Extracellular Matrix Regulates Apoptosis in Mammary Epithelium through a Control on Insulin Signaling

Primary mammary epithelial cells were prepared from 14.5–18.5-d pregnant ICR mice as previously described (Pullan et al., 1996). After culture for 3 days, cells were trypsinized, passed through a 70-μm nylon mesh (Falcon) to remove large cell aggregates, and plated at 1 × 10⁵ cells/cm² in DMEM/F12 medium containing 5% fetal calf serum (Advanced Protein Products) and 5 ng/ml epidermal growth factor (Harlan Sera-Lab) overnight, then washed and cultured in serum-free DMEM/F12 medium containing appropriate combinations of 2.8 nM hydrocortisone, 880 nM insulin, 150 nM prolactin (Sigma), with or without inhibitors of the insulin signaling pathway. Some experiments included IGF-I or IGF-II (R&D systems) in place of insulin.

The ECM substrata used for these studies were rat tail collagen I and reconstituted BM matrix prepared from Engelbreth-Holm-Swarm (EHS) tumor (Pullan and Streuli, 1996). Collagen I was coated onto tissue culture plates overnight at 80 μg/ml and washed extensively before use, or in some experiments commercially precoated Biocoat dishes were used (Becton Dickinson). A factor-reduced preparation of the BM matrix was prepared using sequential ammonium sulfate precipitations (Taub et al., 1990) and coated onto tissue culture plastic at 14 mg/ml. In some experiments, factor-reduced BM matrix was diluted into the culture medium (final concentration 100 or 400 μg/ml) and overlaid onto first passage cells plated on collagen I (Streuli et al., 1995).

To analyze DNA integrity, first passage cells were cultured in medium containing appropriate hormones and inhibitors for two days, then washed over a period of 2 h to remove the accumulated dead cells and debris. Fresh medium was added to cells and after 4 h newly apoptotic cells were collected from the medium and pooled with any remaining attached cells. DNA was extracted from samples (Pullan et al., 1996), its OD was measured at 260 nm, and separated on a 1% agarose gel to confirm DNA loading after staining the gel with ethidium bromide. Equal amounts of DNA were separated by conventional agarose gel electrophoresis, southern blotted, and apoptotic DNA ladders were visualized by hybridization with a digoxigenin-labeled total mouse genomic DNA probe (Boehringer Mannheim).

Single cell cultures were prepared by straining first passage cells through a 20-μm mesh, pelleting and resuspending in factor-reduced BM matrix at 2–4 × 10⁶ cells/ml, as described (Streuli et al., 1991). 100 μl of the mammary cell suspension was gelled for 60 min on dishes precoated with a thin layer of BM, then cultured in serum-free DMEM/F12 medium containing appropriate hormones, inhibitors or antibodies for 2 d before fixation in 2% paraformaldehyde. 25-μm cryosections were stained with 0.5 μg/ml Hoechst 33258 and apoptotic nuclei were counted in the single cell population. Each experiment was repeated two to three times and in each experimental condition, >600 single cells were scored for apoptosis. Polyclonal rabbit anti–β1 integrin antibody was prepared in this laboratory (Edwards and Streuli, 1998). It was shown to inhibit the adhesion of first passage mouse mammary epithelial cells to laminin, collagen IV, but not fibronectin or vitronectin. Purified rat monoclonal antibody to α6 integrin (GoH3) and its rat IgG2a isotype control were obtained from Serotec. Rabbit polyclonal anti-laminin antibody was raised against purified laminin isolated from EHS tumor (a kind gift of P. Yurchenco, Robert Wood Johnson Medical School, Piscataway, NJ). This laminin had previously been shown to contain laminin A and B chains but not collagen IV or nidogen (Streuli et al., 1995). A protein A–derived IgG fraction from the rabbit serum was subsequently purified on an EHS affinity column. This antibody significantly inhibited the adhesion of first passage mouse mammary epithelial cells to laminin at 5 μg/ml and completely inhibited adhesion at 30 μg/ml (data not shown).

A quantitative assay for apoptosis based on measuring cellular detachment from different substrata was previously described (Pullan et al., 1996). Cells detaching into the medium of first passage cultures over periods of 4 h were pelleted and counted using either a hemocytometer or a Coulter counter. Detached cells were confirmed to have apoptotic morphology by fluorescence microscopy after staining with 0.5 μg/ml Hoechst 33258, and the remaining monolayers were also stained with Hoechst 33258 to confirm that apoptotic cells did not accumulate there. Quantitative data for detachment assays are all relative to the number of cells that attached to each substrata after plating.

