Epithelial vimentin plays a functional role in mammary gland development

The mammary gland is a highly dynamic organ that develops through branching morphogenesis in puberty, evolves during the menstrual cycle and undergoes terminal differentiation/dedifferentiation during pregnancy, lactation and involution. Increasing evidence suggests that the mammary epithelium in both humans and mice comprises a hierarchy of cells, spanning from bipotent mammary gland stem cells (MaSCs) to differentiated luminal and basal epithelial cells (Rios et al., 2014; Van Keymeulen et al., 2011; Visvader and Stingl, 2014). The MaSCs and the basal and luminal progenitor cells are considered particularly important for ductal elongation during pubertal growth and for lobulo-alveolar expansion during pregnancy (Tiede and Kang, 2011). Several marker proteins have been indicated for stem and progenitor cells in the mammary epithelium (Visvader and Stingl, 2014). The current view suggests that the transcription factors Sox9 and Slug (Snai2) (Guo et al., 2012), several integrin adhesion receptors (Rangel et al., 2016; Taddei et al., 2008), and molecules involved in Notch and Wnt signalling pathways (Bouras et al., 2008; Zeng and Nusse, 2010) are expressed in the MaSCs and committed progenitor cells that localize to the basal mammary epithelial niche (Shackleton et al., 2006).

Vimentin is a cytoskeletal type III intermediate filament (IF) protein widely used as a marker for mesenchymal cells (Coulombe and Wong, 2004). Despite the widespread expression of vimentin, the phenotype of vimentin knockout mice (Vim^tm1Cba, hereafter called Vim^−/−) is mild (Colucci-Guyon et al., 1994). Nevertheless, defects in motor coordination (Colucci-Guyon et al., 1999), wound healing (Cheng et al., 2016; Eckes et al., 2000) and endothelial function (Nieminen et al., 2006; Antfolk et al., 2017) accompany vimentin loss in these animals. Other IF proteins, such as nestin and GFAP, have been implicated in the self-renewal of neural stem cells (Park et al., 2010). Although vimentin is expressed in basal mouse mammary epithelial cells (bMMECs), including the normal tissue reconstituting MaSCs (Soady et al., 2015; Ye et al., 2015), the functional role of vimentin in mammary gland development and stemness has not been previously studied. Here, we report that Vim^−/− mouse mammary epithelium has delayed ductal outgrowth and reduced regenerative capacity in vivo. Congruently, we find that vimentin is an important regulator of gene expression and proliferation in bMMECs and contributes to cancer stem cell-linked phenotypes in vitro.

Supporting previous observations (Soady et al., 2015; Ye et al., 2015), vimentin was found to be expressed in alpha smooth muscle actin (Acta2)-positive basal mammary epithelial cells, and in stromal cells in human and mouse mammary gland tissue sections (Fig. 1A,B, Fig. S1A). Vimentin expression was lower in the epithelial compartment than in the stroma, although some basal cells particularly in the mouse mammary gland exhibited high vimentin levels (Fig. 1B). In addition, quantitative PCR analysis of vimentin mRNA expression in mouse mammary gland cell populations [basal epithelial cells, luminal progenitor and mature luminal epithelial cells and stromal cells, sorted by fluorescence-activated cell sorting (FACS) based on CD24 and ICAM1 surface expression (Di-Cicco et al., 2015)] revealed notable vimentin levels in stromal cells and to a lesser extent in bMMECs (Fig. 1C). As expected, luminal progenitor cells and mature luminal epithelial cells did not express vimentin (Fig. 1C).

Earlier reports describing vimentin expression in the mouse MaSC compartment (Soady et al., 2015; Wang et al., 2015; Ye et al., 2015) led us to investigate the role of vimentin in mouse mammary gland development. To this end, we compared the mammary glands of wild-type and Vim^−/− mice (lacking vimentin expression also in the mammary gland; Fig. S1B). Mammary gland whole-mount analyses, performed at the same phase of the oestrus cycle, revealed normal ductal branching pattern in both 10- and 15-week-old Vim^−/− mice (Fig. 2A). However, the mammary ductal outgrowth was stunted in adult virgin 10-week-old female Vim^−/− mice (Fig. 2A,B) compared with the wild-type animals. In laboratory mice, sexual maturity and mammary gland maturation is reached by 8 weeks after birth (Green and Witham, 2009), suggesting that the incomplete fat-pad invasion, observed in 10-week-old Vim^−/− mice, could be due to delayed outgrowth. Accordingly, in Vim^−/− mice, the mammary glands appeared to fully fill the fat pad by 15 weeks of age (Fig. 2A,C). Mammary ductal outgrowth and active cell division occur at the tips of the ducts termed terminal end buds (TEBs) (Hinck and Silberstein, 2005). In normally developing mature mammary glands, TEBs are no longer evident after ∼10 weeks (Fig. 2D,E). However, and in line with delayed outgrowth, the Vim^−/− mammary glands were demonstrated to contain significantly more TEBs at 10 weeks of age (Fig. 2D) with some TEBs still visible at 15 weeks of age (Fig. 2E). Taken together, these data demonstrate that mammary ductal outgrowth is significantly delayed but not terminally impaired in the absence of vimentin.

