BAD-LAMP defines a subset of early endocytic organelles in subpopulations of cortical projection neurons

Neurons are polarized cells specialized to carry out regulated secretion of storage vesicles when an appropriate stimulus is applied. Furthermore, synapse formation, stabilization and maintenance require the delivery of transport vesicles to the site of initial contact between axons and dendrites. These vesicles, containing the different proteins necessary for proper establishment and function of synapses, are the results of complex interplay between the secretion and the endocytic membrane transport pathways (Kennedy and Ehlers, 2006).

Another layer of complexity is introduced by the existence of ordered lipid domains in the plasma membrane (Maxfield and Tabas, 2005). In neurons, several types of microdomains have been shown to be distinguishable by the partitioning of different membrane-associated proteins such as thymus cell antigen 1 (THY1) or the prion protein (PrP) (Sunyach et al., 2003), which are found in different, albeit often closely adjacent domains (Madore et al., 1999). These differences in surface localization are reflected in the different trafficking and functions of these proteins. THY1 is slowly internalized and inhibits the activity of Src family kinases, whereas PrP is rapidly endocytosed and induces axonal outgrowth via the activation of fyn-related kinases (Santuccione et al., 2005). Vesicular transport and lipid microdomain organization, therefore, play key roles in neuronal development and function.

The LAMP family is composed of proteins bearing sequence and structural homology with the canonical LAMP1 and LAMP2 molecules. LAMP molecules harbor an endosomal and lysosomal addressing signal within their short cytoplasmic tail, and contain several conserved cysteine residues, which allow the formation of particular structural loops known as `LAMP folds'. Although the structure, subcellular localization and interaction partners of LAMP1 and LAMP2 have been extensively characterized, their physiological function is still elusive (Eskelinen et al., 2003). Lamp1-deficient mice are viable and show a mild astrogliosis in the brain (Andrejewski et al., 1999), whereas Lamp2 mutants show increased postnatal lethality and massive accumulation of autophagic vesicles in different tissues (Tanaka et al., 2000; Eskelinen et al., 2002). Interestingly, LAMP2 deficiency in humans induces Danon disease, a lysosomal glycogen storage disorder characterized by cardio- and skeletal myopathy and a variable degree of mental retardation (Saftig et al., 2001).

We identified a new member of the LAMP protein family in mice. Brain-associated LAMP-like molecule (BAD-LAMP) is expressed after birth in cortical neurons of particular layers, where it is enriched in defined zones along the neuronal projections. BAD-LAMP mainly accumulates in distinct intracellular vesicles, which do not contain any known markers of classical intracellular transport pathways. BAD-LAMP-containing vesicles have a remarkable clustered organization mirroring at the neuronal surface the presence of THY1-containing microdomains, but not of N-CAM and the ganglioside GM1-enriched microdomains. Interestingly the phosphoepitopes present on microtubule-associated protein 1B and neurofilament H also define BAD-LAMP-containing vesicle positioning in neurons. BAD-LAMP has the ability to be endocytosed, but is not targeted to the late endosomal/lysosome compartments (Gruenberg and Stenmark, 2004). The spatiotemporal specificity of BAD-LAMP expression and its distribution reveal therefore a new level of interplay involving unconventional endocytic compartments and membrane microdomains in specific cortical neurons.

During a bioinformatics search to identify lysosomal-associated molecules, several overlapping nucleotide sequences were identified. After PCR cloning of the full-length cDNA from mouse cortex and extensive sequencing, we identified a potential open reading frame coding for a new putative member of the LAMP family. The new ORF codes for a protein of 280 aa (PI 6.42 and molecular mass of 31.7 kDa) predicted to contain a transmembrane domain (aa 236-256) and a cytoplasmic tail of 24 residues (Fig. 1A). This cytoplasmic domain contains a YKHM (aa 276) motif corresponding to a classical YxxΦ internalization and endosomal targeting signal. The sequence also contains four highly conserved cysteine residues separated by a fixed number of amino acids and is likely to form characteristic internal di-sulfide bonds required for a classical `LAMP fold'. The protein was also predicted to contain three consensus N-glycosylation sites. The nucleotide sequence shares 45% identity with LAMP1 and LAMP2, the founding members of the family, whereas alignments at the protein level displayed 25% similarity (19% identity) (see supplementary material Fig. S1). Thus, the protein was placed on an evolutionary classification tree between LAMP1 and DC-HIL sequences, clearly identifying it as a new member of the LAMP family (Fig. 1B). The tree indicates that DC-HIL (15.5% of similarity), a dendritic cell specific molecule functioning as an integrin ligand (Shikano et al., 2001), shared a common ancestor molecule after diverging away from the LAMP1/CD68 evolutionary axis. The molecule is extremely conserved, since it is found in worm, fly, fish, chicken, rodent and human (see supplementary material Fig. S1). The degree of identity at the amino acid level is close to 85% among mammals and 45% between mouse and fugu. This very high level of conservation across species suggests that the molecule performs a conserved cellular function, not accommodating many variations of its tertiary structure.

