Bacteroides fragilis toxin stimulates intestinal epithelial cell shedding and γ-secretase-dependent E-cadherin cleavage

Bacteroides fragilis are common colonic commensals that colonize the majority of humans (Moore and Holdeman, 1974). Despite comprising a minority (∼0.1%) of the total fecal flora, this organism emerges as the leading anaerobic pathogen in bacteremia and is found in the vast majority, if not all, intraabdominal abscesses (Redondo et al., 1995; Polk and Kasper, 1977). Molecular genetic studies indicate that B. fragilis are highly heterogeneous and a subset of strains are termed enterotoxigenic B. fragilis (ETBF) (Franco et al., 1999). ETBFs are associated with diarrheal disease in children, adults and domestic livestock as well as active inflammatory bowel disease (IBD) (Sears, 2001; Basset et al., 2004; Prindiville et al., 2000). However, ETBFs also asymptomatically colonize up to 35% of humans (Basset et al., 2004). To date, the virulence of ETBF strains is ascribed to secretion of a 20-kDa metalloprotease toxin termed the B. fragilis toxin (BFT) (Sears, 2001).

ETBFs produce BFT as a 44 kDa preproprotein toxin that is processed by the organism and secreted as a mature 20 kDa protein (Franco et al., 2002). BFT induces marked changes in intestinal epithelial cell (IEC) function in vitro and in vivo, which are initiated when BFT binds specifically to a IEC receptor that is not E-cadherin (Wu et al., 2006). Biological activity and specific binding of BFT to IECs is mitigated by mutation of the protease motif (Franco et al., 2005; Wu et al., 2006). After IEC binding, BFT stimulates multiple host cell changes including: morphological changes in IECs as well as in polarized renal and pulmonary cell lines (Sears, 2001); reduced barrier function and increased ion transport in intestinal cell monolayers and/or human colonic epithelium (Chambers et al., 1997; Riegler et al., 1999); IEC proliferation dependent on E-cadherin expression and, in part, β-catenin–T-cell-factor (TCF) signaling (Wu et al., 2003); and cytokine secretion by IECs, most notably interleukin-8 (IL-8) (Wu et al., 2004; Kim et al., 2001; Sanfilippo et al., 2000). Consistent with these in vitro results, histopathology of animal intestinal tissue treated with ETBF or BFT reveals disruption of the epithelial layer with a mixture of acute and chronic inflammatory cells in the lamina propia (Obiso, Jr et al., 1995; Sears et al., 1995). Human intestinal histopathology from ETBF disease is not yet published.

E-cadherin is a 120 kDa glycosylated Type I transmembrane protein critical to the formation of intercellular adhesion junctions (zonula adherens in the intestinal epithelium) and to cellular signaling, proliferation and differentiation (Nelson and Nusse, 2004). The cytoplasmic domain of E-cadherin associates with β-catenin, which is in turn tethered to α-catenin and actin (Jou et al., 1995). At the cellular level, BFT induces rapid cleavage (within 1 minute) of E-cadherin yielding sequentially 33 kDa and 28 kDa cell-associated E-cadherin fragments in the human colonic carcinoma cell line HT29/C1 (Wu et al., 1998). E-cadherin cleavage by BFT releases β-catenin to the cytoplasm resulting in β-catenin nuclear localization and stimulation of β-catenin–TCF-dependent cellular proliferation (Wu et al., 2003).

