Functional roles of MMP14 and MMP15 in early postnatal mammary gland development

During postnatal mammary gland development, epithelial ducts mount a cohesive cell invasion program that allows them to penetrate a periductal interstitial matrix populated by fibroblasts and a surrounding adipocyte-rich mesenchyme (Ewald et al., 2008, 2012; Watson and Khaled, 2008). Similar to other epithelial organ systems, branching morphogenesis begins between embryonic day (E) 12 and E15 with the formation of primary buds that undergo reiterative branching into the underlying mesenchyme (Ewald et al., 2008, 2012; Hogg et al., 1983; Watson and Khaled, 2008). Between E16 and E20, mammary epithelial cords give rise to polarized, bilayered tubules composed of inner-facing luminal epithelial cells and basal-oriented myoepithelial cells (Hogg et al., 1983; Sun et al., 2010). In tandem with epithelial development, the surrounding adipose-rich tissue, termed the mammary fat pad, likewise undergoes morphogenesis (Hovey and Aimo, 2010; Inman et al., 2015). Beginning during late gestation (E14 to E18) and continuing through early postnatal development, the fat pad is eventually dominated by committed adipocytes that support epithelial morphogenesis and tissue homeostasis (Wang et al., 2015). At birth, the mammary gland rudiment is a small, simply branched structure that is believed to lie dormant until the onset of puberty (∼3 weeks of age) (Ewald et al., 2008, 2012; Hogg et al., 1983; Watson and Khaled, 2008).

Recent studies emphasize crucial roles for proteolytic enzymes belonging to the matrix metalloproteinase (MMP) family in the tissue remodeling events that are activated during branching morphogenesis, with specific focus on two closely related membrane-anchored MMPs termed MMP14 (MT1-MMP) and MMP15 (MT2-MMP) (Alcaraz et al., 2011; Bonnans et al., 2014; Mori et al., 2009, 2013; Rebustini et al., 2009; Weaver et al., 2014). Consistent with their >50% homology at the amino acid level, both MMP14 and MMP15 are capable of hydrolyzing a wide variety of substrates ranging from cell surface molecules and growth factors to extracellular matrix (ECM) components (Barbolina and Stack, 2008; Itoh, 2015; Rowe and Weiss, 2009). With particular regard to mammary gland morphogenesis, MMP14 has been reported to control branching events by both proteolytic mechanisms, whereby epithelial cells dissolve confronting ECM barriers, as well as proteinase-independent mechanisms by acting as a scaffolding hub for signal transduction cascades that control cell motility and cell sorting (Alcaraz et al., 2011; Mori et al., 2009, 2013; Simian et al., 2001; Weaver et al., 2014). Similarly, MMP15 has been reported to regulate branching morphogenesis and associated proliferative responses as a consequence of its ability to proteolytically remodel the basement membrane and regulate growth factor expression (Hotary et al., 2000; Rebustini et al., 2009). Nevertheless, conclusions regarding the functions of MMP14 and MMP15 in branching morphogenesis largely derive from in vitro and ex vivo models, and the roles that these enzymes play in vivo within the developing mammary gland remain undefined.

Here, we describe a mammary gland branching program that occurs during the first 10 days of early postnatal development and gives rise to an organized network of polarized ductal structures that penetrate a mature adipocyte-populated stroma, offering a robust platform in which to characterize the protease-dependent and protease-independent roles of MMP14 and MMP15 in vivo. Using Mmp14^−/− mice as well as a newly generated Mmp15^−/− mouse line, we provide the first full-scale analysis of the differential functions of these MT-MMPs in early postnatal mammary gland development.

At birth [postnatal day (P) 0], mammary gland rudiments of female mice appear as primitive ductal trees, whose ill-defined branches are found immediately adjacent to the nipple (Fig. 1A,B). In H&E-stained transverse sections, the mammary epithelium is shown to be organized into cohesive structures containing a mixture of solid cords and hollow ducts (Fig. 1C). Immunohistochemical analyses further reveal that epithelial structures are dominated by cells expressing the luminal marker cytokeratin 8 (CK8; or KRT8), with only low levels of CK18, in tandem with a heterogeneous distribution of cells expressing the basal marker CK14 and the absence of CK5 (Fig. 1D) (Ewald et al., 2008, 2012; Sun et al., 2010; Watson and Khaled, 2008). Epithelial cells expressing the myoepithelial marker α-smooth muscle actin (αSMA) (Ewald et al., 2008, 2012; Hogg et al., 1983; Watson and Khaled, 2008) are confined to the basal compartment, but in a scattered fashion that outlines the rudiments (Fig. 1D). In marked contrast, between P5 and P10, the mammary gland undergoes a morphogenic program wherein an established network of hollow ducts extend their length through the surrounding mesenchyme by 2-fold while branch number increases 3-fold and the surface area covered by ductal structures expands 4-fold (Fig. 1A-C). Coincident with this maturation process, expression of both CK8 and CK18 is upregulated in the luminal compartment, while CK14 is confined primarily to basal epithelial cells in association with the continuous circumferential expression of αSMA (Fig. 1D). This early developmental program coincides with epithelial cell proliferation, as evidenced by Ki67 expression in the CK8-positive compartment at P0 as well as P5 (Fig. S1A,B). These changes in epithelial cell and ductal tree organization are complemented by striking changes in the mammary gland fat pad, wherein small lipid-laden adipocytes and adipocyte precursors mature into large, unilocular adipocytes (Fig. 1C). Although the P0-P10 mammary ducts are consistently invested by a laminin- and type IV collagen- rich basement membrane (Fig. 1E), a periductal network of organized type I collagen fibrils does not fully develop until P5-P10 (Fig. 1F). Hence, during the early postnatal period, both the mammary epithelia and stroma activate a morphogenetic program that generates a well-organized network of ductules that predate those more frequently associated with pubertal development (Huebner and Ewald, 2014; Inman et al., 2015; Macias and Hinck, 2012; Watson and Khaled, 2008).

