Register for Mechanobiology 2016!


Similar to most differentiated cells, both neurons and epithelial cells elaborate distinct plasma membrane domains that contain different membrane proteins. We have previously shown that the axonal cell-adhesion molecule L1/NgCAM accumulates on the axonal surface by an indirect transcytotic pathway via somatodendritic endosomes. MDCK epithelial cells similarly traffic NgCAM to the apical surface by transcytosis. In this study, we map the signals in NgCAM required for routing via the multi-step transcytotic pathway. We identify both a previously mapped tyrosine-based signal as a sufficient somatodendritic targeting signal, as well as a novel axonal targeting signal in the cytoplasmic tail of NgCAM. The axonal signal is glycine and serine rich, but only the glycine residues are required for activity. The somatodendritic signal is cis-dominant and needs to be inactivated in order for the axonal signal to be executed. Additionally, we show that the axonal cytoplasmic signal promotes apical targeting in MDCK cells. Transcytosis of NgCAM to the axon thus requires the sequential regulated execution of multiple targeting signals.


Epithelial cells and neurons differentiate and maintain distinct plasma membrane domains, such as an apical and a basolateral surface, or dendrites and axons. The distinct membrane composition is maintained by diffusion barriers found either at the tight junctions in epithelial cells (Shin et al., 2006) or in the axon initial segment in neurons (Winckler et al., 1999). Additionally, transmembrane proteins are secreted in a polarized manner from the trans-Golgi network (TGN) during biosynthetic delivery, or from recycling endosomes after internalization. The molecular bases for polarized membrane trafficking are still being uncovered. It is clear that at least some of the signals and the machinery involved in polarized trafficking are conserved between epithelial cells and neurons (Dotti and Simons, 1990), but cell-type specific mechanisms have also been described.

Most of our current understanding about polarized sorting has come from studying the kidney epithelial cell line MDCK and much less is known about sorting in neurons. For example, both basolateral and somatodendritic sorting signals frequently rely on tyrosine-based sorting signals encoded in the cytoplasmic tail of transmembrane proteins (West et al., 1997; Folsch, 2005; Jareb and Banker, 1998; Silverman et al., 2005). Tyrosine-based signals are recognized by cytosolic adaptor protein complexes (AP-1 through AP-4) (Folsch, 2005; Nakatsu and Ohno, 2003; Rodriguez-Boulan et al., 2005). Frequently, tyrosine-based signals bind more than one adaptor complex and are therefore active as basolateral and as endocytosis signals (Folsch, 2005). Columnar epithelial cells, but not neurons or hepatocytes, express a subclass of AP-1 complexes, AP-1B that is important for correct sorting of basolateral cargo proteins from recycling endosomes (Folsch et al., 2003; Gan et al., 2002; Ohno et al., 1999). In addition, AP-4 may sort proteins directly from the TGN to the basolateral membrane during biosynthetic delivery (Fields et al., 2007; Simmen et al., 2002).

Sorting to the apical and axonal domains is less well understood. Apical and axonal sorting signals have been mapped to the extracellular or transmembrane domains of proteins. Additionally, the ability to partition into lipid raft domains correlates with correct sorting for some, but not all, apical or axonal proteins (Chang et al., 2006; Galvan et al., 2005; Jacob and Naim, 2001; Ledesma et al., 1998; Paladino et al., 2004). In 1998, a cytosolic signal for apical targeting was identified in rhodopsin (Chuang and Sung, 1998), and several more have been identified since (e.g. Hodson et al., 2006; Takeda et al., 2003). Surprisingly, some of these apical signals share characteristics with basolateral consensus sequences, such as tyrosine-based motifs or motifs taking on a beta turn secondary structure. The machinery recognizing these cytosolic apical targeting signals is largely unknown.

Several axonal signals map to cytoplasmic domains as well (for reviews, see Arnold, 2007; Lai and Jan, 2006), but no consensus sequence has emerged. Rather, axonal targeting signals are diverse, and the mechanisms of their action are not yet well understood. These diverse signals might operate at distinct steps on the pathway to the axon. For example, an `axonal targeting' signal could promote lateral enrichment into axonally destined vesicles in the TGN, which may then associate with axonal motors. Subsequently, axonal delivery might be regulated by competence for fusion with axonal secretion sites. Additionally, signals may regulate anchoring and restricted diffusion in axons, or retrieval by endocytosis from inappropriate sites (Winckler, 2004). There are a number of examples of axonal targeting where some mechanistic insights have been gleamed. For example, targeting of the K+ channels Kv1.x to the axonal compartment is achieved via the tetramerization motif (Gu et al., 2003; Rivera et al., 2005). For several other axonal proteins, axonal targeting is endocytosis dependent and the axonal targeting motifs map to endocytosis signals. Therefore, some axonal proteins may become enriched in the axon primarily owing to the preferential endocytosis of `misplaced' somatodendritic receptor pools coupled to specific retention/anchoring in the correct axonal domain, rather than from polarized secretion from the TGN to axons (Garrido et al., 2001; Sampo et al., 2003; Xu et al., 2006).

Previously, we have described a different endocytosis-dependent pathway to the axon, namely transcytosis (Wisco et al., 2003): during biosynthetic delivery, the axonal cell-adhesion molecule L1/NgCAM is first inserted into the somatodendritic domain, then internalized into somatodendritic endosomes, and finally sorted to the axon from the endosomal system. Similarly, NgCAM travels to the apical domain of epithelial MDCK cells via transcytosis (Anderson et al., 2005). In both cell types, a sufficient axonal/apical targeting signal is present in the extracellular domain (Anderson et al., 2005; Sampo et al., 2003; Wisco et al., 2003). Therefore, the cytoplasmic tail is dispensable for axonal/apical accumulation per se. However, signals in the cytoplasmic tail of NgCAM are required for trafficking via the transcytotic route (Wisco et al., 2003), including a signal encompassing tyrosine 33. Likewise, in MDCK cells, the initial basolateral targeting is dependent on tyrosine 33 and on the epithelial cell-specific adaptor complex AP-1B (Anderson et al., 2005). In this work, we undertook a fine-mapping analysis of the cytoplasmic tail signals in NgCAM.