Establishment of correct epithelial polarity is crucial for normal mammary gland development and is regulated by the coordinated actions of adhesion receptors and the cell cytoskeleton. Given that vimentin interacts with integrins and regulates focal adhesion turnover (Kim et al., 2016; Kreis et al., 2005; Liu et al., 2015), we wanted to investigate whether the lagging mammary gland outgrowth is linked to perturbed organization of the mammary bilayer in the Vim^−/− mice. Although normal mammary epithelial polarization was observed in adult Vim^−/− mammary ducts based on the distribution of established markers: keratin 8 (Krt8; luminal) and keratin 14 (Krt14; basal) (Fig. S1C), the lumens of Vim^−/− ducts appeared to be larger compared with those of age-matched 15-week-old wild-type mice (Fig. 2F,G). For comparable quantification, the measurements were restricted to the relative area of the lumen in perpendicular cross-sections of the smaller ducts at corresponding areas distal to the inguinal lymph node (Fig. 2G).

Next, terminal differentiation of the Vim^−/− mammary gland was investigated at the 15th day of pregnancy (P15) and the first day of lactation (L1) (Fig. 2H,I). As expected, as Vim^−/− mice are able to lactate and feed their offspring, vimentin deficiency did not impair lobulo-alveologenesis induced in response to pregnancy (Fig. 2H,J) and lactation (Fig. 2I,K). Quantification of the epithelial content in wild-type and Vim^−/− tissue sections of the L1 mammary gland demonstrated no difference in alveolar density (Fig. 2L). These observations indicate that the functional maturation of the mammary gland is not affected in the absence of vimentin.

To investigate whether basal epithelial vimentin expression (Fig. 1) and the lagging ductal outgrowth in the Vim^−/− mammary gland (Fig. 2) are related to epithelial MaSC/progenitor cell functions, experiments addressing epithelial regeneration capacity were conducted. Both human and mouse MaSCs can form mammospheres in suspension (Dontu et al., 2003; Liao et al., 2007) and the ability of cells to generate mammospheres reflects the number of self-renewing, regenerative MaSCs within the cell population (Liao et al., 2007). We isolated mouse mammary epithelial cells (MMECs) from 15-week-old Vim^−/− and wild-type female mice and investigated the ability of MMECs to grow as mammospheres. To this end, equal numbers of Vim^−/− and wild-type MMECs were seeded at low density in low attachment conditions. Although the size of Vim^−/− mammospheres was comparable with wild type, the number of mammospheres formed by Vim^−/− MMECs was significantly lower (Fig. 3A-C). These data suggested that primary Vim^−/− MMECs may contain fewer MaSC cells compared with their wild-type counterparts.

To evaluate how epithelial vimentin affects mammary gland regeneration in vivo, we grafted adult wild-type or Vim^−/− mammary gland pieces into epithelium-free fat pads of prepubertal wild-type recipient mice. Vimentin deficiency did not influence the outgrowth from primary transplants when the growth area was quantified 10 weeks post-transplantation (Fig. 3D,E). These data could not be directly compared with differential rates of normal mammary gland development in wild-type and Vim^−/− mice at the age of 10-15 weeks (Fig. 2A-C) as outgrowth from the transplants was much slower. Interestingly, when the primary transplants with noticeable epithelial growth (based on whole-mount analysis of the adjacent fat pad; Fig. S2C) were further cut into pieces and re-transplanted in secondary recipients, Vim^−/− transplants demonstrated a substantial loss in growth capacity 15 weeks post-transplantation (Fig. 3F,G). Additionally, the take on rate demonstrating the fraction of transplants that initiated growth was only compromised at the second round of transplantation (Fig. 3H). Successful secondary transplantation was verified in both wild-type and Vim^−/− samples by the presence of carmine-stained tissue in the cleared fat pads of the recipient. The data presented here suggest that vimentin plays a functional role in the regulation of regenerative capacity in the mammary epithelium in vivo.

