Introduction

One of the risk factors for breast cancer is an increased number of menstrual cycles, especially for every year younger at menarche [1]. The molecular basis for this remains unknown, but it likely reflects the accumulation of mutations due to a higher total number of (stem) cell divisions that underlie recurrent changes in tissue organization. The mouse provides a good model system to study the dynamics of mammary gland development and homeostasis as, similar to the human breast, it undergoes major changes in growth, differentiation and morphology during puberty and the adult reproductive cycle under the influence of ovarian hormones and locally produced paracrine signaling factors [2].

Ductal morphogenesis in puberty is mainly stimulated by estrogen [3, 4]. Dynamic growth and differentiation in the adult are most obvious during pregnancy and lactation, when extensive side-branching occurs and secretory alveoli develop in response to progesterone and prolactin [5, 6]. More subtle morphological changes, however, are known to occur during the estrous cycle, which repeats itself every 4 to 5 days in mice [7,8,9,10,11]. This short cycle is divided into four stages: proestrus, estrus, metestrus and diestrus, which correspond to the follicular (proestrus/estrus) and luteal (metestrus/diestrus) phases of the human menstrual cycle, with corresponding fluctuations in estrogen and progesterone (Supplementary Fig. 1) [12,13,14].

Alternating estrogen and progesterone signaling activity are thought to cause continuous cycles of growth and regression of lobuloalveolar outgrowths as well as sustained tertiary side-branching of epithelial ducts over time [7, 9,10,11,12,13,14,15]. However, the molecular basis of the morphological changes that occur during the estrous cycle remains poorly understood. Moreover, discrepancies exist between studies investigating these proliferative changes. Some observed an increase of epithelial cells mainly during late proestrus/estrus [7, 15]. Others reported that outgrowth of lobuloalveolar structures mainly occurs in diestrus compared to other stages [9, 10, 12]. More recent work showed that proliferative expansion of the epithelial ducts does not occur in every cycle in every mouse, highlighting the complexity of proliferative heterogeneity during the estrous cycle [11].

The outgrowth and regression of epithelial ducts that has been reported, is presumably the result of an indirect effect of steroid hormones on mammary stem cell (MaSC) regulation [9, 16,17,18]. One of the candidate factors responsible for controlling mammary stem cell behavior downstream of progesterone is Wnt4 [9, 19, 20], a member of the Wnt gene family that encodes ligands that play a pivotal role in stem cell self-renewal and tissue homeostasis in multiple different tissues [21], including the mammary gland [22,23,24,25,26,27,28,29]. Studies investigating the link between steroid hormones and Wnt genes or MaSC activity are often performed using ovariectomized mice receiving treatment with progesterone and/or estrogen pellets or injections [9, 10, 16, 19, 30]. Therefore, a need still exists for studies that examine the direct molecular consequences of hormonal fluctuations in the mammary gland under physiological conditions. Furthermore, as most studies that try to elucidate the underlying transcriptional changes either do not separate different mammary gland cell types or focus exclusively on epithelial cells [19, 30, 31], the contribution of adipocytes and other stromal cells has been largely overlooked so far. Here we perform genome-wide gene expression analysis of epithelial and stromal mammary cells isolated from wildtype FVB/N mice via differential digestion and fluorescence activated cell sorting (FACS), either during puberty or at the different stages of the adult estrous cycle. We observe complex but robust patterns of Wnt gene expression in basal, luminal, fat and fibroblast cells. In addition, we detect a cell proliferation signature in basal epithelial cells. The most prominent gene expression changes during the estrous cycle occur in non-epithelial cells, however. These data and findings complement existing studies.

Results

Isolation of Pubertal and Adult Estrous-Staged Mammary Glands

To investigate gene expression in specific cell populations of pubertal and adult estrous-staged mammary glands, we wanted to isolate mammary gland tissue at defined time points. The stage of the estrous cycle can be determined by collecting vaginal swabs and scoring the relative proportion of nucleated epithelial cells, cornified epithelial cells and leukocytes in the sample (Fig. 1a-b) [13, 32,33,34]. In total, we tracked the estrous cycle of 32 adult (14–16 weeks old) female FVB/N mice every day for at least 10 days (i.e. more than one cycle) to ensure that the mice selected for analysis were showing regular progression through the estrous cycle (examples in Fig. 1c). When a smear did not allow us to assign a clear estrous cycle stage, it was scored as a transition smear (e.g. proestrus to estrus), also considering the smear of the preceding day for that particular mouse. To confirm the interpretation of the vaginal smears and increase confidence of our estrous cycle scoring, we also evaluated H&E stained histological sections of vaginal tissue isolated at the time of sacrifice (Fig. 1d), since the vaginal epithelium also shows different characteristics at each of the four stages [35, 36]. Our analysis of the estrous cycle stage by assessing the tissue sections corresponded with our staging based on vaginal smears (see Supplementary Fig. 2 for a comprehensive overview of the matching cytology and histology images for every mouse that showed a regular estrous cycle). Of the 32 tracked mice, 7 showed an irregular cycle and were therefore excluded from further analysis (examples in Fig. 1e).