Fig. 1.

NgCAM contains a sufficient somatodendritic sorting signal in its cytoplasmic tail. (A,B) Cultured hippocampal neurons were transfected with DsRed (red) and either the LDLR-based chimera LexNct1-42 (A) or LexNct1-42(Y33A) (B). The cell-surface population of the chimeras was stained with an antibody against an extracellular epitope of LDLR (anti-Lex; green). In A,B, the top row shows the soma region containing the dendrites (arrowheads), whereas the bottom row shows the distant distal portions of the axons (arrows). (C-E) The extent of surface polarity for the three constructs LΔCT, LexNct1-42 and LexNct1-42(Y33A) (shown diagrammatically in C) was determined by measuring the average fluorescence intensity along distal axons and dendrites. The axon/dendrite polarity index (PI) is the ratio of axon/dendrite intensities. The polarity index was determined for 20-25 cells and the percentage of cells displaying preferential axonal accumulation (PI>2), preferential somatodendritic accumulation (PI<0.5) or uniform distribution is plotted in D. The average PI for 20-25 cells from three independent experiments is shown in E. **P<0.001, *P<0.01, Student's t-test.

Trafficking along the transcytotic pathway requires that multiple targeting signals are read and executed sequentially, presumably in different endomembrane compartments. Transcytosis therefore necessitates a system of multiple hierarchically executed signals in which initially a somatodendritic/basolateral signal is active and cis-dominant over an axonal/apical signal. After somatodendritic/basolateral delivery and endocytosis, the axonal/apical signal becomes active in endosomes, whereas the somatodendritic/basolateral signal is turned off. Furthermore, the transcytotic model predicts that the `recessive' axonal signals may be executed throughout the biosynthetic pathway if the somatodendritic/basolateral signal is deleted or mutated. In this work, we determined that the previously identified basolateral signal of NgCAM also acts as a sufficient somatodendritic targeting signal. In addition, we identified a second axonal targeting signal located in the cytoplasmic tail. This cytoplasmic axonal targeting signal also promotes apical trafficking of NgCAM in MDCK cells. Transcytotic routing is therefore dependent on the presence and hierarchical regulation of multiple sorting signals in the cytoplasmic tail of NgCAM.


The basolateral targeting signal of NgCAM is also sufficient for somatodendritic targeting

We have previously shown that NgCAM reaches the axonal plasma membrane by an indirect transcytotic route via the somatodendritic surface (Wisco et al., 2003; Yap et al., 2008). Therefore, NgCAM should contain a bona fide somatodendritic targeting signal for initial delivery to the somatodendritic surface. In MDCK cells, a tyrosine at position 33 of the cytoplasmic tail serves as a basolateral signal. Indirect evidence using a point mutation of tyrosine 33 suggested that somatodendritic targeting was mediated by the same signal (Wisco et al., 2003). To test more directly whether the same tyrosine or other tail-signals are important in trafficking of NgCAM in neurons, we generated chimeric proteins containing the extracellular and transmembrane domains of the low-density lipoprotein receptor (LDLR) plus either the proximal half (CT1-42) or distal half (CT45-114) of the NgCAM tail (Fig. 1C, Fig. 2C). This was necessary because the ectodomain of NgCAM contains a sufficient axonal sorting signal, which makes it almost impossible to identify cytoplasmic signals when the extracellular domain is present. As a control, we used LDLR lacking its cytoplasmic tail (LΔCT). The constructs were expressed transiently in cultured hippocampal neurons using Lipofectamine 2000. DsRed was co-transfected to delineate the entire arborization of the neuron. In order to quantify the extent of polarized expression, we determined the average pixel intensity along the distal part of axons and along the dendrites, and calculated a polarity index by dividing the average axon intensity by the average dendrite intensity [A/D PI; as in Wisco et al. (Wisco et al., 2003)]. LΔCT was uniformly expressed on axons and dendrites and had an A/D PI of 0.9 (±0.11 s.e.m.; n=25 cells), indicating close to equal intensities on axons and dendrites (Fig. 1D,E). When the proximal half of the NgCAM cytoplasmic tail (CT1-42) was added to LΔCT to generate LexNct1-42 (Fig. 1C), the expressed protein was highly enriched on soma and dendrites (soma panel; arrowheads, Fig. 1A; D) and was largely undetectable on the axon (Fig. 1A, axon panel; arrows). LexNct1-42 had an average A/D polarity index of 0.3 (±0.06 s.e.m.; n=10 cells), indicating over threefold higher signal intensity on dendrites than on axons (Fig. 1E).

Fig. 2.

The cytoplasmic tail of NgCAM contains an axonal sorting signal. (A,B) Cultured hippocampal neurons were transfected with DsRed (red) and either the LDLR-based chimera LexNct45-114 (A) or LexNct1-114 (B). The cell-surface population of the chimeras was stained with an antibody against an extracellular epitope of LDLR (green). In order to appreciate the difference in staining intensity on distal axons and on dendrites, single green channel insets are shown at the bottom. (C) Diagram of LDLR- or CD4-based chimeras carrying either the entire NgCAM cytoplasmic tail (Nct1-114) or the C-terminal half of the NgCAM cytoplasmic tail (Nct45-114). (D) Axon/dendrite polarity index (A/D PI) and the percentage of cells showing uniform, axonal or somatodendritic surface accumulation are shown for the constructs depicted in C (20-25 cells were analyzed). ***P<0.0001; statistically significant differences from the ΔCT backbones.

We then tested whether tyrosine 33 was necessary for somatodendritic targeting of LexNct1-42 by introducing a Y to A substitution at position 33 of the chimera (Fig. 1C). Expression of this construct (LexNct1-42Y33A) showed uniform distribution (A/D PI=0.9±0.1 s.e.m.; n=27 cells; Fig. 1B,D,E) similar to LΔCT, demonstrating directly that the NgCAM cytoplasmic tail contains a sufficient somatodendritic targeting signal that requires tyrosine33 for activity. The demonstration of a bone fide somatodendritic signal argues that the initial somatodendritic delivery of NgCAM observed in our previous kinetic experiments is in fact signal mediated.

