MAPK scaffolding by BIT1 in the Golgi complex modulates stress resistance

The organization of biological membranes into microdomains is now believed to have a key role in cellular processes such as protein targeting and trafficking, and cell signaling (Brown and London, 1998; Ikonen and Simons, 1998). Microdomains from the plasma membrane (PM), which are isolated based on resistance to detergent solubilization, have been identified in almost all mammalian cell types, although they have not been well characterized in other eukaryotes, with the exception of Saccharomyces cerevisiae (Kubler et al., 1996) and Tetrahymena (Zhang and Thompson, 1997). In the baker's yeast, membrane microdomains detected in the endoplasmic reticulum (ER) were shown to be crucial for the trafficking of several specific proteins such as Gas1p or Pma1p (Bagnat et al., 2000). In addition, in mammalian cells, specific membrane microdomains are essential for retrograde trafficking of Shiga-like toxin from the PM to the ER and cytosol (Smith et al., 2006). Together, these data indicate that the existence of membrane microdomains in the ER might be significantly relevant, at least to membrane and/or protein trafficking. At the PM, the existence of microdomains has been linked to numerous transmembrane receptors signaling and trafficking [e.g. insulin receptor or tumor necrosis factor receptors (TNFRs)] (Bickel, 2002; Doan et al., 2004; Harder, 2004; Legler et al., 2003; Miljan and Bremer, 2002). With regards to the presence of similar structures in the ER, it is of particular interest to underline that TNFR signaling displays some similarities with an ER stress-specific signaling pathway mediated by inositol requiring enzyme 1 (IRE1), because both proteins recruit TNFR-associated factor 2 (TRAF2) upon activation (Schneider-Brachert et al., 2004; Schroder and Kaufman, 2005). Since TNFR signaling and trafficking are regulated by the presence of lipid microdomains (Legler et al., 2003), we postulated that the existence of similar structures in the ER might also have a role in connecting trafficking and signaling in the early secretory pathway.

In the present study, we aimed to identify new components of ER microdomains and characterize their biological function, as well as their potential role in such membranous structures. To this end, we purified ER microdomains, which were then subsequently analysed using proteomics. This led us to demonstrate the presence of the Bcl-2 inhibitor of transcription (BIT1) in ER microdomains. Although this protein was previously reported to localize to mitochondria (Jan et al., 2004), our localization studies revealed the presence of BIT1 in the secretory pathway, and especially its accumulation in the Golgi complex. Interestingly, forced localization of BIT1 in the ER led to the constitutive activation of the ERK-MAPK pathway and subsequent alterations of cell phenotypes.

Our results suggest a role for BIT1 in the early secretory pathway as a key regulator of the ERK-MAPK pathway, which impacts cell adaptation to stress and resistance to death. In addition, these observations highlight the potential role of the Golgi complex as a signaling platform that regulates the life and death decisions of a cell.

Previous reports demonstrate that highly specialized microdomain structures are partitioned in the ER membrane similarly to those observed at the plasma membrane. We therefore attempted to enrich ER microdomains from dog pancreatic rough microsomes that were resistant to solubilization by detergents. Dog pancreatic rough ER microsomes were solubilized using either Tween20 (TW) or Triton X-100 (TX) at 4°C. ER microdomains were then isolated by flotation on two successive discontinuous sucrose gradients (Fig. 1A), as reported previously (Bagnat et al., 2000). The resulting fractions were resolved by SDS-PAGE and stained with Coomassie blue R-250. Treatment of ER microsomes with TX led to the almost complete solubilization of the membrane fraction, because no specific protein pattern was observed following Coomassie blue staining of the gel (Fig. 1B). By contrast, solubilization with TW resulted in a specific protein-distribution pattern, suggesting the existence of TW-resistant ER microdomains. Each individual stained band obtained from the TW solubilization was then excised from the gel, reduced, alkylated and trypsin digested before mass spectrometry sequencing (Bell et al., 2001). Among the 103 proteins identified in the ER structures that were resistant to solubilization by TW, nine functional groups were generated (Fig. 1B and supplementary material Table S1) and were represented as follows: (1) translocation (13.7%); (2) protein folding and early post-translational modifications (8.8%); (3) trafficking (7.8%); (4) cytoskeleton organization (5.9%); (5) metabolism (9.8%); (6) signaling (12.7%); (7) cargo proteins (3.9%); (8) translation (4.9%); (9) ribosomal proteins (33.3%). The final group was interesting because membrane microdomains purified by floating were associated with heavy ribosomal moieties (probably the 80S fraction). Out of these 103 proteins, 41.2% were identified as resident in the ER or the ER-Golgi intermediate compartment (ERGIC), whereas 49% were cytosolic (but membrane associated), and 2.5% were proteins from the late secretory pathway. Finally, nine novel proteins or proteins of unknown function were identified, eight of which were predicted to localize in the ER by computational analysis.

