Transcriptome analyses of mouse and human mammary cell subpopulations reveal multiple conserved genes and pathways

A single cell suspension of mammary cells was prepared from freshly harvested mammary glands and sorted by flow cytometry, as previously described [3]. Mice were on a pure FVB/N background. All experiments were approved by the WEHI Animal Ethics Committee, and the care of animals was in accordance with institutional guidelines. Experiments using human tissue obtained from the Royal Melbourne Hospital Tissue Bank were approved by the Human Research Ethics Committees of The Walter and Eliza Hall Institute of Medical Research and Melbourne Health.

Unless otherwise specified, antibodies for flow cytometry were obtained from BD Pharmingen. Antibodies against mouse antigens were PE-conjugated antibody to CD24, FITC-conjugated antibody to CD29 (clone HMbeta1-1 from H. Yagita) [16], biotin-conjugated antibodies to CD31, CD45, and TER119, and APC-conjugated antibody to CD61 (Caltag). Antibodies used for human antigens have previously been described [7]. The Alexa Fluor 647 anti-human CD24 antibody (Biolegend) was used for analysis of human breast epithelial subsets. Antibody staining and cell sorting was as previously described [3]. Data were analyzed by using WEASEL software [17].

Total RNA was isolated from primary mammary cell subpopulations with the RNeasy Micro kit (Qiagen). Reverse transcription by using oligo(dT) primer and Moloney murine leukemia virus reverse transcriptase (Invitrogen) was according to the manufacturer's protocol. Quantitative RT-PCR was carried out by using a Rotorgene RG-6000 (Corbett Research) and SensiMix (dT) DNA kit (Quantace) under the following conditions: 10 min at 95°C followed by 35 cycles consisting of 15 seconds at 95°C, 20 seconds at 62°C, and 20 seconds at 72°C. Gene expression was determined with the Rotor-Gene software (version 1.7). The primer sequences used are listed in Supplementary Methods in Additional file 1.

Total RNA was purified from sorted cell populations by using the RNeasy Micro kit (Qiagen). RNA quality was assessed with the Agilent Bioanalyzer 2100 (Agilent Technologies) by using the Agilent RNA 6000 Nanokit (Agilent Technologies) according to the manufacturer's protocol. Up to 500 ng of RNA was labeled with the standard Total Prep RNA amplification kit (Ambion), and complementary RNA (1.5 μg) was hybridized to Illumina MouseWG-6 v2.0 BeadChips. After washing, the chips were coupled with Cy3 and scanned by using an Illumina BeadArray Reader. Unnormalized summary probe profiles, with associated probe annotation, were output from BeadStudio.

Microarray data were analyzed by using the limma package of the Bioconductor open-source software project [18, 19], as described in more detail later.

Freshly sorted cell subpopulations (> 90% purity) were prepared from mouse mammary glands for gene profiling analysis. These subpopulations corresponded to the MaSC-enriched (CD29^hiCD24^loCD61⁺), luminal progenitor (CD29^loCD24⁺CD61⁺), mature luminal (CD29^loCD24⁺CD61^-), and stromal cell (CD29^loCD24^-) fractions. Representative FACS dot plots depicting the four mouse cellular subsets and comparison with the analogous subpopulations found in human breast tissue [7] are shown in Figure 1. For human mammary cells, the subpopulations include the MaSC-enriched (CD49f^hiEpCAM^-), luminal progenitor (CD49f⁺EpCAM⁺), mature luminal (CD49f^-EpCAM⁺), and stromal cell (CD49f^-EpCAM^-) fractions. Although CD24 is expressed in all epithelial subsets in mouse mammary epithelium [3‐5], within human breast tissue, it marks luminal progenitor and mature luminal cells (Figure S1 in Additional file 2).

The Illumina mouse WG-6 v2.0 platform was used for arraying the four murine cell subpopulations, incorporating five biologic replicates for the three epithelial populations and three replicates for the stromal subset. Importantly, the RNA was not subjected to an amplification step before preparation of labeled cRNA, to avoid potential skewing of expression data. Unsupervised clustering revealed that the four subpopulations had distinct gene expression profiles (Figure S2 in Additional file 3). Gene expression signatures were derived for the four murine cell subpopulations, by using a method we applied previously to the analogous human subpopulations [7]. In brief, signature genes were chosen that were consistently up- or downregulated in that subpopulation (with fold-change at least 1.5 and FDR < 0.05) versus each of the other populations (Table 1). This selects a set of signature genes that strongly characterize each subpopulation by their unusually high or low transcriptional activity.