The cytoplasmic tail of NgCAM contains an axonal targeting signal

Next, we tested whether the distal half of the NgCAM tail (Nct45-114) contained any additional targeting information by generating a chimera with LΔCT (LexNct45-114) (Fig. 2C). LexNct45-114 was found at low levels on soma and dendrites (arrowheads) and at higher levels along the axonal surface (arrows; Fig. 2A green). The distal region of the axon was particularly brightly stained. Determination of A/D PI showed two- to threefold enrichment on the axonal surface (2.6±0.2 s.e.m.; n=21 cells; Fig. 2D). To confirm the axonal localization of LexNct45-114, we double-stained cultures expressing LexNct45-114 with either a somatodendritic marker, MAP2, or a marker for the initial segment of the axon, ankyrin G (Boiko et al., 2007). LexNct45-114 was enriched in the MAP2-negative process that contained ankyrin G (data not shown), identifying it as the axon. The cytoplasmic tail of NgCAM therefore contains sufficient targeting information to enrich a non-polarized membrane protein on the axon.

This observation was unexpected as a chimera of CD8α containing the entire cytoplasmic tail of NgCAM showed no axonal enrichment (Sampo et al., 2003). One possible explanation for the discrepancy is that the axonal targeting signal might be masked (and therefore inactive) in the context of the entire cytoplasmic tail. Therefore, we created a chimera of the entire NgCAM cytoplasmic tail (CT1-114) and LΔCT (LexNct1-114) (Fig. 2C). When LexNct1-114 was expressed in cultured hippocampal neurons, significant axonal enrichment was again observed (arrows; Fig. 2B,D; A/D PI=3.8±0.6 s.e.m.; n=23 cells). Given the differences between the capacity of the NgCAM cytoplasmic tail to redirect LDLR or CD8α ectodomain chimeras to the axon, we created an additional chimera using the frequently used CD4 as a backbone (Garrido et al., 2001; Garrido et al., 2003b; Gu et al., 2003; Xu et al., 2006). CD4 without its cytoplasmic tail (CD4ΔCT) is found uniformly along axons and dendrites (Fig. 2C,D; A/D PI=1.1±0.1 s.e.m.; n=14 cells). We added NgCAM CT1-114 or CT45-114 to the CD4 extracellular and transmembrane domain to create CD4exNct1-114 and CD4exNct45-114 (Fig. 2C). Similar to the LDLR-based chimeras, the CD4-NgCAM chimeras were enriched on the axon (A/D PI=2.7±0.17 s.e.m., n=18 cells; or 2.9±0.3 s.e.m., n=18 cells; Fig. 2D), albeit not as strongly as full-length NgCAM [A/D PI of over 5 (Wisco et al., 2003)]. In summary, we showed that the cytoplasmic tail of NgCAM indeed contains both sufficient somatodendritic and axonal targeting information.

As the extracellular domain of NgCAM is sufficient for axonal sorting without the cytoplasmic tail, we wondered whether we could determine any additive benefit of having a second axonal signal in the cytoplasmic tail. We previously found that NgCAM without the cytoplasmic tail [NgCAMCT3 (Chang et al., 2006)] reaches the axonal surface with a kinetic delay of several hours compared with full-length NgCAM. In order to determine more precisely what the role of the second axonal signal might be, we used the previously described NgCAMCT43 (Anderson et al., 2005), which is truncated after amino acid 43 of the cytoplasmic tail, thereby removing the cytoplasmic axonal signal, but retaining the extracellular axonal signal and the cytoplasmic somatodendritic signal. This construct targets to the basolateral domain in MDCK cells dependent on AP-1B (Anderson et al., 2005). By contrast, NgCAMCT43 was still found axonally enriched. When the localization of full-length NgCAM and NgCAMCT43 were compared on parallel coverslips, the A/D PI of NgCAM was 6.6±0.45 s.e.m. (n=23 cells), whereas the A/D PI of NgCAMCT43 was 4.7±0.4 s.e.m. (n=32 cells). Student's t-test showed a significant difference between these values at P=0.003. The surface expression of the two constructs was not significantly different (data not shown) and kinetic differences of surface expression were not noted. The cytoplasmic axonal signal therefore improves axonal polarity of NgCAM.

Cytoplasmic tail signals are sufficient for transcytotic routing

For proper transcytotic routing, multiple targeting signals must be read and executed sequentially, probably in different intracellular compartments. Based on work in MDCK cells, we have previously proposed that the basolateral signal in the cytoplasmic tail is executed first in the biosynthetic pathway, then inactivated by phosphorylation, thus allowing execution of the lumenal apical targeting signal (Anderson et al., 2005). As the cytoplasmic tail contained both a second axonal signal and a somatodendritic signal (Figs 1, 2), we asked whether the NgCAM cytoplasmic tail was sufficient to route a reporter protein to the axon by the multi-step transcytotic route or whether the extracellular axonal signal of NgCAM was required for transcytotic trafficking. We therefore determined the pathway taken by LexNct1-114 to the axon by performing kinetic analysis, as carried out previously (Wisco et al., 2003). For this assay, we infected neuronal cultures with a recombinant adenovirus encoding LexNct1-114 for 4 hours, then blocked transit through the Golgi by treatment with Brefeldin A (BFA) to accumulate LexNct1-114 intracellularly, and then reversed the transport block by washing out BFA. This assay thus allowed us to follow a pulse of LexNct1-114 from the Golgi to the surface (see Materials and Methods for details). Initially, only cells with intracellular staining against LexNct1-114 without surface staining were observed (Fig. 3A; diamonds) demonstrating the efficiency of the BFA block. The first population of cells with detectable surface staining showed somatodendritically enriched localization of LexNct1-114 (Fig. 3A; circles). This somatodendritically expressing population was replaced over time with a population expressing LexNct1-114 on the axonal surface (Fig. 3A; squares). The observed progression of surface expression from initially somatodendritically enriched and later axonally enriched was similar to that observed for full-length NgCAM itself (Wisco et al., 2003). Interestingly, the kinetics of surface appearance was significantly slowed compared with wild-type NgCAM where first surface appearance could be seen as early as 2 hours after BFA washout. Nevertheless, LexNct1-114 was capable of transcytosis in the absence of the NgCAM extracellular domain.