Northern blot analysis of the identified mRNA using different mouse tissues indicated that it is expressed almost exclusively in the adult brain, with a close to background signal in the E14 embryo (Fig. 1C). The detected mRNA corresponds to a unique transcript of around 1.8 kb with no apparent alternative spliced forms. Based on its relationship to the LAMP family and its restricted pattern of expression, the molecule was named BAD-LAMP, for brain-associated-LAMP.

To investigate further BAD-LAMP distribution and function, we raised antibodies against peptide epitopes present in its cytoplasmic tail. These antibodies were characterized by immunoblot of HeLa cells transfected with the cDNA coding for mouse BAD-LAMP (Fig. 2A). Several bands were detected in extracts from transfected cells, whereas control extracts remained non-reactive. Brain extracts also displayed several bands, mostly corresponding to those observed in transfected cells, confirming the existence of several isoforms of BAD-LAMP. The detected proteins had a significantly higher molecular mass than the one predicted from the primary sequence of BAD-LAMP (31.7 kDa). In order to define the nature of these post-translational modifications and in absence of immunoprecipitating antibodies, we transfected an N-terminally tagged form of BAD-LAMP (FLAG-BAD-LAMP), allowing efficient immunoprecipitation and treatment with endoglycosidase H (Endo H) and N-glycosidase F (N-gly F). Immunoprecipitated FLAG-BAD-LAMP was shown to be heavily glycosylated (Fig. 2B). The major form of the protein (38 kDa, gH) remained Endo H sensitive, thus reflecting endoplasmic reticulum (ER) retention due to over-expression. The two higher additional bands (47 and 53 kDa, gF) were Endo H resistant but remained N-gly F sensitive, as indicated by the accumulation of a fully trimmed 31 kDa protein (g0) after treatment. Transfected BAD-LAMP is therefore heavily glycosylated on at least two of its three acceptor sites, a situation likely to be shared with endogenous BAD-LAMP detected in brain.

Although glycosylation was in support of BAD-LAMP membrane association, we demonstrated the membrane-bound nature of BAD-LAMP by submitting mouse cortex post-nuclear supernatants to high speed ultracentrifugation, in which BAD-LAMP was found associated with the membranes pellets similar to other membrane-associated molecules such as RAB3a, syntaxin 6 and syntaxin 13 (Fig. 2C). Thus BAD-LAMP, is a glycosylated LAMP-like molecule associated with cortical membranes.

Analysis of mouse brain extracts by immunoblotting revealed that levels of BAD-LAMP increased strongly after birth (P0) reaching its maximum level at adulthood, but being already strongly expressed at P10-P12 (Fig. 2D). We used in situ hybridization to investigate in detail the expression pattern of BAD-LAMP in the developing mouse forebrain. The first expression of BAD-LAMP was found at P2 in the cingulate cortex, in a thin band of intermediate cells (Fig. 2E). At P5, expression extended ventrally into the cortical plate. Furthermore, the caudate putamen showed a punctuate expression of Bad-lamp transcripts. This expression pattern was maintained at P7, when an additional broad band of large and strongly Bad-lamp-positive cells appeared in superficial parts of the cortical plate. Although the cortical expression intensified until P9, no major regional changes in Bad-lamp expression were obvious during this period. At P12, expression of Bad-lamp in the striatum ceased, while expression further intensified in the cortex. This expression pattern was stable until adulthood. Altogether, this expression pattern indicates that BAD-LAMP does not function in the early steps of brain development, such as neurogenesis and cell migration, but potentially during terminal steps of neuronal differentiation and neuronal function.

Within the adult cortex, the homogenous staining in outer regions of the cortical plate, as well as in a more restricted band of cells localized centrally, was suggestive of an expression in neurons of specific cortical layers. We used well-known markers for cortical layers to further characterize the respective populations. Comparison of the expression of Bad-lamp to that of Cux2, a marker for layers II-IV, showed that the BAD-LAMP domain is included in the CUX2 domain and confined to its outer part (Fig. 3A,B). Thus, BAD-LAMP is expressed in the upper layers II and III of the neocortex, but is excluded from layer IV. Furthermore, there was a perfect overlap with the layer V marker ER81 (Fig. 3A,B) demonstrating that the deeply positioned Bad-lamp-positive population is located in layer V.