Mammary cells were cultured for 2 d, then washed and serum-free DMEM/F12 medium with or without insulin was replaced for 15 min. The cells were washed twice with 1 mM Na₃VO₄, 10 mM NaF in PBS and scraped into NET lysis buffer (50 mM Tris-Cl, 150 mM NaCl, 1% NP-40, 2 mM EDTA, 10 mM NaF, 1 mM Na₃VO₄, 10 μg/ml aprotinin, 20 μM leupeptin, 1 mM PMSF, pH 7.5). Cells were lysed by vortexing and then rotating for 30–60 min at 4°C, and detergent insoluble proteins were cleared by centrifugation at 7,000 g. Aliquots of samples were subjected to SDS-PAGE on 6% gels, and the gels were stained with Coomassie Brilliant Blue to normalize protein levels. Equal amounts of protein were immunoprecipitated either with anti–IRS-1 antibody (UBI) or anti-insulin receptor β antibody (Transduction Laboratories) and 20 μl protein A–Sepharose beads (Zymed Laboratories), before being separated on 6% gels under reducing conditions and transferred to Immobilon-P membranes (Millipore). Phosphotyrosine blots were blocked in 1% BSA in TBST, (50 mM Tris-Cl, 150 mM NaCl, 0.1% Tween 20, pH 7.5) and incubated overnight with monoclonal antibody 4G10 (UBI) at 1:1,000 for insulin receptor β or 1:3,000 for IRS-1. Interaction of the p85 subunit of PI 3-kinase with IRS-1 was examined by probing IRS-1 immunoblots with anti–PI 3-kinase (UBI). For each experiment, equal amounts of protein were confirmed for IRS-1 by blocking IRS-1 immunoblots in 3% nonfat dry milk/PBS and probing with 1:3,000 anti–IRS-1 antibody; for insulin receptor β, whole cell lysate blots were blocked with 3% BSA/TBST and probed with 1:250 anti-insulin receptor β. Protein kinase B (PKB, otherwise known as c-Akt) was analyzed by immunoblotting equal amounts of cell protein with either an anti-PKB antibody or an antibody specific for phosphorylated PKB (both at 1:2,000; New England Biolabs). Proteins were visualized by enhanced chemiluminescence (Amersham).

Mammary epithelial cells require appropriate cell–ECM interactions for survival since they undergo apoptosis in the absence of a BM (Pullan et al., 1996). To determine the mechanism by which BM suppresses apoptosis, we asked whether BM was able to act as a survival ligand by itself or whether additional factors were required to prevent cell death. First passage mammary epithelial cells isolated from midpregnant mice were plated on substrata of collagen I or a reconstituted BM matrix derived from the EHS tumor. Culture on a BM matrix in the presence of lactogenic hormones, insulin, prolactin, and hydrocortisone, suppressed apoptosis, whereas cells died on collagen I (Fig. 1 A). However, removal of lactogenic hormones from the medium resulted in apoptotic death of the cells, even those cultured on the BM matrix (Fig. 1 A). This indicated that survival of mammary cells required two signals, one from the BM and a second signal from the hormonal milieu.

To determine which of the three hormones acted in a survival capacity, cells plated on BM were incubated with combinations of insulin, prolactin, and hydrocortisone and the extent of cell death was measured by DNA fragmentation. Apoptosis occurred only when the cells were cultured without insulin (Fig. 1 B). We also used an alternate assay for apoptosis, where single mammary cells were cultured within a BM gel, but apart from each other so that they were unable to form cell–cell interactions (Streuli et al., 1991). In this assay, apoptosis was measured by determining nuclear morphology after culture for 48 h (Fig. 1 C). 20–30% of single cells were apoptotic providing insulin was present in the medium. However, in the absence of insulin, cell death increased significantly (P < 0.01) to 60– 70% (Fig. 1 D). These two different approaches collectively demonstrate that BM does not act alone to suppress apoptosis in primary cultures of mammary epithelia. Instead, it regulates survival in combination with signals elicited by insulin. Furthermore, the data show survival is not dependent on mammary differentiation, since removal of prolactin and hydrocortisone, which are required for milk protein gene expression, did not result in apoptosis.