Recently, basal and agonist-induced (such as by staurosporine or ionomycin) E-cadherin processing has been reported to involve sequential steps that include cleavage of the E-cadherin ectodomain by host matrix metalloproteinase 7 (MMP-7 or matrilysin) or ADAMs (adisintegrin and metalloprotease) (Marambaud et al., 2002; Ito et al., 1999; Maretzky et al., 2005; Reiss et al., 2005; McGuire et al., 2003; Noe et al., 2001; Steinhusen et al., 2001) followed by processing of the intracellular E-cadherin domain, a 38 kDa protein (in mouse embryos and A431 human epithelial cells) by γ-secretase and caspase-3 (Marambaud et al., 2002). ADAMs, a family of at least 32 transmembrane zinc-dependent host proteases, are implicated in the ectodomain shedding of various membrane-bound proteins (Blobel, 2005). Presenilin-1 or γ-secretase (termed γ-secretase hereafter) is a 48 kDa transmembrane protein that associates directly and independently with both E-cadherin and β-catenin and regulates the stability and function of the cadherins or catenin adhesion complex (Baki et al., 2001; Murayama et al., 1998; Kang et al., 2002). γ-Secretase is an aspartyl protease of the intramembrane cleaving protease family (termed iCLiPs) and appears to be a promiscuous enzyme that cleaves various Type I transmembrane domain proteins after they have undergone ectodomain shedding (Kopan and Ilagan, 2004; Wolfe and Kopan, 2004; Chyung et al., 2004). Given these new insights into cadherin biology, the goal of this study was to investigate in greater detail the mechanism by which BFT induces E-cadherin cleavage.

Using a polyclonal antibody to the C-terminal 25 amino acids of the cytoplasmic domain of E-cadherin (E2 antibody), Fig. 1A,B demonstrates that, consistent with previous data (Wu et al., 1998), BFT treatment of HT29/C1 cells stimulates rapid cleavage of intact 120 kDa E-cadherin yielding 33 kDa and 28 kDa cell-associated fragments of E-cadherin that are degraded over time. To determine whether the extracellular domain of E-cadherin is shed after BFT treatment, HT29/C1 cells were treated with BFT for 30 minutes to 3 hours followed by concentration of the proteins in the culture supernatant by TCA precipitation. Western blot analysis of the cell culture supernatant proteins using an antibody to the extracellular region of E-cadherin (H-108 antibody) revealed the release of an 80 kDa protein band in the cell supernatants of BFT-treated cells over time compared with a faint signal in the control samples (Fig. 1A,B). Using the H-108 antibody, the 80 kDa band was the only protein band detected in BFT-treated cell supernatants. To determine whether E-cadherin was the only cell surface protein released after BFT treatment of HT29/C1 cells, total cell membrane protein shedding was examined by biotinylation of cell membrane proteins prior to BFT treatment. As shown in Fig. 1C, unexpectedly, multiple biotinylated proteins as well as the E-cadherin ectodomain (80 kDa) were released from the HT29/C1 cell surface after 1 hour of BFT treatment when compared to untreated control cells. These results demonstrate that BFT induces shedding into the cell supernatant of a 80 kDa E-cadherin fragment as well as other, as yet unknown, membrane proteins.

To assess whether BFT was responsible for the release of the 80 kDa ectodomain of E-cadherin, HT29/C1 cells were treated with partially purified culture supernatants of recombinant B. fragilis strains secreting wild-type BFT or biologically inactive mutant BFT-H352Y and E-cadherin cleavage fragments were assessed in both HT29/C1 cell supernatants and lysates. Treatment of HT29/C1 cells with bacterial culture supernatants containing wild-type BFT released the E-cadherin 80 kDa ectodomain and cleavage of intact 120 kDa E-cadherin in cell lysates similar to that observed in response to purified BFT (Fig. 1D). In contrast, neither release of the 80 kDa E-cadherin ectodomain nor cleavage of E-cadherin in HT29/C1 cell lysates was detected in HT29/C1 cells treated with bacterial culture supernatants containing mutant BFT-H352Y (Fig. 1D). These results indicate that biologically active BFT is required to stimulate E-cadherin cleavage.

MMP-7 as well as ADAM10, but not ADAM17, mediates E-cadherin ectodomain shedding in selected epithelial cells (McGuire et al., 2003; Maretzky et al., 2005; Noe et al., 2001). In other cells, ADAM10 and ADAM17 appear to exhibit overlapping substrate profiles (Blobel, 2005). Because mutations in the protease domain of BFT ablate binding of BFT to IECs and hence the biological activity of BFT (Wu et al., 2006; Franco et al., 2005), we tested whether MMP-7 and/or ADAM10 or ADAM17 mediate E-cadherin cleavage after BFT cellular binding. We initially tested the role of MMP-7 by determining whether recombinant active MMP-7 (3.3 μg/ml) mimicked the biological activity of BFT by stimulating E-cadherin cleavage and/or cell shape changes in HT29/C1 cells over 24 hours. MMP-7-treated HT29/C1 cells developed a spindle morphology distinct from the cell rounding and separation observed in BFT-treated cells and no cleavage of E-cadherin was detected at 3 or 24 hours suggesting that MMP-7 does not mimic or mediate BFT-initiated E-cadherin cleavage. Using RNA interference to directly deplete HT29/C1 cell MMP-7 also did not inhibit BFT-initiated E-cadherin cleavage. However, HT29/C1 cells depleted of MMP-7 did not grow as well as control cells and displayed less E-cadherin (Fig. 2A). We also observed that, although expression of MMP-7 was induced by BFT after 24 hours, the active form of the protein was not detected by western blot in control or BFT-treated HT29/C1 cells, further suggesting that MMP-7 is not activated by BFT (data not shown).