Next we tested the phenotype of Vim^−/− isolated epithelial cells in vitro. MMECs cultured in laminin-rich reconstituted basement membrane (rBM) display features of the in vivo mammary epithelium, including the formation of acini-like organoids with a hollow lumen and apicobasal polarity. Vim^−/− MMECs seeded as single cells in three dimensional (3D) rBM formed organoids with normal epithelial polarity (Fig. 4A). There was no significant increase in MMEC cell death observed in vitro in 3D organoid cultures (Fig. S3C, lower panel) or by flow cytometry of isolated bMMECs (Fig. S3E). In contrast, proliferation of Vim^−/− MMECs was reduced in vitro (Fig. S3C, upper panel, Fig. S3D), in accordance with the delayed ductal outgrowth rate observed in vivo (Fig. 2A,B). Interestingly, the number of Slug-positive bMMECs (Guo et al., 2012) was significantly reduced in the organoids (Fig. 4B,C) and in adult Vim^−/− mammary gland tissue sections in vivo (Fig. 4D,E). These data suggest that vimentin could regulate the basal MaSC/progenitor cell population in the mouse mammary gland and the lack of vimentin could be reflected in an unbalance of the basal and luminal cell populations.

To evaluate the basal to luminal MMEC ratio more quantitatively, we isolated MMECs from 15-week-old wild-type and Vim^−/− mice, and surface labelled single-cell suspensions to detect CD24 expression in combination with CD29 (integrin beta 1) or CD49f (integrin alpha 6). From lineage-negative cells (CD31^negCD45^neg), the proportion of basal (CD24^intCD29⁺, CD24^intCD49f⁺) and luminal (CD24^hiCD29^neg, CD24^hiCD49f^neg) MMECs was then quantified by flow cytometry (Fig. 4F,G) (Taddei et al., 2008). Vimentin deficiency led to a significantly (P<0.05) or almost significantly (P=0.05) reduced proportion of bMMECs with both the employed labelling strategies (Fig. 4F,G), as well as with an alternative basal/mature luminal epithelial labelling against CD61 (integrin beta 3) in combination with CD29 and CD49f (Fig. S3A,B). Thus, the abundance of bMMECs harbouring the MaSC/progenitor cells is diminished in the absence of vimentin.

The functional outcome of a diminished basal to luminal cell ratio in the Vim^−/− epithelial cells was also evident in the organoids. The Vim^−/− MMECs formed organoids with a lumen significantly more frequently than did the wild-type organoids (P=0.0008; Fig. 4H,I). This observation is likely related to the reduced proportion of basal epithelial cells in the Vim^−/− mammary gland. Luminal cells predominantly form hollow acini in rBM, whereas the basal cell population tends to form a more heterogeneous array of structures, including ductal forms and solid spheroids (Shackleton et al., 2006; Stingl et al., 2006). Interestingly, this in vitro phenotype resembles the enlarged mammary epithelial lumen in Vim^−/− tissue sections in vivo (Fig. 2F,G). Together, our data demonstrate that vimentin expression is supportive of the basal mammary epithelial cell compartment and therefore lack of vimentin is reflected in reduced regeneration capacity of the mouse mammary epithelium.

The existence and regulation of stem cell populations has been a matter of intense research in the context of breast cancer. Previously, vimentin expression was shown to increase during serial tumoursphere culture of several breast cancer cell lines (Borgna et al., 2012), and increased vimentin expression was demonstrated in a low adherent and more metastatic side population of MDA-MB-231 human breast cancer cells (Morata-Tarifa, 2016). To investigate whether vimentin regulates the stem cell-associated features of the triple-negative and mesenchymal-like MDA-MB-231 breast cancer cells, vimentin was silenced with an siRNA smartpool (Fig. 5A), previously reported to be specific for vimentin with no evidence of off-target effects using gold standard controls (Virtakoivu et al., 2015). CD49f and CD61 have been described as markers of breast cancer stem cells (Brooks et al., 2016; Vaillant et al., 2008). Interestingly, vimentin knockdown resulted in significant downregulation of CD49f and CD61 cell surface expression (Fig. 5B,C). Importantly, in low attachment assays, MDA-MB-231 tumoursphere formation was reduced in the absence of vimentin, based on the number of spheres (Fig. 5D,E) and on total protein content of spheres generated by equal numbers of seeded vimentin-silenced or control cells (Fig. 5F). These results indicate that vimentin influences tumoursphere formation capacity and the expression of integrin adhesion receptors in breast cancer cells in vitro.

The above data suggest that vimentin expression is coupled to supporting the bMMEC compartment in vitro and in vivo. To specifically address the role of vimentin in bMMECs, MMECs isolated from 15-week-old wild-type or Vim^−/− mice were sorted by FACS as in Fig. 4C (CD31^negCD45^negCD24^intCD49f⁺), dead cells excluded and the resulting bMMEC population cultured as 3D organoids. Vim^−/− bMMECs demonstrated reduced organoid size (Fig. 6A,B) and number (Fig. 6A,C) compared with wild-type bMMECs, while epithelial polarity appeared unaltered (Fig. 6D). Together, these data further demonstrate the requirement for vimentin in the growth of bMMECs.