Fig. 1
figure 1

Determining the mouse estrous cycle stage by vaginal cytology. A Cartoon depicting the relative abundance of cell types present, based on descriptions from the literature [32, 33]. Horizontal axis: estimate of the duration of each stage, with the total estrous cycle taking 4–5 days. Dashed lines represent transition into the next stage. B Brightfield microscopy images of vaginal swabs after staining with Giemsa. Images are from the day of mammary gland isolation for the mice depicted in C). Scale bar = 100 µm. In proestrus, mostly nucleated epithelial cells are present, together with cornified epithelial cells. A small number of leukocytes may be detected in early proestrus stage. When the female mouse is in estrus, mostly cornified epithelial cells are present, together with a small number of nucleated cells (in early estrus) or some leukocytes (in late estrus). Metestrus presents as a mixture of cornified epithelial cells, nucleated epithelial cells and leukocytes. Swabs containing mostly leukocytes indicate that the mouse is in diestrus. C Graphs depicting monitoring of the estrous cycle. One example is depicted for each endpoint stage (i.e. the time of isolation). From left to right: endpoint proestrus, estrous, metestrus, diestrus. Each graph represents a single mouse that was followed over multiple days, corresponding to the mice depicted in B). D Brightfield microscopy images of H&E stained formalin fixed paraffin-embedded (FFPE) tissue sections of the vaginal epithelium at different stages of the estrous cycle. One example is depicted for each endpoint stage, corresponding to the mice in B-C). Proestrus can be recognized by a relatively thick epithelial layer and an outer layer consisting of nucleated cells. In estrus the epithelial layer is at its thickest and the outer later is now comprised of cornified cells. Metestrus is characterized by a thinner epithelium, the loss of the cornified cell layers and the transepithelial migration of leukocytes. Diestrus samples show the thinnest layer of epithelium, solely containing non-cornified epithelial cells. E Two examples of mice with irregular estrous cycles. These mice were excluded from further analysis

Full size image

For the adult stages, we ultimately isolated mammary glands from 5 adult female mice in proestrus, 5 in estrus, 4 in metestrus and 4 in diestrus. For each animal, the 3rd and 4th mammary glands were isolated, pooled and processed. For the puberty stage, we selected female FVB/N mice of 34–36 days old (P34-P36, where P stands for postnatal day), since the mammary epithelium is actively undergoing branching morphogenesis at this time point [37] (Supplementary Fig. 2). For each day, the 3rd and 4th glands of 4 individual mice from 3 different litters were isolated, pooled per mouse and processed.

Number and proportion of cell types remains similar throughout the estrous cycle

To profile gene expression in different cell types, we processed all samples for further analysis, taking care to separate luminal, basal, fat and non-adipose stromal mammary cells. Following tissue digestion, we first isolated the adipose fraction of the fat pad (hereafter referred to as adipocytes) by centrifugation. The remainder of the sample was processed to single cells. Using FACS we isolated luminal (lin−/EpCAMhigh/CD49fmed), basal (lin−/EpCAMmed/CD49fhigh) and non-adipose stromal (lin-/EpCAM−/CD49f−) cells (Fig. 2a) [38]. Supplementary Fig. 3 shows the complete gating strategy, as well as a direct comparison of our EpCAM/CD49f staining with the often-used CD24/CD29 staining. A larger variability in absolute cell numbers (and relative percentages) was observed in adult staged mice compared to puberty. We did not find consistent differences in epithelial cell numbers between different estrous cycle stages (Fig. 2b). Furthermore, the percentages of basal, luminal and non-adipose stromal cells (hereafter referred to as fibroblasts) were comparable across different stages (Fig. 2c).

Fig. 2
figure 2

Number and proportion of cell types remains similar throughout the estrous cycle. A FACS plot depicting the gating strategy for luminal, basal and non-adipose stromal cells (hereafter referred to as fibroblasts) using EpCAM-PE/CD49f-FITC staining. This example is from an adult virgin in proestrus. B Absolute cell numbers of the sorted populations. Note that the numbers for pubertal and adult mice cannot be compared directly, as for every pubertal replicate sample 4 mice were pooled, whereas all adult mammary glands were sorted individually. Error bars represent mean with SD. C Same samples as B), but here the relative proportions of the different cell types are plotted. Lin- cells = lineage negative cells

Full size image

To confirm this observation, we visualized the morphology of epithelial ducts in mammary glands of stably cycling mice by carmine staining. As we had used all the available tissue for digestion and FACS sorting, we monitored the estrous cycle of a second cohort of FVB/N mice using vaginal cytology (Supplementary Fig. 4). A total of 6 mice were used for this analysis, of which 3 were sacrificed in estrus and 3 in diestrus. No clear difference was visible in epithelial morphology between the two stages (Supplementary Fig. 5). From these same mice, we used another gland for RNA isolation and qRT-PCR. Expression levels of the progesterone receptor (Pgr) were increased in diestrus compared to estrus (Fig. 3). This fits with earlier studies [9, 13, 39] and confirms that the mice were cycling. In contrast, Wnt4 expression levels between estrus and diestrus were similar, indicating that any changes in the proposed progesterone-mediated induction of Wnt4 expression are subtle at best [9, 19, 20, 40] and below the detection limit of our qRT-PCR analysis when using whole mammary gland RNA as input. Summarizing, expansion of the epithelial network is not detected at the tissue-wide level (Figs. 2 and 3) during every estrous cycle in every mouse, in line with a previous report [11].