Next, we asked whether there is something unique about the basolateral/somatodendritic signal in NgCAM that allows for transcytosis or whether another basolateral/somatodendritic signal could also support transcytotic routing of a protein. We used the well-characterized basolateral signals from LDLR to answer this question. LDLR possesses two basolateral signals in its cytoplasmic tail (Matter et al., 1992). The proximal signal (between residues 5 and 27) is tyrosine based and supports both rapid endocytosis and AP-1B-dependent basolateral targeting, whereas the distal signal supports only basolateral targeting but not rapid endocytosis (Matter et al., 1992; Fields et al., 2007). Simultaneous inactivation of both signals abrogated somatodendritic targeting of LDLR in hippocampal neurons (Jareb and Banker, 1998). We used a truncated LDLR, which lacked the distal signal but contained the proximal signal (Lct27) as a backbone for another set of NgCAM tail chimeras (Fig. 4B). Lct27 by itself did not efficiently polarize to the somatodendritic domain at steady state (Fig. 4B,C). When Lct27 was expressed from a recombinant adenovirus, somatodendritic accumulation was similarly observed in only 20% of cells, with 72% of cells showing uniform distribution. In MDCK cells, the proximal signal was found to be entirely dependent on the epithelial-specific adaptor complex AP-1B for targeting to the basolateral surface (Fields et al., 2007). As AP-1B is expressed at very low levels in neurons (Ohno et al., 1999) when compared with epithelial cells, the poor somatodendritic restriction of Lct27 might be due to saturation of the sorting machinery at longer times of expression (see also below).

Fig. 3.

The presence of multiple cytoplasmic tail signals is required for transcytotic routing of NgCAM chimeras. The site of initial surface delivery of LexNct1-114 (A), Lct27Nct45-114 (B) and LexNct45-114 (C) was determined using the Brefeldin A (BFA) block release assay. The percentage of neurons displaying the chimera on the surface uniformly (triangles), axonally enriched (squares) or somatodendritically enriched (circles) was determined for multiple time points after release from BFA. Additionally, the percentage of cells was scored in which no surface staining was detectable, but intracellular staining was present (broken line). The chimera LexNct45-114, which contains the cytoplasmic axonal signal of NgCAM but lacks the somatodendritic and endocytosis motifs, is delivered directly to the axonal plasma membrane (C). If the somatodendritic/endocytic signal surrounding Tyr33 is present [as in LexNct1-114 (A)], the first surface appearance was frequently somatodendritically enriched, suggestive of transcytosis. Inclusion of the LDLR-derived somatodendritic/endocytic signal in Lct27Nct45-114 (B) also led to initial somatodendritic surface appearance, suggesting transcytotic routing. One representative experiment (of a minimum of three) is shown for each construct.

We then added residues 45-114 of the NgCAM cytoplasmic tail to Lct27 to generate Lct27Nct45-114. Interestingly, this construct was axonally enriched (Fig. 4A, arrows; A/D PI=3.2±0.6 s.e.m.; n=14 cells; Fig. 4C). In order to determine whether Lct27Nct45-114 traveled to the axon by transcytosis, we performed kinetic analysis using BFA block and release, as above, using adenoviral expression of Lct27Nct45-114. Similar to NgCAM and LexNct1-114, Lct27Nct45-114 first appeared enriched on the somatodendritic surface after BFA release (Fig. 3B), before appearing enriched on the axonal surface, suggestive of transcytosis.

Active somatodendritic signals are dominant for sorting in the TGN

One possibility for the delayed execution of the axonal signal during transcytosis is that the cytoplasmic machinery necessary for its execution might be spatially restricted to later compartments of the transcytotic route, such as endosomes, but missing from the biosynthetic pathway (presumably the TGN). We showed previously that a mutant NgCAM with a Y33A substitution did not transcytose but rather traveled directly from the TGN to the axonal surface (Wisco et al., 2003). We therefore wondered whether LexNct45-114 (which lacked the somatodendritic and endocytic motifs surrounding tyrosine 33) reached the axon by transcytosis or via a direct pathway. Again, we used the BFA block/release assay after adenoviral expression to study the kinetics of surface delivery. Unlike LexNct1-114, the first detectable surface expression of LexNct45-114 was on the axon (Fig. 3C, squares) and somatodendritically enriched surface expression was not observed (Fig. 3C, circles). This observation indicates direct rather than transcytotic delivery, and suggests that the axonal signal on its own could be read and executed in the biosynthetic pathway. However, when both a somatodendritic and an endocytosis signal (such as the YRSLE motif of NgCAM) were present in the cytoplasmic tail in addition to the axonal signal, the transcytotic route predominated. Therefore, it appeared that the somatodendritic signal was dominant over the axonal signal in the biosynthetic pathway, and was preferentially read and executed in the TGN.

To further investigate this signal dominance, we combined the distal basolateral signal of LDLR (which lacks rapid endocytosis information) with the axonal NgCAM tail signal. Given the rather poor somatodendritic capacity of the proximal basolateral signal (see above), we first tested how well the distal signal worked as a somatodendritic signal in neurons. We deleted the proximal signal located between residues 5 and 27 of the LDLR cytoplasmic tail to create LΔct5-27 (Fig. 4B). LΔct5-27 showed robust somatodendritic enrichment (A/D PI=0.3±0.06 s.e.m.; n=20; Fig. 4C). Therefore, the distal signal of LDLR was active as a somatodendritic signal in neurons. We then added the axonal signal residues 45-114 of the NgCAM tail (LΔct5-27Nct45-114) (Fig. 4B). This construct therefore contains both a somatodendritic and an axonal signal, but lacks a rapid endocytosis signal. Interestingly, LΔct5-27Nct45-114 showed somatodendritic enrichment (A/D PI=0.5±0.1 s.e.m.; n=15 cells; Fig. 4C), similar to the backbone LΔct5-27 alone. In order to test whether silencing of the somatodendritic signal would allow the execution of the axonal signal, we inactivated the distal basolateral signal by introducing two point mutations in the crucial tyrosine residues at positions 35 and 37 (Matter et al., 1994) to create LΔct5-27YAYA Nct45-114 (Fig. 4B). We find that LΔct5-27YAYA Nct45-114 was no longer somatodendritically enriched, but became axonally enriched instead (A/D PI=2.4±0.3 s.e.m.; n=16 cells; Fig. 4C). This finding indicates that: (1) somatodendritic sorting information is cis-dominant for sorting at the TGN; and (2) that axonal cytoplasmic signals are only executed in the TGN in the absence of active somatodendritic signals.