Based largely on in vitro evidence, MMP14 has been proposed to regulate branching morphogenesis by both proteinase-dependent and independent mechanisms (Alcaraz et al., 2011; Mori et al., 2009, 2013; Weaver et al., 2014). Using Mmp14^lacZ/+ mice (Yana et al., 2007), β-galactosidase activity is detected in the branching ductal tree as well as the surrounding mesenchyme at birth and P5 (Fig. 2A). By P10, lacZ-expressing cells delineate and circumscribe the entire neonatal mammary gland, from the ducts closest to the nipple to their terminal ends (Fig. 2A), with transverse sections highlighting MMP14 expression in luminal epithelial cells and myoepithelial cells as well as the surrounding stroma (Fig. 2B). lacZ expression also correlates with an increase in Mmp14 levels in the developing mammary tissue (Fig. 2C).

To first assess the role of mammary epithelial cell-derived MMP14 in vitro, Mmp14^lacZ/+ mammary epithelial organoids were embedded in 3-dimensional (3D) type I collagen hydrogels (Chun et al., 2004; Haslam et al., 2008; Sabeh et al., 2009; Simian et al., 2001). As shown, Mmp14 is expressed both at time 0 and during the formation of an arborized branching network generated in the presence of FGF2 after a 5 day culture period (Fig. 2D) (Haslam et al., 2008; Simian et al., 2001). Consistent with the premise that MMPs are able to drive type I collagen-invasive activity (Alcaraz et al., 2011; Chun et al., 2004; Mori et al., 2009, 2013; Sabeh et al., 2009; Simian et al., 2001; Tang et al., 2013; Weaver et al., 2014), branching morphogenesis is inhibited completely when organoids are cultured in the presence of the pan-specific MMP inhibitor BB-94 (Chun et al., 2004; Sabeh et al., 2009) (Fig. 2E). More importantly, organoids recovered from Mmp14-deleted mice display a complete loss of branching activity (Fig. 2E), a result consistent with the role of MMP14 as the dominant pericellular type I collagenase operative in mammalian cells (Chun et al., 2004; Rowe and Weiss, 2009; Sabeh et al., 2009; Tang et al., 2013).

Nevertheless, an MMP14-dependent block in branching morphogenesis through native type I collagen matrices could also result from protease-independent effects on epithelial cell sorting, motility or metabolism (Mori et al., 2009, 2013; Sakamoto et al., 2014). As such, Mmp14^+/+ and Mmp14^−/− mammary epithelial organoids were alternatively embedded in 3D hydrogels of Matrigel (Ewald et al., 2008, 2012; Lo et al., 2012), a mixture of basement membrane-associated proteins whose proteolytic remodeling is not required to support cell invasion programs (Rowe and Weiss, 2008). As reported previously (Ewald et al., 2008, 2012; Lo et al., 2012), wild-type mammary organoids embedded in Matrigel hydrogels form budding lobular structures in response to FGF2-mediated signaling (Fig. 2F). Inconsistent, however, with the contention that MMP14 regulates morphogenesis via proteinase-independent mechanisms (Mori et al., 2009, 2013), neither Matrigel-embedded organoids cultured in the presence of BB-94 nor organoids recovered from Mmp14^−/− mice display defects in branching activity (Fig. 2F). Hence, in the presence of a permissive ECM environment, MMPs are not required to support morphogenic programs.