Fig. 3
figure 3

Whole mammary glands in estrus and diestrus show similar Wnt4 expression. Graph showing Wnt4 and Pgr mRNA expression levels, measured by qRT-PCR on whole mammary gland samples (n = 3 mice in estrus and n = 3 mice in diestrus). Experimental samples (#4 mammary glands) are from the same mice as those depicted in A. Ct values of Wnt4 and Pgr were normalized to Krt8 expression, which is a marker for luminal cells. Horizontal lines represent the mean

Full size image

Gene expression changes during the estrous cycle mainly occur in non-epithelial cells

For an unbiased overview of changes in gene expression across different developmental time points and cell types, we performed genome wide expression analyses. Enzymatically digested, FACS-sorted samples from multiple pubertal or multiple adult mice were pooled to obtain sufficient RNA for bulk RNA-seq analysis (see methods for details on how samples were processed and pooled). Two independent RNA sequencing runs were performed, one for the puberty and one for the adult samples. To measure the level of similarity between samples we conducted a multidimensional scaling (MDS) analysis (Fig. 4a). As expected, the different samples clustered by cell type and epithelial and non-epithelial cells are separated in the first dimension. With the exception of pubertal basal cells, the two replicates from corresponding developmental stages (i.e. puberty or adult samples) did not consistently cluster together within each cell type. Thus, MDS analysis reveals major differences in gene expression between mammary gland cell types, but not between the selected developmental or estrous cycle stages. Expression analysis of individual cell-type specific markers (luminal: Krt18 [41]), basal: Krt14 [41], adipocyte: Adipoq [42], fibroblast: Col1a1 [43]) (Fig. 4b) and hierarchical clustering (Supplementary Fig. 6) confirms the correct isolation of defined cell populations.

Fig. 4
figure 4

Clustering of different mammary gland cell populations. A MDS plot of all puberty and adult RNA-seq samples. Distances on the plot correspond to the biological coefficient of variation (BCV) between samples. B Plot depicting expression of cell type specific marker genes in every sample as Reads Per Kilobase per Million mapped reads (RPKM). Error bars represent mean with SD. In both A) and B) colors represent different cell types and shapes represent different developmental stages. Source data for B) available via https://osf.io/xv83g/ (rpkm_values_allsamples.xlsx)

Full size image

To compare the different stages of the estrous cycle, we performed differential gene expression analysis (Fig. 5a). The luminal cell population, which contains the estrogen- and progesterone-receptor positive cells [44], showed the smallest number of differentially expressed genes in pairwise comparisons of two different estrous cycle stages (21 unique genes in total, FDR < 0.05). A total of 255 unique genes were differentially expressed in basal cells, mainly in pairwise comparisons involving the estrus stage (Fig. 5a, FDR < 0.05). Statistically significant gene expression changes in adipocytes and fibroblasts were more prominent. In adipocytes, all pairwise comparisons involving diestrus showed > 1000 upregulated genes (Fig. 5a, FDR < 0.05). In fibroblasts, the estrus stage stood out the most, with most differentially expressed genes being downregulated. These results suggest that the non-epithelial cell populations are also hormone responsive, either directly or indirectly, and that reciprocal signaling likely occurs between the different cellular compartments.

Fig. 5
figure 5

Estrous-cycle dependent changes in gene expression are most prominent in adipocytes and fibroblasts. A Graph depicting the number of differentially expressed genes for each pairwise comparison between different stages of the estrous cycle (FDR < 0.05). Yellow bars: upregulated genes. Blue bars: downregulated genes. On the X-axis are different pairwise comparisons of different estrous cycle stages. P = proestrus, E = estrus, M = metestrus, D = diestrus. B Graphs depicting functional enrichment analysis of all differentially expressed genes (DE genes) in basal cells, adipocytes and fibroblasts (FDR < 0.01) in DAVID (version 6.8) [45, 46]. Per cell type the top 10 most enriched GO terms are shown. Color intensity indicates Benjamini corrected p-value. CC = cellular component, BP = biological process, MF = molecular function. Source data available via https://osf.io/xv83g/ (Fig. 5A_sourcedata_numbers.xlsx, Fig. 5A_sourcedata_FDR005 and Fig. 5B_sourcedata_FDR001)

Full size image

Functional gene ontology (GO) classification of differentially expressed genes (cut off FDR < 0.01, Fig. 5b) [47, 48] revealed that differentially expressed genes in adipocytes are mostly linked to cell membrane and cell adhesion, while the differentially expressed gene list in fibroblasts is enriched for genes associated with nucleus and regulation of transcription. The most prominent enrichment for specific GO terms was observed in basal cells and is related to cell cycle/cell division (see Supplementary Table 1 for a selection of differentially expressed genes with the highest fold change (logFC of >|5|) in basal cells and a logCPM of > 1). A cell cycle gene signature containing 65 out of the total 85 genes that are differentially expressed in basal cells (FDR < 0.01) showed elevated expression in basal and luminal cells, but not in non-epithelial cells in estrus compared to other estrous cycle stages (Supplementary Fig. 7). However, this does not translate to clear changes in the number of cells after sorting specific cell types during at diestrus (Fig. 2), nor in readily apparent side-branching (Supplementary Fig. 5).