In addition to the ability of insulin to interact with its own receptor, this hormone can also bind to the IGF receptor although with a 100–1,000-fold lower affinity (Parrizas et al., 1997). IGF-I or IGF-II have been shown to act as survival factors in other cell types (Rodriguez-Tarduchy et al., 1992; Harrington et al., 1994; Kulik et al., 1997), and we therefore tested whether they could directly suppress mammary apoptosis. At physiological concentrations, both IGF-I and IGF-II significantly inhibited the DNA fragmentation exhibited by cells cultured on a BM matrix in the absence of other hormones (Fig. 1 E). Moreover, IGF-I suppressed apoptosis (P < 0.01) in single mammary cells cultured within the BM gel to the same extent as insulin (Fig. 1 F). The survival of mammary epithelia can therefore be regulated by signals from IGF-I and IGF-II in addition to those from insulin, and these signals act coordinately with BM.

Mammary epithelial cells plated onto a BM substratum form multicellular, alveolar-like structures (Barcellos-Hoff et al., 1989; Aggeler et al., 1991). Thus, it was important to ask whether this complex three-dimensional structure was involved with mammary cell survival, or if signals from the BM–integrin interactions were sufficient. Therefore, we cultured mammary cells as monolayers on collagen I and incubated them with the EHS BM preparation diluted into the culture medium. Under these conditions, the cells did not form alveoli but remained as monolayers on the culture dish with some matrix proteins precipitating over the cells (Streuli et al., 1995). Dilution of the BM preparation 140-fold to 0.1 mg/ml resulted in significant suppression of apoptosis, both in a quantitative assay (Fig. 2 A) and in DNA fragmentation studies (Fig. 2 B). BM proteins diluted 35-fold to 0.4 mg/ml suppressed apoptosis to virtually the same extent as in the cells cultured as alveoli on top of a BM substratum (Fig. 2, A and B). Thus, the three-dimensional multicellular structure is not a primary determinant of mammary cell survival.

BM contains several ECM proteins and one component, laminin, has previously been shown to be a survival ligand for other cell types (Kim et al., 1994; Vachon et al., 1996; Bozzo et al., 1997). To determine whether laminin was a survival factor for mammary epithelial cells we developed a function-blocking anti-laminin antibody that inhibited adhesion of mammary cells to laminin (data not shown). We assessed whether this antibody was able to interfere with the survival of mammary cells that had been cultured as single cells within BM gels (Fig. 3 A). We found that when the anti-laminin antibody was included in the gel, mammary cells underwent apoptosis even when they were cultured in the presence of insulin. In contrast, an adhesion-blocking antibody directed against fibronectin had no effect (Fig. 3 A), neither did the inclusion of an RGD peptide in the assay (Fig. 3 B). These results demonstrated that within the context of a three-dimensional BM gel, laminin has a survival-promoting function. Indeed, they show that both insulin and laminin can act as survival ligands for mammary epithelial cells, although they do not rule out the possibility that other soluble factors or ECM proteins may have a similar function.

Major cellular receptors for BM proteins, including laminin, are the transmembrane heterodimeric integrin receptors. To confirm that BM acts directly to suppress apoptosis, and that the survival signals were mediated by integrins, we assessed whether function-blocking anti-integrin antibodies interfered with the mammary cell survival using the single cell assay. In experiments where an adhesion-blocking polyclonal anti–mouse β1 integrin antibody was included in the gel, insulin failed to suppress apoptosis. Indeed, the levels of death were similar to those observed in single mammary cells in the absence of insulin (Fig. 4 A). One receptor for laminin is α6β1 integrin, and we therefore performed parallel experiments using a function-blocking anti–α6 integrin antibody (Fig. 4 B). The cellular response was similar to that for the anti–β1 integrin antibody, providing evidence that both the α6 and β1 integrin subunits can transduce survival signals in normal mammary epithelial cells. Suitable antibodies to other mouse integrin subunits were not available to test for specificity. Therefore, we performed similar experiments using function-blocking anti–human integrin antibodies and demonstrated that α6 and β1 integrin, but not α2 integrin, acts as survival receptors in primary cultures of human mammary cells (Oliver, J., M. O'Hare, and C.H. Streuli, manuscript in preparation). In addition, our results with anti-fibronectin antibodies and RGD peptides do not support a role for αv integrins in mammary cell survival.