The size of the Bad-lamp-positive cells in the respective cortical layers was suggestive of neuronal cells. To confirm this observation we investigated the expression of Bad-lamp in Scrambler mice. These animals show a well described inversion of the layers of cortical projection neurons, with upper layer neurons (layers II-IV) positioned deeply whereas deep layer neurons (layers V and VI) are positioned superficially (Rice and Curran, 1999). The organization of B Bad-lamp-positive cells in the Scrambler cortex was strikingly altered (Fig. 3B). Small, lighter stained cell bodies were displaced towards the ventricular side, whereas larger and more strongly labelled cells were merely found at the pial side of the cortex. This pattern was in agreement with an inversion of the position of Bad-lamp-positive cells and suggestive of a projection neurone identity. Furthermore, all Bad-lamp-positive cells in the cortex were co-expressing the neuronal marker NeuN, again confirming the neuronal identity of labelled cells (Fig. 3C), which were found also to express BAD-LAMP protein (Fig. 3D).

BAD-LAMP distribution in cortical brain sections was monitored by confocal microscopy (Fig. 3E). BAD-LAMP was found in vesicles mostly located in cell bodies as delineated by the MAP2 staining. In addition, BAD-LAMP accumulated in defined domains along cellular projections (Fig. 3E arrowheads). It could be detected at the plasma membrane and in vesicles present in the cell bodies, but was also enriched in vesicles clustered in defined domains of dendrites. BAD-LAMP mostly accumulated within the boundaries of specific neuronal areas in vivo.

In order to confirm the relevance of these observations, embryonic cortical neurons were explanted and BAD-LAMP sub-cellular distribution was investigated after 3 days in culture. Owing to the particular clustered distribution of BAD-LAMP, we also investigated the distribution of proteins known to partition in different cellular domains, such as lipid microdomain-associated proteins (Madore et al., 1999). Using confocal microscopy we found that semi-ordered lipid microdomain residents such as PrP, N-CAM, as well as the ganglioside GM1 (stained with cholera toxin, CT) were enriched in zones excluding BAD-LAMP vesicles (Fig. 4A-C). This observation was particularly striking with CT staining and N-CAM, which accumulated almost exclusively in areas negative for BAD-LAMP (Fig. 4A,C). By contrast, at this low magnification, THY1, a molecule representing ordered lipid microdomain-associated proteins, displayed an overlapping distribution with BAD-LAMP (Fig. 4D). However, at higher magnification, no direct co-localization of THY1 and BAD-LAMP molecules could be observed. Instead, accumulation of BAD-LAMP-containing vesicles was revealed directly underneath THY1-enriched areas at the plasma membrane (Fig. 4D, arrowheads). BAD-LAMP-containing vesicles therefore accumulate in cellular zones, defined by the presence of THY1 at the plasma membrane, whereas they are segregated from the detergent-resistant microdomains containing most of the PrP, GM1 and N-CAM (Madore et al., 1999).

BAD-LAMP distribution and microdomain organization seemed to be closely linked. Cholesterol depletion efficiently affects lipid microdomains and was therefore tested for its ability to influence BAD-LAMP distribution. Cortical neurons were treated for 2 hours with cholesterol-esterase prior to immunostaining and confocal microscopy visualization (Fig. 4E). As expected, cholesterol depletion had a potent effect on GM1 distribution at the plasma membrane. Moreover, BAD-LAMP vesicular staining was also deeply affected, displaying an extensive co-localization with GM1, which was never observed in normal conditions. Thus, microdomain organization at the membrane and clustering of BAD-LAMP-positive vesicles appeared to be directly linked.

This particular organization was likely to be maintained with the active participation of the cytoskeleton and/or associated proteins. In order to test this hypothesis, several candidate molecules were followed by confocal microscopy in cortical neurons. Surprisingly, BAD-LAMP containing-vesicles clustered within punctate zones delimited by staining with the Smi31 antibody (Fig. 5A). Smi31 detects phosphorylated epitopes present in neurofilament H and mostly in the microtubule-associated molecule 1B (MAP1B) (Fischer and Romano-Clarke, 1990). Although the precise function of MAP1B phosphorylation is still debated, experimental evidence suggests a role in regulating microtubules and actin dynamics as well as being necessary for axonal growth (Dehmelt and Halpain, 2004; Del Rio et al., 2004). The perfect overlapping distribution of BAD-LAMP and Smi31 strongly suggested that microtubules or actin are likely to play in important role in the organization and the clustering of BAD-LAMP-positive vesicles, however BAD-LAMP distribution is not dependent on the neuronal polarity.

