Mammary-Enriched Transcription Factors Synergize to Activate the Wap Super-Enhancer for Mammary Gland Development

Super-enhancers are large clusters of enhancers critical for cell-type-specific development. In a previous study, 440 mammary-specific super-enhancers, highly enriched for an active enhancer mark H3K27ac; a mediator MED1; and the mammary-enriched transcription factors ELF5, NFIB, STAT5A, and GR, were identified in the genome of the mammary epithelium of lactating mice. However, the triggering mechanism for mammary-specific super-enhancers and the molecular interactions between key transcription factors have not been clearly elucidated. In this study, we investigated in vivo protein–protein interactions between major transcription factors that activate mammary-specific super-enhancers. In mammary epithelial cells, ELF5 strongly interacted with NFIB while weakly interacting with STAT5A, and it showed modest interactions with MED1 and GR, a pattern unlike that in non-mammary cells. We further investigated the role of key transcription factors in the initial activation of the mammary-specific Wap super-enhancer, using CRISPR-Cas9 genome editing to introduce single or combined mutations at transcription factor binding sites in the pioneer enhancer of the Wap super-enhancer in mice. ELF5 and STAT5A played key roles in igniting Wap super-enhancer activity, but an intact transcription factor complex was required for the full function of the super-enhancer. Our study demonstrates that mammary-enriched transcription factors within a protein complex interact with different intensities and synergize to activate the Wap super-enhancer. These findings provide an important framework for understanding the regulation of cell-type-specific development.


Introduction
Mammary gland tissue undergoes a cycle of development and involution processes that occur in sequential stages, starting with the establishment of the ductal tree during puberty and proceeding through the formation of alveoli during pregnancy, lactation, and tissue remodeling upon weaning [1,2]. Mammary gland epithelium, which plays a key role in temporal regulatory dynamics during development, is composed of basal myoepithelial cells and milk-secreting luminal alveolar cells. The lactating alveolar epithelium is formed during pregnancy and becomes remodeled when lactation is stopped. Differentiation of the luminal lineage is primarily regulated by the cytokine prolactin (PRL), which is secreted during pregnancy and lactation [3,4] and leads to the activation of a number of mammary-specific genes [5]. A previous study reported that mammary-specific genes are regulated by 440 super-enhancers that are activated in the lactating mammary gland [6]. A super-enhancer is a cluster of enhancers spanning a long region of genomic loci that is associated with genes involved in cell identity [7][8][9]. Mutations in super-enhancers are also closely associated with a variety of diseases, including cancer, diabetes, and autoimmune diseases [7,[10][11][12][13][14]. Integrated chromosome immunoprecipitation sequencing (ChIP-seq) analyses have demonstrated that mammary-specific super-enhancers are cooccupied by histone H3 acetylation at lysine 27 (H3K27ac), a histone mark characteristic of transcriptionally active enhancers, as well as several mammary-enriched transcription Previous ChIP-seq analyses have revealed that a super-enhancer associated with the Wap gene consists of three constituent enhancers: the proximal E1 enhancer (−0.7 kb from the transcription start site (TSS)), the distal E2 enhancer (−1.4 kb from the TSS), and the E3 enhancer (−5.6 kb from the TSS) (Figure 1a) [6]. Like other super-enhancers, the Wap superenhancer is sequestered by the insulator CTCF, and the promoter of the Wap gene is highly enriched with H3K4me3 histone marks [25,26]. Each constituent enhancer is co-occupied by transcription factors essential for mammary gland development (STAT5, GR, NFIB, and ELF5), the mediator MED1, and transcription-activating H3K27ac enhancer marks. To investigate whether co-occupancy of each transcription factor within the Wap superenhancer reflects independent chromatin binding of individual factors or is attributable to interactions with other transcription factors, we first determined the cellular localization of each transcription factor in mammary epithelial cells. To this end, murine full-length ELF5, NFIB, GR, STAT5A, and MED1 were tagged with green fluorescent protein (GFP) and their cellular localizations in both mammary epithelial HC11 cells and non-mammary NIH3T3 cells were visualized by fluorescence microscopy (Figure 1b, Supplementary Figure S1). ELF5 and NFIB were exclusively localized to the nucleus, whereas MED1 was localized to both the nucleus and cytoplasm. GR was mostly limited to the cytoplasm but gained entrance to the nucleus after stimulation with hydrocortisone (HC). STAT5A was localized to both the nucleus and cytoplasm but translocated to the nucleus following stimulation with prolactin (PRL). These data indicate that the transcription factors constituting the Wap super-enhancer can independently enter the nucleus in both mammary epithelial and non-mammary cells. super-enhancer reflects independent chromatin binding of individual factors or is attributable to interactions with other transcription factors, we first determined the cellular localization of each transcription factor in mammary epithelial cells. To this end, murine full-length ELF5, NFIB, GR, STAT5A, and MED1 were tagged with green fluorescent protein (GFP) and their cellular localizations in both mammary epithelial HC11 cells and nonmammary NIH3T3 cells were visualized by fluorescence microscopy (Figure 1b, Supplementary Figure S1). ELF5 and NFIB were exclusively localized to the nucleus, whereas MED1 was localized to both the nucleus and cytoplasm. GR was mostly limited to the cytoplasm but gained entrance to the nucleus after stimulation with hydrocortisone (HC). STAT5A was localized to both the nucleus and cytoplasm but translocated to the nucleus following stimulation with prolactin (PRL). These data indicate that the transcription factors constituting the Wap super-enhancer can independently enter the nucleus in both mammary epithelial and non-mammary cells.  Representative fluorescence images from three independent experiments are shown (800×). After serum starvation, cells were left untreated or were treated with hydrocortisone (HC) or prolactin (PRL) to activate GR or STAT5A, respectively. Scale bars, 10 µM.