Our combined data using function-blocking antibodies to laminin and to integrin subunits show that suppression of apoptosis in mammary epithelia depends on interactions of the cells with BM through the α6 and/or β1 integrin receptors, and that laminin can act as a survival ligand.

Since both insulin and laminin had been identified as mammary cell survival factors, it is possible that they trigger separate pathways required for suppressing apoptosis. However, recent work on the control of cell cycle and differentiation indicates that ECM affects the ability of growth factors/cytokines to trigger their downstream signaling kinases (Lin et al., 1997; Renshaw et al., 1997; Edwards et al., 1998). Thus an alternative mechanism to explain survival signaling in mammary cells is that the insulin- and BM-triggered pathways converge. To determine whether this was indeed the case, the extent of tyrosine phosphorylation in the proximal components of the insulin signaling pathway was measured (Fig. 5).

Insulin interacts with its receptor, causing receptor oligomerization and activation of the kinase domain. This results in tyrosine phosphorylation of the receptor β subunit. The adaptor protein insulin receptor substrate-1 (IRS-1) is recruited to the insulin receptor, then becomes tyrosine phosphorylated creating SH2 domain interaction sites for downstream enzymes such as PI 3-kinase (Myers et al., 1994; Sun et al., 1992). The ability of insulin to signal in primary cultures of mammary cells plated on BM was compared with those on collagen I. Tyrosine phosphorylation of the insulin receptor β subunit occurred within 15 min of insulin treatment and was independent of cell– ECM interactions (Fig. 5 A). However, the ability of IRS-1 to become tyrosine phosphorylated in response to insulin was strongly dependent on cell interactions with BM, and virtually no IRS-1 phosphorylation was observed in cells cultured on collagen I (Fig. 5 B).

It has been shown in several cell types that PI 3-kinase is required for preventing apoptosis, and that in neuronal cells and fibroblasts IGF-I suppresses apoptosis through PI 3-kinase (Yao and Cooper, 1995; Minshall et al., 1996; Dudek et al., 1997; Kennedy et al., 1997; Kulik et al., 1997; Miller et al., 1997; Parrizas et al., 1997). PI 3-kinase is activated in response to insulin or IGF-I following the interaction of its p85 subunit with IRS-1, and therefore we assessed whether cell–ECM interactions regulated the association of the p85 subunit of PI 3-kinase with IRS-1 (Myers et al., 1994). PI 3-kinase did not associate with IRS-1 in mammary cells cultured on either substratum in the absence of insulin (Fig. 5 C). However, within 15 min of insulin treatment PI 3-kinase bound IRS-1, but the extent of the interaction was strongly dependent on cell adhesion to BM; quantitation of the data indicated that threefold more PI 3-kinase bound to IRS-1 in cells cultured on a BM substratum than in cells cultured on collagen I (Fig. 5 D).

One downstream target of PI 3-kinase signaling pathway is PKB, which has previously been implicated in the suppression of apoptosis (Dudek et al., 1997; Kauffmann-Zeh et al., 1997; Kulik et al., 1997; Crowder and Freeman, 1998). This enzyme is recruited to the plasma membrane following PI 3-kinase phosphorylation of membrane lipids, and then activated phosphorylation on serine/threonine residues (Downward, 1998). Using an antibody specific for phosphorylated PKB, we found that maximal activation only occurred when mammary cells treated with insulin had been cultured on BM (Fig. 6). Wortmannin completely abrogated insulin-induced PKB phosphorylation. Since many of the assays to measure mammary apoptosis in our work were carried out over several days, we also measured PKB phosphorylation status in cells cultured with insulin for 3 days. PKB was only phosphorylated in cells cultured on BM, even though its level remained constant whether the cells were on collagen I or BM substrata. These results are an indirect confirmation that PI 3-kinase activity depends on coordinate signals from insulin and BM. Most importantly, they demonstrate that the PKB pathway, already shown to be a critical determinant of survival in other cell systems, relies on converging insulin and ECM signals in mammary epithelia.

Thus, the ECM dependence of insulin signaling correlates with the ability of BM, but not collagen I, to deliver survival cues.

The experiments described above suggested that the mechanism through which cell–ECM interactions regulate apoptotic fate in mammary epithelial cells is through a control on insulin-mediated PI 3-kinase signaling. To determine whether the PI 3-kinase pathway, or related pathways, were required for survival, the effects of the pharmacological inhibitors LY 294002 and wortmannin were examined in two independent assays for mammary apoptosis.