BAD-LAMP-positive vesicles were found in the close vicinity of the microtubule network, mirroring, by their accumulation, the intensity of the tubules bundling (Fig. 5B). A treatment with the microtubule depolymerizing agent nocodazole was thus carried out (Fig. 5B). Nocodazole induced a strong redistribution of BAD-LAMP-containing vesicles and a loss of BAD-LAMP staining intensity in cortical neurons. Thus the microtubule network influences the positioning of BAD-LAMP vesicles. Lipid microdomain organization and BAD-LAMP distribution in cortical neurons are therefore linked, use the microtubule network and possibly depend on MAP1B phosphorylation for their regulation.

The unusual distribution of BAD-LAMP vesicles led us to investigate their relationship with other types of sub-cellular compartments. We focused primarily on endocytic organelles, likely to be relevant to a transmembrane molecule bearing a YxxΦ motif in its cytoplasmic tail. BAD-LAMP could not be detected in classical endosomal compartments as judged from its lack of co-localization with LAMP2 (late endosomes and lysosomes) and internalized transferrin-FITC (sorting and recycling endosomes) (Fig. 6A,B) or syntaxin 13 (Hirling et al., 2000). BAD-LAMP was not found in more specialized endocytic compartment such as TI-VAMP-positive vesicles (Coco et al., 1999) (Fig. 6C). Co-labeling with synaptic vesicle proteins such as synaptotagmin 1, RAB3a, VAMP2 revealed some level of co-localization with BAD-LAMP in the growth cones (Fig. 6D-F). Interestingly, co-localization was not observed in other cellular areas and a similar overlapping distribution in the growth cone was observed with TI-VAMP, which is not found enriched in synaptic vesicles (Coco et al., 1999). Thus, this co-distribution in the growth cone probably reflects the difficulty of segregating, at this optical resolution, individual carrier vesicles congregating in the same area of the cone, rather than a true co-localization in the same vesicles. Pre-and post-synaptic transport carriers are derived from trans-Golgi network (TGN) vesicles, which aggregate at initial contacts between axons and dendrites (Sytnyk et al., 2002). We, therefore, examined the possible association of BAD-LAMP with other known vesicular markers of these pathways, such as syntaxin 6 or N-CAM (Sytnyk et al., 2004) (Fig. 6E). We failed to detect co-localization of BAD-LAMP with any of these markers (see supplementary material Fig. S2), suggesting that the molecule is sorted in an uncharacterized type of vesicles, which can accumulate in the growth cone of developing axon, as well as in defined and organized domains along the cellular processes.

BAD-LAMP distribution at the plasma membrane as well as in localized intracellular vesicles suggested a possible shuttling of the molecule between the cell surface and the vesicles. The co-localization of GM1 and BAD-LAMP upon cholesterol depletion suggests that BAD-LAMP vesicles are accessible to plasma membrane constituents under specific conditions. To address this issue, cortical neurons were surface biotinylated at 4°C prior to incubation at 37°C. Biotinylated surface proteins could either diffuse, or be internalized, and their intermixing with BAD-LAMP-positive compartments was evaluated at different time points by confocal microscopy (Fig. 7). Biotinylated proteins were detected rapidly co-localizing with BAD-LAMP after 5 minutes of internalization. This significant overlapping distribution decreased after 45 minutes, suggesting that BAD-LAMP-containing organelles could represent a subset of early endocytic vesicles, rapidly accessible from the neuronal surface and serving as an intermediate step for the intracellular sorting of specific surface molecules present in developing neurons.

To further investigate the distribution of BAD-LAMP, we generated N-terminally FLAG-tagged and C-terminally GFP-tagged BAD-LAMP constructs and monitored their behavior by microscopy in co-transfection experiments of cortical neurons (Fig. 8). Surprisingly, endogenous BAD-LAMP expression and domain organization were strongly inhibited in electroporated neurons. Nevertheless, transfected FLAG-tagged BAD-LAMP was found enriched in vesicles clustered in specific zones along the neurites. Clearly, the tagged protein is not addressed in conventional endo/lysosomes as judged by its lack of co-localization with LAMP2 (Fig. 8A), internalized transferrin (not shown) or cholera toxin (see supplementary material Fig. S3). The exact location of FLAG-BAD-LAMP in the cell body was difficult to establish since its over-expression induced an accumulation of the molecule in the ER and Golgi network. Surprisingly, the C-terminally GFP-tagged, BAD-LAMP (BAD-GFP) was found accumulating in LAMP2-positive lysosomal compartments (Fig. 8A, arrowheads in Z1). Therefore, the BAD-LAMP cytoplasmic tail contains a cryptic lysosomal retention motif, which is revealed by the addition of the GFP moiety. This observation also suggests that the cytoplasmic tail of BAD-LAMP is actively interacting with, or modified by, molecules that promote its sorting away from traditional endocytic compartments. Co-localization of BAD-GFP and FLAG-tagged BAD-LAMP was observed in discrete vesicles in neurites (Fig. 8A, Z2 arrowheads), despite the fact that BAD-GFP was found mostly accumulating in large lysosomes in the cell body. This demonstrates that a small fraction of BAD-GFP can be sorted normally.