Protein Interactions between Mammary-Enriched Transcription Factors in Mammary Epithelial Cells Differ from Those in Non-Mammary Cells
A previous study using mutant mice revealed that ELF5 plays an important role in activating the Wap super-enhancer [6]. Mutating a single STAT5 binding site or both STAT5 and NFIB binding sites in the E1 enhancer was not sufficient to disrupt Wap superenhancer activity, but mutating all three STAT5, NFIB, and ELF5 binding sites completely disabled super-enhancer function. To determine whether ELF5 can interact with the four other transcription factors-STAT5A, NFIB, GR, and MED1-constituting the Wap superenhancer, we investigated in vivo protein-protein interactions in mammary epithelial cells using the BiFC assay, which is designed to assess interactions between two fluorescencetagged proteins in vivo [19][20][21]. Briefly, when the target protein linked to the N-terminal domain of a fluorescent protein and the counterpart molecule linked to the C-terminal domain of the same fluorescent protein come sufficiently close to each other to molecularly interact, the N-and C-terminal domains of distinct fluorescent proteins bind to each other to form a complex, subsequently emitting intracellular fluorescence. To apply this strategy, we linked the N-terminal domain of the Venus fluorescent tag (VN) to ELF5 and the C-terminal domain of the Venus fluorescent tag (VC) to STAT5A, GR, NFIB, or MED1 ( Figure 2a). Prior to BiFC assays, we confirmed that each fusion protein was properly expressed in cells (Figure 2b). We next investigated protein-protein interactions between ELF5 and the other four transcription factors, STAT5A, GR, NFIB, and MED1, in mammary epithelial HC11 cells and non-mammary NIH3T3 cells. Since GR and MED1 are known to interact with each other [27], we used the GR-VN and MED1-VC pair as a positive control for the BiFC assay. As expected, the positive control (GR-VN and MED1-VC pair) emitted strong Venus fluorescence in both HC11 and NIH3T3 cells, whereas the negative control (JUN-VN and ELF5-VC pair) did not emit a fluorescence signal in either cell type (Supplementary Figure S2). Although ELF5 paired with each of the four other transcription factors induced fluorescence in both HC11 and NIH3T3 cells (Figure 2c,d), the patterns of molecular interactions in mammary epithelial cells and non-mammary cells were completely different (Figure 2e,f). In HC11 cells, the percentage of BiFC-positive cells for protein pairs followed the rank order ELF5-NFIB, ELF5-MED1, ELF5-GR, and ELF5-STAT5A (Figure 2e), indicating that ELF5 can interact with all four transcription factors, but with slightly different affinities. Non-mammary NIH3T3 cells showed different BiFC patterns compared with mammary epithelial cells, with ELF5 exhibiting the strongest interaction with MED1 and the weakest interaction with GR ( Figure 2f). Collectively, these data suggest that the pattern of molecular interactions within a mammary-enriched transcription factor complex differs between mammary epithelial cells and non-mammary cells, a difference that may form the basis of the unique features of mammary-specific super-enhancers.
To further confirm the molecular interactions of ELF5 with NFIB, MED1, GR, and STAT5A, we performed in situ PLAs in the mammary epithelium of lactating mice. PLAs, which integrate immunoassay and fluorescence microscopy techniques, are useful for investigating protein-protein interactions in unmodified cells or tissue [22][23][24]. In application, the target protein and its counterpart in tissues are immunoprecipitated with the corresponding primary antibodies and PLA probe-linked secondary antibodies. If two proteins come sufficiently close to molecularly interact, a specific DNA strand attached to the PLA probe is capable of initiating rolling circle DNA synthesis. The DNA circle is subsequently amplified several hundred fold and binds to a fluorescently labeled complementary oligonucleotide probe, achieving a high concentration of fluorescence that can be visualized by fluorescence microscopy. Before conducting PLA assays, we used quantitative reverse transcription polymerase chain reaction (RT-qPCR) to confirm the endogenous expression of Elf5, Stat5a, Gr, Nfib, and Med1 in mouse mammary epithelium obtained on day 1 of lactation (L1) (Figure 3a). Immunofluorescence assays also revealed the expression of ELF5, STAT5A, GR, NFIB, and MED1 proteins in lactating mouse mammary epithelium ( Figure 3b). Consistent with BiFC results, ELF5 showed stronger interactions with NFIB than with phosphorylated STAT5A (pSTAT5A) in mouse epithelium ( Figure 3c). The GR and MED1 positive control pair exhibited strong PLA signals, whereas untreated cells showed no fluorescence signal. The relative percentage of PLA signals per cell was higher for the ELF5 and NFIB pair than for the ELF5 and pSTAT5A pair (Figure 3d). Taken together, in vivo BiFC assays and in situ PLAs demonstrated that ELF5 can form a protein complex with NFIB, GR, MED1, and pSTAT5A, but with different affinities. ELF5 interacts more closely with NFIB than with STAT5A in mammary epithelium, whereas ELF5 is more closely associated with MED1 in non-mammary cells. 3c). The GR and MED1 positive control pair exhibited strong PLA signals, whereas untreated cells showed no fluorescence signal. The relative percentage of PLA signals per cell was higher for the ELF5 and NFIB pair than for the ELF5 and pSTAT5A pair ( Figure  3d). Taken together, in vivo BiFC assays and in situ PLAs demonstrated that ELF5 can form a protein complex with NFIB, GR, MED1, and pSTAT5A, but with different affinities. ELF5 interacts more closely with NFIB than with STAT5A in mammary epithelium, whereas ELF5 is more closely associated with MED1 in non-mammary cells.