As a small number of potential mitochondrial proteins were identified in our proteomics screen, we first monitored the quality of dog pancreatic rough microsomes used in this study. Mitochondrial contamination was ruled out in the present study because rough ER microsomes used for purification of membrane microdomains showed negligible contamination (<3%) by smooth vesicles (devoid of ribosomes), as demonstrated by a morphological analysis (Fig. 2A). Results obtained through the proteomics analysis were then further validated using an immunoblotting approach. Indeed the presence of four ER-resident or ER-associated proteins (PTP-1B, calnexin, PDI and ORP150) and one ribosomal protein (L4) was monitored in rough ER microsomes and TX- or TW-resistant microdomains. Although in all cases the proteins tested were not or were barely detectable in TX-resistant fractions (Fig. 2B), calnexin and PTP1B were enriched in TW-resistant fractions, whereas no specific enrichment for PDI or L4 and exclusion of ORP150 were observed in those fractions (Fig. 2B). In addition, when the fractions were probed using anti-Tom20 antibodies, no signal was detected in ER (total) microsomes or in detergent-extracted fractions compared with the total particulate (TP) sample (Fig. 2B). Tom-20 was undetectable in our ER microsome preparation, thus reinforcing the above reported EM analysis. Recently, an extensive proteomic analysis of the rat liver secretory pathway has been reported (Gilchrist et al., 2006). Interestingly, we found that more than 55% of the proteins identified (57/103) in our study were also described by these authors (Gilchrist et al., 2006). Moreover, 73% of the remaining proteins identified (n=46) were clearly reported to localize to the ER or ERGIC. Finally, the remaining 17% mostly accounted for pancreas-specific cargo proteins. In an attempt to further evaluate the biological relevance of the protein list identified in the TW-resistant fraction, we reconstituted a functional network using the STRING program suite as described in the Materials and Methods. This led to the construction of a highly connected network (74 nodes and 1483 edges) including 72% of the proteins identified through the proteomics screen (Fig. 2C). This indicates that enrichment of specific proteins in TW-resistant microdomains might be relevant for their functional activity in the ER.

Among the unexpected proteins found in the ER microdomains, we identified BIT1 (PTHR2) (Fig. 2C), a protein that was previously reported to localize in the mitochondria and functions as an apoptosis regulator inhibiting the transcription of Bcl-2 (Jan et al., 2004). Since numerous apoptosis regulators are known to shuttle between the ER and mitochondria (Berridge, 2002; Germain and Shore, 2003; Rudner et al., 2002), the presence of BIT1 in the ER detergent-resistant membranes (DRMs) might be predictive of a trafficking specificity of this protein, as previously demonstrated in S. cerevisiae for the role of ER DRMs (Bagnat et al., 2000).

BIT1 mRNA derives from a single-exon gene and encodes a protein of 179 residues with a predicted transmembrane domain (residues 15-40) and a predicted cytosolic domain (residues 41-179). Alignment of BIT1 sequences from several organisms showed that this protein is well conserved and displays a particularly conserved region homologous to the peptidyl-tRNA hydrolase PTH2 from S. cerevisiae (supplementary material Fig. S1A,B). Although Jan and co-workers (Jan et al., 2004) reported a mitochondrial localization for BIT1, mitochondrial contamination was ruled out in the present study (Fig. 2A). To further characterize the distribution of BIT1 in our samples, dog pancreatic ER microsomes and the ER microdomains isolated as in Fig. 1 were immunoblotted using chicken anti-BIT1 antibodies. Two immunoreactive species of ~28 kDa and ~16 kDa were detected (Fig. 2D). The 16 kDa entity was enriched in the sub-ER fractions resistant to TW solubilization and was consistent with the molecular mass of the band in which we initially identified BIT1 (supplementary material Table S1; DPT24). This observation is also consistent with the molecular mass of BIT1 reported by Jan and colleagues (Jan et al., 2004). Both the 28 kDa and 16 kDa forms were confirmed to be BIT1 by mass spectrometry following immunoprecipitation using BIT1 antibodies; the 28 kDa entity was further identified as a potential SDS-resistant dimer of BIT1 because it disappeared upon treatment with 6 M and 8 M urea and heating at 60°C before resolution by SDS-PAGE (Fig. 2E). These data indicated that BIT1 might also localize in the early secretory pathway. This was also reinforced by the enrichment of BIT1 mRNA in the membrane fractions using a transcriptomic approach (Stitziel et al., 2004). Indeed mRNAs encoding ER-resident and secretory proteins were found to accumulate at their site of translation, most likely by binding ER-bound polysomes.

To evaluate the significance of BIT1 localization to the ER, subcellular localization studies were carried out in cultured cells. First, Fisher Rat 3T3 fibroblasts (FR3T3) were homogenized as described in the Materials and Methods and the post-nuclear supernatant was fractionated into P10 and S10 (pellet and supernatant collected after centrifugation at 10,000 g for 30 minutes). The latter was further fractionated into P100 and S100 (pellet and supernatant collected after centrifugation at 100,000 g for 30 minutes). The presence of BIT1 in these fractions was evaluated by immunoblotting. The 28 kDa entity distributed predominantly in the P100, whereas the 16 kDa entity localized exclusively in the P10 fraction (Fig. 3A). As shown by the distribution of the mitochondrial marker Tom20, the Golgi marker MG-160 and the ER marker calnexin, P100 mainly contained ER and some Golgi, whereas P10 contained ER, Golgi and mitochondria (Fig. 3A). BIT1 was never detected in the cytosolic fraction (S100), thus indicating that BIT1 is always membrane associated. Second, an immunocytochemistry analysis was carried out in the human liver cancer cell line HuH7 using anti-BIT1 antibodies (Fig. 3B). This analysis unexpectedly revealed that in whole cells, BIT1 did not localize to the mitochondria (using Mitotracker as a mitochondrial marker) (Fig. 3B, top panels). Moreover, BIT1 localization to the ER was not detectable using this approach with calnexin as an ER marker (Fig. 3B, middle panels). Finally, except for an intense nuclear staining, which was nonspecific (as shown in supplementary material Fig. S2), BIT1 did not localize in the cytoplasm, but was enriched in the Golgi complex, as indicated by the overlapping staining observed with the Golgi marker giantin (Fig. 3B, bottom panels). To further evaluate the localization of BIT1 in the secretory pathway, we purified ER and Golgi fractions from rat liver, as previously described (Bell et al., 2001; Gilchrist et al., 2006). These fractions were either directly resolved by SDS-PAGE or membrane microdomains were extracted using Tween solubilization before SDS-PAGE and immunoblotting with anti-BIT1, anti-MG160 and anti-CNX antibodies. This revealed that BIT1 was present in membrane microdomains in the Golgi complex and that most BIT1 was in monomeric conformation (Fig. 3C).