We next evaluated the internalization dynamics of BAD-LAMP by using the N-terminally FLAG-tagged construct and by monitoring FLAG antibody uptake after cold binding (Fig. 8B). The antibody was rapidly endocytosed after 5 minutes at 37°C. Inside the cell, it was detected in a different compartment from conventional endo/lysosomes as shown by the absence of co-localization with co-transfected BAD-GFP, LAMP2 (Z3 and arrows) and internalized cholera toxin (supplementary material Fig. S3A). After 30 minutes of synchronous uptake (Z4 and arrowheads), co-localization of the antibodies with BAD-LAMP-GFP and LAMP 2 indicated that BAD-LAMP can reach conventional endocytic compartments, after being internalized from the surface. Surprisingly, this co-localization was more evident in the more discrete LAMP2-positive organelles present in the neurite (late endosomes, arrowheads) than in the large lysosomes observed in the cell body (Fig. 8B).

We next investigated the contribution of tyrosine 276 to BAD-LAMP trafficking by introducing a mutational change to alanine at this position (Tyr276Ala). The FLAG-tagged mutant was also found accumulating in the ER and Golgi network of transfected neurons. However, the fraction of the mutant that exited these organelles accumulated at the surface of the neurites in a manner very distinct from the normal molecule (wild type), which was mostly found in intracellular vesicles (supplementary material Fig. S3B). Similar results were obtained with a construct lacking the entire cytoplasmic tail of BAD-LAMP (not shown). Thus, tyrosine 276 is directly involved in intracellular addressing of BAD-LAMP and allows its internalization from the surface. FLAG antibody uptake after binding in the cold was performed in neurons expressing FLAG-BAD-LAMP Tyr276Ala. Transfected cells remained mostly antibody-decorated at the surface 30 minutes after warming at 37°C (supplementary material Fig. S3C). Thus, BAD-LAMP is probably cycling between the plasma membrane and a subset of endocytic vesicles.

In order to further dissect the molecular mechanisms governing BAD-LAMP endocytosis, we studied the distribution and transport of transfected BAD-LAMP in a cell type easy to manipulate, such as HeLa and mouse NIH 3T3 cells. In HeLa cells, FLAG-BAD-LAMP was found at the cell surface and in internalized transferrin-containing vesicles distributed in the vicinity of the plasma membrane, whereas the Tyr276Ala mutant accumulated only at the cell surface (Fig. 9A). No co-localization was found in LAMP1-positive late endosomes or lysosomes, nor with co-transfected DC-LAMP tagged with GFP (Fig. 9B), another lysosomal resident of the LAMP family (de Saint-Vis et al., 1998). These observations were confirmed after Percoll density gradient subcellular fractionation of transfected HeLa cells (supplementary material Fig. S4). BAD-LAMP was mostly detected in the low density fractions of the gradient containing plasma membrane, ER and early endosomes, but it was absent from the high density fractions containing lysosomes, as indicated by β-hexosaminidase activity. Thus, most of transfected BAD-LAMP was found on the cell surface contrasting with transfected neurons in which BAD-LAMP mostly accumulated intracellularly, underlining the specificity of its sorting even when over-expressed.

Anti-FLAG antibody uptake in transfected cells saturated with FITC-transferrin (FITC-TF), confirmed that BAD-LAMP could be internalized rapidly in sorting endosomes (Fig. 9C). Interestingly, 15 minutes after uptake BAD-LAMP was found present in transferrin-positive recycling endosomes clustered around the microtubule organizing center, suggesting that BAD-LAMP could recycle to the plasma membrane after internalization. This hypothesis was supported by the poor co-localization of the antibody with LAMP1 after 45 minutes of uptake, indicating that the molecule does not efficiently reach late endocytic compartments. This underlines again a difference with neurons, in which the antibodies could be detected in discrete LAMP1-positive compartment 45 minutes after uptake.