Fig. 4.

Axonal targeting by the cytoplasmic axonal targeting signal in NgCAM requires the inactivation of the somatodendritic signal. (A) Lct27Nct45-114 (green) localizes preferentially to the axonal surface (arrows) of a transfected hippocampal neuron. DsRed was co-transfected to delineate all neuronal processes (red). Arrowheads indicate dendrites. (B) Diagrammatic depiction of LDLR-based NgCAM cytoplasmic tail chimeras. Chimeras contain either the proximal (amino acid residues 5-27) basolateral sorting signal of LDLR, which is co-linear with the endocytosis motif, or the distal (amino acid residues 28-50) basolateral signal of LDLR, which does not contain an endocytosis motif. Point mutations at Tyr35 and 37 inactivate the distal basolateral sorting signal. (C) The axon/dendrite polarity index was determined for the chimeras depicted in B. **P<0.001, ***P<0.0001; statistically significant differences between bracketed constructs.

The axonal cytoplasmic information mapped to a small region of NgCAM

LexNct1-114 and LexNct45-114 were then used as starting templates to further map the axonal targeting region in the cytoplasmic tail of NgCAM. Successive truncations from the C terminus were generated in LexNct45-114, terminating at positions 97, 78 and 59 of the cytoplasmic tail. All of these constructs showed similar axonal enrichment (Fig. 5A). The axonal enrichment of the smallest region mapped (LexNct45-59) is shown in Fig. 5B (green; arrowheads). A chimera containing residues 66-114 of the NgCAM cytoplasmic tail (LexNct66-114), however, lost axonal enrichment (A/D PI=1.4±0.13 s.e.m.; n=15 cells; Fig. 5A,C) and was easily detected throughout the dendrites (arrows; Fig. 5C) and the axon (arrowheads). To further confirm the axonal targeting capacity of the mapped domain, residues 43-78 were added onto the somatodendritic LexNct1-42 to generate LexNct1-78. As expected, this construct again showed axonal enrichment (A/D PI=3.3±0.5 s.e.m., n=17; Fig. 5A). Therefore, a glycine-rich 15 amino acid residue stretch (SASGSGAGSGVGSPG) in the NgCAM cytoplasmic tail was necessary and sufficient for axonal targeting of LDLR chimeras.

Glycines, but not serines, are required for the targeting capacity of the glycine-rich axonal signal

In order to determine whether the glycine residues or the serine residues, or both, were necessary for axonal targeting capacity of the glycine-rich motif, we introduced point mutations into LexNct45-59 either replacing serines with alanines (S-A: SAAGAGAGAGVGAPG) thereby removing potential phosphorylation sites or glycines with arginines (G-R: SASRSRSRSRVRSPG) to introduce charged residues to create LexNct45-59(S-A) and LexNct45-59(G-R). Total surface expression of the mutants was the same as for LexNct45-59 (not shown), but axonal targeting of LexNct45-59(G-R) was abolished (Fig. 6). The A/D PI of LexNct45-59(S-A) was not significantly different from LexNct45-59. Surprisingly then, the serine residues are not essential for the activity of the axonal targeting motif; however, the signal does not tolerate positive charges.

Fig. 5.

Residues 45-59 of the cytoplasmic tail of NgCAM contain a necessary and sufficient axonal targeting signal. (A) The location of the cytoplasmic axonal targeting signal of NgCAM was mapped by sequential truncations and deletions. The constructs depicted diagrammatically were expressed in neurons and the axon/dendrite polarity index of the surface pool determined. ***P<0.0001, *P<0.01; significantly different from LΔCT. (B,C) Examples of one axonally enriched construct LexNct45-59 (B) and one uniform construct LexNct66-114 (C) are shown. Surface distribution of the chimeras was detected with an anti-Lex antibody without permeabilization (green). DsRed was co-expressed (red) to delineate all neuronal processes. The soma region containing the dendrites (arrows) are shown in the top row, whereas the distal regions of the long axons (arrowheads) are shown in the bottom row for both B and C.

The axonal cytoplasmic signal of NgCAM is recognized as an apical signal in MDCK cells

When NgCAM is expressed in MDCK cells, it localizes to the apical surface (Anderson et al., 2005) (Fig. 7B) and reaches it overwhelmingly by transcytosis (Anderson et al., 2005; Hua et al., 2006). We therefore wondered whether the cytoplasmic axonal signal we identified in NgCAM was active as an apical signal in MDCK cells as well. As LΔCT was apically enriched in MDCK cells on its own (Hunziker et al., 1991), chimeras based on LΔCT could not be used for trafficking studies in MDCK cells. Therefore, we tested two chimeras made with Lct27 as the backbone. We added either NgCAM cytoplasmic tail 45-59 (which still possesses axonal targeting information) or NgCAM cytoplasmic tail 66-114 (which does not contain axonal targeting information) to generate Lct27Nct45-59 and Lct27Nct66-114 (Fig. 7A), and expressed them by transient transfection in MDCK cells. As described before (Matter et al., 1992), Lct27 localized to the basolateral surface (Fig. 7C). By contrast, Lct27Nct45-59 accumulated highly at the apical domain (Fig. 7D), with lower levels detectable on the lateral surfaces. By contrast, Lct27Nct66-114 was restricted to the basolateral domain, indistinguishable from Lct27 by itself (Fig. 7E). Together, these observations suggested that the axonal targeting signal (cytoplasmic tail residues 45-59) of NgCAM also promoted apical targeting.