RNA interference was also used to test whether ADAM10 or ADAM17 mediate BFT-initiated E-cadherin cleavage. Fig. 2B,C shows that active ADAM10 and ADAM17 were nearly completely depleted without modifying either basal levels of E-cadherin or BFT-initiated E-cadherin cleavage. Similar results were obtained with dual ADAM10 and ADAM17 depletion by RNA interference (data not shown).

Our previous data suggested that the initial generation of the 33 kDa C-terminal E-cadherin fragment by BFT was not dependent on cellular ATP, whereas the subsequent degradation of this 33 kDa E-cadherin fragment was dependent on cellular ATP (Wu et al., 1998). This suggested that degradation of the 33 kDa E-cadherin fragment was probably attributable to cellular proteases. Recent data has identified E-cadherin as one potential substrate for γ-secretase (Marambaud et al., 2002). Thus, we tested whether degradation of the C-terminal E-cadherin fragments generated by BFT treatment of HT29/C1 cells was dependent on γ-secretase using the selective γ-secretase inhibitor, L-685,458. As shown in Fig. 3A, compared with HT29/C1 cells treated only with BFT, pretreatment of HT29/C1 cells with L-685,458, but not the β-secretase inhibitor, AEBSF, inhibits degradation of the BFT-generated 33 kDa E-cadherin fragment to a 28 kDa E-cadherin fragment. This result was verified using RNA interference to directly deplete HT29/C1 cell γ-secretase (Fig. 2D).

Previous data suggest that γ-secretase forms a protein complex binding directly and independently (via distinct amino acid residues) to both E-cadherin and β-catenin (Baki et al., 2001; Murayama et al., 1998; Kang et al., 2002). We previously observed that BFT cleavage of E-cadherin in HT29/C1 cells results in nuclear translocation of a portion of the cellular β-catenin with proteasomal degradation of the remaining β-catenin by 3 to 6 hours after BFT treatment of the cells (Wu et al., 2003) (our unpublished data). To further examine the effect of γ-secretase inhibition on E-cadherin cleavage and β-catenin distribution in intact cells, HT29/C1 cells were treated with BFT for 1 to 6 hours in the presence or absence of L-685,458 and assessed by immunofluorescent confocal microscopy (Fig. 3B). In untreated control cells and cells pretreated with L-685,458 for 30 minutes, E-cadherin is located at the cell membrane and colocalizes with β-catenin (Fig. 3B, panels a,e). In HT29/C1 cells treated with BFT (5 nM), E-cadherin and β-catenin co-association is lost by 1 hour, with diffusion of the signals to the cytoplasm over time (panels b-d). By contrast, in HT29/C1 cells pretreated with L-685,458, BFT treatment also results in loss of E-cadherin and β-catenin co-association; however, a portion of the cellular β-catenin remains distinctly anchored on the cell membrane for at least 6 hours after BFT treatment (panels f-h). However, by 6 hours, membrane-associated β-catenin in HT29/C1 cells treated with BFT and L-685,458 is focal rather than uniform on the HT29/C1 cell membrane as at earlier time points. By contrast, a portion of the C-terminal E-cadherin fragment appears to lose its membrane staining and becomes cytoplasmic during the time course of BFT treatment even in the presence of L-685,458. Additional time course analyses indicated that total cellular β-catenin significantly declines over 6 hours in BFT-treated cells, consistent with our previously reported data (Wu et al., 2003), whereas β-catenin levels are maintained in HT29/C1 cells treated with BFT in the presence of L-685,458 (Fig. 3C,D). These data suggest that γ-secretase regulates BFT-induced cytoplasmic release of membrane-associated β-catenin as well as β-catenin degradation.