To investigate this further, FACS-sorted bMMECs from wild-type and Vim^−/− mice were subjected to global gene expression analysis by mRNA sequencing. Expression of basal-specific genes (Krt14, Krt5) could clearly be detected, while luminal maker gene expression (Krt18, Krt8) was absent (Fig. 6E). Vimentin deficiency of Vim^−/− bMMECs was also confirmed by mRNA and protein analysis (Fig. S3F,G). Ingenuity pathway analysis of the normalized gene expression dataset identified several differentially expressed canonical pathways, with the most significant ones regulating cell adhesion and diapedesis, embryonic stem cell pluripotency, cancer cell signalling and growth, and Wnt/β-catenin signalling (Fig. 6F). Of the 69 differentially expressed (DE) genes between wild-type and Vim^−/− bMMECs (FC>2, P>0.01) 32 were downregulated and 37 upregulated (Fig. 6G, Table S1). The top KEGG terms enriched in the DE list included Cell-adhesion molecules, Pathways in cancer, Leukocyte transendothelial migration, Hedgehog signalling pathway and Wnt signalling pathway (Table S2). Thus, vimentin significantly influences multiple cell fate regulatory pathways either directly or indirectly, and the integrated effect of these alterations is likely to be reflected in the functional deficiencies observed in the Vim^−/− mouse mammary gland in vivo. This would be fully concordant with the established role of vimentin as a hub for multiple signalling pathways (Eriksson et al., 2009; Ivaska et al., 2007).

Here, we have investigated for the first time the functional consequences of vimentin loss for mammary gland biology. These studies were prompted by the established role of vimentin in epithelial-to-mesenchymal transition (EMT) (Mendez et al., 2010), and the strong links between EMT-inducing factors and the self-renewing capacity of the mammary epithelium (Asiedu et al., 2014; Ye et al., 2015). We have earlier demonstrated that vimentin regulates gene expression and is required for EMT induced by Ras, Slug and TGFβ in cancer cells (Virtakoivu et al., 2015; Vuoriluoto et al., 2011). In addition, we have defined a signal-integrating function for vimentin on the ERK-Slug-EMT axis where vimentin contributes to EMT progression (Virtakoivu et al., 2015). Here, we identify vimentin as a regulator of mammary gland development. In the absence of vimentin the outgrowth of mouse mammary ducts is significantly delayed in vivo. The Vim^−/− mammary gland is characterized by fewer basal epithelial cells, and global gene expression profiling on the affected bMMEC compartment reveals a connection between vimentin expression and many key cellular signalling pathways linked to development and stemness. Furthermore, we demonstrate that vimentin also supports the formation of breast cancer tumourspheres, indicative of a role in regulating cancer stem cell capacity (Fig. 5D-F).

Vimentin, like the other IF proteins, provides different cell types with necessary mechanical stability, and mutations in IF genes are linked to many tissue-specific diseases (Toivola et al., 2015). However, it is becoming increasingly evident that the function of IFs extends well beyond structural support. For example vimentin has been implicated in regulation of signalling, protein trafficking and gene expression (Antfolk et al., 2017; dos Santos et al., 2015; Ivaska et al., 2005; Virtakoivu et al., 2015). In addition, especially dynamic biological processes linked to cell motility, such as actin dynamics (Jiu et al., 2015, 2017), cell spreading and polarization (Leduc and Etienne-Manneville, 2017; Mendez et al., 2010), are supported by vimentin in different cell types and tissues. Therefore, the observed phenotypes in Vim^−/− mice are likely to be the net outcome of several slightly defective processes normally supported by vimentin.

In this study, we focused on investigating the previously unknown role of vimentin in basal mammary epithelium. As vimentin deficiency in the mammary gland stroma may also contribute to the mammary gland phenotype of Vim^−/− mouse, experiments were conducted with epithelial transplants in vivo and isolated MMECs in vitro. We show that proliferation of Vim^−/− bMMECs is reduced in vitro and the number of basal epithelial cells is significantly reduced in Vim^−/− mammary epithelium. These phenotypes could be linked to the role of vimentin as a signal-integrating hub through its interactions with phosphorylated kinases and even transcription factors (Eriksson et al., 2009; Ivaska et al., 2007; Virtakoivu et al., 2015). The altered wiring of signalling networks is reflected in the gene expression changes detected in Vim^−/− bMMECs. Interestingly, signalling pathways linked to adhesion and pluripotency were affected in Vim^−/− bMMECs. Although we currently lack the technical capability to image the collective migration of MMECs in the developing mammary gland in vivo, it would be interesting to investigate whether the established functions of vimentin in cell migration and adhesion turnover are also contributing to the delayed ductal outgrowth in vivo.