First, we used the signature genes to identify relations between tumor cells and normal epithelial cell types. Genetically engineered mouse models of mammary tumorigenesis have been previously described and include the mouse mammary tumor virus (MMTV)-Wnt-1, MMTV-Neu, MMTV-PyMT, WAP-Myc, WAP-Int3 (Notch-1) transgenic, and p53-null mouse models [27]. We interrogated the expression profiles of whole mammary tumors isolated from these mouse models [14] for the expression signatures characteristic of our mouse MaSC-enriched, luminal progenitor, mature luminal, and stromal subpopulations. These analyses were carried in an analogous manner to that described for comparison of human mammary cells with the different breast cancer subtypes [7]. In brief, the signature genes for each subpopulation were used to construct an index of transcriptional activity characteristic of that subpopulation. These indices, or expression signatures, were then computed for each tumor sample. The MaSC transcriptional signature was found to be highest in MMTV-Wnt-1 and p53^-/- tumors (Figure 2a). Robust results were obtained even though the p53^-/- tumors are on a different background compared with the other tumor types (BALB/c versus FVB/N). In contrast, the luminal progenitor signature was highest in MMTV-Neu and MMTV-PyMT tumors. The mature luminal signature was highest in MMTV-Myc tumors. The tumors arising in MMTV-Int3 mice did not correspond to a specific subset within the mammary epithelial hierarchy. As anticipated, the mouse stromal signature was not apparent in any of the mammary tumor profiles, thus reflecting the epithelial content of the tumors. Figure 2b summarizes potential relations between normal epithelial cell types and commonly used models of mammary tumorigenesis.

We previously reported the expression profiling of human mammary epithelial cell subpopulations, by using freshly sorted cells, unamplified material, and the Illumina platform [7]. The human and mouse gene expression profiles were first compared in a multidimension scaling (MDS) plot analysis, an unsupervised two-dimensional display of differences between profiles. As expected, samples separated clearly by species (Figure 3a). After normalizing for species differences, however, the samples clustered clearly by cell subtype (Figure 3b), showing that relative expression patterns across the cell subtypes are largely conserved between the two species. Dimension 1 distinguishes stromal cells from luminal cells, whereas dimension 2 distinguishes stem cells from others. The luminal progenitor and mature luminal subpopulations shared the greatest similarity, especially in the case of mouse.

To relate the two species more closely, we examined the transcriptional activity scores of the mouse subtype-specific signature genes in each of the human RNA samples. This demonstrated a conserved expression profile for each epithelial cell subtype across the two species. For each subpopulation, the mouse transcriptional signature was consistently highest in the corresponding human subtype for every biologic replicate (Figure 3c). The two luminal subtypes showed intermediate cross-over transcriptional activity with each other, whereas the transcriptional activity of the MaSC-enriched subset was relatively specific (Figure 3c).

Next we looked for genes in common between the mouse and human signatures for each epithelial subpopulation. For this analysis, we used a more comprehensive set of signature genes, by loosening slightly the FDR criteria, as described in Methods (Tables S1-S3 in Additional files 4, 5, and 6). Of a total of 8,451 mouse probes that were signature probes for the three epithelial subpopulations, 4,758 unique mouse genes with human orthologues were found, of which 1,204 (25%) were found to be signature genes for the corresponding human subpopulation (Table 2). As expected, the MaSC-enriched subpopulation had the largest number of signature genes and the highest conservation rate between species, with 489 shared upregulated and 428 shared downregulated genes (Table 2), indicating strong conservation in basal lineage genes. The lower degree of conservation evident in the luminal progenitor and mature luminal cell signatures in part reflects the closer relation between these two subpopulations but also suggests that they may be more heterogeneous than previously anticipated.