To further study the role of BIT1 and investigate its previously described pro-apoptotic functions (Jan et al., 2004) in the secretory pathway, we generated a fusion protein in which the BIT1 cytosolic domain replaced the cytosolic domain of the ER-resident transmembrane protein Trapα (which was also found in the DRM fraction in this proteomics approach, supplementary material Table S1). The resulting fusion protein, C-terminally Myc-tagged Trap.BIT1, consisted of the putative BIT1 cytosolic domain fused downstream to the luminal and transmembrane domains of Trapα and was expressed from pcDNA3.myc with a predicted molecular weight of 43 kDa (supplementary material Fig. S3). Trapα luminal and transmembrane domains (TrapLT, without the cytosolic domain) cloned in pcDNA3.myc and full-length pcDNA3.BIT1 were used as controls. The expression of the Trap.BIT1 transgene was first assessed in COS7 cells transiently transfected with pcDNA3.Trap.BIT1 for 24 hours and 48 hours, followed by immunoprecipitation from the precleared lysates using anti-myc antibodies. Immunoblotting with the anti-BIT1 antibodies revealed the presence of a protein at the predicted molecular weight (~43 kDa), which increased over time and which was absent in the mock-transfected cells (supplementary material Fig. S3), thus confirming that the anti-BIT1 antibodies are specific to BIT1.

Trap.BIT1 proteins stably expressed in FR3T3 cells were separated in P10, P100 and S100 fractions after homogenization of individual clones of each construct, followed by immunoblotting with anti-BIT1 antibodies. The distribution of BIT1 in FR3T3 cells expressing stably BIT1 (Fig. 4A,B) was similar to that of non-transfected cells with a predominant localization of the 28 kDa entity in the P100 and an exclusive localization of the 16 kDa form in the P10 fraction. Unexpectedly, when FR3T3 cells were stably transfected with Trap.BIT1, Trap.BIT1 was found exclusively in the P10 fraction (Fig. 4A,B). In both cell types, Trapα and calnexin were found to distribute in both P100 and P10 fractions, whereas the Golgi marker MG160 mostly distributed in the P10 fraction, with some remnants in the P100 fraction (Fig. 4A,B). Most interestingly, upon stable expression of Trap.BIT1 in these cells, even at levels lower than that of endogenous BIT1, the 16 kDa form of endogenous BIT1 was barely detectable in the P10 fraction (Fig. 4B). This result suggests that Trap.BIT1 prevents BIT1 from exiting the ER under monomeric configuration.

To investigate whether Trap.BIT1 prevents the exit of endogenous BIT1 from the ER and consequently block its functions in the Golgi, we first tested whether expression of Trap.BIT1 led to the expected phenotypes observed in BIT1-silenced cells (Jan et al., 2004). Cells expressing stably the different transgenes were grown in suspension for 24-96 hours and apoptosis was measured. FR3T3 cells stably transfected with pcDNA3 alone or pcDNA3.TrapLT showed a moderate increase in apoptotic cells (20% and 13%, respectively, after 48 hours). By contrast, detachment induced a higher level of apoptosis in BIT1 expressing cells (45%, after 48 hours). Notably, the expression of Trap.BIT1 led to a significantly lower level of apoptosis than in cells expressing BIT1 (6% vs 45%, respectively, after 48 hours) (Fig. 4C). These results confirmed that overexpression of wild-type BIT1 has a proapoptotic potential, as previously shown (Jan et al., 2004), whereas Trap.BIT1 expression might lead to a reduced sensitivity towards anoikis (Fig. 4C). This observation was further validated when the expression of BCL2 mRNA was evaluated. In cells overexpressing wild-type BIT1, the expression of BCL2 mRNA was dramatically reduced under basal conditions, confirming the results previously reported (Jan et al., 2004); however, such a reduction was not observed in cells expressing Trap.BIT1 (Fig. 4D and quantification in Fig. 4E). These results suggest that Trap.BIT1 prevents BIT1 function, most likely by affecting its exit from the ER under monomeric conformation.

Cells expressing BIT1 and Trap.BIT1 exhibited a significantly slower growth rate than the other cell types tested (Fig. 5A). Because the observed impact on anoikis might correlate with the initial findings that apoptosis mediated by a loss of cell attachment to the extracellular matrix involves BIT1 (Jan et al., 2004), we investigated the consequences of absence of substrate on these four cell types (Fig. 5B; supplementary material Fig. S4). Cells expressing TrapLT or full-length BIT1 were similarly unable to grow on soft agar, similarly to cells transfected with the empty vector. In stark contrast, Trap.BIT1-expressing cells were able to grow in an anchorage-independent manner (Fig. 5B). To elucidate the potential consequences of constitutive expression of BIT1 in the ER, pCDNA3 TrapLT, BIT1 and Trap.BIT1 cell lysates were immunoblotted using antibodies raised against various proteins involved in major survival and death signaling cascades (see Materials and Methods). Interestingly, the constitutive activation of MEK-1 and ERK1/2 was observed in Trap.BIT1 cells (Fig. 5C). This result correlates nicely with the previously reported observation that BIT1-knockout mice display high levels of ERK1 phosphorylation (Kairouz-Wahbe et al., 2008). Interestingly, when the phosphorylation of RAF1 and B-RAF was monitored in these cells, a weak phosphorylation of B-RAF was detected only in Trap.BIT1 cells (Fig. 5D). The expression and phosphorylation levels of other signaling intermediates (MKK7, JNK, SMAD2, AKT) were also tested in the four cell types under basal conditions (Fig. 5E,F). No significant change in either phosphorylation or expression was observed. As a consequence, and to test whether ERK activation correlated with the expression of Trap.BIT1, HuH7 cells were transiently transfected with Trap.BIT1 expression vector and the expression and localisation of this protein was evaluated by immunofluorescence using anti-MYC antibodies, anti-CNX (for ER staining) and anti-Giantin (for Golgi staining). Trap.BIT1 localized in the ER as expected (Fig. 5G, top and middle panels). In addition, when we monitored ERK phosphorylation in those cells, we observed that increased phosphorylation of ERK was present exclusively in Trap.BIT1-transfected cells (Fig. 5G, bottom panel). To confirm the hypothesis that the absence of BIT1 could lead to enhanced MAPK activation, a siRNA-based approach was carried out. BIT1 protein expression was reduced to approximately 25-35% of its initial levels using previously described siRNA sequences (Jan et al., 2004) (Fig. 5H) and this led to enhanced phosphorylation and activation of the MAPK pathway (Fig. 5I), thus agreeing with the Trap.BIT1- and BIT1-knockout data (Kairouz-Wahbe et al., 2008). These data demonstrate that the impairment of BIT1 expression or localization leads to the enhancement of MAPK signaling.