Mammary cells were cultured on a BM matrix with 10– 100 nM IGF-I, with or without 0.1 or 1 μM LY 294002. In the absence of soluble factors the cells underwent apoptosis (Fig. 7 A, lane 2), and this was suppressed by IGF-I (Fig. 7 A, lanes 3–6). However, IGF-I failed to suppress apoptosis when LY 294002 was included in the cultures (Fig. 7 A, lanes 7–10). We observed low levels of apoptosis at 0.1 μM and very extensive DNA fragmentation at 1 μM LY 294002, a concentration previously shown to inhibit IGF-I signaling and PI 3-kinase activity (Vlahos et al., 1994; Parrizas et al., 1997). Similar results were obtained with wortmannin (data not shown).

The effect of the kinase inhibitors on suppression of apoptosis by insulin was also examined in the single cell assay. As before, mammary cells cultured within BM matrix were not able to survive in the absence of insulin. Insulin rescued the cells from apoptosis, but low levels of wortmannin (1 μM) and LY 294002 (0.1 μM) blocked survival signaling (Fig. 7 B). The target specificity for wortmannin is fairly broad but includes PI 3-kinase whereas LY 294002 is specific for PI 3-kinase (Vlahos et al., 1994; Ui et al., 1995). Thus, these two experiments indicate that the PI 3-kinase class of enzymes are necessary for mammary cell survival.

Together our results suggest that mammary epithelial cell survival depends upon a functional PI 3-kinase signaling pathway which is normally triggered by insulin when the cells are cultured with a BM substratum. Restricting the activity of this pathway by culture on collagen I results in cell death.

The mechanisms that regulate death and survival decisions in epithelial cells are not understood. Yet these are fundamental cellular processes and are essential not only for deciphering the control of normal development, but also for designing strategies to tackle progressive diseases such as breast cancer.

In this study, we have addressed the mechanism whereby BM suppresses apoptosis in primary cultures of mammary epithelial cells and found that this type of ECM controls the ability of insulin to deliver survival signals. We have demonstrated that insulin is necessary to prevent death of mammary cells through the PI 3-kinase pathway, and furthermore that insulin-mediated interaction of PI 3-kinase with IRS-1 and phosphorylation of PKB, a key enzyme downstream in the PI 3-kinase pathway, is influenced by cell interactions with BM. This provides new evidence for the importance of ECM in the growth factor–mediated regulation of cell phenotype. Since α6 and β1 integrins are necessary for mammary cell survival, our work supports the notion that one function of integrins is to regulate growth factor signaling.

Our previous studies, and those of others, have demonstrated that mammary epithelial cells depend on BM for survival (Boudreau et al., 1995; Pullan et al., 1996). We have now found that they require a further signal from soluble ligands to remain alive. Of three hormones tested, insulin, but not hydrocortisone or prolactin, was found to be necessary for suppression of apoptosis (Fig. 1). This survival effect of insulin could be mimicked by IGFs.

Insulin and IGFs have been shown to rescue apoptosis in several cell lines following interleukin-3 or serum withdrawal (Rodriguez-Tarduchy et al., 1992; Harrington et al., 1994; Parrizas et al., 1997). However, although IGFs have been proposed to act as survival factors in mammary gland, this has not been demonstrated directly. For example, overexpression of des(1-3)hIGF-I or IGF-I in mammary glands of transgenic mice delayed involution and the glands showed lower levels of apoptosis, but the target cells were not identified (Hadsell et al., 1996; Neuenschwander et al., 1996). In the experiments performed in our study, purified populations of mouse mammary epithelial cells survived in the presence of insulin or IGFs, indicating that the factors suppress apoptosis by a direct action on the epithelial cells themselves.

Glucocorticoids have also been demonstrated to rescue cells from apoptosis in the mammary gland, although it was not determined whether the hormone acts directly on epithelial cells or through an indirect mechanism (Feng et al., 1995; Lund et al., 1996). In purified mammary epithelial cells hydrocortisone had a minimal effect on survival, and therefore its function in vivo would appear indirect. Thus, although in vivo studies have indicated a role for both IGFs and glucocorticoids in suppressing apoptosis in mammary gland, our study with isolated cells shows that only IGFs act directly on the epithelial cells themselves. IGFs are normally only synthesized by stromal cells in the mammary gland, but their expression is regulated by a synergy between growth hormone and estradiol (Yee et al., 1989; Ruan et al., 1995). Therefore, one explanation for the inhibition of apoptosis by hydrocortisone in vivo is that this steroid can act on stromal cells to alter IGF expression. Alternatively, since IGFBP-5 has been suggested to induce mammary involution, glucocorticoids may inhibit apoptosis in vivo by decreasing IGFBP-5 synthesis, as occurs in cultured fibroblasts (Conover et al., 1995; Tonner et al., 1997).