A Complex Wnt Gene Expression Landscape in the Mammary Gland

Because the expression of Wnt4 and other Wnt genes has been linked to steroid hormones in different tissues [9, 19, 20, 49,50,51,52], we zoomed in on the expression of individual Wnt genes. Neither basal nor luminal epithelial cell types showed differential expression of any Wnt gene across the estrous cycle. We did, however, observe differential expression of 6 Wnt genes in non-epithelial cells (Supplementary Table 2, cut off: FDR < 0.05 in pairwise stage comparisons). In adipocytes, Wnt4, Wnt5b, Wnt7b and Wnt10a are upregulated in diestrus. In fibroblasts, Wnt6 and Wnt5a are downregulated in diestrus compared to estrus and metestrus, respectively. We do point out that the absolute expression levels of these Wnts in fibroblasts and adipocytes are often close to background levels (Fig. 6). Note that Wnt expression itself is not inherently low: Multiple Wnt genes are expressed at decent levels in luminal and basal cells – and although Wnt4 does not meet our cut off for statistical significance, we do detect an ~ 1.7 fold difference in luminal Wnt4 expression between proestrus (mean RPKM:17.2) and diestrus (mean RPKM: 10.1, Fig. 6).

Fig. 6
figure 6

Expression levels of differentially expressed Wnt genes in different mammary gland cell types during the estrous cycle. RNA-sequencing results from different mammary gland cell types and different stages of the estrous cycle. Dot plots depict the expression of 6 individual Wnt genes that were found to be differentially expressed (FDR < 0.05) in pairwise comparisons of different estrous cycle stages (Table 3). RPKM = Reads Per Kilobase per Million mapped reads. Inset in the top left panel shows Pgr expression, which rises and peaks prior to Wnt4 expression. Colors represent different cell types and symbols represent different stages of the estrous cycle. Source data available via https://osf.io/xv83g/ (rpkm_values_allsamples.xlsx)

Full size image

Finally, we performed unsupervised hierarchical clustering of our pubertal and estrous cycle staged samples (Fig. 7). This revealed two interesting features. First, epithelial and non-epithelial mammary cell populations cluster according to cell type based solely on the expression of different Wnt genes. Second, the expression patterns of individual Wnt genes vary between cell type rather than developmental time point. In puberty, Wnt4 and Wnt7b expression is restricted to luminal cells, while Wnt6 and Wnt10a expression is restricted to basal cells. Wnt5a and Wnt5b are expressed in both luminal and basal epithelial cells. This is in line with earlier work that showed expression of these genes in the pubertal mammary epithelium using microarray gene expression analysis [53]. Wnt4, Wnt5a, Wnt5b and Wnt7b remain expressed in the adult luminal cell population. This fits with published microarray and transcriptomic data [53,54,55], and with studies using RNA in situ hybridization or qRT-PCR analysis [19, 20, 25, 56, 57]. In the epithelium, expression of Wnt6 and Wnt10a remains restricted to the adult basal layer, also corresponding to previous studies [54, 55]. Wnt5a, Wnt5b, Wnt9a, Wnt10a and Wnt11 are also expressed in basal cells, in line with published scRNA data [55]. Although expression of Wnt2 and Wnt6 in stromal cells has been reported [57, 58], most genome wide expression analyses have so far focused on mammary epithelial cells. We find that Wnt2 is indeed mainly expressed in stromal cells, in agreement with RNA in situ hybridization experiments [59]. In adipocytes, Wnt gene expression was relatively low, although these cells showed some expression of Wnt2, Wnt10b, and Wnt16. Fibroblasts express multiple Wnt genes in addition to Wnt2, including some Wnt4 and Wnt6 (Fig. 6), with only Wnt9b and to a lesser extent Wnt2b expression being exclusive for the fibroblast cell population. The precise regulation and function of these Wnt gene expression patterns warrant further research.

Fig. 7
figure 7

RNA-sequencing reveals complex spatiotemporal expression patterns of individual Wnt genes. Unsupervised hierarchical clustering of Wnt gene expression in different mammary gland cell types at different developmental time points. Labels for different samples: * = puberty samples, P = proestrus, E = estrus, M = metestrus, D = diestrus. Numbers “1” and “2” indicate replicate samples (see methods for details). RPKM = Reads Per Kilobase per Million mapped reads. Values were normalized for each gene across all samples using Z-scores. Source data available via https://osf.io/xv83g/ (rpkm_values_allsamples.xlsx)

Full size image

Discussion

Few studies to date have sampled the basal, luminal, adipocyte and fibroblast cell populations in pubertal and adult mice. The specific aim of this study was to better understand the transcriptional response of the epithelial and non-epithelial mammary cell populations to physiologically changing hormone levels, with a specific focus on estrous-cycle staging. Given the role of WNT signaling in mammary gland development and tissue homeostasis, including a prior link to mammary stem cell maintenance and steroid hormone responsiveness, we paid specific attention to Wnt gene expression patterns and changes therein.