Previously we suggested that L1/NgCAM follows a transcytotic route to the axonal/apical domains (Wisco et al., 2003; Yap et al., 2008; Anderson et al., 2005). In this work, we identify the cytoplasmic tail signals that orchestrate trafficking along the successive steps of the transcytotic pathway, including a sufficient somatodendritic signal of the YxxØ class and a cytoplasmic axonal targeting signal in a novel class. Additionally, we establish that the same signal can mediate axonal and apical transport in neurons and MDCK cells, respectively.

Mapping of a somatodendritic targeting signal

The transcytotic model proposes that NgCAM first becomes targeted to the somatodendritic domain. We identified the sequences surrounding the alternatively spliced exon RSLE as a sufficient somatodendritic sorting signal, confirming our previous indirect results using a tyrosine 33 point mutant of NgCAM (Wisco et al., 2003). The tyrosine preceding the RSLE motif was found to be necessary for the activity of this somatodendritic signal (Fig. 1). The same sequence mediates basolateral sorting in MDCK cells (Anderson et al., 2005). The YRSLE motif corresponds to the classical tyrosine-based sorting motif (consensus: YxxØ where x is any amino acid and Ø is a bulky hydrophobic amino acid). Significantly, the YRSLE motif has previously been shown to act as an endocytosis motif and bind the clathrin adaptor AP-2 (Kamiguchi et al., 1998). Both somatodendritic targeting and endocytosis of NgCAM are therefore mediated by the same motif.

Fig. 6.

Glycine but not serines are required for activity of the CT45-59 axonal signal. (A) The A/D PI of NgCAM (n=23), LΔCT (n=24), LexNct45-59 (n=38) and its mutants LexNct45-59(G-R) (n=41) and LexNct45-59(S-A) (n=30) (as indicated on left) were determined. Error bars correspond to s.e.m. ***P<0.0001. (B,C) The localization of LexNct45-59(G-R) (B) and LexNct45-59(S-A) (C) is shown. Surface distribution of the chimeras was detected with an anti-Lex antibody without permeabilization (green). DsRed was co-expressed (red) to delineate all neuronal processes. The soma regions containing the dendrites are shown in the top row, whereas the distal regions of the long axons are shown in the bottom row for both B and C.

We also investigated the somatodendritic sorting capacity of the two basolateral sorting motifs in LDLR. Both of these basolateral signals are tyrosine-dependent signals, but fall into distinct classes from YxxØ. The proximal signal falls into the FxNPxY class, whereas the distal signal is atypical and requires two tyrosines for activity (QDGYSYPSR). A triple mutant replacing the three tyrosines with alanines in both motifs loses somatodendritic targeting (Jareb and Banker, 1998). Unexpectedly, the proximal signal on its own had poor capacity for somatodendritic targeting (Fig. 4). The proximal signal is strictly dependent on the epithelial adaptor AP-1B in MDCK cells (Fields et al., 2007). AP-1B is highly expressed in many, but not all, epithelia and only expressed at very low levels in embryonic brain (Ohno et al., 1999). It is therefore likely that neurons express insufficient levels of AP-1B for efficient somatodendritic targeting via the proximal motif. The distal signal, however, was a potent somatodendritic signal (Fig. 4). Adaptors other than AP-1B, possibly AP-4 (Yap et al., 2003), therefore could mediate somatodendritic targeting of YxxØ and GYSY somatodendritic signals. The FxNPxY motif, however, is a poor somatodendritic signal in neurons and apparently does not make efficient use of neuronal adaptor complexes.

Mapping of a cytoplasmic axonal targeting signal

Two domains of NgCAM contain sufficient information for axonal transport, the extracellular domain and the cytoplasmic tail (Figs 2, 5). The extracellular axonal signal was mapped to the FNIII repeats (Sampo et al., 2003). It has previously been reported that the cytoplasmic tail of NgCAM was dispensable for targeting because a chimera between the extracellular domain of CD8α and the cytoplasmic tail of NgCAM was not targeted preferentially to the axon (Sampo et al., 2003). By contrast, we observed axonal targeting of chimeras constructed from the ectodomain of LDL receptor and the cytoplasmic tail of NgCAM or from the ectodomain of CD4 and the cytoplasmic tail of NgCAM (Fig. 2). It is unclear why the CD8α chimera does not polarize to the axon, but folding or accessibility of tail signals might be influenced by the nature of the extracellular domain. None of the chimeras accumulated as highly on axons as did full-length NgCAM, suggesting that the NgCAM extracellular domain contributes to the efficiency of axonal targeting. Likewise, the cytoplasmic axonal signal improves axonal targeting (NgCAM versus NgCAMCT43). We mapped the location of the cytoplasmic axonal signal to CT residues 45-59 (Fig. 5), encompassing the glycine-rich region SASGSGAGSGVGSPG. As the signal contains also numerous serine residues, it seemed possible that the signal might be regulated by phosphorylation. However, as mutating all serine residues to alanines had no effect, phosphorylation does not seem to play a role. Instead, disruption of the hydrophobic nature of this segment by introducing charged residues in places of glycine residues inactivated the signal (Fig. 6). Importantly, we show here that the same cytoplasmic sequence in NgCAM promotes both axonal and apical accumulation (Fig. 7), pointing to conservation of signals and possibly machinery for the transcytotic pathway in the two cell types.

Fig. 7.

NgCAM tail 45-59 is sufficient for sorting to the apical membrane in MDCK cells. (A) Diagrammatic depiction of LDLR-based NgCAM cytoplasmic tail chimeras. Chimeras contain the proximal (amino acid residues 5-27) basolateral sorting signal of LDLR that is co-linear with the endocytosis motif and part of the NgCAM cytoplasmic tail as indicated. (B) Fully polarized MDCK cells were transiently transfected with cDNA encoding NgCAM and incubated at 37°C for 30 hours. Cells expressing NgCAM were surface stained with anti-NgCAM antibody (8D9). Subsequently, cells were fixed, permeabilized and incubated with secondary antibodies labeled with Alexa 594. The locations of the apical, basal and lateral domains are indicated by arrows for easier orientation. (C-E) Polarized MDCK cells were transiently transfected with cDNAs encoding Lct27 (C), Lct27Nct45-59 (D) and Lct27Nct66-114 (E), and incubated at 37°C for 30 hours. The transfected constructs were visualized by surface staining with anti-LDLR antibodies (C7). After surface staining, cells were fixed, permeabilized and stained for an endogenous basolateral marker protein using the anti-gp58 antibody. Cells were then incubated with secondary antibodies labeled with Alexa 488 (gp58) and Alexa 594 (Lct27 constructs). Specimens were analyzed by confocal microscopy, and representative x-z sections are shown.