The localization of Trap.BIT1 in HuH7 cells was further monitored in stably transfected cells using immunofluorescence and confocal microscopy (Fig. 6). As expected, Myc signals were not detected in mock-transfected cells (Fig. 6A); by contrast, in Trap.BIT1-transfected cells, a strong ER staining was observed with anti-MYC antibodies, which strongly co-localized with the ER marker calnexin (Fig. 6B, middle panel). Interestingly, a weak co-localization of Trab.BIT1 with the Golgi marker giantin was also observed (Fig. 6B, bottom panels) and no colocalization with the mitochondrial marker mitotracker was detected (Fig. 6B, top panel). These results indicate that Trap.BIT1 molecules mostly localize in the ER, which could lead to the subsequent alteration of MAPK signaling.

The above-mentioned results suggested that BIT1 could interfere with MAPK signaling owing to its subcellular localization. We consequently tested the potential interaction of BIT1 with the MAPK module using a GST-BIT1 pull-down approach. This was carried out using liver extracts from mice subjected to ER stress, because this was previously shown to affect both the secretory pathway and MAPK activation (Nguyen et al., 2004). The N-glycosylation inhibitor tunicamycin was selected as a stress inducer and liver extracts were prepared from three mice treated with this antibiotic and three mice treated with DMSO. GST-BIT1 was then carried out, and the association of BIT1 with the ERK-MAPK pathway, namely RAF1, MEK1 and ERK1/2, was evaluated by immunoblot (Fig. 7A). This revealed that BIT1 was able to bind mainly to ERK1, independently of MEK1 and RAF1, only under basal conditions (Fig. 7A). We further investigated the BIT1-ERK1 complex in liver organelles. To this end, subcellular fractionation was carried out on total cell extracts using isopycnic sucrose gradient sedimentation. Fractions corresponding to total microsomes, ER/mitochondria and Golgi or endosomes were solubilized, followed by immunoprecipitation using anti-BIT1 antibodies. Immunoprecipitates were then resolved by SDS-PAGE, transferred to nitrocellulose and immunoblotted using both anti-ERK1/2 and anti-BIT1 antibodies (Fig. 7B, top and bottom panels respectively). These results confirmed that BIT1 formed a complex with ERK1 in the Golgi and that this complex was disrupted upon ER stress. To further evaluate the contribution of BIT1 in the activation of the MAPK pathway, BIT1 expression was silenced in HuH7 cells, which were then treated either with brefeldin A (BFA) or tunicamycin (Fig. 7C). Our results showed that ERK activation was induced in control and siBIT1 cells upon tunicamycin treatment (Fig. 7C, bottom panel). By contrast, ERK activation upon BFA treatment was observed only in BIT1-silenced cells, not in control cells (Fig. 7C, top panel). Our results indicate that BIT1 participates in the regulation of MAPK activity in the Golgi. Our next step was to assess the physiological relevance of BIT1 functionality and MAPK activation.

As the activation of the MAPK module is triggered from various spatial locations including the Golgi (Inder et al., 2008; Lunin et al., 2004), we postulated that BIT1 localization and/or oligomeric status and/or BIT1 expression represent a regulatory parameter in this activation process. To further test the hypothesis that BIT1 localization regulates the activation of the MAPK module in intracellular organelles, HuH7 mock-transfected and Trap.BIT1-tranfected cells were then analyzed by immunofluorescence microscopy for the localization of phosphorylated ERK. This was found to be localized in the cytoplasm, but was also enriched in the periphery of the nucleus – an area that was co-labelled with anti-giantin antibodies, thus showing a Golgi localization in both cell types (Fig. 8A). When a confocal microscopy analysis of these cells was carried out, we observed that phosphorylated ERK and giantin co-localized in cells stably expressing Trap.BIT1, but not in mock-transfected cells (Fig. 8B, left panels). This colocalization disappeared upon BFA treatment, but ERK phosphorylation remained more intense in Trap.BIT1 cells (Fig. 8B, right panels), thus agreeing with data shown in Fig. 7C.

In an attempt to quantify this observation, the percentage of cells presenting phosphorylated ERK and giantin co-localization was determined. Interestingly, levels of co-localization of these proteins in Trap.BIT1 cells were 2- to 3-fold greater than levels observed in control cells (Fig. 8C, left panel). This suggested that the normal localization of BIT1 in the Golgi prevents the activation of the ERK-MAPK module from this compartment. In addition, we noticed that in Trap.BIT1 cells, the Golgi was approximately half the size of that in control cells (Fig. 8C, right panel). Similar data were also observed upon silencing of BIT1 expression (Fig. 8D, right and left panels respectively). These data demonstrate that BIT1 negatively regulates activation of MAPK when present in the Golgi.