Determining Estrous Cycle Stage By Vaginal Cytology

Different methods exist for assessing the stage of the estrous cycle in adult female mice. While it can be done by assessing the overall appearance of the vaginal opening [60], metestrus and diestrus are difficult to distinguish using only visual inspection. It is therefore recommended to assess the ratio of different cell types using vaginal smears [33]. For this study, we confirmed our stage determination by vaginal cytology with histological analysis of the vaginal epithelium (Supplementary Fig. 2). While sampling our cohort, we found that 7/32 mice were not cycling reliably, despite being housed in the vicinity of male mice in the same open cages as their cycling littermates (Fig. 1e). This highlights the importance of monitoring the cycle for multiple days in a row: only collecting swabs on the day of tissue isolation or only collecting samples once or twice a week is insufficient to determine the precise stage and progression of the estrous cycle [31]. As shown previously, proper estrous cycle stages requires samples to be collected every day, at the same time, for at least one week [9, 10, 12, 61]. Indeed, daily monitoring for at least 8 days allowed us to identify animals with irregular cycles and exclude these mice from further analysis. Hormonal testing in blood plasma could still be added to enhance the reliability of testing.

Epithelial Outgrowth Does Not Occur in Every Mouse in Every Cycle

Studies investigating morphological changes of the mammary epithelium during the estrous cycle have reported conflicting results [7, 9,10,11,12, 15]. Although findings concerning the timing of apoptosis are consistent (diestrus [9, 11, 12, 15]), it remains unclear whether side branching occurs in proestrus/estrus [11, 15] or diestrus [9, 10, 12]. Assuming that outgrowth of the epithelium requires cell division, one would expect a burst in cell proliferation to precede the appearance of tertiary branches. Our RNAseq data show a cell division/mitosis signature in basal and luminal cells at estrous (Fig. 5, Supplementary Table 1 and Supplementary Fig. 7), suggesting cell proliferation or a mitotic arrest, but we did not observe accompanying or subsequent changes in cell numbers (Fig. 2) or morphology (Supplementary Fig. 5) across the stages of the estrous cycle. Tissue level analyses at single cell resolution, including intravital imaging approaches [62], will be required to understand how, when and where fluctuations in systemic hormone levels are translated into a locally controlled cell division, differentiation and migration response. Altogether, we and others [11] find that proliferation of epithelial cells during the estrous cycle is not as black and white as reported in some studies. Rather, proliferative heterogeneity exists. From our set up and analysis we cannot exclude confounding factors such as environmental conditions (e.g. housing and diet) or effects arising from sampling different anatomical locations (e.g. #3 versus #4 mammary gland). It is likely, however, that true differences exist between strains, between individual animals, and possibly between different regions of the tissue. Our findings suggest that cell proliferation is a local, rather than a global event, in the same way that some terminal end buds (TEBs) during puberty are more proliferative than others [63].

Gene Expression Changes Between Different Estrous Cycle Stages

While we were able to detect estrous-cycle dependent changes in gene expression by both RNAseq (Supplementary Fig. 7) and qRT-PCR (Fig. 3), other differences may have been missed due to our experimental setup. First, we isolated the bulk luminal cell population. As such, the effect of fluctuating steroid hormone levels might be masked, since the estrogen and progesterone receptor (ER and PR) are only expressed in a subset of luminal cells in adult mice [44]. Either enriching for ER + and PR + mature luminal cells via FACS sorting [64,65,66], or single-cell rather than bulk RNAseq [55], would facilitate gene expression analysis of hormone receptor expressing cells specifically. Second, while we harvested glands from 4–5 mice for each stage, we had to pool isolates to obtain sufficient RNA. This unfortunately reduced the number of biological replicates in our analysis, which also prevents further statistical testing (e.g. in Fig. 6). Third, it is also possible that rapid changes in morphology as previously observed, are not caused by changes in gene expression. In fact, other mechanisms may enable faster and/or more dynamic changes in cell behavior. This includes alternative splicing, which several studies have shown to be important in vertebrate tissue development and homeostasis [67, 68]. Changes to the poly(A) tail of mRNA can also play a regulatory role by changing translation efficiency, mRNA stability and degradation [69]. Post translational modifications such as phosphorylation, acetylation or ubiquitination can change protein functionality and add another level of regulation [70].

Non-epithelial Cells Should be Included in Studies of Mammary Gland Biology

Pairwise comparisons of the different estrous cycle stages revealed that most differentially expressed genes were found in non-epithelial rather than in epithelial cell populations (Fig. 5a). While hormone receptor positive cells are indeed also found outside of the mammary epithelium [30, 71], the difference in absolute numbers of differentially expressed genes between the epithelial and non-epithelial mammary cell types, is remarkable. This adds to the evidence that molecular cues that control epithelial cell behavior during the estrous cycle are not only coming from the epithelial cells itself, but also from adipocytes and fibroblasts [72], either in direct response to hormones or in response to hormone-induced paracrine signals. It also fits with a large and increasing body of work that demonstrates the importance for epithelial-stromal interactions in mammary tissue remodeling at other developmental stages, as well as in the context of breast cancer.