The ability of prolactin to suppress apoptosis directly in purified mammary epithelial cells was also tested. In agreement with previous studies which showed that prolactin could not rescue mammary apoptosis in vivo, we demonstrated that it also had a negligible effect on epithelial cell survival in culture (Feng et al., 1995; Marti et al., 1997). Prolactin regulates intracellular signaling through a Jak/Stat pathway, which is required for mammary differentiation. Although other cytokines have been proposed to suppress apoptosis through Stat transcription factors, it appears that this pathway is not directly involved in mammary cell survival (Fujio et al., 1997).

Together our data show that, of the hormones tested, insulin is sufficient for the survival of mammary cells cultured on BM. Since prolactin and hydrocortisone, in addition to insulin, are necessary for lactation, the results indicate that mammary cell survival does not depend on differentiation. Moreover, as cells on collagen I undergo apoptosis even in the presence of all three hormones, the survival mechanism would appear to have more to do with cell–ECM interactions than with differentiation.

Cell–ECM interactions are known to suppress apoptosis in epithelial cells (Meredith et al., 1993; Frisch and Francis, 1994; Khwaja et al., 1997). However, such studies are based on comparing the survival potential of suspension-cultured epithelial cell lines with those adhering to ECM in short term assays. We have developed a model where primary mammary cells isolated directly from tissue are maintained on physiological substrata in longer term assays. These cells have not been selected for their ability to form cell lines and therefore retain an apoptotic potential that is as close as possible to that in vivo. They undergo apoptosis specifically after culture for several days on tissue culture plastic or collagen I, but not on BM (Pullan et al., 1996). The cells plated on a BM substratum form multicellular structures resembling alveoli in vivo (Barcellos-Hoff et al., 1989; Aggeler et al., 1991). However, this three-dimensional architecture did not appear to play a role in suppression of apoptosis (Fig. 2). Under conditions where the cells remained as monolayers, apoptosis was efficiently suppressed by diluted BM proteins precipitating over the cells.

Further verification that BM had a direct survival signaling role came from studies using function-blocking antibodies. Using a single cell assay developed previously for analysis of differentiation (Streuli et al., 1991), we found that although insulin was able to significantly suppress apoptosis in cells cultured within a three-dimensional BM gel, it was not able to do so if anti-laminin antibodies were included (Fig. 3). Laminin has previously been shown to act as a survival ligand for other cell lineages including fibrosarcoma and neuroblastoma cells and has been implicated in preventing apoptosis during myogenesis (Kim et al., 1994; Vachon et al., 1996; Bozzo et al., 1997; Vachon et al., 1997). Our results extend these previous findings and show that laminin also has a survival function in mammary epithelium. However, laminin may not be a survival ligand for all cells, since endothelial cells underwent apoptosis on laminin substrata but not on fibronectin or vitronectin (Wary et al., 1996). Thus, distinct cell types have a different requirement for ECM to rescue them from apoptosis. It remains to be determined whether or not the survival response to different ECM ligands is mediated through similar signaling pathways.

In several cell systems, integrins have been shown to be required for mediating a survival response (Frisch and Ruoslahti, 1997). We tested the possibility that integrins delivered survival signals in mammary cells by examining the survival of single cells in BM gels after treatment with function-blocking antibodies, and found that both α6 and β1 integrins were necessary to prevent the cells from undergoing apoptosis. Although β1 integrins have been shown to be required for mammary epithelial cell survival, our results augment these studies by providing the first direct evidence that α6 integrin functions in such a pathway (Boudreau et al., 1995; Howlett et al., 1995; Fig. 4). These findings have recently been confirmed in our experiments with primary human breast epithelia, where anti–α6 and anti–β1 integrin antibodies induced cell death, although antibodies to the α2 subunit did not. The latter result indicates a specificity of response as mammary cells express significant levels of α2 integrin, but it also suggests that the α2β1 integrin, a receptor for collagen, is not involved with suppressing mammary apoptosis.