We next investigated the molecular mechanisms involved in BAD-LAMP endocytosis. Experiments performed in cells co-transfected with wild-type GTPase dynamin II or dominant-negative mutant A44K, indicated that BAD-LAMP internalization is mediated in a dynamin-dependent manner, since antibody internalization was abolished in cells expressing dynamin A44K (Fig. 9D and control, supplementary material Fig. S4C). In order to further define the endocytic pathway used by BAD-LAMP to enter the cell, we used an RNA inhibition approach to reduce the expression of molecules involved in protein triage from the surface, such as the clathrin adaptor AP2 (Dugast et al., 2005; McCormick et al., 2005). Antibody uptake was monitored by immunostaining and FACS detection after binding at 4°C and internalization at 37°C. Cells co-transfected with FLAG-BAD-LAMP and control RNAi plasmid showed rapid internalization of the antibody (Fig. 9E), whereas RNAi depletion of AP2 clearly inhibited BAD-LAMP internalization as well as transferin uptake (supplementary material Fig. S4C). In cells depleted for AP2, higher surface levels of BAD-LAMP were also consistently detected (not shown), suggesting that BAD-LAMP is internalized constantly through a dynamin/AP2-dependent endocytic pathway.

Interestingly, monitoring of surface anti-FLAG antibody by FACS also indicated that the molecule was rapidly internalized between 5 and 7.5 minutes after warming (supplementary material Fig. S4B). Surface levels of antibodies then re-increased after 10 minutes, to be diminished again but with a relatively slower internalization rate. These observations confirm that BAD-LAMP and associated antibodies constantly recycle to the plasma membrane with a relatively high efficiency.

BAD-LAMP sequence analysis clearly indicates that it represents a new member of the LAMP family. However, its expression pattern and intracellular distribution are unconventional compared to other LAMP family members, which show a widespread expression and specifically accumulate in the lysosomes.

Our observations on BAD-LAMP intracellular distribution are clearly indicative of a strong regulation of its trafficking in a subset of early endosomes. Although we have not been able to identify molecular markers able to identify these organelles, the absence of transferrin or synaptotagmin 1, as well as late endosomal markers such as LAMP1 suggests that these vesicles represent a distinct class of neuronal endosomes. The kinetics of biotinylated proteins and antibody uptake indicate that they can serve as sorting platforms, prior to transport to other organelles, which are positive for LAMP1, but only represent a minor fraction of the neuronal organelles containing LAMP1.

We have shown that BAD-LAMP, through an interaction with its YKHM domain, requires dynamin and AP2 to be internalized and sorted towards the early endocytic recycling pathway of transfected HeLa cells. LAMP1 has also been shown to require the AP2 adaptor, but its sorting is directed towards lysosomes (Janvier and Bonifacino, 2005). Interestingly, modification of the BAD-LAMP C-terminal domain by GFP deeply affects its transport in neurons and demonstrates the existence of an active sorting pathway in these cells, which normally prevents the accumulation of BAD-LAMP in the lysosomes. The YKHM domain is a relatively weak consensus endosomal/lysosomal addressing signal (Bonifacino and Traub, 2003), although it is also found in CTLA-4, a molecule known to recycle upon activation of T cells (Linsley et al., 1996). As suggested by its early endosomes distribution, we could show that BAD-LAMP also recycles in transfected HeLa cells. Whether this is the case in neurons remains to be further investigated, although it clearly indicates that the `YKHM' domain is not normally used as a lysosomal addressing signal.

One of the features of BAD-LAMP-containing organelles is their clustered distribution. This distribution mirrors the organization of the different microdomains at the cell surface. Whether BAD-LAMP-containing organelles participate in the maintenance of this organization within the neuritic plasma membrane remains to be proved. Nevertheless their sensitivity to cholesterol-depleting drugs suggests that microdomains and BAD-LAMP-containing vesicles are functionally linked. Strikingly, the clustering the BAD-LAMP-containing vesicles is also defined by the distribution of the phosphorylated epitopes (SMI31) found on the microtubule-associated protein MAP1B or neurofilament H (Fischer and Romano-Clarke, 1990). MAP1B in the cortex has been strongly implicated in synapse formation and function (Kawakami et al., 2003). Such a role has been recently functionally demonstrated through the observation that mice lacking the phosphorylated form of MAP1B specifically in the hippocampus, show deficits in long-term potentiation in the Schaeffer collaterals pathway (Zervas et al., 2005). Therefore, it is conceivable that MAP1B is implicated in the positioning and transport of BAD-LAMP vesicles at sites of postsynaptic densities on the dendrites of cortical neurons, and that this process could be essential for stabilization, function and plasticity of cortical synapses. Indeed, BAD-LAMP expression is temporally and spatially restricted in cortical neurons of layers II, III and V. Whereas the generation and migration of cortical neurons in rodents is an embryonic process, synaptogenesis in the cortex occurs in the postnatal animal with a peak between P10 and P15 to approach adult values (Micheva and Beaulieu, 1996). This increase in functional synapses in the cortex is strikingly mirrored by the expression of BAD-LAMP during corticogenesis. Thus, it appears very possible that BAD-LAMP, together with MAP1B, is involved in the terminal maturation steps and/or function of defined cortical neurone populations.