Mammary-Enriched Transcription Factors Synergize to Activate the Wap Super-Enhancer for Mammary Gland Differentiation
A previous study using CRISPR-Cas9-targeted mice revealed that mutations of STAT5, NFIB, and ELF5 binding sites in the E1 enhancer decommission the Wap super-enhancer [6]. However, whether all three transcription factors are required for the establishment of a super-enhancer complex or whether ELF5 plays a pioneering role in initiating the activation of a super-enhancer has not been clearly elucidated. To address this issue, we established CRISPR-Cas9-targeted mice carrying the following deletion mutations in the E1 enhancer of the Wap super-enhancer: deletion of a single ELF5 binding site (∆E1A); combined deletion of ELF5 and STAT5 binding sites (∆E1B); combined deletion of STAT5 and NFIB binding sites (∆E1C); and combined deletion of ELF5, STAT5, and NFIB binding sites (∆E1D) (Figure 4a and Supplementary Figure S3). We first assessed Wap gene expression levels in these E1 mutant mice (Figure 4b). Consistent with previous results [6], deletion of all three ELF5, NFIB, and STAT5 binding sites (∆E1D) completely silenced Wap gene expression in mice. Although expression levels of the Wap gene in mice with a single ELF5 binding site deletion (∆E1A) were comparable to those in wild-type mice, deletion of ELF5 and STAT5 binding sites (∆E1B) reduced Wap gene expression levels by 99.9% compared with wild-type controls, and deletion of STAT5 and NFIB binding sites (∆E1C) decreased Wap gene expression levels by 88% compared with wild-type mice. Immunofluorescence analysis also confirmed that WAP proteins were rarely expressed in mammary alveolar cells of ∆E1B and ∆E1D mice at L1 (Figure 4c). WAP proteins were abundantly expressed in lactating mammary alveoli in ∆E1A mice and modestly expressed in ∆E1C mice. Expression levels of Csn2, a gene encoding another milk protein, were not affected by mutations at E1 enhancers in mice (Supplementary Figure S4). Collectively, our findings indicate that mutation of a single transcription factor binding site at E1 is not sufficient to disable the Wap super-enhancer function. Instead, the bindings of all three transcription factors-ELF5, STAT5, and NFIB-at the E1 enhancer synergize to constitute the full function of the Wap super-enhancer.