To investigate the protein-protein associations of γ-secretase, β-catenin and E-cadherin in HT29/C1 cells, γ-secretase was immunoprecipitated from HT29/C1 cell lysates treated or not with BFT (5 nM) and/or L-685,458 (1.5 μM) and co-precipitated E-cadherin, β-catenin and γ-secretase was examined by western blot. As shown in Fig. 3D, in untreated cells or cells pretreated for 30 minutes with L-685,458, both full-length E-cadherin (120 kDa) and β-catenin are associated with γ-secretase. Treatment of HT29/C1 cells with BFT over time leads to near complete loss of association of γ-secretase with E-cadherin initially and then β-catenin (Fig. 3D, left 4 lanes). By contrast, γ-secretase remains associated with the 33 kDa E-cadherin fragment and β-catenin in cells treated with BFT and L-685,458 (Fig. 3D, four right-hand lanes). Consistent with these results, Western blot analysis of Triton-X-100-soluble and -insoluble cell fractions revealed that a portion of the 33 kDa E-cadherin fragment and β-catenin are present in the Triton-X-100-insoluble cell fraction after 3 hours of treatment with BFT and L-685,458 but is not detectable in either fraction after BFT treatment alone (data not shown). Together these data suggest that γ-secretase regulates BFT-induced E-cadherin cleavage in HT29/C1 cells and that γ-secretase co-associates with a portion of the E-cadherin cytoplasmic domain and β-catenin. A portion of this complex may dissociate from the cell membrane over time (Fig. 3B, panel h).

We previously reported that BFT stimulates not only E-cadherin cleavage but also nuclear localization of β-catenin and β-catenin–TCF-dependent cellular proliferation (Wu et al., 2003). To evaluate the impact of γ-secretase inhibition on BFT-induced β-catenin–TCF nuclear signaling, HT29/C1 cellular proliferation was measured using [³H]thymidine incorporation into parental HT29/C1 and HT29/C1 TCF-dominant negative (HT29/C1::TcfΔN31) cell lines. Because β-catenin nuclear signaling requires co-association with TCF, TCF dominant negative HT29/C1 cells disrupt β-catenin-dependent nuclear signaling (Wu et al., 2003). Time course analysis indicated that L-685,458 inhibited the degradation of the 33 kDa C-terminal E-cadherin fragment generated by BFT treatment of HT29/C1 cells for at least 48 hours and did not alter cell viability as assessed by trypan blue exclusion (data not shown). Fig. 4 shows that, consistent with previous data, BFT alone stimulates a significant increase in [³H]thymidine uptake versus control parental HT29/C1 cells (lane 1, P<0.001) that is not diminished when the parental HT29/C1 cells are treated with BFT in the presence of L-685,458 (lane 3 vs lane 1, P=NS). Unexpectedly, L-685,458 alone significantly reduced [³H]thymidine uptake in parental HT29/C1 cells (lane 2, –34±13%, L-685,458-treated cells vs control, P<0.001).

In TCF dominant negative HT29/C1 cells, as previously reported (Wu et al., 2003), BFT stimulates a significant increase in [³H]thymidine incorporation in BFT-treated cells compared with the control (Fig. 4, lane 4, P<0.001) but the [³H]thymidine uptake stimulated by BFT in HT29/C1 cells lacking functional TCF is ∼40% lower than in BFT-treated parental cells (comparison between lanes 1 and 4, P<0.05). These results indicate that β-catenin–TCF nuclear signaling accounts for a portion, but not all, BFT-initiated HT29/C1 cell proliferation. Similar to the parental HT29/C1 cells, L-685,458 did not inhibit BFT stimulation of [³H]thymidine uptake in TCF dominant negative HT29/C1 cells (lane 6, BFT + L-685,458 vs lane 4, BFT alone, P=NS). By contrast, the ability of L-685,458 to inhibit [³H]thymidine incorporation in parental HT29/C1 cells is obliterated in TCF dominant negative HT29/C1 cells (comparison between lanes 2 and 5, P<0.01). These results suggest that γ-secretase does not regulate BFT-induced HT29/C1 cell proliferation but does regulate basal TCF-dependent HT29/C1 cell proliferation.