The β1 and β3 integrin subunits have previously been shown to be survival receptors (Brooks et al., 1994; Boudreau et al., 1995; Howlett et al., 1995; Zhang et al., 1995; Scatena et al., 1998). RGD is a peptide substrate for αvβ3 integrin but we found that it had no effects on mammary cell survival. Furthermore, in preliminary immunostaining experiments with anti–αv integrin antibodies, we have only detected this subunit in the epithelial cells of virgin mammary gland and it appeared to be completely absent from pregnant and lactating tissue. Therefore, our data would not support the possibility that it is a survival receptor for mammary cells isolated from pregnant mammary gland, even though it does have this role in endothelia (Brooks et al., 1994; Scatena et al., 1998). In addition to the β1 integrin subunit, mammary cells also express β4 integrin, and both of these can partner the α6 subunit. Therefore, our data are not yet sufficient to confirm that laminin suppresses apoptosis through a direct cell interaction with the α6β1 integrin heterodimer, neither do they exclude the possibility that other integrin subunits are involved in survival. However, they do demonstrate that laminin is a ligand for survival and that the α6 and β1 subunits can act as survival receptors.

The metastatic breast cancer cell line, MDA-MB-435 expresses α6β1 integrin. Tumors induced by a derivative cell line in which α6β1 was functionally ablated were much smaller those of the parental cells, and showed a sixfold higher apoptotic/mitotic index (Wewer et al., 1997). These results suggest that α6 integrin provides an anti-apoptotic signal in vivo as well as in culture and together with the present study, they may explain why higher levels of α6 integrin in human breast carcinomas correlate with an increased likelihood of patient morbidity (Friedrichs et al., 1995). Moreover, they suggest that α6 integrin–ligand interactions might represent a target for therapeutic intervention in breast disease, especially in combination with agents which inhibit growth factor–mediated survival signals.

Since both BM and insulin were required for sustained prevention of apoptosis in mammary cells, it was possible that these ligands triggered either parallel or convergent survival pathways. An abundance of studies indicate that PI 3-kinase is essential for suppressing apoptosis in other cell systems, possibly through PKB which has been implicated as a downstream regulator of survival (Yao and Cooper, 1995; Minshall et al., 1996; Dudek et al., 1997; Kauffmann-Zeh et al., 1997; Kennedy et al., 1997; Khwaja et al., 1997; Kulik et al., 1997; Parrizas et al., 1997; Crowder and Freeman, 1998). Our experiments using inhibitors also demonstrated a requirement for PI 3-kinase in mammary cell survival (Fig. 7). Therefore, we examined whether this pathway was independently regulated by the two separate ligands or if BM controlled the ability of insulin to trigger its phosphorylation cascade, by measuring the levels of phosphotyrosine in its proximal signaling proteins.

Insulin rapidly induced tyrosine phosphorylation of its receptor. However, the signaling events downstream of the insulin receptor resulting in IRS-1 tyrosine phosphorylation and its association with PI 3-kinase were only propagated in cells cultured on BM and not on collagen I (Fig. 5). These results contrast with a study using CHO cells overexpressing insulin receptor, which showed that transient adhesion to fibronectin >20 min enhanced the insulin-induced phosphorylation of both the receptor and IRS-1 (Guilherme et al., 1998). Our data support the conclusion that ECM amplifies insulin signaling, but extends it by showing that once nontransfected primary epithelial cells have become established within an ECM environment over several days, they develop a selective sensitivity to growth factor signaling. Thus, the cellular environment determines whether the insulin signal can be propagated efficiently, and this only occurs when the cells contact a BM but not when they are on collagen I. Indeed, the results indicate the existence of an ECM-dependent restriction point in insulin signaling, which occurs downstream of insulin receptor phosphorylation.

The conclusion that BM and insulin cooperate to drive insulin signaling was confirmed in further experiments where we examined PKB phosphorylation, using both short and long term treatments with insulin (Fig. 6). Cell– matrix interactions had a similar effect of enhancing EGF signaling in MDCK cells, although in that study the PKB activation was examined in cells briefly attached to an ECM in comparison with those detached from a substratum (Khwaja et al., 1997). Together, these complementary results indicate that both the short and long term activation of enzymes in the PI 3-kinase signaling pathway requires adherent cells to be in the correct ECM microenvironment, which contributes to both EGF and insulin signaling.