Interestingly, the vimentin null glands were defective only during the ductal outgrowth phase. Terminal differentiation of the mammary epithelium during pregnancy and lactation were not impaired in the absence of vimentin. This observation is in agreement with the capacity of the Vim^−/− mice to lactate and feed their offspring, and reflects the high redundancy involved in securing this vital function. One of the obvious phenotypes was the presence of enlarged lumens in the adult Vim^−/− mouse mammary glands. This could be related to the reduced basal cell amount and regenerative potential in Vim^−/− mammary gland (Fig. 3). The enlarged lumen in Vim^−/− mouse mammary ducts resembles a benign breast condition called ductal ectasia, the dilation of mammary ducts (Rahal et al., 2011) that in humans and mice is related to ageing. In normal mice, ductal ectasia occurs at about two years of age, when progenitor activity is already reduced (Jackson et al., 2015). Increased levels of mammary stem/progenitor cells inhibit ectasia in aged mice (Jackson et al., 2015), suggesting that the accelerated occurrence of ectasia in the Vim^−/− mice could also be related to reduced self-renewal capacity and MaSC content. Previous publications have described MaSC gene expression profiles for the adult mouse mammary gland (Lim et al., 2010; Soady et al., 2015; Stingl et al., 2006). Interestingly, a fairly large fraction (22%) of the DE genes in Vim^−/− bMMECs (Abca8a, Ankrd1, Aqp1, Cdh5, Cldn5, Emcn, Fabp4, Foxj1, Gfra2, Gimap6, Ntn4, Pamr1, Pecam1 and Sfrp1) were among the MaSC-enriched genes recently reported by Soady et al. (2015). Together with the reduced MaSC-enriched bMMEC population, reduced mammary epithelial regenerative capacity in vitro and in vivo, and altered mammary gland development in Vim^−/− mice, these gene expression changes provide strong evidence for vimentin being a regulator of mouse mammary gland stem/progenitor cells.

In breast carcinomas, vimentin expression is most prevalent in the triple-negative subtype (61% vimentin positive) and vimentin positivity correlates strongly with nuclear Slug expression (Virtakoivu et al., 2015). Accordingly, vimentin expression is typically associated with high tumour invasiveness and resistance to chemotherapy (Korsching et al., 2005; Yamashita et al., 2013). It has also been suggested that vimentin-expressing breast carcinomas could be derived from the bipotent breast progenitor cells (Korsching et al., 2005). Our data demonstrating that vimentin influences tumoursphere formation in breast cancer cells supports this hypothesis. Recently, vimentin knockdown in orthotopic breast cancer allografts in a hyperinsulinemia mouse model was shown to reduce pulmonary metastasis (Zelenko et al., 2017). Together, these findings suggest that vimentin contributes to breast cancer initiation and progression.

Our data demonstrate that vimentin regulates the rate of mammary ductal outgrowth, a process involving mammary gland stem/progenitor cells, during mouse puberty, and affects the morphology of mature mammary ducts in vivo. Importantly, vimentin appears to be a central node in cell signalling pathways operating in cancer and normal stem cells. Understanding how vimentin contributes to mammary stem cell regulation can be of importance in developing therapeutic strategies against breast cancer.

Vimentin-deficient homozygote (Vim^−/−) (Colucci-Guyon et al., 1999) and wild-type mice in a mixed 129/Sv×C57BL/6 background were generated through heterozygote and homozygote mating. The PCR genotyping method was used to analyse the genotype of the mice. Age-matched female mice were used in all experiments. Mice were synchronized for oestrus cycle in experiments where individual mice were examined. Otherwise, a minimum of two mice were pooled per MMEC isolation. For timed mating, mice were examined for mating plug appearance and sacrificed on P15 or L1. For qPCR experiments, mammary glands were collected from adult virgin wild-type BALB/cByJ female mice. Mice were housed under standard conditions (12 h light/dark cycle) with food and water available ad libitum. All animal studies were ethically performed and authorized by the National Animal Experiment Board and in accordance with The Finnish Act on Animal Experimentation (Animal licence numbers 7522/04.10.03/2012, ESAVI-5588-04.10.07-2014, ESAVI-9339-04.10.07-2016).

The fourth mammary gland was dissected and placed on an object glass, left to adhere for few minutes, fixed in Carnoy's medium (60% ethanol, 30% chloroform, 10% glacial acetic acid) overnight (o/n) at +4°C, followed by hydration in decreasing ethanol series and staining with carmine alum (0.2% carmine, 0.5% aluminium potassium sulphate dodecahydrate) overnight at room temperature. Samples were dehydrated and cleared in xylene for 2-3 days. Mounting was carried out in DPX Mountant for histology (Sigma) and images were taken using a Zeiss SteREO Lumar V12 stereomicroscope (NeoLumar 0.8× objective, Zeiss AxioCam ICc3 colour camera). All images per gland were combined automatically into a mosaic picture with PhotoShop. Ductal outgrowth was analysed in ImageJ by measuring the distance of the ductal tree from the inguinal lymph node into the fat pad (adult mice) or by measuring the area covered by the ductal tree (transplantation).