Bacteroides fragilis are commensals colonizing the majority of humans and are the most frequent Bacteroides species identified at the mucosal surface of the colon where they probably contribute to intestinal homeostasis (Moore and Holdeman, 1974; Namavar et al., 1989; Hooper and Gordon, 2001; Coyne et al., 2005). Virulence of B. fragilis is ascribed, in part, to the unparalleled diversity of the polysaccharide capsule (Krinos et al., 2001) and, for ETBF strains, to the production of a zinc-dependent metalloprotease toxin termed BFT (Sears, 2001). We previously reported that BFT induces E-cadherin cleavage on intestinal and other polarizable epithelial cell lines resulting in adherens junction disassembly (Chambers et al., 1997; Wu et al., 1998), which initiates IEC proliferation dependent on TCF–β-catenin signaling (Wu et al., 2003).

We have recently demonstrated that BFT binds to a specific HT29/C1 cell membrane receptor; that specific BFT binding requires BFT protease activity; and that the BFT receptor is unlikely to be either E-cadherin or one of the described protease-activated receptors (PAR1-4) (Wu et al., 2006). The data reported here extend our understanding of the mechanism by which BFT acts to induce disruption of the adherens junction and also suggest that signaling from the adherens junction in IECs is more complex than previously reported. Our results indicate that the proteolytic activity of BFT is necessary for shedding of the 80 kDa E-cadherin ectodomain but do not define whether BFT directly cleaves or indirectly stimulates a host cell protease to cleave the E-cadherin ectodomain. Because E-cadherin alone is insufficient to permit specific IEC binding of BFT, we propose a model of BFT action (Fig. 5) in which BFT binds to a specific host cell receptor activating a host cell protease directly or by signal transduction yielding shedding of the E-cadherin ectodomain as well as other cell membrane proteins. This model in which the host cell is proposed to mediate BFT-initiated E-cadherin cleavage is consistent with recent data indicating that E-cadherin undergoes stepwise cleavage in response to activation of cellular Ca²⁺-dependent signal transduction with release of the E-cadherin ectodomain (Marambaud et al., 2002; Ito et al., 1999; Steinhusen et al., 2001). Further dissection of the initial steps in BFT interaction with the cell and the role(s) of the BFT protease domain and/or cellular proteases in the mechanism of action of BFT has been limited because we have not identified a protease inhibitor that specifically inhibits eukaryotic proteases, but not BFT. Similarly, mutational analyses of BFT suggest it is a highly conserved protein without, as yet, a defined binding domain distinct from its protease domain (Franco et al., 2005; Sears et al., 2006).

The host cell proteases involved in ectodomain cleavage of E-cadherin have been examined in select cell lines with a role identified for MMP-7 in lung epithelium (McGuire et al., 2003) and ADAM10, but not ADAM17, in fibroblast and keratinocyte cell lines (Maretzky et al., 2005). A similar role for ADAM10 in shedding of the N-cadherin ectodomain has been reported (Reiss et al., 2005). Additional data, however, suggest that ADAM10 and ADAM17 (also known as tumor necrosis factor-α converting enzyme or TACE) may, in some cells, modify similar substrates (Blobel, 2005). Thus, we tested, but did not identify any evidence that MMP-7, ADAM10 or ADAM17 mediates BFT-initiated E-cadherin cleavage. In particular, siRNA depletion of MMP-7, ADAM10 or ADAM17 was nearly complete but did not modify the time course of BFT-initiated E-cadherin cleavage (Fig. 2 and data not shown). In addition, ADAM10 has been proposed to regulate physiological E-cadherin cleavage in selected cells (Maretzky et al., 2005). Thus, we expected, but did not observe, increased basal intact E-cadherin levels in ADAM10-depleted cells. Our results suggest that, in contrast to other cell lines, neither ADAM10 nor ADAM17 are critical to basal or BFT-stimulated E-cadherin cleavage in an IEC model cell line, HT29/C1 cells. Additional studies using, for example, stable inducible knockouts of these enzymes in intestinal epithelium will be necessary to definitively evaluate the role of ADAMs in IEC E-cadherin processing.