Although in vitro models provide useful tools for dissecting structure/function relationships, mounting evidence suggests that isolated epithelial cell-based model systems are unlikely to recapitulate the morphogenic programs occurring in vivo where the functional properties of the surrounding mesenchyme play active roles in branching (Nelson and Larsen, 2015; Varner and Nelson, 2014). As such, we next sought to directly assess requirements for MMP14 during mammary gland morphogenesis in vivo. Although the morbid status of Mmp14^−/− mice and their failure to undergo sexual maturation preclude efforts to analyze late stage mammary gland development (Holmbeck et al., 1999; Zhou et al., 2000), the newly described P0-P10 program potentially provides a unique window of opportunity for such analyses. Interestingly, mammary glands harvested from P0 Mmp14^−/− mice generate branching ductal networks comparable to those of wild-type and Mmp14^+/− littermates, as visualized by whole-mount Carmine staining and quantified by ductal length and branch point number (Fig. 3A,B). More remarkably, at P5 and P10, Mmp14^−/− mice mount a mammary gland branching program similar, if not identical, to those observed in Mmp14^+/+ or Mmp14^+/− littermates (Fig. 3C-F). Further immunohistological assessments of Mmp14^−/− mammary gland sections revealed no differences in the expression or organization of CK18, CK14 or αSMA from P0 through P10 (Fig. 4A,B). Likewise, epithelial cell organization, as assessed by the apical distribution of the tight junction protein zonula occludens 1 (ZO-1; or TJP1) or the adherens junction marker E-cadherin (Ewald et al., 2008, 2012; Inman et al., 2015), is indistinguishable between Mmp14^+/+, Mmp14^+/− and Mmp14^−/− mice at P10 (Fig. 4B). Although MMP14 has previously been implicated in the proliferation-associated signaling cascades that drive branching morphogenesis (Gutierrez-Fernandez et al., 2015; Riggins et al., 2010), neither Ki67 expression, TUNEL staining nor senescence is affected in Mmp14^−/− glands (Fig. 4C,D). Finally, deleting Mmp14 did not discernably alter basement membrane or interstitial matrix assembly, as reflected in comparable type IV collagen and laminin staining (Fig. 4E) or levels of type I collagen as assessed by immunofluorescence and Sirius Red staining (Fig. 4F).

Neonatal mammary gland branching morphogenesis also coincides with the increased expression of two secreted members of the MMP family that have previously been implicated in mammary gland branching and development, namely MMP2 (gelatinase A) and MMP3 (stromelysin 1) (Correia et al., 2013; Kessenbrock et al., 2013; Lochter et al., 1997; Wiseman et al., 2003; Witty et al., 1995) (Fig. S2A,C). Further, MMP14 can activate MMP2 in vivo, a proteinase thought to mediate MMP14 function and to compensate partially for Mmp14 deficiency (Itoh, 2015; Oh et al., 2004). Nevertheless, as previously described (Wiseman et al., 2003), Mmp2^−/− mammary glands develop comparable branching structures to those of littermate controls in pre-pubertal mice (Fig. S2B). Further, although recent reports have identified crucial protease-dependent as well as protease-independent roles for MMP3 in mammary gland development (Correia et al., 2013; Kessenbrock et al., 2013; Lochter et al., 1997; Wiseman et al., 2003; Witty et al., 1995), Mmp3^−/− mammary glands establish branching structures that are comparable in length and complexity to those of Mmp3^+/+ and Mmp3^+/− littermates, with a normal distribution of luminal epithelial and myoepithelial cell populations (Fig. S2D,E).

Membrane-anchored MMP15 is structurally related to MMP14, is enriched in the mammary epithelium (Szabova et al., 2005) and has recently been implicated in branching morphogenesis ex vivo (Rebustini et al., 2009). Using Mmp15^lacZ knock-in mice (Fig. 5A), Mmp15 expression can be detected throughout the mammary epithelial cell compartment at birth as well as through P10 (Fig. 5B,C), a result further corroborated by upregulated Mmp15 levels during this period (Fig. 5D). Mmp15-targeted mice were therefore employed to address the requirement for Mmp15 in branching morphogenesis, wherein exons 4 and 5 of the catalytic domain were flanked by loxP sites (Fig. S3A) and deleted by breeding Mmp15^flox/flox mice with EIIA-Cre mice. Crosses between Mmp15^+/− mice generated Mmp15^+/+, Mmp15^+/− and Mmp15^−/− mice in normal Mendelian ratios, yielding a total of 68 Mmp15^+/+, 146 Mmp15^+/− and 70 Mmp15^−/− mice among 284 total offspring; and Mmp15^−/− mice remained similarly viable throughout adulthood. Importantly, germline deletion of Mmp15 did not affect Mmp14 expression (Fig. 5E), ruling out any compensatory interactions between these closely related MMP family members.