At the current time we do not know whether PKB is absolutely required for mammary cell survival, but given the results from other cell systems, this seems likely. Candidates for the downstream link between PKB and the apoptotic machinery include glycogen synthase kinase-3, procaspase-9, and the pro-apoptotic protein Bad (Datta et al., 1997; del Peso et al., 1997; Cardone et al., 1998; Pap and Cooper, 1998). Although the phosphorylation of PKB and Bad have been dissociated from each other, primary cultures of mammary epithelial cells do express Bad, which therefore remains a potential target for regulation by PKB (Scheid and Duronio, 1998; our unpublished data).

Adhesion to ECM has previously been shown to suppress apoptosis in several cell types (Meredith et al., 1993; Frisch and Francis, 1994; Kim et al., 1994; Vachon et al., 1996; Wary et al., 1996; Bozzo et al., 1997; Khwaja et al., 1997). In addition, cell–ECM interactions trigger integrin-mediated downstream phosphorylation cascades involving MAP kinase and PI 3-kinase, providing possible mechanisms for ECM control of survival (Khwaja et al., 1997; King et al., 1997; Lin et al., 1997). However, in the mammary gland model we have now demonstrated that although integrin signaling is necessary for survival, sustained cell interactions with a BM substratum alone are not sufficient to suppress apoptosis. In addition, insulin or IGFs are also required. Similar conclusions have been reached in studies of the cell cycle where serum factors and ECM are both required for MAP kinase activation, cyclin E–cdk2 activation, and retinoblastoma phosphorylation (Zhu et al., 1996; Renshaw et al., 1997). Dual signals from soluble factors and ECM are also implicated in the control of mammary differentiation, where prolactin and BM are both necessary for triggering the prolactin signaling cassette, Stat5 DNA binding, and milk protein gene transcription (Streuli et al., 1995; Edwards et al., 1998).

Although our work shows that both laminin and insulin act as survival ligands for mammary cells, we cannot rule out the possibility that other types of ligand are important in regulating survival. Insulin, for example, was only able to suppress apoptosis of single mammary cells cultured within BM by about twofold, suggesting that other factors may play a role in mammary cell survival in vivo. Preliminary studies with primary cultures of human mammary epithelial cells, have demonstrated that apoptosis in the single cell assay can be reduced by insulin, but is further suppressed by EGF. But with mouse mammary cells, EGF does not have an additional survival effect over and above insulin (Oliver, J., and C.H. Streuli, unpublished data). Therefore, it may be that additional cytokines can suppress apoptosis still further than insulin and BM in the mouse system. Even if other mammary survival factors remain to be identified, our data are still sufficient to demonstrate that both insulin and integrin contribute to the survival of mammary cells.

In summary, these studies indicate that an important function of cell–ECM interactions is to modulate growth factor and cytokine responses. This suggests that current thinking about growth factor signaling in adherent cells should include a component from the ECM. One mechanism might be via direct or indirect associations between integrin and growth factor receptors, as has been demonstrated for αvβ3 integrin and the insulin and PDGF receptors, or with integrin and downstream components in the insulin signaling pathway, as shown with αvβ3 integrin and IRS-1 (Vuori and Ruoslahti, 1994; Schneller et al., 1997). The role of integrins and the cytoskeleton in signal transduction may be to provide a scaffold whereby components of growth factors cassettes are assembled into discreet domains within the cell, so that they can propagate signals efficiently. Indeed, a modular nature for membrane signaling complexes has been proposed, which may explain the recruitment of signaling proteins to cytoskeletal structures such as adhesion plaques (Miyamoto et al., 1995; Plopper et al., 1995; Pawson and Scott, 1997; Simons and Ikonen, 1997). Our current goal is to address this issue by determining whether the macromolecular organization of insulin signaling proteins and integrins in mammary cells cultured on BM are different to those in cells on collagen I.

The authors are grateful in particular to Dr. Andrew Gilmore and also to Drs. Neil Anderson, Gwynneth Edwards, David Garrod, Teresa Klinowska, and John Hickman for critical review of the manuscript.

BM

basement membrane

ECM

extracellular matrix

EHS

Engelbreth-Holm-Swarm tumor

IGF

insulin-like growth factor

IRS

insulin receptor substrate

PI 3-kinase

phosphatidylinositol 3-kinase

PKB

protein kinase B