Cleared fat pad transplantation was conducted as described previously (Peuhu et al., 2017). Briefly, mammary gland pieces (∼1 mm³) from 15-week-old wild-type or Vim^−/− female donor mice were transplanted under isoflurane anaesthesia and analgesia (Temgesic, Rimadyl) to the fourth fat pad of 3-week-old wild-type hosts (wild-type and Vim^−/− transplants on each side of the host) after clearing the fat pad up to, and including, the lymph node. The removed part of the fat pad was fixed and stained to confirm complete removal of the recipient mouse mammary epithelium. Growth of transplants was analysed after 10 weeks by Carmine Alum staining of the fourth mammary gland whole mounts. For secondary transplantation, one mouse with wild-type and Vim^−/− primary transplants in each fat pad was chosen randomly, both fat pads were excised, and an area close to the site of primary transplantation in each fat pad was cut into five pieces. These pieces were further transplanted in the fourth fat pads of 3-week-old wild-type recipient mice (wild-type and Vim^−/− transplants on each side of the host). Growth of transplants was analysed after 15 weeks by Carmine Alum staining of the fourth mammary gland whole mounts. Growth take on rate was calculated from all the transplant samples.

Female mice (15-18 weeks old) were sacrificed and the inguinal lymph node was removed. Tissues were minced with a scalpel into small pieces and the homogenate was collected into a collagenase solution [DMEM/F12 media containing 2.5% FCS, 5 μg/ml insulin, 50 μg/ml gentamicin and 200 mM glutamine and 20 mg of collagenase XI (Sigma) for 1 g of tissue] and incubated at 37°C with agitation (120 rpm) for 2-3 h. The resulting cell suspension was centrifuged at 400 g and resuspended in DMEM/F12 medium (containing 1% of penicillin/streptomycin and 50 μg/ml gentamicin) and DNase I (20 U/ml) for 3 min. The sample was then subjected to five pulse centrifugation steps at 400 g; each time the supernatant was carefully removed and the pellet was resuspended in 10 ml of DMEM/F12 medium. Next, organoids were incubated with Accutase (StemCell Technologies) to obtain a single-cell suspension and cells were pipetted through a 70 µm cell strainer (BD Biosciences). Cells were used for organoid cultures, mammosphere assays and flow cytometry.

MDA-MB-231 cells were cultured in DMEM supplemented with 10% FBS, 1% L-glutamate and non-essential amino acids. Lipofectamine 3000 transfection reagent (Invitrogen) was used for siRNA transfections and the transfection was carried out according to the manufacturer's instructions and as previously described (Virtakoivu et al., 2015). Silenced cells were used for experiments 48-72 h after transfection. Specific silencing of vimentin by single siRNA oligos in the vimentin siRNA Smartpool was previously validated by western blotting (Virtakoivu et al., 2015).

The following antibodies were used in the study. For immunofluorescence and western blotting, Acta2 (alpha smooth muscle actin, clone 1A4, Sigma, 1:500), Krt8 (keratin 8, clone Troma-I, Hybridoma Bank, 1:1000), Krt14 (keratin 14, Covance, 1:500), Slug (Snai2, C19G7, Cell Signalling, 1:50), GAPDH (5G4 Mab 6C5, Hytest, 1:3000) and vimentin (VIM, V9, Santa Cruz, 1:100; D21H3, Cell Signalling, 1:100). For flow cytometry of mouse cells, CD45-Pacific Blue (clone 30-F11, 1:100), CD31-Pacific Blue (clone 390, 1:50), CD29-A488 (Itgb1, clone HMβ1-1, 1:50), CD49f-A488 (Itga6, clone GoH3, 1:20) (all from BioLegend) and CD24-APC (clone M1:69; BD Biosciences, 1:100) antibodies were used. For flow cytometry of human cells, CD61 (ITGB3, MCA728, Serotec, 1:50) and CD49f (ITGA6, MCA699, clone GoH3, Serotec, 1:50) antibodies were used. AlexaFluor-conjugated (Life Technologies, 1:300) and HRP-linked (GE Healthcare, 1:5000) secondary antibodies against rat, rabbit and mouse IgG were used in immunolabelling, flow cytometry and western blotting.