In vivo assessments of Mmp15^+/+, Mmp15^+/− and Mmp15^−/− mammary glands revealed correspondingly similar branching structures, as assessed by whole-mount Carmine staining at birth as well as P10 and quantified by ductal penetration or total branch points (Fig. 5F-I). Similarly, explanted Mmp15^−/− and Mmp15^+/+ mammary epithelial cell organoids in 3D type I collagen or Matrigel mounted comparable in vitro branching programs (Fig. 5J,K). Further, as observed in Mmp14^−/− mice, Mmp15^−/− mammary epithelial ducts display an unaltered pattern of CK18, CK14, αSMA and E-cadherin expression (Fig. S3B-D). Although MMP15 has been reported to regulate the proliferative and apoptotic responses required for epithelial morphogenesis (Abraham et al., 2005; Rebustini et al., 2009), Ki67 levels are unchanged in the absence of Mmp15, with apoptosis assessments revealing no significant differences across genotypes (Fig. S3E). Similarly, although MMP15 has been identified as a crucial basement membrane remodeling enzyme during branching morphogenesis ex vivo (Rebustini et al., 2009) and a type I collagenolytic enzyme in vitro (Chun et al., 2004; Hotary et al., 2000), neither type IV collagen, laminin nor type I collagen levels are affected in the absence of Mmp15 (Fig. S2F).

In order to probe for unanticipated functional roles of MMP14 and MMP15 in the early postnatal mammary gland in an unbiased fashion, we analyzed the transcriptome of the ductal networks and associated stroma in wild-type versus null mice. In Mmp14^−/− mice, gene expression profiling revealed >250 differentially expressed genes relative to Mmp14^+/+ mammary tissue (using a 2-fold enrichment cutoff), with functional annotation clusters bridging 16 different gene ontology (GO) categories, including many associated with the epithelial cell compartment, such as cell adhesion, endocytosis, responses to endogenous stimuli, and G protein-coupled signaling cascades [DAVID bioinformatics resource; P-value (EASE score) <0.05; Fig. 6A]. Of note, gene expression profiling did not uncover significant changes in the proliferation-associated or senescence-associated signaling cascades highlighted in earlier work (Gutierrez-Fernandez et al., 2015; Riggins et al., 2010), despite undetectable levels of Mmp14 expression in Mmp14^−/− mammary gland tissue. However, given the number of genes associated with epithelial cell adhesion complexes and cytoskeletal organization [e.g. calmin, plakophilin 2, folliculin (Arimoto et al., 2014; Khabibullin et al., 2014; Loo et al., 2013)], we examined mammary ducts recovered from P10 Mmp14^+/+ and Mmp14^−/− mice by transmission electron microscopy (TEM). Interestingly, Mmp14^−/− ducts form more intimate cell-cell junctions, with a marked increase in the number of apical microvilli (Fig. S4A,B). By contrast, the transcriptome of Mmp15^−/− mammary tissues uncovered only ∼30 differentially expressed genes whose classification lacks any major associations with mammary epithelium-related GO categories (Fig. S5A). Accordingly, TEM analyses of Mmp15^−/− ducts appear normal (Fig. S5B,C). Hence, although these data confirm active roles for MMP14 and MMP15 in early postnatal mammary gland development, the transcriptional changes recorded do not significantly impact on the ability of the organizing ductal network to penetrate the surrounding mesenchymal tissues.

Independent of the roles played by MMP14 or MMP15 in the mammary gland epithelial compartment, analyses of differentially expressed transcript profiles indicate that both MT-MMPs exert more significant effects on the developing mammary gland stroma. In Mmp14^−/− mice, the downregulated GO categories with the highest enrichment scores are tightly linked with the mammary gland stromal fat pad, with decreased expression of multiple transcripts associated with sterol metabolism, lipid biosynthesis, fatty acid metabolism and adipocyte maturation (Fig. 6B). In tandem with these changes, mammary gland adipocytes display a marked 50% reduction in overall size, as evidenced by H&E staining and TEM analyses, and as quantified by mean adipocyte diameter (43.3±1.01 µm in Mmp14^+/+ versus 22.8±1.60 µm in Mmp14^−/− mammary glands; n=3 per genotype, P=0.0004) (Fig. 6C, Fig. S6A). In an almost diametrically opposed fashion, the metabolic genes repressed in Mmp14^−/− tissue are either upregulated or unchanged in Mmp15-targeted mice, as most clearly evidenced by the differential expression of the adipokine leptin (Fig. 6D), an adipocyte-derived secreted molecule responsible for regulating tissue metabolism and energy homeostasis (Wang et al., 2015; Wu et al., 2013). Unexpectedly, transcriptome analysis of Mmp15^−/− mammary tissue further reveals a marked increase in a core suite of positive regulators of brown or induced brown (beige) adipocytes (Fig. S5A) – thermogenic adipocytes that can be identified based on their restricted expression of uncoupling protein 1 (UCP1) (Wang et al., 2015; Wu et al., 2013). Indeed, whereas adipocyte size is unaffected in many regions of the Mmp15^−/− mammary glands, large clusters of small, multilocular adipocytes – the hallmark of beige/brown fat – are found throughout the knockout tissues in four of six animals characterized (Fig. 6E, Fig. S6B). As confirmed by qPCR, Mmp15^−/− mammary tissues display a 4- to 5-fold increase in Ucp1 transcript levels, as well as increased expression of multiple genes required for mitochondrial biogenesis, adipocyte thermogenesis and Ucp1 induction, including Dio2, Fabp3, Prdm16 and Pgc1a (Ppargc1a) (Cohen et al., 2014; Marsili et al., 2011; Vergnes et al., 2011; Wu et al., 2013) (Fig. 6F). Further, transverse sections of Mmp15^−/− mammary glands document increased UCP1 protein levels as well as a marked increase in mitochondrial number, size and cristae (Fig. 6G, Fig. S6C-E), hallmarks of beige/brown adipocytes (Gouon-Evans and Pollard, 2002; Harms et al., 2014; Liesa and Shirihai, 2013). Finally, corroborating the divergent roles of MMP14 and MMP15 in mammary fat pad development, brown/beige-associated transcripts and protein levels remain unchanged or are markedly reduced in Mmp14^−/− glands (Fig. 6G,H). Hence, whereas MMP14 promotes white fat-associated adipogenesis in the developing mammary gland, MMP15 serves as an endogenous suppressor of beige/brown fat production.