Some caution is warranted when interpreting these results. First, we isolated adipocytes via differential centrifugation after the initial enzymatic digestion step, possibly resulting in less pure samples than for the other cell types, which were isolated by FACS sorting. While our quality controls do not show any evidence of impurity (Fig. 4 and Supplementary Fig. 5), we cannot exclude some contamination. Second, the FACS sorted fibroblast population is not well defined. While we made sure to exclude doublets, dead cells, hematopoietic and endothelial cells (Supplementary Fig. 3), our sorting of non-adipose mammary stromal cells was not based on the presence of specific markers, but rather on the absence of epithelial markers (Fig. 2a). Because fibroblasts are the main component of the mammary stroma and we did not actively exclude them in any step prior to sorting, we expect these cells to be the main component of our sorted non-adipose stromal cell population. This is supported by the specific expression of Col1a1 in this population (Fig. 4b). In view of the above, it will be interesting to further dissect cell-specific changes in more purely sorted populations of fibroblasts, macrophages and potentially other cells.

Wnt Genes Show Defined Spatial Gene Expression Patterns in the Mammary Gland

We did not detect statistically significant changes in Wnt4 expression in luminal cells (Fig. 6), where it has been reported to act downstream of progesterone [9, 19, 20, 40]. In some of these prior studies, Wnt4 expression was measured after treating ovariectomized mice with 17β-estradiol and progesterone. The fact that we measure gene expression under physiological conditions might be one explanation for the discrepancy, since endogenous hormone fluctuations, and their downstream effects, are likely to be much smaller. This is supported by the fact that we did not detect a correlation between Pgr and Wnt4 gene expression in whole mammary gland mRNA (Fig. 3), and by the fact that we were able to detect subtle, non-statistically significant changes in Pgr (~ 1.5 fold difference between diestrus/proestrus versus estrus/metestrus, mean RPKM of 58.4 and 40.1, respectively) that preceded equally subtle changes Wnt4 expression (~ 1.7 fold difference between proestrus and diestrus, mean RPKM of 17.2 and 10.1, respectively) in our FACS sorted, and thus partially enriched, samples (Fig. 6).

As far as we know, spatiotemporal Wnt gene expression differences across luminal, basal, fibroblast and adipocyte populations during puberty and across the four stages of the estrous cycle has not been examined before. Previous transcriptomic analyses that distinguished luminal cells based on their hormone receptor expression status [54, 55] showed that Wnt4, Wnt5a and Wnt7b are mainly (but not exclusively) expressed in hormone receptor positive luminal cells. We can conclude that multiple Wnt genes show well defined spatial gene expression patterns, resulting in a complex Wnt gene expression landscape in basal, luminal, adipocyte and fibroblast cell populations in the postnatal mammary gland.

Conclusion and Outlook

Although a link between the reproductive cycle and breast cancer development is known to exist, the molecular basis of this interaction remains incompletely understood. This work contributes to understanding the effects of physiological hormone fluctuations on gene expression and the molecular mechanisms underlying dynamic morphological changes in the mammary gland during the reproductive cycle. It shows that endogenous gene expression changes in response to physiological changes in steroid hormone levels are subtle. Our dataset should be a useful resource for future studies investigating gene expression patterns in different cell types of the mammary gland across different developmental timepoints. It also informs future studies, which should ideally include a higher number of replicates to capture sample-to-sample heterogeneity and should continue to consider early and late timepoints of each estrous cycle stage, while at the same time isolating hormone-receptor positive and negative cells for both epithelial and non-epithelial cell populations to reduce intra-sample heterogeneity. Ideally, these experiments would be performed at single-cell resolution. Ultimately, a better comprehension of the molecular mechanisms underlying the dynamic and heterogenous changes in mammary cell behavior in response to systemic hormones and paracrine growth factors will increase our understanding of reproductive cycle-related breast cancer risks. In the long run, these insights have the potential to contribute to the prevention and personalized treatment of breast cancer in premenopausal women [73].

Materials and Methods

Mice

Wildtype, inbred FVB/NHan®Hsd mice (referred to as FVB/N in the main text) were purchased from Envigo. Breeding and estrous cycle monitoring of the experimental cohort was performed in-house. All mice were maintained under standard housing conditions in open cages, with 12 h light/dark cycle and ad libitum access to food and water. Experiments were performed according to institutional and national guidelines and regulations (see study-specific approval). Pubertal mice were 34–36 days old. N = 4 mice from n = 3 different litters were pooled to obtain sufficient cells for RNA-sequencing according to Table 1. Adult mice were 14–16 weeks old and were sacrificed at the appropriate stages of the estrous cycle according to Supplementary Fig. 2. Samples from n = 2 or n = 3 different mice were pooled according to Table 1 to obtain sufficient material for RNA-sequencing.

Table 1 Mammary glands of different mice were pooled per stage
Full size table

Vaginal Cytology

Vaginal swabs were collected using plastic Pasteur pipettes and PBS. From the tip of the vaginal opening, the vagina was flushed 2–3 times with a few drops of PBS. The sample was transferred to a glass slide, air-dried at 37 °C, stained with Giemsa (Sigma-Aldrich cat. #48900) for 30 s and rinsed with PBS. Animals of the right stage were sacrificed as soon as possible after the samples had been stained and assessed under the microscope. Brightfield images of the stained cytology samples were taken at 20 × magnification using a Zeiss Axio Vert.A1 microscope.