Protein-Protein Interactions within a Mammary-Enriched Transcription Factor Complex Are Essential for Activation of Mammary-Specific Wap Super-Enhancers
We further investigated whether the limited Wap gene expression in ΔE1B and ΔE1D mice was attributable to the inactivation of all three constituent Wap super-enhancers, E1, E2, and E3. To this end, we performed ChIP assays to detect modifications in H3K27ac, a histone mark typically enriched at active enhancer regions. Consistent with observed Wap gene expression levels, H3K27ac levels at the E1 enhancer of ΔE1A mice were comparable to those of wild-type controls and were reduced by ~40% in ΔE1C mice (Figure 5a). On the other hand, H3K27ac levels at the E1 enhancer were reduced by ~70% in ΔE1B mice and by ~90% in ΔE1D mice. H3K27ac signals on E2 and E3 enhancers revealed that E2 and E3 were modestly activated in ΔE1A mice, whereas a dysfunctional E1 enhancer suppressed the activation of distal E2 and E3 enhancers in ΔE1B, ΔE1C, and ΔE1D mice.
In addition to investigating the molecular interactions between ELF5 and the transcription factors NFIB, GR, MED1, and pSTAT5A, we also evaluated protein interactions between each transcription factor pair to further estimate the molecular proximity within the protein complex recruited to the Wap super-enhancer (Figure 5b-d). NFIB interacted with ELF5, pSTAT5A, and MED1 to a similar degree and showed slightly weaker interactions with GR ( Figure 5e); MED1 showed the strongest binding to GR and the weakest