Most of our observations point towards a link between BAD-LAMP and endocytosis. The transformation of a transient contact between two neurons into a stable and functional synapse requires major changes in the membrane composition of the respective neuronal surface areas. Endocytic processes have been implicated in the regulation of synaptic function and plasticity in vertebrates (Vissel et al., 2001) and in Drosophila (Dickman et al., 2006). For example, NMDA receptors are subject to constitutive (Roche et al., 2001) as well as agonist-induced (Vissel et al., 2001) internalization through clathrin-mediated endocytosis. Interestingly, in situ hybridization for NMDAR1 resulted in strong cellular labeling in neurons of layers II/III, V and VI (Rudolf et al., 1996), resembling the pattern we found for BAD-LAMP in the postnatal cortex. The BAD-LAMP-containing endocytic compartment could therefore play a regulatory role in these events by maintaining specific zones in the neuronal projections.

The BAD-LAMP protein sequence ID in Ensembl database is ENSMUSP00000061180. All LAMPs sequences were aligned using CLUSTALW package (EBI) and results were treated with TreeView for phylogeny. Image analysis was performed with the Image J software and the plugin JacoP.

All animals were treated according to protocols approved by the French Ethical Committee. CD1 mice (Iffa-Credo, Town?, France) were used to determine the Bad-lamp expression pattern. Disabled 1 deficient Scrambler mice were purchased from Jackson Laboratories. The day of the vaginal plug appearance was considered as embryonic day (E)0.5 and the day of the birth as postnatal day (P)0. For in situ hybridization and immunohistochemistry, postnatal and adult brains were collected after the animals were anaesthetized with a lethal dose of Rompun/Imalgen 500 and intracardially perfused with 4% paraformaldehyde (PFA). Brains were further fixed in 4% PFA overnight. Adult brains were sectioned at 80 μm on a vibratome whereas P2-P12 brains were cryoprotected in 20% sucrose/PBS, frozen in OCT compound and sectioned at 16 μm on a cryostat. Sections collected on Superfrost slides were treated as described below.

Northern blot analysis was done with FirstChoice Northern Blot Mouse Blot I (Ambion) using a probe corresponding to exons 4, 5 and 6 of BAD-LAMP (clone IMAGE 2588577). 2 mg of Trizol extracted total mouse cortex RNA was used for reverse transcription with oligo(dT) primers. The cDNAs coding for BAD-LAMP were amplified after 30 cycles of PCR using Taq polymerase. Sense primer was ACC GGC CAC TTT GAG GGA and antisense GGG GCG GCC TTT GCA GCA (1.5 kb). PCR products were cloned into pGEM-Teasy plasmid (Promega). BAD-LAMP-GFP fusion construct was constructed using pEGFP-NI vector (Clontech). FLAG-BAD-LAMP was constructed using pTEJ-8-HA-FLAG plasmid (Didier Marguet, Marseille, France). A tyrosine mutant of BAD, FLAG-BAD-Tyr-276-Ala was produced by targeted PCR mutagenesis. FLAG-BAD-LAMP cDNA were transferred into pCX-MCS2 plasmid, a pCAAGS derived plasmid with an extended cloning site (a kind gift from Xavier Morin, Marseille, France). Dynamin-GFP wt plasmid and dynamin-GFP A44K were kindly given by M. McNiven, Rochester, MN. RNAi constructs pSUPER AP2 μ2 and pSUPER control were a gift from Philippe Benaroch, Paris, France.

IMAGE clone 2588577 was used to make an antisense RNA probe. Antisense RNA probes for Bad-lamp, Cux2 (Zimmer et al., 2004) and ER81 (Lin et al., 1998) were generated using the Dig-RNA labelling kit (Roche). Single in situ hybridization and combined in situ hybridization with immunohistochemistry were described previously (Tiveron et al., 1996; Zimmer et al., 2004) for all probes and the NeuN monoclonal mouse IgG (MAB377; Chemicon).