For organoid and mammosphere cultures, cells were counted and equal amounts of wild-type and Vim^−/− cells were plated in 96-well low-attachment plates (Corning) and allowed to proliferate for 4-9 days. For organoid assays the following medium was used: Epicult Base medium (StemCell Technologies) supplemented with Epicult supplements, 10 ng/ml EGF, 4 µg/ml heparin, 10 ng/ml bFGF, 5 µM Y-27632 ROCK inhibitor, 5% matrigel and 5% FCS. For mammosphere assays, the same medium was used without ROCK inhibitor, Matrigel and FCS.

When using vimentin or control siRNA-silenced MDA-MB-231 cells for the tumoursphere assays, more siRNA mix was added to the cells on day 3 when the cells were plated on low-adhesion plates in serum-free medium. Formed tumourspheres were manually counted and images taken using an EVOS Cell Imaging System (ThermoFisher Scientific). All wells/cell type/cell amount were pooled together and lysed for western blotting. Organoids were stained with Calcein AM cell-permeant dye (Life Technologies) according to the manufacturer's instructions or were fixed and stained with DAPI, after which the samples were imaged with a Nikon Eclipse wide-field microscope and the number and size of organoids quantified in ImageJ.

For immunofluorescence staining, the organoids were gently collected and centrifuged at 90 g for 5 min, resuspended into 10-20% matrigel in PBS and plated for 3-4 h at 37°C on matrigel-coated eight-well Ibidi µ-slide wells (Ibidi GmbH). Next, organoids were fixed and permeabilized with 1% PFA/0.4% Triton-X in PBS for 15 min at 37°C, washed with PBS and blocked with 10% horse serum, and 0.1% Tween20 in TBS (TBST) for 1.5-2 h at room temperature. Organoids were then incubated with Alexa488-conjugated phalloidin (Sigma-Aldrich) and primary antibodies diluted in blocking buffer overnight at 4°C, washed three times with TBST and labelled with fluorochrome-conjugated secondary antibodies in blocking buffer for 3 h at room temperature. Nuclei were stained with Hoechst or DAPI. Organoids were washed with TBST, and either left in PBS or mounted with Vectashield and dried for 1 h at 37°C. Imaging was carried out using a Carl Zeiss LSM780 laser-scanning confocal microscope or 3i CSU-W1 spinning disk confocal microscope (Intelligent imaging innovations) with Hamamatsu CMOS Orca Flash 4, and quantification with ImageJ software.

Formalin-fixed paraffin wax-embedded human mammary gland tissues were collected from the archives of the Department of Pathology, Helsinki University Central Hospital, Helsinki, Finland. An Institutional Review Board of the Helsinki University Central Hospital approved the study. Mammary gland tissues destined for formalin or Carnoy's medium fixation and paraffin wax embedding were collected from 10- to 15-week-old female mice of the indicated genotype. Tissue sections were deparaffinized and rehydrated. Epitope unmasking was performed in citrate buffer using 2100 Antigen retriever (Aptum, UK). Samples were blocked with 0.5% FCS in PBS for 45 min and incubated with primary antibodies overnight in blocking buffer, followed by washing with PBS and incubation with fluorochrome-conjugated secondary antibodies for 2 h at room temperature. Samples were washed, stained with DAPI (Life Technologies) and mounted in Mowiol containing DABCO (Sigma) anti-fading reagent. Imaging was conducted using a Zeiss Axiovert 200 M inverted wide-field microscope (Haematoxylin-Eosin), a 3i CSU-W1 spinning disk confocal microscope with Hamamatsu CMOS Orca Flash 4 or with a Carl Zeiss LSM780 laser scanning confocal microscope.

Cells were suspended into Tyrodes buffer [10 mM HEPES-NaOH (pH 7.5), 137 mM NaCl, 2.68 mM KCl, 1.7 mM MgCl₂, 11.9 mM NaHCO, 3.5 mM glucose, 0.1% BSA] and ∼5×10⁵ cells were used per staining. In the case of MMEC cells, directly conjugated antibodies were added and incubated at 4°C for 30 min, washed twice and fixed with 4% PFA. For staining the vimentin-silenced MDA-MB-231 cells, cells were first fixed with 4% PFA, washed with PBS and suspended into PBS containing 1% BSA. Primary antibodies were incubated for 30 min followed by washes and incubation with Alexa Fluor 488/647 secondary antibodies (1/400). Samples were run with BD LSR Fortessa flow cytometer (BD Biosciences) and analysed with Flowing software (Cell Imaging Core, Turku Centre for Biotechnology).