Histology

Vaginal tissue samples were fixed in 4% PFA for 24 h, dehydrated through ascending grades of ethanol, cleared in Histo-Clear II (National Diagnostics cat. #HS-200) and embedded in paraffin. Five µm thick tissue sections were cut and mounted on glass slides. Slides were baked at 60 °C for 45 min, deparaffinized in Histo-Clear II, rehydrated through descending grades of ethanol, stained with 50% Mayer’s Hematoxylin (Sigma-Aldrich) for 30 s, rinsed for 5 min in tap water, washed in PBS for 3 min and 70% ethanol for 5 min, stained with Eosin Y (Sigma-Aldrich, cat. #HT110132) for 2 min, dehydrated (through 70% ethanol, 100% ethanol and 100% isopropanol), cleared in Histo-Clear II (National Diagnostics cat. #HS-200) and mounted with a coverslip. Brightfield images of the histology samples were taken at 20 × magnification using a Zeiss Axio Vert.A1 microscope.

Mammary Gland Digestion

The 3rd and 4th mammary glands of FVB/N mice were isolated (the #4 lymph node was removed prior to this), manually minced and enzymatically digested in an orbital shaker for 2 h at 37 °C in digestion mix (9.2 ml DMEM/F12, 5% FCS, 1% Penicillin/Streptomycin, 25 mM HEPES (Gibco, cat. #15630056) and 300 U/ml Collagenase IV (Gibco, cat. #17104019), with 10 ml digestion mix per mouse for 4 glands per mouse). After centrifugation, the fat fraction was obtained by pipetting and transferring the top layer to TRIzol LS (Invitrogen, cat. #10296028). Samples were stored at -80 °C. The remainder of the sample was processed for single cell isolation. For this, cell pellets were resuspended in HBSS (Gibco, cat. #11540476 supplemented with 2% FBS (Gibco, cat. #11573397) and ACK solution (Gibco, cat. #A1049201) (1:3) and incubated at room temperature (RT) for 5 min to lyse red blood cells. To dilute and inactivate the ACK buffer, 13 ml of HBSS was added and cells were spun down for 5 min, 1000 rpm, 4 °C, and brake set to 1. Cell pellets were resuspended in 2 ml pre-warmed 0.05% Trypsin–EDTA (Gibco, cat. #11590626) and incubated for 5 min at 37 °C, after which 3 ml of pre-warmed serum-free DMEM and 1 µg/ml DNAseI was added to the solution. After mixing, 8 ml of DMEM with 10% FCS was added to stop trypsinization. Cells were filtered through a 40 µm mesh into a fresh tube.

Antibody Staining and FACS

Cells were resuspended in 200 µl HBSS supplemented with 10% FBS and a cocktail of the following antibodies: EpCAM-PE (1:100, eBioscience, 12–5791-82, clone G8.8), CD49f-FITC (1:100, eBioscience, 11–0495-82, clone GoH3), CD45-Bio (1:100, eBioscience, 13–0451-82, clone 30-F11), CD31-Bio (1:100, eBioscience, 13–0311-81, clone 390), Ter119-Bio (1:100, eBioscience, 13–5921-81, clone TER-119). After incubating in the dark on ice for 20 min, cells were washed twice with HBSS supplemented with 2% FBS and incubated in 200 µl HBSS supplemented with 10% FBS containing Streptavidin-APC (1:200, eBioscience, 17–4317-82). After antibody staining, cells were stained with DAPI (1:5000 (Invitrogen cat. #D1306) and filtered through a 50 μm mesh. Cells were sorted using a BD FACS Aria III. FITC was excited with a 488 nm laser and emission was filtered using a 530/30 nm bandpass filter. PE was measured using a 561 nm laser and 582/15 nm bandpass filter. DAPI was excited with a 407 nm laser and emission was filtered using a 450/50 nm bandpass filter. APC was measured using a 633 nm laser and 660/20 nm bandpass filter. Cells were sorted with a plate voltage of 2500 V using the 4-Way Purity precision mode. Sorted cells were collected in TRIzol LS (Invitrogen, cat. #10296028). Post-sort purity checks were performed after every sort and were always > 90%. Samples were stored in TRIzol LS at -80 °C.