A polyclonal rabbit anti-BAD-LAMP was raised in rabbit against two peptides of the BAD-LAMP cytoplasmic tail, KMTANQVQIPRDRSQC and KQIPRDRSQYKHMC. Anti-synaptotagmin 1 and anti-RAB3a/b antibodies were obtained from P. Di Camilli, New Haven, CT, anti-FLAG M2 antibody and anti-β-tubulin-Cy3 were obtained from Sigma, anti-VAMP2 from SYSY, anti-syntaxin 6 from BD Transduction Laboratories; Aanti-PrP (6H4) was from Prionics (Schlieren, Switzerland), anti-syntaxin 13 from Stressgen (Ann Arbor, MI); anti-Thy1 from Michel Pierres, Marseille, France; human Alexa Fluor 568-Tf from Molecular Probes; mouse FITC-Tf from Rockland (Gilbertsville, PA), Cy3-β-tubulin from Sigma, FITC-cholera toxin B subunit (GM1 staining) from Sigma; anti-NCAM H28 from C. Goridis, Paris, France; Anti-Ti-VAMP from T. Galli, Paris, France; Rat anti-mouse LAMP2 from I. Mellman (New Haven, CT) and anti-human LAMP1 from Abcam. All FITC and Cy3-5 secondary antibodies were from Jackson ImmunoResearch. All Alexa secondary antibodies were from Molecular Probes. Immunofluorescence and confocal microscopy was performed with a Zeiss LSM 510 microscope as described previously (Cappello et al., 2004). Vibratome adult brain sections were immunostained with rabbit anti-BAD-LAMP and mouse anti-MAP2.

HeLa cells were grown in DMEM containing 10% FCS. Cortical neurons were prepared from E15.5 embryonic cortices. Cortices were dissected out in HBSS, treated for 15 minutes at 37°C in Trypsin/EDTA-HBSS (Invitrogen), washed once in NeuroBasal medium (NB; Invitrogen) complemented with 10% horse serum to block trypsin activity and washed once more in NB alone. Cortical neurons were dissociated, plated on glass coverslips in NB with B27 complement, 2 mM L-glutamine and 50 μg/ml penicillin/streptomycin (Invitrogen) and cultured for 3 days at 37°C, 5% CO₂. Coverslips were coated overnight with poly-L-lysine (10 μg/ml).

Neurons were electroporated using Amaxa Nucleofactor Kit according to the manufacturer's instructions. HeLa cells were grown on coverslips and transfected using Lipofectamine 2000 (Invitrogen) using the manufacturer's protocol. After 8-24 hours of transfection the HeLa cells were processed to study internalization kinetics or fixed using 3% paraformaldehyde. Internalization assays were performed using FITC-conjugated transferrin or unconjugated antibodies. The cells were first incubated for 20 minutes at 37°C in DMEM/100 mM HEPES to eliminate endogenous transferrin. Cells were incubated for 15 minutes at 4°C with ligand and/or antibody and washed twice in ice-cold PBS before incubation with DMEM, 1% BSA, 100 mM Hepes at 37°C, for different times prior to fixation and immunocytochemistry. Neurons were processed identically in NB medium. Cortical neurone biotinylation was performed using EZ-Link Sulfo-NHS-Biotin kit (Pierce) with a 15-minute reaction time at 4°C, followed by three washes with ice-cold PBS containing 10 mM glycine. Cells were incubated for 5 and 45 minutes at 37°C to allow endocytosis of biotinylated membrane proteins, prior to fixation and immunostaining.

1% Triton X-100 cell extracts complemented with protease inhibitors cocktail (Roche) were immunoblotted after separation by 12% or a 7-17% gradient SDS-PAGE. Immunoprecipitation with anti-FLAG antibody and N-glycosidase F or endoglycosidase H (Calbiochem) treatment were performed as described previously (Cappello et al., 2004).

This work was supported by grants to P.P. from CNRS-INSERM, the Ministère de la Recherche et de la Technologie (ACI), La Ligue Nationale Contre le Cancer and the Human Frontier of Science Program. A.D. is supported by the MRT and ARC. P.P. is part of the EMBO Young Investigator Program. H.C. was supported by the French Fondation pour la Recherche sur le Cerveau (FRC), the Association Francaise contre le Myopathies (AFM) and the European Community through the NOE NeuroNE. We thank the PICsL imaging core facility for expert technical assistance. We are grateful to Vilma Arce for expert technical advice.