Interestingly, our data show that BFT induces shedding of multiple cell membrane proteins. Although the release of cellular ligands and receptors by proteases may serve to amplify cell signaling induced by an agonist (Koon et al., 2004), a specific role of the E-cadherin ectodomain as a cell signaling agonist has not yet been described (Kopan and Ilagan, 2004). Nonetheless, it seems likely that one or more of the proteins released by BFT may act to modify the cellular response. We are presently defining the identity of the BFT-released cell membrane proteins.

After BFT stimulates shedding of the E-cadherin ectodomain, processing of the residual E-cadherin cytoplasmic domain appears to be mediated by γ-secretase, similarly to previously reported data in mouse embryos and A431 cells (Marambaud et al., 2002). Prior reports also indicate that cleavage of E-cadherin with dissociation of the adherens junction by BFT, Ca²⁺ signaling or staurosporine results in cytoplasmic redistribution of the nuclear signaling protein, β-catenin (Wu et al., 2003; Marambaud et al., 2002; Ito et al., 1999). γ-secretase has been proposed to inhibit cytoplasmic, but not membrane-associated, β-catenin signaling and this regulatory role has been reported in embryonic cells as distinct from the proteolytic activity of γ-secretase (Murayama et al., 1998; Kang et al., 2002). Consistent with this, loss of γ-secretase was reported to enhance β-catenin signaling and skin tumor development in vivo (Xia et al., 2001). In contrast to these results, our data suggest that γ-secretase activity positively regulates a membrane- and cytoskeleton-associated pool of β-catenin that acts in concert with TCF to contribute to basal HT29/C1 cell replication. Consistent with our results, in vivo treatment of rodents with γ-secretase inhibitors nearly halted intestinal epithelium proliferation, modified IEC differentiation and altered small intestine morphology (Searfoss et al., 2003; Wong et al., 2004; van Es and Clevers, 2005). The impact of in vivo γ-secretase inhibition on colonic morphology was not reported. However, γ-secretase inhibition did not alter BFT-induced cellular proliferation in parental or TCF dominant negative HT29/C1 cells. Together these results reveal, to our knowledge, the first evidence for two differentially regulated or heterogeneous pools of β-catenin associated with the adherens junction of IECs. BFT-induced cellular proliferation in HT29/C1 cells, although dependent on E-cadherin expression and partially dependent on β-catenin–TCF signaling (Wu et al., 2003), is independent of γ-secretase-mediated E-cadherin cleavage whereas basal cell proliferation is dependent on γ-secretase-regulated β-catenin–TCF signaling (Fig. 4).

Previous data indicate that β-catenin associates with E-cadherin and that both proteins bind to γ-secretase at distinct sites (Murayama et al., 1998; Baki et al., 2001; Kang et al., 2002). Thus, one possibility to explain our results is that BFT treatment of cells leads to a biochemical modification of a portion of E-cadherin-associated β-catenin contributing to its cytoplasmic and nuclear redistribution whereas unmodified β-catenin remains regulated by γ-secretase. Initial studies using antibodies to mono- or triphosphorylated β-catenin have not identified differences in these β-catenin species between HT29/C1 cells treated with BFT in the presence or absence of L-685,458 (S.W. and C.L.S., unpublished data). However, additional work is necessary to explore this possibility. Unlike other γ-secretase-regulated pathways, such as proteolysis of Notch, where the released cytoplasmic domain of the target protein acts as a transcriptional regulator, the E-cadherin cytoplasmic domain has not yet been identified to act as a transcriptional regulator (Kopan and Ilagan, 2004; Wolfe and Kopan, 2004). Our preliminary data suggest BFT does not upregulate expression of Notch-regulated genes (S.W. and C.L.S., unpublished data). Although mutations in γ-secretase account for the majority of early-onset familial Alzheimer's disease (Brunkan and Goate, 2005), it is unknown whether mutations in γ-secretase contribute to IEC signaling disturbances resulting in intestinal pathology or disease.