Protein-Protein Interactions within a Mammary-Enriched Transcription Factor Complex Are Essential for Activation of Mammary-Specific Wap Super-Enhancers
We further investigated whether the limited Wap gene expression in ∆E1B and ∆E1D mice was attributable to the inactivation of all three constituent Wap super-enhancers, E1, E2, and E3. To this end, we performed ChIP assays to detect modifications in H3K27ac, a histone mark typically enriched at active enhancer regions. Consistent with observed Wap gene expression levels, H3K27ac levels at the E1 enhancer of ∆E1A mice were comparable to those of wild-type controls and were reduced by~40% in ∆E1C mice (Figure 5a). On the other hand, H3K27ac levels at the E1 enhancer were reduced by~70% in ∆E1B mice and by~90% in ∆E1D mice. H3K27ac signals on E2 and E3 enhancers revealed that E2 and E3 were modestly activated in ∆E1A mice, whereas a dysfunctional E1 enhancer suppressed the activation of distal E2 and E3 enhancers in ∆E1B, ∆E1C, and ∆E1D mice.  In addition to investigating the molecular interactions between ELF5 and the transcription factors NFIB, GR, MED1, and pSTAT5A, we also evaluated protein interactions between each transcription factor pair to further estimate the molecular proximity within the protein complex recruited to the Wap super-enhancer (Figure 5b-d). NFIB interacted with ELF5, pSTAT5A, and MED1 to a similar degree and showed slightly weaker interactions with GR ( Figure 5e); MED1 showed the strongest binding to GR and the weakest interaction with pSTAT5A ( Figure 5f); and GR bound most strongly to MED1 and least strongly to NFIB and ELF5 (Figure 5g). These data indicate that all five transcription factors (ELF5, NFIB, GR, MED1, and pSTAT5A) can interact with each other, but their differential binding intensities suggest differences in molecular proximity within the protein complex. BiFC results shown in Figures 3 and 5 suggest that MED1 mediates protein interactions between the ELF5-NFIB complex and the GR-pSTAT5A complex.

Discussion
Super-enhancers were first discovered in embryonic stem cells about a decade ago [7,8]. Since then, super-enhancers have been identified in a variety of cell types, including mammary epithelial cells, immune cells, chondrocytes, and hair follicle cells [6,13,28,29]. Accumulating evidence also shows that abnormal forms of super-enhancers are causally linked to various diseases [7,[11][12][13]. However, the detailed molecular interactions within a transcription factor complex recruited on super-enhancers have not been elucidated. Although several super-enhancer studies have identified the pioneering factor that establishes superenhancers in specific cell types [29], the initial activation mechanism of mammary-specific super-enhancers has not been investigated. Here, we used both in vivo BiFC assays and in situ PLAs to unravel the protein interactions between mammary-enriched transcription factors that constitute mammary-specific super-enhancers. Our results clearly revealed that the pattern of molecular interactions of these transcription factors differs between mammary epithelial cells and non-mammary cells, underscoring the unique characteristics of cell-type-specific super-enhancers. We further created several CRISPR-Cas9-targeted mice for use in identifying key transcription factors essential for the initial activation of mammary-specific Wap super-enhancers. By disrupting binding sites for key transcription factors at the E1 pioneering enhancer, we demonstrated that the mammary-specific Wap super-enhancer is activated by synergy among key transcription factors co-localized at the pioneering enhancer rather than by a single molecule (Supplementary Figure S5). The present study also revealed that disruption of all three binding motifs, ELF5, NFIB, and STAT5, in the E1 enhancer markedly reduced the acetylation of H3K27 in the distal E2 and E3 enhancers, whereas deletion of the ELF5 and STAT5 binding motifs from the E1 enhancer preferentially inhibited the activity of the E2 enhancer. Deletion of STAT5 and NFIB from the E1 enhancer modestly reduced activation of the E2 and E3 enhancers, whereas deletion of a single STAT5 motif had minimal effect on these distal enhancers. Additional studies are required to determine whether these differential effects are caused by chromatin looping between the proximal E1 enhancer and the distal E2 and E3 enhancers.
Several studies have shown that super-enhancers are typically bound by lineagespecific transcription factors, and a single pioneering factor is essential for their activation [8,29]. Oct-4 and Sox9 play a pioneering role in the activation of super-enhancers in embryonic stem cells and hair follicle cells, respectively. Unlike these super-enhancers, the mammary-specific Wap super-enhancer appears to be activated by cooperation among the key transcription factors ELF5, STAT5, NFIB, GR, and MED1, which are essential for mammary gland development. Therefore, the cell-type-specific molecular interactions between major transcription factors might be critical for the unique properties of mammary-specific super-enhancers. Our results also suggest that ELF5 and STAT5 play a critical role in the pioneering E1 enhancer of the mammary-specific Wap super-enhancer. Although we could estimate the molecular proximity between transcription factors that constitute mammaryspecific super-enhancers using in vivo BiFC assays and in situ PLAs, these techniques may not be sufficient to investigate the molecular interplay within a native protein complex bound to a super-enhancer. A future study using ChIP-mass spectrophotometry may be necessary to accurately determine the molecular interactions in intact super-enhancer complexes [30]. Recent studies have also reported evidence for the formation of phase-separated condensates-microscopically detectable transcriptional regulatory complexes [31,32]-in super-enhancer regions. An intriguing question is whether such phase-separated condensates can also be found in mammary epithelial cells. Because mammary gland epithelium undergoes repeated alveolar formation and degeneration, it will be an ideal model for understanding super-enhancer assembly and disassembly processes, which would explain the reversible feature of super-enhancers. Moreover, these findings and experimental approaches may be useful for studying the molecular mechanisms of super-enhancer activity in different breast cancer subtypes. ChIP-seq analyses found that super-enhancers in estrogen receptor (ER)-positive breast cancer cells co-localize with ELF5, FOXA1, and ER [33]. In contrast, super-enhancers found in triple-negative breast cancer (ER − , PR − , HER2 − ) co-localized with the transcription factors FOXC1, MET, MYC, and ANLN [34,35]. It will be interesting to determine whether mammary-enriched ELF5, STAT5, NFIB, and GR are also recruited in these super-enhancers and whether differential molecular recruitment to super-enhancers determines the different breast cancer subtypes. Our findings offer basic insight into the regulatory mechanisms underlying mammary-specific developmental processes and further suggest how this knowledge might be applied to cell-type-specific regeneration and treatment of abnormal super-enhancer-driven diseases.