MMP14 and MMP15 have been reported to play diverse roles in a range of morphogenesis-associated processes, including cell adhesion and motility, invasion, proliferation, energy metabolism, senescence, the proteolytic processing of growth factors, cytokines and chemokines, as well as ECM remodeling (Ager et al., 2015; Alcaraz et al., 2011; Barbolina and Stack, 2008; Bonnans et al., 2014; Chun et al., 2004; Fahlman et al., 2014; Fu et al., 2013; Itoh, 2015; Kajita et al., 2001; Koshikawa et al., 2000; Koziol et al., 2012; Mori et al., 2009, 2013; Rebustini et al., 2009; Sabeh et al., 2009; Sakamoto et al., 2014; Shimizu-Hirota et al., 2012; Simian et al., 2001; Tang et al., 2013; Taylor et al., 2015; Weaver et al., 2014; Yana et al., 2007). Consistent, in part, with previous reports linking MT-MMPs with epithelial cell branching programs (Alcaraz et al., 2011; Bonnans et al., 2014; Mori et al., 2009, 2013; Rebustini et al., 2009; Weaver et al., 2014), we find that MMP14, but not MMP15, is required for mammary epithelial organoid branching through type I collagen barriers in vitro. However, in contrast to earlier findings (Mori et al., 2009), we find that MMP14-expressing mammary epithelial cells in 3D culture lack a competitive sorting advantage relative to Mmp14-null epithelial cells and are instead distributed throughout the branching networks. Further, when Mmp14^−/− or Mmp15^−/− epithelial organoids are embedded in Matrigel (Ewald et al., 2008, 2012; Lo et al., 2012), morphogenesis proceeds in an unperturbed fashion, even in the presence of a pan-specific MMP inhibitor. Given these divergent outcomes, and the inability of model systems to recapitulate mammary gland architecture or stromal cell interactions (Nelson and Larsen, 2015), we turned to the in vivo setting.

Unlike in in vitro culture conditions, primordial mouse mammary ducts are encased in a coat of type IV collagen and laminin that is surrounded by low levels of type I collagen. Nevertheless, given the broad substrate repertoire of MMP14 and the multiple biological activities that have been assigned to its functions (Ager et al., 2015; Alcaraz et al., 2011; Barbolina and Stack, 2008; Bonnans et al., 2014; Chun et al., 2004; Fahlman et al., 2014; Fu et al., 2013; Itoh, 2015; Kajita et al., 2001; Koshikawa et al., 2000; Koziol et al., 2012; Mori et al., 2009, 2013; Rebustini et al., 2009; Rowe and Weiss, 2009; Sabeh et al., 2009; Sakamoto et al., 2014; Shimizu-Hirota et al., 2012; Simian et al., 2001; Tang et al., 2013; Taylor et al., 2015; Weaver et al., 2014; Yana et al., 2007), we were surprised to find that deleting the proteinase in vivo exerted little, if any, effect on morphogenesis. Likewise, despite an ex vivo requirement for MMP15 in submandibular gland proliferation, matrix remodeling and branching (Rebustini et al., 2009), mammary glands from Mmp15-targeted mice assembled normal ductal networks through P10. Taken together, the overall similarities of the ductal networks generated in Mmp14^−/− or Mmp15^−/− mice suggest that any in vivo defects related to tissue-invasive activity or branching are, at best, subtle. Of note, a small cohort of the Mmp14^−/− mice (3 of 15 total) displayed an unusually reduced body size with partial mammary gland defects noted at P10 (the mammary ducts branched comparably, but appeared more primitive in their architecture). However, given the reduced feeding of this cohort and the known influence of nutrient deprivation on tissue morphogenesis (Londhe et al., 2013), we did not include these animals in our analyses.