Fig. 8.

Sequence alignment of the cytoplasmic tail of chick NgCAM with chick neurofascin and chick NrCAM. Identical sequences are boxed and shaded. The NgCAM axonal targeting motif is underlined. The somatodendritic targeting signal is in red.

The L1 family members NrCAM and neurofascin are found restricted to axon initial segments and nodes of Ranvier, rather than localized all along axons (Hedstrom and Rasband, 2006; Salzer, 2003). Sequence alignment of the cytoplasmic tails of chick NgCAM, NrCAM and neurofascin revealed that both NrCAM and neurofascin in chick share the YRSLE motif with NgCAM, but only NgCAM contains the glycine-rich axonal motif (Fig. 8). It is thus likely that different L1 family members use different targeting pathways and machinery from L1/NgCAM to localize to their correct domains (Boiko et al., 2007; Dzhashiashvili et al., 2007).

Cytoplasmic axonal signals: a comparison

Other axonal targeting signals have been mapped in several other transmembrane proteins. These signals can be found in the extracellular domain in some proteins (βAPP) (Simons et al., 1995), in the cytoplasmic domain [K+ channels (Chung et al., 2006; Gu et al., 2003; Rivera et al., 2005)] or in both [agrin (Neuhuber and Daniels, 2003); NgCAM, this work and Sampo et al. (Sampo et al., 2003)]. In the case of the Kv1 class of K+ channels, the axonal targeting motif maps to the binding site of its auxiliary subunit Kvβ which provides the link to microtubule binding proteins (EB1) and motors (KIF3). It is possible, but we think unlikely, that other axonal proteins also bind to Kvβ for axonal targeting. Rather, each axonal protein might use distinct signals to bind to unique adaptors that enable complexing with common sets of microtubule-regulating proteins and motors. Multiple KIFs other than KIF3 have been implicated in axonal cargo transport, for example KIF5 [βAPP (Nakata and Hirokawa, 2003); Kv1.3 (Rivera et al., 2007)] and KIF4 [L1 (Peretti et al., 2000)]. However, the adaptors linking L1 to its presumptive motor KIF4 are not known and no candidate genes were identified in a yeast two-hybrid screen using the glycine-rich axonal signal of NgCAM as bait (C.C.Y., unpublished).

In several proteins, the axonal signal maps to an endocytosis signal that is active preferentially in the somatodendritic domain. Axonal accumulation for these proteins is therefore achieved by selective removal from the somatodendritic surface by endocytosis and selective retention on the axon (Garrido et al., 2003a). The axonal signal identified here in the NgCAM cytoplasmic tail does not map to the known endocytosis signal (YRSLE) in NgCAM and does not act as a somatodendritic endocytosis signal (C.C.Y., unpublished).

Signal hierarchy

As the NgCAM cytoplasmic tail is sufficient to direct a heterologous protein to the axon via transcytosis (LexNct1-114; Fig. 3), it contains all the signals needed to execute all three steps: a somatodendritic and endocytosis signal co-linear with YRSLE and an axonal targeting determinant in CT45-59. Therefore, axonal localization of NgCAM does not depend on a single `axonal targeting motif', but on the sequential read-out and execution of multiple sorting motifs in different intracellular compartments.

The transcytotic pathway raises important questions of how a multi-signal protein traverses a multi-step pathway in a sequential fashion. The best studied transcytosing receptor is the pIgR, which carries dimeric IgA from the basolateral to the apical side of epithelial cells (Mostov et al., 2003). For pIgR, ligand binding stimulates transcytosis several-fold in epithelial cells (van Ijzendoorn et al., 2002) and in neurons (de Hoop et al., 1995). Phosphorylation of a serine residue in the basolateral sorting motif might silence it and permit apical delivery from endosomes (Casanova et al., 1990; Luton et al., 1998). Alternatively, kinetic studies combined with modeling suggested that pIgR might enter the transcytotic pathway by bulk-flow (Sheff et al., 1999), rather than by a signal-mediated process.

We mapped an axonal targeting signal in the cytoplasmic tail of NgCAM that can provide access to the transcytotic pathway. Additionally, we show that the axonal targeting information in CT45-114 can be read and executed not only in the endosome (i.e. transcytotic pathway, LexNct1-114), but also in the TGN (i.e. direct pathway) if somatodendritic signals are removed (LexNct45-114; Fig. 3). In MDCK cells, NgCAM remains basolateral if the activity of tyrosine kinases is inhibited (Anderson et al., 2005). Regulated phosphorylation of the YRSLE motif might thus silence this sorting motif in one compartment and keep it active in another. Indeed, work by Lemmon's group has demonstrated that the YRSLE motif can be phosphorylated by a src family kinase, which leads to inhibition of AP-2 binding (Schaefer et al., 2002). Furthermore, the YRSLE motif is bound by additional proteins (Dickson et al., 2002), the binding of which might also be regulated by phosphorylation. Thus, phosphorylation probably plays a role in regulating the activity of signals in NgCAM.

Previously, we suggested that NgCAM tail contains crucial binding sites for src kinases in the cytoplasmic stretch following the YRSLE motif as NgCAMCT43 was not phosphorylated in vitro and remained expressed at the basolateral membrane (Anderson et al., 2005). Interestingly, we found in this study that NgCAMCT43 is axonal in neurons. The exact role of phosphorylation in regulating the activity of targeting signals in NgCAM in neurons therefore requires additional work in the future.

Multiple regulated pathways to the axon?