ChIP-Seq Data
All ChIP-seq data were obtained from datasets deposited in the Gene Expression Omnibus (GEO) under accession number GSE74826 [6] and were visualized with an Integrated Genomics Viewer (IGV version 2.8.9., Broad Institute and the Reagents of the University of California, CA, USA).

Cell Culture
The mouse mammary epithelial HC11 cell line was purchased from the American Type Culture Collection (ATCC, Manassas, VA, USA), and the mouse embryonic fibroblast NIH3T3 cell line was kindly provided by Dr. Young Bong Kim (Konkuk University, Seoul, Korea). HC11 cells were maintained in Roswell Park Memorial Institute 1640 medium (RPMI 1640; Capricorn Scientific, Ebsdorfergrund, Germany) supplemented with 10% fetal bovine serum (FBS; Merck, Kenilworth, NJ, USA) and 1% penicillin/streptomycin (Capricorn Scientific). NIH3T3 cells were cultured in Dulbecco's Modified Eagle's Medium (DMEM; Capricorn Scientific) supplemented with 10% FBS and 1% penicillin/streptomycin. Both cell lines were grown in a humidified incubator at 37 • C under 5% CO 2 conditions.
For Western blot analysis, proteins were transferred to a nitrocellulose membrane and then incubated first with an anti-GFP primary antibody (#ab6556; Abcam, Cambridge, UK) and then with a horseradish peroxidase (HRP)-conjugated anti-mouse IgG (#A10551; Invitrogen, Waltham, MA, USA) or HRP-conjugated anti-rabbit IgG (#G21234; Invitrogen) secondary antibody as appropriate. Immunoreactive proteins were detected using an enhanced chemiluminescence system.

Bimolecular Fluorescence Complementation (BiFC) Assay
Cells were plated on glass coverslips in 12-well plates containing growth medium 1 day before transfection. When cells reached~70% confluence, they were transfected with BiFC constructs using the Lipofectamine 3000 reagent (Invitrogen) according to the manufacturer's instructions. After 24 h of serum starvation, cells were treated with prolactin (PRL) or hydrocortisone (HC) for 1 h to activate STAT5 or GR, respectively. Cells were washed with phosphate-buffered saline (PBS) and fixed with ice-cold 4% paraformaldehyde in PBS for 15 min. Fixed cells were mounted on glass slides with ProLong Gold Antifade Mountant containing DAPI (Invitrogen). GFP-and Venus-tagged proteins were visualized with a Zeiss LSM 800 confocal microscope using a 40× oil-immersion objective (Zeiss, Oberkochen, Germany). GFP and Venus were excited at 509 nm using an argon laser, and emission was collected using a 400-650 nm filter. DAPI was excited at 465 nm using an argon laser, and emission was collected using a 400-495 nm filter. Images were captured using the ZEN image browser (ZEN blue version 2.6; Zeiss).