Bacterial toxins have long served as tools to investigate normal and pathological cellular function. BFT shares sequence homology with the eukaryotic MMPs of the metzincin family and has been proposed to be the ancestral origin of the eukaryotic MMPs (Massova et al., 1998). Further studies to detail the mechanism of action of BFT will contribute to our understanding of IEC signaling pathways and cell function.

HT29/C1 cells are cloned cells derived from a human colon carcinoma (obtained from Dr Daniel Louvard, Institut Pasteur, Paris, France) (Huet et al., 1987; Montrose-Rafizadeh et al., 1991). The HT29/C1::TCF-4ΔN31 cell line is a derived polyclonal stable TCF dominant negative HT29/C1 cell line (Wu et al., 2003). Both cell lines were grown subconfluently in Dulbecco's minimum essential medium (DMEM) containing streptomycin (0.1 mg/ml), penicillin (100 U/ml), G418 (0.4 mg/ml for HT29/C1::TcfΔN31 cell line only), human transferrin (10 μg/ml, Sigma, St Louis, MO) and 10% fetal bovine serum (HyClone, Logan, Utah). All culture media and reagents were purchased from GIBCO BRL Life Technologies (Rockville, MD) unless otherwise stated.

BFT was purified from culture supernatants of B. fragilis recombinant strain I1345(pFD340::P-bft) as previously described (Wu et al., 2002). In some experiments, filter-sterilized, concentrated culture supernatants of B. fragilis recombinant strains 9343(pFD340::P-bft, secretes wild-type BFT) and 9343(pFD340::P-bftΔH352Y, secretes mutant biologically inactive BFT owing to a single nucleotide point mutation in the BFT metalloprotease domain) (Franco et al., 2005) were used for the treatment of HT29/C1 cells. Similar quantities of wild-type and mutated inactive BFT were present in culture supernatants as assessed by western blot analysis using anti-BFT antibody (data not shown).

HT29/C1 cells were washed once with Hank's balanced salt solution before treatment with purified BFT at specified concentrations in DMEM lacking serum. Cells were incubated with the γ-secretase inhibitor (L-685,458; Calbiochem, San Diego, CA) or β-secretase inhibitor [(4-(2-aminoethyl) benzenesulfonylfluoride (AEBSF), Sigma] for 30 minutes before the addition of BFT and then continuously during the experiment unless otherwise described. Active human recombinant matrilysin (MMP-7; specific activity ⩾3000 U/mg) was obtained from Calbiochem.

E2: polyclonal anti-C-terminal of E-cadherin antibody (provided by James Nelson, Stanford University, Palo Alto, CA); C36: monoclonal anti-C-terminal E-cadherin antibody (BD Biosciences, Palo Alto, CA); H108: polyclonal anti-extracellular E-cadherin domain antibody (Santa Cruz Biotechnology, Santa Cruz, CA); CAT-5H10: monoclonal anti-β-catenin antibody (Zymed Laboratories Inc., San Francisco, CA); polyclonal anti-α-presenilin-1-loop antibody (Calbiochem); monoclonal anti-presenilin-1-loop antibody (Chemicon International, Temecula, CA); AC-40 or AC-15: monoclonal anti-actin antibody (Sigma); monoclonal anti-biotin antibody (Jackson ImmunoResearch Laboratories, West Grove, Pa); polyclonal anti-cytoplasmic ADAM10 or ADAM17 domain antibody (Axxora, San Diego, CA); polyclonal anti-active/latent MMP-7 antibody (Chemicon).

For western blot analyses, proteins were probed using the primary antibodies as indicated then horseradish peroxidase-coupled secondary antibodies (Jackson ImmunoResearch Laboratories) after 10% SDS-PAGE separation and nitrocellulose membrane transfer (Sambrook et al., 1989). Immunoreactive proteins were detected using Super Signal West Pico Chemiluminiscent Substrate (Pierce, Rockford, IL) and quantified by densitometry using AlphaEase™ (version 5.1; Alpha Innotech Corporation, San Leandro, CA).

