouse TEBs compared to the stroma. These findings support an alternative explanation that the reduced expression of this stroma-enriched protein following treatment with E (76) is driven by a change in tissue composition that accompanies ductal elongation and TEB proliferation. Indeed, when we treated OVX mice with physiological doses of E for only 2.5 or 3.5 days, there was no change in the expression of Ppara within the MG (36).

To extend these insights, we also combined our analysis of hormone-dependent responses with the phenotypic regression analyses to identify gene sets having levels of expression that correlated with TB frequency and that were also regulated by E, a known stimulant of TB/TEB development (2). This approach excluded genes associated with TB frequency as the result of a change in glandular composition. A top E-regulated candidate was AREG, for which expression was positively correlated with number of TBs, %PGR⁺, %BrDU⁺, and %TDLU2, and negatively correlated with %TDLU1. In mice, AREG is a well-recognized E-inducible growth factor expressed in TEBs that stimulates MEC proliferation (77). Others have demonstrated that expression of Areg and the development of TEBs in mice is also induced by P, which occurs in the presence of endogenous PRL (25). By contrast, Areg mRNA abundance was unchanged in the MGs of OVX pigs treated with P, raising the question of which hormone(s) regulates its expression and/or the extent of its function during TDLU development in the human breast. Another biologically relevant E-induced gene from our analysis was OXTR, which is expressed by myoepithelial cells (78), and is also E-inducible in MCF-7 cells, rat uterus and hypothalamus (79), mouse MG (36), although interestingly, not rat MG (80). Notably, myoepithelial cells in the MGs of mice arise from pluripotent cap cells at the leading edge of advancing TEBs (2), although their origins and distribution in the MGs of pigs (11) or in the elongating ducts in the developing human breast (81) remain poorly defined. Here the expression of OXTR was positively correlated with number of TBs, %PGR⁺, %BrDU⁺, %TDLU2, and negatively correlated with %TDLU1. Of note, oxytocin stimulates myoepithelial cell proliferation and differentiation in the MG of mice (82), where our data suggest a link between OXTR expression and E-regulated development of TBs and the progression of TDLU-2 structures.

Comparative transcriptomics across species can also reveal conserved mechanisms underlying E-induced growth in the normal MG. Among the 1760 E-regulated genes in pigs (see Supplementary Table 1A) (54), 278 were also regulated by E in the MGs of OVX mice following E-treatment for 2.5 or 3.5 days (36), with 11 of these also being altered by a main effect of E in the MGs of heifers (83). The E-regulated genes common to all 3 studies were CKAP4, GREB1, PGR, STC1, and TIPARP (see Supplementary Table 1D) (54). The identification of PGR in this list was not surprising given its critical role in the MG and its regulation by E (55), nor was the identification of GREB1 given that its expression is E regulated and is associated with ESR expression in breast cancer (84, 85). CKAP4 is a microtubule-binding protein that maintains the structure of the endoplasmic reticulum (86) and acts as a ligand-activated receptor for tissue plasminogen activator, antiproliferative factor, and surfactant protein (87). The induction of CKAP4 expression by E in the MGs of pigs, mice, and heifers (36, 83) and the correlation between its expression and epithelial proliferation and the development of TB, TDLU-1, and TDLU-2 (Supplementary Table 16) (54), similar to CKAP2, lends strong support to it having a crucial role during hormone-regulated MG development. The expression of STC1 has similarly been associated with ESR expression in breast cancer cells (88). However, STC1 expression in the MGs of pigs herein was inhibited by E, whereas its expression was stimulated by E in OVX heifers (83), lactating cows (89), and OVX mice (36). In the same way, TIPARP was upregulated by E in the MGs of pigs and heifers (83) and in MCF-7 cells (90), but was downregulated by E in the MGs of OVX mice (36). Whether these discordant responses for STC1 and TIPARP across species reflect details such as study design, tissue sampling, and/or treatments, or whether there are clear species-specific factors underlying their regulation, remains to be determined. Indeed, similar across-species differences may explain the expression of HAPLN1, which had the greatest induction by E in the MGs of pigs (see Supplementary Table 1A) (54), and was also interactively regulated by E + P (see Supplementary Table 4) (54). By contrast, Hapln1 expression in the MGs of mice was not regulated by E (36), nor did its expression differ between the TEBs and stroma in this species (35). Likewise, the expression of HAPLN1 was not regulated by E in the MGs of OVX or ovary-intact heifers (83). HAPLN1 is an integral component of the ECM both in neurons (91) and cartilage, where it binds and stabilizes aggregates of aggrecan and hyaluronic acid (92) and functions as a growth factor to upregulate the synthesis of aggrecan and type II collagen (93). The expression of HAPLN1 is also associated with cell survival, breast tumor progression, and stem/progenitor cell formation (94). Given similarities between the pig MG and the human breast, our data suggest there may be a candidate role for E-induced HAPLN1 during breast development, especially during ECM remodeling.

Of the 14 genes having expression that associated positively with measures of MEC proliferation, TB frequency, and TDLU advancement, 4 genes (KPNA2, LMNB2, MAGED1, and MTHFD2) have not been implicated in regulating mitosis. That said, KPNA2 is described as an oncogene in breast cancer (95), and MAGED1 inhibits proliferation of breast cells (96), likely via its interaction with BRCA2 (97). Likewise, LMNB2 expression is elevated in metastatic breast cancer from tamoxifen-resistant tumors (98), and expression of the metabolic enzyme MTHFD2 is associated with a poor prognosis in breast cancer (99, 100), where it is upregulated during cancer cell proliferation, promotes proliferation, and localizes with newly synthesized DNA (101). The only gene that was negatively associated with proliferation, TB, and TDLU advancement was tetraspanin 7 (TSPAN7), which is X-linked both in humans and pigs. TSPAN7 inhibits myeloma tumor development in vivo (102), and its expression is associated with a favorable outcome for patients with renal cancer (103). Another member of the tetraspanin or transmembrane 4 superfamily, CD151 (TSPAN24), interacts with laminin-binding integrins and inhibits tertiary branching and TEB formation in the MGs of mice (104). These findings support the potential that various TSPANs participate in the hormone-regulated transition of the MG from a proliferative to a differentiated state, as occurs during gestation.