Carmine Staining

Freshly isolated 3rd mammary glands of estrous cycle monitored mice were flattened between two glass slides, incubated on a nutator at RT for 4 h in a 50 mL tube containing 12.5 ml 100% EtOH and 12.5 ml acetic acid. Glands were removed from the glass slides, incubated in 70% EtOH for 1 h on a nutator at RT and rinsed in demi-water and stained overnight (O/N) in carmine solution (1 g carmine (Sigma, cat. #C1022), 2.5 g aluminum potassium sulphate (Merck, cat. #101047), 500 ml water, boiled for 20 min and filtered). After staining, glands were washed with 100% EtOH for 4 h. Glands were cleared and stored in Histo-Clear II at RT. Images were taken on a Leica stereomicroscope at 5 × magnification.

qRT-PCR

Total RNA from the 4th mammary glands was isolated from TRIzol (Invitrogen, cat. #15596018) according to manufacturer’s guidelines. After DNAse treatment, cDNA synthesis was performed from 4 µg RNA using SuperScript IV Reverse Transcriptase (Invitrogen, cat. #18090200) and Random Hexamers (Invitrogen, cat. #N8080127) according to manufacturer’s guidelines. During the reverse transcriptase reaction, RiboLock RNase Inhibitor (Thermo Scientific, cat. #EO0328) was added. After completion of cDNA synthesis, samples were diluted 10 × and qRT-PCR reactions were performed using a QuantStudio 3 Real-Time PCR System (Applied Biosystems). For each reaction, 5 µl of diluted cDNA was added to a mix of 4 µl 5X HOT FIREPol EvaGreen qPCR Mix Plus (ROX) (Solis Biodyne, cat. #08–24-00008), 1 µl primers (0.5 µl forward and 0.5 µl reverse, from a 10 µM stock) and 10 µl nuclease-free water. Reactions were performed in triplicate in a 96 × 0.2 ml plate (BIOplastics, cat. #AB17500). Thermal cycling reactions included the following stages: 2 min. at 50.0 °C and 15 min. at 95.0 °C, then 40 cycles of 15 s. at 95.0 °C and 1 min. at 60.0 °C, followed by the melting curve stage. The following primers were used:

  • Wnt4 forward: ACTGGACTCCCTCCCTGTCT, Wnt4 reverse: TGCCCTTGTCACTGCAAA,

  • Pgr forward: TGCACCTGATCTAATCCTAAATGA, Pgr reverse: GGTAAGGCACAGCGAGTAGAA,

  • Krt8 forward: AGTTCGCCTCCTTCATTGAC, Krt8 reverse: GCTGCAACAGGCTCCACT,

RNA-sequencing

RNA extraction, purification and sequencing, as well as data processing until read count calculation was done at the NKI Genomics Core facility. For the puberty samples (see Table 1), the following samples were used based on quality control: Replicate 1: Basal P34, Luminal P34, Stromal P34, Fat P34; Replicate 2: Basal P35, Luminal P35, Stromal P35, Fat P36. Adult samples were pooled according to Table 1. RNA was extracted from TRIzol LS and purified using the Qiagen RNeasy column purification kit. RNA quality was checked with a Bioanalyzer (Agilent), after which polyA + stranded RNA library preparation was performed using the Illumina TruSeq stranded RNA prep kit. RNA-sequencing was performed on a HiSeq 2500 (Illumina) System at the NKI Genomics Core Facility.

Bioinformatics

Single-end reads (65 bp) were aligned to reference genome GRCm38/mm10 with Tophat version 2.1 and Bowtie version 1.1.0 [74]. Expression values were determined by HTSeq-count [75]. Raw gene-level count tables were processed further using edgeR (version 3.28.0) [76, 77] and limma (version 3.42.1) [78] packages with R (version 3.6.2) [79]. A cutoff of CPM > 1 in at least 2 libraries was applied to filter out genes with low counts prior to trimmed mean of M-values (TMM) normalization. To visualize distances between gene expression profiles based on the biological coefficient of variation, the plotMDS function was used with method set to “bcv”. RPKM values were generated with the RPKM function after calculating exonic region lengths from non-overlap** exons per transcript ID (genome version mm10). The edgeR glmQLFit function was used to estimate the dispersion trend before testing for differential expression between different estrous cycle stages. Unless otherwise noted, a false discovery rate (FDR) of 5% was considered as a threshold for significance. The heatmap of Wnt gene expression in different mammary gland cell types was created with the heatmap.2 function (gplots) with Z-score scaling along rows and dendrograms computed using default settings. The heatmaps for Supplementary Fig. 5 and 6 were made with the pheatmap package (version 1.0.12) [80].

Software and Online Resources

Differential gene expression analysis for Figs. 6 and 7 was performed in R and R Studio as described under Bioinformatics. Gene set enrichment analysis for Fig. 7 and Supplementary Fig. 7 was performed using DAVID (https://david.ncifcrf.gov/ [81]) and MSigDB (https://www.gsea-msigdb.org/gsea/msigdb [82, 83]. Plots and graphs were made in R Studio and Graphpad Prism. Figure 1 was drawn in Adobe Illustrator. All figures were assembled in Adobe Illustrator. Tables were made in Microscoft Word. Spreadsheets were made in Microscoft Excel.

Data availability

RNA-seq source data for puberty and adult samples have been uploaded to NCBI GEO and can be accessed via GSE213451 as raw counts, with links to the SRA read archive to access the original.fastq files. Source data for the figures and tables are provided via the Open Science Framework and can be accessed through https://osf.io/xv83g/ (identifier: https://doi.org/10.17605/OSF.IO/XV83G). This includes processed RNAseq data (provided as RPKM values in both.txt and.xlsx format), as well as spreadsheets with source data for Fig. 5, Supplementary Fig. 1, 6 and 7 and Supplementary Table 1.