To detect released E-cadherin fragments in HT29/C1 cell culture supernatants, proteins in the culture supernatants were precipitated with 10% trichloroacetic acid followed by western blot using the H108 antibody. To detect cell membrane proteins shed from cells treated with BFT, HT29/C1 cell surface proteins were biotinylated using EZ-Link Sulfo-NHS-Biotin Reagents (Pierce, Rockford, IL) as per the manufacturer's instructions. Proteins shed into the culture supernatants were retrieved using streptavidin beads, eluted with 0.05 M glycine (pH 2.5) and detected by western blot using an anti-biotin antibody.

Cell proteins were extracted with 1% Triton X-100 in phosphate-buffered saline for 10 minutes at 4°C. The supernatant (Triton X-100 soluble fraction) was collected and the residual portion (Triton X-100 insoluble fraction) was solubilized with 1% SDS-lysis buffer (Nathke et al., 1994). The proteins from both fractions were analyzed by western blot as described.

Small interfering RNA (siRNA) duplex oligoribonucleotides against human ADAM10, ADAM17, MMP-7 or γ-secretase were selected and synthesized by Invitrogen (Stealth™ RNAi, Carlsbad, CA). The sequences were as follows: (i) ADAM10 sense 5′-GAGGAAAUACCAGAUGACUGGUGUA-3′, antisense 5′-UACACCAGUCAUCUGGUAUUUCCUC-3′, (ii) ADAM 17 sense 5′-GGAAGCUGACCUGGUUACAACUCAU-3′, antisense 5′-AUGAGUUGUAACCAGGUCAGCUUCC-3′; (iii) MMP-7 sense 5′-CCCGCGUCAUAGAAAUAAUGCAGAA-3′, antisense 5′-UUCUGCAUUAUUUCUAUGACGCGGG-3′; (iv) γ-secretase sense 5′-GCUCAGGAGAGAAAUGAAACGCUUU-3′, antisense 5′-AAAGCGUUUCAUUUCUCUCCUGAGC-3′. Control siRNA (Cat.No. 12935-300) was purchased from Invitrogen. Two hours after cell plating (5×10⁴ cells/well using 24-well plate), the siRNA oligoribonucleotides were transfected into HT29/C1 cells using DharmaFECT™ 4 reagent (Dharmacon, Lafayette, CO) according to the manufacturer's protocol followed by BFT treatment 3 days later. After washing with cold PBS, cells were lysed with RIPA buffer [Sigma; 50 mM Tris-HCl (pH 7.2-8.0), 150 mM sodium chloride, 1.0% NP-40, 0.5-1% sodium deoxycholate, 0.1% SDS] supplemented with protease inhibitors (Roche Diagnostics, Indianapolis, IN) followed by western blot analysis.

HT29/C1 cells were lysed in RIPA buffer and incubated with the desired antibody (1-5 μg) overnight at 4°C. Protein A-Sepharose beads (CL-4B, Amersham Biosciences, Piscataway, NJ) were added and incubated at 25°C for 1 hour. Proteins were eluted from the beads in 50 μl of 0.1 M citric acid (pH 2.5).

HT29/C1 cells grown on eight-well chamber slides were fixed with 4% paraformaldehyde (Sigma) and permeabilized with 1% Triton X-100. Specific proteins were immunostained with antibodies as indicated followed by incubation with Alexa Fluor 488-labeled or Alexa Fluor 568-labeled secondary antibody (Molecular Probes, Eugene, OR). The signal was observed by single or dual channel confocal microscopy (Zeiss, LSM410). The images were processed using Metamorph (Universal Imaging, West Chester, Pa) or Photoshop 8.0.

HT29/C1 or HT29/C1::TCF-4ΔN31 cells in a 96-well plate treated with or without BFT in the presence or absence of L-685,458 were exposed to [³H]thymidine (1 μCi/ml) for 3 hours prior to the designated end of the experiment 48 hours after initiation of BFT treatment as previously described (Wu et al., 2003).

Data are expressed as means ± s.d. GraphPad Instat Version 3.05 was used for Student's t-test or ANOVA determinations to examine for significant statistical differences considered to be P values less than 0.05.

The authors thank Dwight Derr for assistance with tissue culture and the laboratory of Anirban Maitra for assessment of Notch-regulated gene expression. Supported by grant RO1 DK 45496 (to C.L.S.) and R24 DK 64388 (PI: Mark Donowitz).