ESR1 and PGR both are critical during MG development and are important prognostic indicators for breast cancers (105), although the physiological regulation of their levels and distribution during the entire course of normal breast development lacks definition. The incidence of ESR1 in the MGs of pigs changed markedly across the range of hormone treatment combinations, where ESR1 frequency was suppressed by E, P, PRL, and their interactions, leading to the near absence of ESR1 in females administered E + P + PRL, even though there was a correspondingly high rate of MEC proliferation. Similarly, whereas E increased %PGR⁺ an average of 20-fold, the frequency of PGR⁺ MECs was reduced by P or PRL and even more by their combination. These responses represent an entire range of physiologically relevant MG phenotypes, spanning from a state where ESR1 and PGR both were elevated in the presence of E alone, through to a state of increased MEC proliferation and differentiation alongside a declining abundance of ESR1 and PGR. These phenotypes highlight that E-induced MEC proliferation can occur independently of ESR1 abundance. The clear inverse relationship between proliferation and ESR1/PGR expression is also consistent with various findings, including that P inhibited the expression of PGR in normal human MECs (106), our finding of a pronounced correlation between the %PGR+ MECs and the %ESR1+ MECs, and the demonstration that steroid receptor–positive cells generally do not proliferate (107). Given that a strong inverse relationship between proliferation and ESR1/PGR expression has, to our knowledge, never been described for the MGs of any species, these data present a unique revelation about how proliferation in normal MECs corresponds to the regulation and potential function(s) of ovarian steroid hormones and their receptors.

An intriguing revelation from the IPA was that E, P, and PGR were all identified as upstream regulators of PRL-responsive gene expression in the MG. A likely explanation here is that some functions that are typically attributed to the actions of E and P may actually reflect a response to PRL, given that few studies experimentally suppress/eliminate this hormone from the endocrine background, as we did here by using bromocriptine. For example, 67 of the PRL-regulated genes were, according to the IPA, regulated by E. In fact, only 12 of these were regulated by E in the MG (ABLIM1, AR, CD55, COL4A5, FHL1, GARS, GPD2, PAQR7, PRSS35, RANBP1, SLC7A8, and TMEM258), a discrepancy that likely reflects the well-known induction of PRL synthesis and secretion in vivo in response to exogenous E (108). This explanation would account for the incorrect assignment of many genes to a so-called E-responsive pathway. Another example of how IPA is limiting during these transcriptomic analyses comes from our data for transcobalamin (TCN1, the B12 binding protein), which is secreted by the MGs of mice (109), pigs (this study), and lactating cows (110). To date, the only literature describing the hormonal regulation of TCN1 indicates its expression by osteosarcoma cells is upregulated by E (111). Instead, our data indicate that TCN1 expression is clearly upregulated by PRL, not E, where its expression was negatively correlated with ESR1 expression, and positively correlated with TDLU-3 development.

Combined, our data provide important insights to how the growth and histomorphogenesis of the developing porcine MG, which mirrors many aspects of human breast development, responds to various combinations of critical mammogenic hormones. The range of phenotypic outcomes within the MG can be associated with unique patterns of gene expression. In particular, our finding that TDLU progression reflects a pronounced synergistic response to E and PRL, alongside a clear inverse relationship between MEC proliferation and ESR1/PGR expression, stands to inform a better understanding of the dynamic changes that occur in response to hormone action on the normal human breast that, in turn, can affect breast cancer risk.

Glossary

Abbreviations

ANOVA

analysis of variance

ATP

adenosine 5′-triphosphate

B-H

Benjamini-Hochberg

BrDU

5-bromo-2-deoxyuridine

Bromo

2-bromo-α-ergocryptine methanesulfonate salt

cDNA

complementary DNA

E

estrogen/17β-estradiol

ECM

extracellular matrix

ESR1

estrogen receptor α

FBS

fetal bovine serum

FDR

false discovery rate

GO

Gene Ontology

IPA

Ingenuity Pathway Analysis

MEC

mammary epithelial cell

MG

mammary gland

mRNA

messenger RNA

OVX

ovariectomized

P

progesterone

PBS

phosphate-buffered saline

PGR

progesterone receptor

PRL

prolactin

pSTAT5

phosphorylated STAT5

qPCR

quantitative polymerase chain reaction

RMA

robust multichip average

RT

reverse transcription

TB

terminal bud

TDLU

terminal ductal lobular unit

TEB

terminal end bud

Financial Support

This work was supported by the United States Department of Agriculture (USDA) National Research Initiative (competitive grant Nos. 2004-35206-14140, 2003-35206-12849, and 2015-06351); the Vermont Genetics Network, the Department of Defense Breast Cancer Research Program (grant No. W81XWH-04-1-0522); a core grant to the Vermont Cancer Center; the UC Davis Genome Center; the University of Vermont Agricultural Experiment Station; and the University of California, Davis Agricultural Experiment Station.

Disclosures

The authors have nothing to disclose.

Data Availability

All data generated or analyzed during this study are included in this published article or in the data repositories listed in “References.”