We first investigated the ability of Trap.BIT1-expressing cells to survive upon tunicamycin-induced ER stress. Indeed, we found that BIT1 expression was upregulated upon tunicamycin-induced ER stress and impairment of BIT1 functionality led to enhanced MAPK activation; we thus predicted that BIT1 could be part of a pro-death program that is induced upon ER stress. Consequently, impairment of BIT1 functionality, by either localization or silencing, might lead to enhanced survival. This was indeed the case, because Trap.BIT1-expressing FR3T3 cells were more resistant to tunicamycin treatment than any of the control cells (transfected with mock vector, Trap or BIT1) (Fig. 9A). Similar results were obtained in HuH7 cells using both the Trap.BIT1 stable cell line and cells transfected with BIT1 siRNA (Fig. 9B), thus validating our initial hypothesis. To further test whether this phenomenon was related to the high MAPK activation level observed under impairment of BIT1 functionality, similar experiments were carried out in the presence of the MEK1 inhibitor PD98059. Interestingly, although PD98059 did not have a major effect on the survival of either control or BIT1-transfected cells, this chemical completely abrogated the resistance of BIT1-silenced cells to tunicamycin-induced death (Fig. 9C). This suggested that BIT1-mediated regulation of the MAPK pathway upon ER stress could significantly contribute to cell death and provides a new role for BIT1 as a pro-apoptotic agent.

In this study, we have used a sub-organellar proteomics approach to enrich new components of this organelle. Our efforts focused on purifying membrane microdomains from the ER. This led us to identify BIT1, a protein previously reported to localize to mitochondria (Jan et al., 2004), which we found to be enriched in ER membrane microdomains, most likely en route to its final destination in the Golgi. This led us to the discovery of a new mechanism for BIT1 in the Golgi, where it negatively regulates MAPK activation, prevents adaptation and promotes cell death upon stress. We discuss below the functional implications of this discovery.

Enrichment of TW-resistant ER microdomains is thought to be similar to what happens at the PM, where signaling or trafficking initiators can be conditioned by the existence of specific membrane microdomains. As a consequence, we searched for proteins in TW-resistant ER microdomains that might function as signaling intermediates. The nature of membrane microdomains in the ER would be expected to significantly differ from those found at the PM, because cholesterol is enriched at the PM and present at very low levels in the ER. We anticipated that ER microdomains would contain less cholesterol that those purified from PM, so we used Tween20 to extract ER microdomains rather that Triton X-100, which is commonly used for PM extractions. Interestingly, besides the tyrosine phosphatase PTP-1B, which was previously found to potentiate IRE1 signaling in the ER (Gu et al., 2004), the only identified protein whose function was directly related to signaling events was the Bcl2 inhibitor of transcription 1 (BIT1). This protein had previously been reported to localize to mitochondria (Jan et al., 2004) and is involved in anoikis by preventing the upregulation of BCL2 transcription. This observation led us to confirm the presence of BIT1 in the ER by assessing the mitochondrial contamination level of our initial dog pancreatic ER preparations. When all the smooth vesicles were counted as contaminants, this amount remained below 3% of total microsomes, thus indicating that our observation was accurate. Moreover, only one mitochondrial protein was identified by mass spectrometry sequencing (supplementary material Table S1). Moreover, other large-scale proteomics analyses of the secretory pathway revealed the presence of BIT1 in the secretory pathway (Gilchrist et al., 2006). Finally, our results were also confirmed by BIT1 localization studies in cultured cells. Indeed, BIT1 was enriched in the Golgi and no co-localization was found when Mitotracker was used as a mitochondrial marker (Fig. 2). This result was in accordance with previously reported observations indicating that lipid microdomains in the ER are important for protein trafficking through the secretory pathway (Bagnat et al., 2000). We were not able to reproduce the observation made by Jan and co-workers (Jan et al., 2004) who showed the almost exclusive localization of BIT1 to mitochondria, even though we analyzed the endogenous BIT1 or a MYC-tagged recombinant protein. This discrepancy is probably due to the biological models used in our study (purified ER and Golgi microsomes, FR3T3 and HuH7 cells) whereas HeLa cells were used by Jan and colleagues.

To measure the functional impact of BIT1 in the secretory pathway, we specifically targeted this molecule to the ER by generating a fusion protein between the luminal and transmembrane domains of Trapα and the cytosolic domain of BIT1. A similar approach had already been successfully used with Bcl-2 to demonstrate its ER-associated functions (Foyouzi-Youssefi et al., 2000). Cells expressing Trap.BIT1 presented an endogenous BIT1 expression pattern that was different to that of control cells (Fig. 4); in addition, we showed that Trap.BIT1 proteins mostly resided in the ER (Fig. 6). Indeed, the 16 kDa entity was not detected in the former, probably because BIT1 was unable to exit SDS-resistant complexes. We consequently postulated that Trap.BIT1 impairs normal localization of endogenous BIT1 function by retaining it in the ER. This hypothesis was in part confirmed by the fact that BIT1 mostly existed as monomeric entities in the Golgi (Fig. 3). Interestingly, Trap-BIT1 cells, although showing minor defects in cell growth, were able to form colonies in soft agar, demonstrating a role for Trap.BIT1 in the activation of pro-survival pathways, either as a consequence of the absence of the 16 kDa form of BIT1 and/or of the presence of the expression of the Trap.BIT1 fusion in the ER.