Despite the retention of intact mammary gland branching programs in Mmp14^−/− or Mmp15^−/− mice, we did not directly rule out a potential compensatory role for Mmp14 and Mmp15 in postnatal morphogenesis. As Mmp14/Mmp15 double-null mice die during embryogenesis (Szabova et al., 2010), we alternatively targeted both proteinases selectively in the mammary epithelial compartment, but these analyses demonstrated that branching morphogenesis nevertheless proceeds in a normal fashion from birth through adulthood (our unpublished observations). Interestingly, a double-knockout system was recently employed to target Mmp14 and Mmp15 in postnatal mice wherein mammary gland involution was unaffected (Szabova et al., 2010), suggesting that adult mammary tissue remodeling likewise proceeds in a MT-MMP-independent fashion. Although MMP14 and MMP15 have previously been shown to endow neoplastic cells with the ability to degrade basement membrane (BM) barriers (Ota et al., 2009), it should be stressed that there is no evidence that mammary epithelial cells dissolve their underlying BM during morphogenesis in vivo. Indeed, during postnatal development, mammary epithelial cells remain separated from the periductal type I collagen meshwork by a patent BM (Ewald et al., 2008, 2012; Williams and Daniel, 1983). More recently, Yamada and colleagues concluded that BMs undergo proteolytic remodeling during branching morphogenesis ex vivo (Harunaga et al., 2014), but efforts to implicate MMPs in this process were confined to the use of synthetic inhibitors that not only target all MMPs, but also ADAM/ADAM-TS family members that play crucial roles in morphogenesis (Rowe and Weiss, 2009). Further, contrary to the conclusion that MMPs participate directly in BM proteolysis, TIMP2, a potent endogenous MMP inhibitor (Chun et al., 2004; Sabeh et al., 2009), did not affect BM remodeling in their studies (Harunaga et al., 2014). Finally, we note that the early postnatal branching programs tracked in our studies were not mediated by other MMP family members implicated in later stages of development, including Mmp2 and Mmp3 (Correia et al., 2013; Kessenbrock et al., 2013; Wiseman et al., 2003). While recent work has also identified more subtle roles for MMP9 and MMP11 in mammary gland branching after the onset of puberty, both Mmp9^−/− and Mmp11^−/− mammary glands undergo normal pre-pubertal branching (Tan et al., 2014; Ucar et al., 2010; Wiseman et al., 2003).

Given the paucity of obvious structural defects in ductal morphogenesis in Mmp14^−/− or Mmp15^−/− mice, we alternatively interrogated the mammary gland transcriptomes of the knockout tissues. Although MMP14 and MMP15 share considerable structural homology (Itoh, 2015), effects on mammary gland gene expression were notably distinct. For example, Mmp14^−/−, but not Mmp15^−/−, mammary glands display increased expression of a number of cell adhesion/cytoskeleton-associated transcripts, with Mmp14^−/− ducts displaying changes in epithelial cell-cell junctions and the number of apical microvilli. Although further work will be needed to define the functional consequences of these changes, the observed structural alterations highlight the fact that MMP14 does, in fact, affect mammary gland development, but not in a fashion that impacts branching-associated programs per se. In contrast to the epithelial cell-associated gene changes observed in Mmp14^−/− mammary glands, the effects of deleting Mmp15 were subtle, with only a small subset of genes affected. We are, however, unable to assess the catalytic activity of MMP15 per se as specific antibodies capable of distinguishing between the pro- and active forms of the enzyme have not been generated. Nevertheless, it is intriguing that the breast cancer-associated transcripts γ-synuclein (Ahmad et al., 2007) and peroxiredoxin 2 (Stresing et al., 2013) are among the most significantly affected (∼5-fold increase and ∼7-fold decrease, respectively).

In the mammary gland, the adipocyte-rich stroma plays important roles in regulating morphogenetic responses (Hovey and Aimo, 2010; Inman et al., 2015). Intriguingly, early in postnatal life, the mammary gland-associated pool of adipocytes contains both white and brown/beige fat (Gouon-Evans and Pollard, 2002; Master et al., 2002), but the factors that control their differential expression have not been defined previously. Independent of the effects of MMP14 and MMP15 on mammary epithelial cells, their more global and differential effects on mammary fat pad development were unanticipated. Mmp14 targeting alone triggers a widespread downregulation of transcripts associated with adipogenesis (Wang et al., 2015). By contrast, an Mmp15-null status exerts little effect on white fat development, but instead strongly enhances the formation of brown and/or beige fat, as evidenced by the increased expression of a cohort of thermogenic genes (Wu et al., 2013). Our laboratory has previously established roles for MMP14 in coordinating adipocyte maturation in vivo, as evidenced by the aborted white adipose tissue development in the dermal and inguinal adipose tissue depots of male Mmp14^−/− mice (Chun et al., 2006), but the role of MMP14 in the mammary gland stromal fat pad had not been previously appreciated.