The conserved upregulated genes in the MaSC-enriched population spanned diverse gene ontology groups, including transcription factors (for example, Irx4, Mef2c, Slug, Egr2, Twist2, Tbx2, Id4, p63, and Sox11), cytokeratins (Krt5, 14, 16), and plasma membrane proteins (for example, Lgr6 and the receptors for Oxytocin, Oncostatin M, and Lif). The Notch ligand Jag2 was highly expressed in this subpopulation, and its product may directly signal through Notch receptors expressed on adjacent luminal cells [28]. The Wnt/β-catenin pathway is anticipated to be active in self-renewing MaSCs, compatible with the observation that Fzd8 and Tcf4 are components of the conserved upregulated gene signature in the MaSC-enriched population. The Wnt-pathway inhibitors Wif1 and Dkk3, however, were also found to be abundantly expressed. These antagonists may be expressed and secreted by mature myoepithelial cells present within this population to attenuate Wnt signal transduction in the basally located MaSCs.

For the luminal progenitor signature, Kit (receptor tyrosine kinase), Elf5 (Ets transcription factor), Cyp24A1 (vitamin D metabolizing enzyme), Lbp, and Cxcr4 were highly expressed in both species. Aldh1a3 was also upregulated in luminal progenitor cells versus other cell types, although another isoform Aldh5a1 was identified in the luminal-restricted population isolated by Raouf et al. [11]. Interestingly, virtually all the ALDH activity in human breast epithelium resides within the luminal progenitor population (unpublished data) rather than the more primitive mammary stem cell subset [29]. In mature luminal cells, highly expressed genes included the transcription factors Foxa1, Myb, estrogen receptor (ER), progesterone receptor (PgR), and Tbx3, as well as the prolactin receptor (Prlr) and Rank ligand (Tnfsf11).

Quantitative RT-PCR analysis was used to validate a number of genes in the conserved signatures, examples of which are shown in Figure 4. In the MaSC-enriched population, the Wnt inhibitory factor Wif1 and transcription factors Δ Np63, Tbx2, and slug (snail2), were preferentially expressed, thus validating the Illumina microarray data. Human NOTCH-4 was most highly expressed in the basal population, compatible with the findings of Raouf et al. [11] but differing from the mouse Notch-4 gene, which was expressed in all epithelial subsets at low levels [28]. C-Kit, Cyp24A1, and Elf5 were predominantly expressed in the luminal progenitor population in both species. Although low levels of KIT mRNA were evident in the human MaSC-enriched population, FACS analysis has shown that KIT protein is selectively expressed by human luminal progenitor cells (data not shown). As expected, Krt18, ERα, and PgR were preferentially expressed in mature luminal cells in both mouse and human, consistent with immunostaining of freshly sorted cells [7, 12]. The differential expression of other genes, including RankL, amphiregulin, Wnt4, and ErbB2 was also confirmed in the mouse and human subsets (data not shown).

To identify pathways and gene networks active in both human and mouse, the conserved signature genes for each epithelial subpopulation were analyzed by using the Ingenuity Pathway Analysis (IPA) software [26]. For each subpopulation, canonic molecular pathways that had greatest overlap with the conserved signature genes were selected. The resulting pathways therefore center on conserved genes characteristic of the various epithelial populations. In the MaSC-enriched subset, several pathways were found to be conserved across species, forming a number of specific nodes that include the ephrin receptor, integrin, interleukin-8, p53, and Wnt/β-catenin signaling pathways (Figure 5). Interestingly, IL-8 has recently been implicated in the cancer stem cell signature of the ALDH⁺ population in several breast cancer cell lines. IL-8 was also shown to enhance mammosphere formation and ALDH activity in these cell lines [30].

For the luminal progenitor cell population, the network was expanded through the addition of neighboring genes (depicted in white, Figure 6), as few connections were evident. Conserved pathways include the Toll-like receptor, vitamin D receptor, and Erk/Mapk signaling pathways. Kit, Elf5, and Cyp24A1 represent highly differentially expressed genes that form key components of the luminal progenitor cell signature. In the mature luminal progenitor subset, the steroid hormone receptor, HER2/erbB2, and Notch signaling networks emerged as conserved pathways across species (Figure 7).