The molecular consequences of BIT1 retention in the ER, although having specific impact on cell phenotypes, remain unclear. In the hypothesis of a Trap.BIT1-mediated impairment of the role of endogenous BIT1, we tested the impact of Trap.BIT1 expression on the previously reported observation that in BIT1^−/− cells, the ERK/MAPK pathway was hyperactivated (Kairouz-Wahbe et al., 2008). As expected, phosphorylation levels of both MEK-1 and ERK1/2 were highly increased in Trap.BIT1 cells compared with the three other cell lines tested. In addition, our data show that BIT1 might act to prevent the activation of the MAPK module from the Golgi (Figs 7, 8 and 9) because the expression of dominant-negative Trap.BIT1, which correlated with altered expression of the 16 kDa entity (monomeric BIT1), led to enhanced Golgi localization of phosphorylated ERK. Interestingly, it was recently demonstrated that the protein kinase PKD1 modulates BIT1 activity (Biliran et al., 2008). As PKD was previously reported to modulate ERK activation (Brandlin et al., 2002; Wong and Jin, 2005), we therefore postulate that PKD1 modulates ERK activation through BIT1. This observation could also fit with a model where this regulatory network, involving ERK, PKD1 and BIT1, controls Golgi structure and size, as previously reported separately for ERK (Inder et al., 2008; Lunin et al., 2004; Mor and Philips, 2006; Morgado-Diaz et al., 2007; Philips, 2004) and PKD (Ghanekar and Lowe, 2005; Hausser et al., 2005; Sanchez-Ruiloba et al., 2006) proteins.

In previous reports, BIT1 was demonstrated to be a crucial apoptosis regulator through its role on the expression of BCL2 expression (Jan et al., 2004), MAPK regulation (Kairouz-Wahbe et al., 2008) or through its role on TAp63γ (Fan et al., 2009). In the present report, we demonstrate that the role of BIT1 on MAPK activation could occur in the secretory pathway and more particularly in the Golgi, as already demonstrated for Sef proteins (Philips, 2004), and thus BIT1 selectively prevents the activation of specific downstream targets. Interestingly, our data suggest that BIT1 could also act through the inhibition of B-Raf (Fig. 5D) and consequently might have a scaffolding role for B-Raf, MEK1 and ERK1/2. Moreover, the role of BIT1 in the Golgi complex could also be further reinforced by the recent observation that PKD1 is a positive regulator of the apoptotic functions of BIT1 (Biliran et al., 2008), because PKD1 has also been shown to have a significant role in Golgi functions.

In this work, we report that the protein BIT1 localizes to the early secretory pathway and is enriched in the Golgi. Constitutive localization of BIT1 in the ER leads, in turn, to the basal activation of ERK signaling and enhanced resistance to anoikis. This observation correlated with enhanced localization of phosphorylated ERK in the Golgi and reduced Golgi size. Our work might therefore reveal a key regulatory mechanism for the balance between pro-survival and pro-apoptotic pathways via BIT1 and its subcellular localization, through the control of MAPK signaling from the Golgi.

Anti-calnexin (CNX) antibodies were a kind gift from John Bergeron (McGill University, Montreal, Canada). Anti-MYC antibodies were purchased from Cell Signaling. HiFi Taq Polymerase and LR recombinase were from Invitrogen. Complete™ Protease inhibitors were purchased from Roche. The predicted cytosolic domain of BIT1 (residues 41-179) was PCR amplified from a human liver cDNA library with HiFi polymerase and the primers 5′-GGG GAC AAG TTT GTA CAA AAA AGC AGG CTT GCC CAA AAG CAA GAC GAG CAA G-3′ and 5′-GGG GAC CAC TTT GTA CAA GAA AGC TGG GTC TAG TAA AGT TTT AGG TGA CCA GTG ACT TTG TC-3′. The amplified fragment was then recombined into pGEX2TK that was previously made GATEWAY^® compatible (Invitrogen) to generate pGEX2TK.BIT1 encoding BIT1 (residues 41-179) fused downstream of GST. Expression of GST.BIT1 by exponentially growing DH5α bacteria transformed with pGEX2TK.BIT1 was induced with 1 mM IPTG for 5 hours, GST.BIT1 was purified from the lysate using glutathione-Sepharose beads (GE Bioscience) and BIT1 was released from GST by thrombin cleavage. After purification, BIT1 in PBS (1 mg/ml) was mixed (v:v) with TiterMax-^RGold adjuvant (CytRX Corporation) and 200 μl of the emulsion was injected intramuscularly in brown leghorn hens. Immunization was boosted 3 weeks after the first injection and eggs were collected the following weeks. Egg yolks were stored diluted tenfold in PBS, 0.2% Tween20, 0.05% NaN₃ until used. Scrambled and BIT1 siRNA were designed, as previously reported (Jan et al., 2004).

Dog pancreatic microsomes were isolated as previously described (Chevet et al., 1999). Dog pancreatic microsomes were solubilized using Tween 20, a high hydrophilic-lipophilic balance (HLB=16.7) (Roper et al., 2000) detergent to generate DRMs. The purification of dog pancreatic ER DRMs was adapted from the method previously described for yeast microsome DRMs (Bagnat et al., 2000; Roper et al., 2000). Briefly, microsomes (500 μg proteins) were solubilized in 50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 5 mM EDTA with a protease inhibitor mixture (1 mM PMSF, 2.5 μg/ml chymostatin, leupeptin, and pepstatin) containing 0.75% Tween-20, on ice for 30 minutes. Sucrose concentration was increased to 40% in the lysate then the resulting mix was overlaid with 500 μl of the same buffer containing 30% sucrose, and 200 μl of the same solution containing 2.5% sucrose. After centrifugation at 50,000 r.p.m. for 140 min in a TL50 rotor, the top fraction was reprocessed the same way. After the second spin, the upper fraction was precipitated with TCA, the 15,000 g pellet was then washed and resuspended in Laemmli sample buffer (Fig. 1A). Subcellular fractionation studies from rat liver were carried out, as previously described (Bell et al., 2001).