Recent studies have emphasized the ability of secreted as well as membrane-anchored MMPs to cleave hundreds of intracellular, cell surface-associated or extracellular substrates (Barbolina and Stack, 2008; Fahlman et al., 2014; Fu et al., 2013; Itoh, 2015; Koziol et al., 2012; Rowe and Weiss, 2009). Given these results, it is surprising, if not perplexing, that global targeting of Mmp14 or Mmp15 – as well as Mmp2 and Mmp3 – did not derail ductal invasion programs or morphogenesis to a more striking degree. Indeed, although MMP14 has been implicated – to varying, but modest degrees – in early lung, kidney and submandibular gland development, these findings do not extend to mammary gland development (Atkinson et al., 2005; Oblander et al., 2005; Riggins et al., 2010). Results such as ours are often countered by the argument that the mouse genome encodes more than 20 MMP family members, but virtually all of these genes have now been targeted and, with the exception of Mmp14, each knockout gives rise to viable mice with normal lifespans (Bonnans et al., 2014; Holmbeck et al., 1999; Itoh, 2015; Rowe and Weiss, 2009; Zhou et al., 2000). The fact that only Mmp14 targeting causes a dramatic increase in the morbidity and mortality of knockout mice (Holmbeck et al., 1999; Zhou et al., 2000) and that dual targeting of Mmp14 and Mmp15 results in an embryonic lethal phenotype (Szabova et al., 2010) have, predictably, catalyzed increased efforts to identify the key functions of these proteinases. Clearly, MMP14 and MMP15 do exert complex effects on mammary gland development, but the observed changes in function are, for the most part, novel and unanticipated from the perspective of current paradigms.

Mmp14^lacZ/+ mice (Yana et al., 2007) were bred onto a C57BL/6J background (Jackson Laboratory, 000664). To generate Mmp15^lacZ/+ mice, C57BL/6J Mmp15^lacZ/+ embryonic stem cell (ESC) clones from the UC Davis KOMP Repository [MGI allele Mmp15^{tm1a (KOMP)Wtsi}, clone EPD0097_3_B09] were introduced into albino C57BL/6J host blastocysts. The resulting chimeras were mated with C57BL/6J mice to test for germline transmission. Mmp15^lacZ/+ mice were bred and maintained on a C57BL/6J background. The Mmp14^+/− mice (Swiss Black background) were obtained from the National Institutes of Health (Holmbeck et al., 1999) and maintained on an outbred Swiss Black background [Charles River, NIHBL(S)]. Mmp3^−/− mice (C57BL/6J background) and Mmp2^−/− mice (Swiss Black background) were genotyped as described (Itoh et al., 1997). To generate Mmp15 global knockout (Mmp15^−/−) mice, 129/SVJ ESCs were electroporated with a targeting vector for Mmp15 containing a neomycin selection cassette flanked by FRT sites, wherein exons 4 and 5 of the catalytic domain were floxed by loxP sites (Fig. S2A). Following neomycin selection, ESC recombinants were injected into E3.5 C57BL/6J blastocysts that were subsequently injected into the uteri of pseudo-pregnant FVB mice. Male chimeras were mated to C57BL/6J females, and heterozygous agouti offspring carrying the Mmp15^flox allele were mated to C57BL/6J FlpE mice to remove the neomycin selection cassette. The resultant Mmp15^fl/fl mice were crossed to C57BL/6J EIIA-Cre mice for germline deletion of Mmp15, and Mmp15^+/− mice were backcrossed with C57BL/6J mice (n>10 generations). Genotyping primers are provided in Table S2. All mouse work was performed with the approval of the Institutional Animal Care and Use Committee of the University of Michigan.

Organoids were prepared from inguinal mammary glands of P10 mice as described (Fata et al., 2007). Mammary epithelial organoids were then embedded in 2.2 mg/ml acid-extracted rat-tail type I collagen (Sabeh et al., 2009) for 3D culture in 24-well plates with DMEM/F12 medium containing 0.1 mM non-essential amino acids (Gibco), 0.3 mg/ml L-glutamine, 1× ITS (Sigma, 13146), 10 µg/ml insulin (Gibco, 12585-014), penicillin and streptomycin, with 50 ng/ml FGF2 (Simian et al., 2001). Medium was changed every 2-3 days for 5-8 day culture periods.

Inguinal mammary glands were harvested, mounted on glass slides, stained with Carmine alum, and processed as described (Lu et al., 2008). Whole-mounts were imaged using a Leica MZFLIII dissecting microscope and Adobe Photoshop and ImageJ used to process images (Lu et al., 2008). The length of ductal penetration was quantified as the average distance of straight lines from the nipple to the terminal ends of the four longest mammary epithelial ducts (Wiseman et al., 2003). The epithelial surface area was quantified as the region occupied by mammary epithelial ducts from the primary stalk to all terminal ends of the branching network. Total branch point number was quantified as the total number of primary, secondary and tertiary branches from the primary stalk of each ductal network.