Mice
Eight-week-old female C57BL/6 mice were purchased from Orient Bio (Seongnam, Korea) and used as wild-type controls. Founder CRISPR-Cas9-targeted mice were generated by Toolgen (Seoul, Korea). Single-guide RNAs (sgRNAs) were designed to specifically target the ELF5 motif or the region spanning the GAS to ELF5 motif at E1 of the Wap superenhancer (Supplementary Table S1). All mice were maintained under specific-pathogen-free conditions with a 12 h light/dark cycle at 20-25 • C (room temperature) and 30-70% relative humidity. All animal procedures were approved by the Institutional Animal Care and Use Committee (IACUC) of Konkuk University (Seoul, Korea).

Generation of Homozygous Mice and Genotyping
Founder (F0) CRISPR-Cas9-targeted mice were bred with C57BL/6 wild-type mice to segregate the mosaicism, resulting in the generation of F1 heterozygous mice. F2 homozygous mice were generated by interbreeding F1 mice. All mice were genotyped by PCR amplification of genomic DNA isolated from mouse tail snips, followed by Sanger sequencing. Details of the PCR primers and sequencing primers are presented in Supplementary  Table S2. The four lines carrying deletions in transcription factor binding sites at E1 (0.7 kb upstream of the Wap TSS) were named ∆E1A (9-bp deletion), ∆E1B (49-bp deletion), ∆E1C (26-bp deletion), and ∆E1D (97-bp deletion). The specific sequences deleted in each mouse are provided in Supplementary Figure S3

In Situ Proximity Ligation Assay (PLA)
PLA was performed with the Duolink detection kit (#DUO92103; Merck, Darmstadt, Germany). The procedure from tissue harvesting to antigen retrieval was the same as that for conventional immunochemistry, described above. Following antigen retrieval, slides were blocked by incubating at 37 • C for 30 min and then were incubated with primary antibody overnight at 4 • C. The primary antibodies used in PLA were the same as those used for histology. After washing, tissues were incubated with Plus and Minus Duolink probes, ligated, and subjected to rolling circle amplification according to the manufacturer's protocol. Samples were mounted using ProLong Gold Antifade Mountant with DAPI (Invitrogen). PLA signals were visualized with a Zeiss LSM 800 confocal microscope and analyzed using the ZEN image browser. PLA signals were detected using the Cy3 filter (excitation, 561 nm; emission, 535-700 nm), and nuclei were visualized with the DAPI filter (excitation, 465; emission, 400-600 nm). PLA signals were counted in cell nuclei in five different fields per slide from three independent replicates.

Quantitative (RT-qPCR) and Conventional RT-PCR
RNA was extracted using the PureLink RNA Mini Kit (Invitrogen) according to the manufacturer's instructions. cDNA was synthesized from total RNA by reverse transcription using SuperScript III Reverse Transcriptase (Invitrogen). qPCR was then performed on a LightCycler 96 Instrument (#05815916001; Roche, Basel, Switzerland) using the SsoAdvanced Universal SYBR Green Supermix (BioRad, Hercules, CA, USA) and gene-specific primers. Sequences of primers used for RT-qPCR are listed in Supplementary Table S3. Wap and Csn2 mRNA levels were normalized to Gapdh levels. Conventional RT-PCR was carried out using specific primers for Elf5, Stat5a, Gr, Nfib, Med1, or Gapdh (internal control). Sequences of primers used for conventional RT-PCR are provided in Supplementary Table S4.