Samples were prepared as previously described (Di-Cicco et al., 2015; Peuhu et al., 2017). Single cells were prepared from inguinal mammary glands taken from virgin adult BALB/cByJ females according to a detailed protocol described previously (Di-Cicco et al., 2015). Freshly isolated cells were incubated at 4°C for 20 min with the conjugated antibodies. Labelled cells were sorted on a FACSVantage flow cytometer (BD Biosciences) and data were analysed using FlowJo software. CD45⁺ immune cells and CD31⁺ endothelial cells were excluded during the cell-sorting procedure. Mammary gland lineage-specific gene expression (Krt5, Krt18 and Pdgfr) in the purified cell populations was checked by quantitative PCR (qPCR) as reported (Di-Cicco et al., 2015; Peuhu et al., 2017). RNA was reverse-transcribed using MMLV H(-) Point reverse transcriptase (Promega) and qPCR was performed by monitoring, in real time, the increase in fluorescence of the SYBR Green dye in a LightCycler 480 Real-Time PCR System (Roche Applied Science). The primers used for qPCR analysis (Vim and Gapdh) were purchased from SABiosciences/Qiagen.

Freshly isolated wild-type and Vim^−/− MMECs were labelled with viability marker (eBioscience Fixable Viability Dye eFluor 780) and antibodies (CD45-Pacific Blue, CD31-Pacific Blue, CD49f-A488 and CD24-APC). Basal MMECs (Live, Lin^neg, CD24^int, CD49f⁺) were sorted with BD FACSAria IIu directly to RNA extraction lysis buffer (Macherey-Nagel). RNA was isolated and DNAse-treated using NucleoSpin RNA kit 740955.10 (Macherey-Nagel). Total RNA samples were pure, intact and all samples had similar quality. RNA quality was determined using Bioanalyzer 2100 (Agilent). RNA-sequencing libraries were prepared using TruSeq Stranded mRNA Sample Preparation Kit (Illumina) and sequenced with HiSeq 3000 (Illumina) using TruSeq v3 sequencing chemistry. 50 bp single-read sequencing was used, followed by 6 bp index run. Twelve samples were run in one lane. The technical quality of the sequencing run was excellent and the cluster amount was as expected. Greater than 80% of all bases were above Q30. The base calling was performed using standard bcl2fastq software v 2.17(Illumina).

The raw sequencing reads were aligned with Tophat (v. 2.0.1) to mm10 reference genome derived from UCSC database and downloaded from Illumina's iGenomes website (https://support.illumina.com/sequencing/sequencing_software/igenome.html). Gene-wise read counting was carried out using HTSeq (v 0.5.4p3) based on RefSeq gene annotations. Downstream data analysis was carried out with R (v. 3.1) and Bioconductor (v. 2.14). Raw count values were normalized using the TMM method of the edgeR package. The normalized expression values were also summarized to the reads per kilobase of exon per million reads mapped (RPKM), although these values were not used for statistical testing. The R package ROTS was used for performing the statistical testing between the groups. Differentially expressed (DE) genes were filtered based on fold-change (logFC>1) and P value (P<0.01). In addition to applying a cutoff for fold-change and  value, it was also required that the difference in the mean RPKM expression values of the two groups was at least 1. A heat map was generated for the DE genes using the Morpheus matrix visualization and analysis platform (https://software.broadinstitute.org/morpheus/). Functional analysis of DE genes was performed against the KEGG database using topGO and GOstats packages in R/Bioconductor. Terms/pathways including more than 15 or at most 200 genes have been included in the analysis. Normalized expression values were also analysed with Ingenuity Pathway Analysis software (P<0.01 and logFC >1) for canonical signalling pathways.

The immunoblotting was carried out using standard western blotting techniques and Odyssey LICOR imaging system.

Sample size for the studies was chosen according to previous studies in the same area of research. A minimum of three mice were analysed for each genotype for comparison, except for Fig. 2H,J, where glands from two mice for wild type were analysed. No samples were excluded from the analysis, and randomization or blinding was not used for animal studies. Data were assumed to follow a normal distribution. The GraphPad software and Student's t-test (paired or unpaired, two-tailed) and Fisher's exact test were used for statistical analyses. Paired Student's t-test was used if samples with unequal variance were compared. Data are presented in column graphs as mean±s.e.m. and P values. Individual data points per condition are shown when n≤15 and n values are indicated in the figure legends.

We thank J. Jukkala, J. Siivonen, P. Laasola and M. Saari for instrumental technical assistance; Dr H. Hamidi for editing of the manuscript; Dr E. Närva, Dr C. Guzman and Dr G. Jacquemet for critical reviewing of the manuscript; and Dr M.-A. Deugnier and A. Di-Cicco for vimentin qPCR analysis. We also thank Prof. J. Eriksson for sharing the vimentin knockout mouse strain and Dr F. Cheng for assistance in sample preparation. Finnish Functional Genomics Centre (FFGC), The Turku Centre for Biotechnology Cell Imaging Core, University of Turku Central Animal Laboratory and Biocenter Finland are acknowledged for services, instrumentation and expertise.