Inguinal mammary gland tissue was flash frozen in liquid nitrogen, homogenized in 1 ml TRIzol (Ambion, Life Technologies). Total RNA was extracted and purified using RNeasy Mini-Kit columns (Qiagen, 74104). RNA quality was confirmed using an Agilent 2100 Bioanalyzer and samples were profiled on Affymetrix Mouse MG–430 PM expression array strips. Expression values for each probe set were calculated using a robust multi-array average (RMA) (Irizarry et al., 2003) and filtered for genes with a greater than 2-fold change. The Affymetrix microarray data have been deposited in the NCBI Gene Expression Omnibus (GEO) database (Edgar et al., 2002) and are available through accession number GSE77679 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE77679).

Complementary cDNA was prepared with the SuperScript II First-Strand Synthesis System (Invitrogen, 11904-018). Quantitative reverse-transcription PCR (qPCR) was performed with SYBR Green (Applied Biosystems) using Arbp (Rplp0) for reference gene expression. Primers are listed in Table S1. Data are reported as mean expression levels with error bars representing s.e.m. P-values were calculated with an unpaired, two-tailed t-test.

Inguinal Mmp14^lacZ mammary glands were fixed at 4°C in PBS containing 2% paraformaldehyde and 0.2% glutaraldehyde. Mmp15^lacZ glands were alternatively fixed in 4% paraformaldehyde. Tissues were either transferred to 30% (w/v) sucrose in PBS for frozen sections, or incubated for 18-24 h in PBS containing 4 mM potassium hexacyanoferrate (III), 4 mM potassium hexacyanoferrate (II) trihydrate, 2 mM MgCl₂, 0.2% NP-40, 0.1% sodium deoxycholic acid and 1 mg/ml X-Gal. Stained tissues and whole-mounts were post-fixed with 10% formalin phosphate.

Mammary gland tissue was fixed in 10% formalin phosphate at 4°C overnight and either dehydrated in an ethanol/paraffin series for paraffin embedding or transferred to 30% sucrose in PBS for embedding in OCT (Fisher Healthcare). Sections were permeabilized with 0.3-0.5% Triton X-100 in PBS, and blocked for 2 h prior to primary antibody addition. Samples were incubated overnight at 4°C with antibodies directed against ZO-1 (Invitrogen, 617300; 2.5 µg/ml), E-cadherin (BD Transduction Laboratories, 610181; 1/400), cytokeratin 14 (Covance, PRB-155P, clone AF64; 1/400), cytokeratin 18 (Abcam, ab668; 1/150), α-smooth muscle actin (Abcam, ab5694; 1/400) or UCP1 (Alpha Diagnostic International, UCP11-A; 200 ng/ml). For ECM staining, frozen sections were incubated in blocking buffer (10% FBS, 1% BSA in PBS) and incubated with primary antibodies directed against type IV collagen (Millipore, AB-769; 1/40) (Nguyen-Ngoc et al., 2012), laminin (Sigma-Aldrich, L9393; 1/250) or type I collagen (Abcam, ab34710; 1/250) overnight at 4°C. Following primary antibody incubations, sections were incubated with Alexa 488- or Alexa 594-conjugated secondary antibodies (Invitrogen Molecular Probes). Nuclear counterstaining was carried out with either Toto3 (1:200; Invitrogen Molecular Probes) or 2 µg/ml 4′,6-diamidino-2-phenylindole (DAPI). All fluorescence images were acquired with an Olympus FluoView FV500 laser-scanning confocal microscope and analyzed with Adobe Photoshop and ImageJ software. For immunohistochemistry, endogenous peroxidase activity was blocked with 3% H₂O₂ and tissue sections developed with the Vectastain ABC Kit (Vector Laboratories, PK-6100) and DAB Kit (Vector Laboratories, SK-4100). ImageJ was used to measure the diameter of n>20 adipocytes per 40× field of Hematoxylin and Eosin (H&E)-stained paraffin transverse sections.

Samples were fixed (1% glutaraldehyde, 1% tannic acid, 0.1 M Sorensen's buffer, pH 7.2) and post-fixed with 1% osmium tetroxide as described (Abrahamson and Perry, 1986) prior to embedding in EMbed 812 epoxy resin (Electron Microscopy Sciences). Thin sections (70 nm) were post-stained with uranyl acetate and Reynolds lead citrate, and imaged with a JE0L JEM-1400 Plus electron microscope at 80 kV. Images were recorded digitally using a Hamamatsu ORCA-HR digital camera system, operated using AMT software (Advanced Microscopy Techniques Corporation). ImageJ was used to quantify the average number of microvilli per µm from n>5 fields per sample and n=3 per genotype, and to quantify the surface area of individual mitochondria in n>10 fields per sample and n=2 per genotype. Results are expressed as mean±s.e.m.

We thank Alan Saltiel (UC San Diego) and David Ginsburg (University of Michigan) for helpful discussions. We acknowledge Dorothy Sorenson (University of Michigan) for assistance with TEM and Craig Johnson (University of Michigan) for assistance with microarray analysis.