Data were analyzed for homologies using BLAST at the NCBI (http://www.ncbi.nlm.nih.gov). Potential transmembrane domains (TMD) were identified using http://www.cbs.dtu.dk/services/TMHMM-2.0 and Pi and molecular masses computed at http://ca.expasy.org/tools/pi_tool.html. BIT1 homolog sequences were retrieved from Homologene database (NCBI, HomoloGene entry no. 5330). Seventeen protein sequences from various species were aligned using the multiple sequence alignment tool Multalin (version 5.4.1) (Corpet, 1988) and the alignment result was presented as a phylogenetic tree using the phylogeny reconstruction algorithm BIONJ (Gascuel, 1997). The protein sequences from six (representative) species were aligned using CLUSTALW (Thompson et al., 1994) and visualized using the program ESPript (Gouet et al., 2003). Functional network analysis was carried out using the STRING suite program (Snel et al., 2000; von Mering et al., 2003).

The quality of the ER microsomes was assessed by electron microscopy, as previously described (Lavoie et al., 2000). Microsomes were quantified in 20 different images (1430 microsomes counted) with rough vesicles counted as ER vesicles, and smooth vesicles (devoid of ribosomes) counted as mitochondrial by default.

Cells were grown in DMEM with 10% FBS (Hyclone). Plasmids pcDNA3.Myc (empty vector), pcDNA3.BIT1, pcDNA3.TrapLT and pcDNA3.Trap.BIT1 were used to transfect FR3T3 cells using Lipofectamine (Invitrogen). Stable transfectants were selected with G418 (400 μg/ml) over a period of 2-3 weeks. Two individual clones from each construct were kept for further studies. Transfection of HuH7 cells was carried out using PromoFectin reagent (PromoKine) following the manufacturer's recommendations.

Cells were seeded in triplicate at 20,000 cells/well in 24-well plates and cultured in DMEM containing 10% FBS. Cells were counted after 4 days in culture using a Neubaer hemocytometer.

Cells were lysed in 20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1% Triton X-100 (TX), 1 mM NaF, 1 mM Na₃VO₄, 1 mM phenylmethylsulfonyl fluoride (PMSF), and 10 μg/ml leupeptin and aprotinin. Precleared lysates were incubated with anti-MYC antibodies (Upstate) for 16 hours at 4°C followed by incubation with protein G-Sepharose beads (Amersham) for 1 hour at room temperature. Beads were washed three times with PBS, mixed with Laemmli sample buffer, boiled for 3 minutes and spun briefly. Protein complexes present in the resulting supernatant were resolved by SDS-PAGE and analyzed by immunoblot using anti-BIT1 antibodies.

Cells grown in 24-well plates (20,000 cells/coverslip) were fixed with 3.7% formaldehyde for 10 minutes, permeabilized with 0.1% Triton-X 100 for 5 minutes, incubated with anti-Myc or anti-calnexin antibodies for 1 hour at room temperature followed by incubation with anti-rabbit or anti-mouse FITC-conjugated antibodies. Protein immunolocalization was carried out using epifluorescence microscopy with a Zeiss ×63 1.4 NA oil-immersion objective, recorded with a digital camera (DVC) and analyzed with Northern Eclipse software (Emprix imaging), or confocal microscopy with a LEICA SP5 confocal microscope. Determination of the size of the Golgi complex was carried out following staining using anti-giantin antibodies and the ‘analyze particles’ function was used after thresholding to yield the sum of each Golgi perimeter on a per-cell basis. Three independent experiments were performed.

Cells (5×10⁶ cells/plate) were washed twice with cold PBS, homogenized in 150 mM KCl, 10 mM Tris-HCl, pH 7.5, 2.5 mM magnesium acetate, 0.25 M sucrose, 4 mM imidazole, pH 7.4, 1 mM PMSF, 5 mM NaF, 1 mM Na₃VO₄, complete protease inhibitors cocktail (Roche), homogenized (30 strokes with a Teflon potter) and centrifuged at 1000 g for 20 minutes at 4°C. The resulting post-nuclear supernatants were fractionated into P10 by centrifugation at 10,000 g for 30 minutes at 4°C, then into P100 by centrifugation of the resulting supernatant at 100,000 g for 30 minutes at 4°C in a Beckman TLA-100.2 rotor. Each pellet was resuspended in Laemmli sample buffer. Proteins were detected by immunoblotting with anti-BIT1, anti-CNX, anti-Trapα, anti-Tom20 or anti-ERK-1 antibodies.

Soft agar assay was done in duplicate using six-well plates. Each well contained 2.5 ml of bottom layer (0.5% agarose in DMEM 10% FBS) and 1.5 ml of top layer (0.2% agarose in DMEM 10% FBS) containing 5000 cells. Cells were fed twice a week by adding 1 ml of top layer without cells. Colonies from ten different fields were counted after two and three weeks in culture by two independent investigators.

500,000 cells were cultured in suspension on a 1% agarose bottom layer in DMEM 10% FBS. After 24, 48 or 96 hours, cells were collected and centrifuged at 3000 r.p.m. for 5 minutes. Cells were resuspended in methanol and spread out on a slide. Apoptotic Tunel assay was performed using the in situ cell death detection kit (Roche) according to the manufacturer's instructions. Nuclei were stained with DAPI. More than 500 nuclei from each cell clone were counted and the percentage of apoptotic nuclei calculated.

The expression of Bcl2, CHOP and S16 was assessed by semiquantitative RT-PCR as described previously (Nguyen et al., 2004). The following oligonucleotides were used: 5′-TGCACCTGACGCCCTTCAC-3′ and 5′-ACACAGCCAGGAGAAATCAAACAG-3′ for Bcl2; 5′-GGCAGCTGAGTCTCT-3′ and 5′-GGTGCAGACTGACCA-3′ for CHOP; 5′-TGCTGGTGTGGATATTCGGG-3′ and 5′-CCTTGAGATGGGCTTATCGG-3′ for S16.

Cells were lysed in 20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1% TX, 1 mM NaF, 1 mM Na₃VO₄, 1 mM PMSF, and 10 μg/ml leupeptin and aprotinin. Lysates were resolved by SDS-PAGE and analyzed by immunoblot using anti-MEK1/2, anti-ERK1/2, anti-PAK1, anti-RAF1 (all from Santa-Cruz Biotechnologies), anti-MEKK-7 and anti-phospho-MEKK-7 (kind gifts from Philip A. Barker, McGill University, Montreal, Canada), anti-phospho-MEK-1/2, anti-phospho-ERK-1/2, anti-phospho-RAF1, anti-phospho-AKT, anti-AKT, anti-phospho-Smad2 (all from Cell Signaling), or anti-Smad2 antibodies (Santa-Cruz).

After resolution of the DRM-associated proteins by SDS-PAGE, and Coomassie R-250 staining, each individual band was cut out and the corresponding slices were dehydrated in acetonitrile and washed by two cycles of 10 minutes in 100 mM (NH₄)₂CO₃ before the addition of an equal volume of acetonitrile. The completely destained gel slices were then treated for 30 minutes with 10 mM dithiothreitol to reduce cystinyl residues and for 20 minutes with 55 mM iodoacetamide to effect alkylation. After an additional round of (NH₄)₂CO₃ and acetonitrile washes, the slices were extracted with acetonitrile at 37°C. They were then incubated with trypsin (6 ng/μl in 50 mM [NH₄]₂CO₃) for 5 hours at 37°C and the peptides were first extracted in 1% formic acid, 2% acetonitrile followed by two further extractions with additions of acetonitrile. All treatments were performed robotically using a MassPrep Workstation (MicroMass). Extracted peptides were applied to a reverse-phase guard column and then eluted in-line to a 10 cm by 75 μm PicoFrit column filled with BioBasic C18. The column was eluted at 200 nl/minute with a linear gradient of 5-70% acetonitrile, 0.1% formic acid. For all peptides, a 30 minute gradient was used. A 2000 V charge was applied to the PicoFrit column such that the eluted peptides are electrosprayed into a Cap liquid chromatography quadrapole time-of-flight MS (MicroMass). When doubly or triply charged ions were detected, the quadrapole was used to select the precursor ion, which was passed into a collision cell where fragmentation was induced by collision with argon gas. The collision energies are determined automatically by the instrument based on the m/z values and charge states of each peptide. Fragmentation data were collected in 1 second scans for up to 5 seconds. The MS/MS data was peaklisted (MassLynx; MicroMass) and submitted to Mascot (MatrixScience) software for database search analysis against the NCBI non-redundant database.

A human liver cDNA library was used as a template to PCR amplify either full-length or subfragments of BIT1 and Trap: BIT1 full-length (BIT1, aa 1-179): 5′-GGG GAC AAG TTT GTA CAA AAA AGC AGG CTT GAT GCC CTC CAA ATC CTT GGT TAT G-3′ and 5′-GGG GAC CAC TTT GTA CAA GAA AGC TGG GTT GTA AAG TTT TAG GTG ACC AGT GAC TTT GT-3′. Trapα luminal and transmembrane (TrapLT): 5′-GGG GAC AAG TTT GTA CAA AAA AGC AGG CTT GAT GAG ACT CCT CCC CCG CTT-3′ and 5′-GGG GAC CAC TTT GTA CAA GAA AGC TGG GCA TCA TTC TGA CTT GAT GTA CCC ATT TCT A-3′ for amplification of the predicted luminal and transmembrane domains of Trapα (aa 1-250). Trap.BIT1: 5′-GGG GAC AAG TTT GTA CAA AAA AGC AGG CTT GAT GAG ACT CCT CCC CCG CTT-3′ and 5′-GGC GCC GGC ATC ATT CTG ACT TGA TGT ACC CAT TTC TAC-3′ for amplification of the predicted luminal and transmembrane domains of Trapα followed by an AGA linker sequence (Trapα, aa 1-250; TrapLT-AGA); 5′-GCC GGC GCC CCC AAA AGC AAG ACG AGC AAG-3′ and 5′-GGG GAC CAC TTT GTA CAA GAA AGC TGG GTT GTA AAG TTT TAG GTG ACC AGT GAC TTT GTC A-3′ for amplification of the predicted BIT1 cytosolic domain (aa 41-179) downstream of an AGA linker sequence (AGA-BIT1cyto). TrapLT-AGA and AGA-BIT1cyto (aa 41-179) fragments were gel-purified, mixed and subjected to a second round of PCR amplification with 5′-GGG GAC AAG TTT GTA CAA AAA AGC AGG CTT GAT GAG ACT CCT CCC CCG CTT-3′ (Trapα Fwd Gateway compatible) and 5′-GGG GAC CAC TTT GTA CAA GAA AGC TGG GTT GTA AAG TTT TAG GTG ACC AGT GAC TTT GTC A-3′ (BIT1 Rev Gateway compatible) to generate a DNA fragment encoding a Trap.BIT1 fusion protein with a Ala-Gly-Ala linker sequence. All PCR products were amplified using HiFi polymerase (Invitrogen) and then inserted into the pDONR201 vector by recombinational cloning (GATEWAY, Invitrogen). The resulting plasmids were then recombined into pCDNA3-Myc His (Cterm) that had been previously made GATEWAY compatible (Invitrogen) to generate C-terminally tagged full-length BIT1, TrapLT and Trap.BIT1.

We thank the Chevet lab for discussions and critical advice. We are grateful to Philip A. Barker (McGill University, Montreal, Canada) for the anti-phospho-MKK7 and anti-MKK7 antibodies. This work was supported by grants from INSERM (AVENIR), Institut National du Cancer, la Fondation pour la Recherche Médicale and the European Community (Marie Curie International reintegration grant) to E.C. and by a Génome Québec/Génome Canada Grant to the Cell Map Project. P.Y. was supported by a ‘poste vert’ from INSERM.