In previous studies by Vance Lemmon and co-workers, accumulation of L1 in neurites of DRG neurons (rather than just in somata) was shown to be dependent on the YRSLE motif (Kamiguchi and Lemmon, 1998) suggesting that the cytoplasmic YRSLE sequence acts as an axonal targeting motif in DRG neurons. In hippocampal neurons, however, NgCAM with a mutation in the YRSLE motif did not accumulate in dendrites but was instead routed directly into the axon (Wisco et al., 2003). Furthermore, the YRSLE motif acts as a sufficient somatodendritic targeting signal in hippocampal neurons (Fig. 1). Sorting differences depending on neuronal cell type have also been reported for some K+ channels (Rivera et al., 2005). Taken together, these observations raise the interesting possibility that different neurons may differ in the ways they process sorting signals of membrane proteins and might be able to route proteins along distinct pathways in a regulated fashion (Winckler, 2004).

Materials and Methods


Hybridomas producing anti-LDLR (C7) antibodies were purchased from American Type Culture Collection. Hybridomas producing antibodies directed against gp58 (6.23.3) were generated in the laboratory of Dr Kai Simons (Max Planck Institute, Dresden, Germany). The 8D9 anti-NgCAM hybridoma was obtained from NIH Hybridoma Bank. Anti-CD4 antibody was a generous gift from Dr Benedicte Dargent (INSERM Marseille).

Generation of LDLR and CD4-based NgCAM cytoplasmic tail chimeras

The CD4 plasmid was a kind gift from Dr Benedicte Dargent (INSERM Marseille), while LDLR plasmid was from Dr Ira Mellman. NgCAMCT43 has been described previously (Anderson et al., 2005). To generate LDLR and CD4-based NgCAM chimeras, a SalI site was inserted downstream of the transmembrane region of LDLR or CD4 or in the cytoplasmic tail of LDLR at position 27 using QuickChange Site-directed Mutagenesis Kit (Stratagene). Various length of NgCAM cytoplasmic tails were amplified by PCR with forward primers that contained a SalI site and reverse primers with a stop codon followed by a XbaI site. The amplified DNAs were ligated with LDLR- or CD4-containing extracellular domain and transmembrane region at the SalI site and subcloned into pCB6BS vector to generate various LDLR and CD4 chimeras. For constructing LΔCT5-27-based plasmids, the cytoplasmic region of LDLR from residues 5 to 27 was looped out using Altered Sites Mutagenesis Kit (Promega), and a SalI site was inserted downstream of the last residue of the cytoplasmic tail for the purpose of ligating in frame with NgCAM cytoplasmic region. All point mutations produced in this study were generated using Site-directed Mutagenesis Kit. The sequences of all PCR products were confirmed by automated sequence analysis. Adenoviruses were prepared for several of the chimeras after subcloning into pShuttle according to the manufacturer's instructions (see Wisco et al., 2003).

Generation of point mutations in the glycine-rich motif

To generate point mutations in the glycine-rich motif of the cytoplasmic region of NgCAM from residue 45 to 59, plasmid LexNct45-59 was used as a template and amplified by PCR with reverse primers containing point mutations, where either all glycines were replaced by arginines or alanines were substituted for serines to create LexNct45-59 (G-R) and LexNct45-59 (S-A). The sequences of both constructs were confirmed bidirectionally by sequencing analysis.

Cell culture

Primary cultures of hippocampal neurons were grown as described by Wisco et al. (Wisco et al., 2003) and cultured for 8-11 days. The data in Fig. 6 were obtained from a changed culture protocol which we adopted late during the course of this work. In this new protocol, glial-conditioned media was used to feed the neuronal cultures. This led to longer lived and more robust cultures that polarize NgCAM and chimeras more efficiently, as manifested in higher A/D PI. MDCK cells were cultured in MEM containing 7% (v/v) fetal bovine serum, 2 mM L-glutamine and 100 μg/ml penicillin/streptomycin. To allow for polarization, cells were seeded on polycarbonate membrane filters at a density of 4×105 cells per 12 mm filter (0.4-μm pore size; Corning-Costar Transwell units) and cultured for 3 days with daily changes of the medium in the basolateral chamber.


The filter-grown MDCK cells were transiently transfected with cDNAs encoding LDLR-CT27, NgCAM, or LDLR-CT27/NgCAM constructs added to the apical chamber using LipofectAMINE (Invitrogen) according to the manufacturer's protocol and incubated at 37°C for 30 hours. Subsequently, cell-surface staining was performed as described previously (Fields et al., 2007). All immunofluorescence preparations were analyzed using a Zeiss confocal microscope (Microsystem LSM 510; Carl Zeiss MicroImaging) with an Axiovert 100 microscope (Carl Zeiss MicroImaging) and a Plan-Apochromat 63× objective. Images were enhanced and combined using Adobe Photoshop.

Neuronal cultures at DIV 9-12 were transfected using Lipofectamine2000 with 1 μg DNA, 3 μl Lipofectamine2000 for 60-90 minutes, washed and incubated for 16-20 hours. A plasmid encoding DsRed was mixed with the chimera-encoding plasmids and co-transfected. The extent of polarization to axons or dendrites (axon/dendrite polarity index) was determined as in Wisco et al. (Wisco et al., 2003) for staining without permeabilization. For some experiments, recombinant adenoviruses were used as described before (Wisco et al., 2003). Hippocampal cultures grown on glass coverslips were moved to 12-well plates in 500 μl conditioned medium in the presence of 1 mM kynurenic acid and 1-5 μl of purified adenovirus was added per well. After 4 hours, 500 μl of conditioned medium was added and cells incubated for 24-36 hours.

Brefeldin A block/release

Brefeldin A block/release experiments were carried out as described (Wisco et al., 2003). Briefly, cells were infected as described above. After 4 hours, Brefeldin A (Epicentre Technologies) was added to 0.5 to 0.75 μg/ml final concentration. After 12-18 hours, coverslips were washed twice in conditioned medium and transferred to fresh dishes. Coverslips were removed and fixed at indicated times and stained by the sandwich technique described below in which surface and intracellular pools of the same protein can be visualized with two different fluorophores. The sandwich staining allowed counting cells that were infected (i.e. positive for intracellular NgCAM), but where the protein had not reached the surface yet (i.e. negative for surface staining). We observe that the kinetics of surface transport are somewhat variable between cultures and appear slower for older cultures (12DIV) than for younger (8 DIV). All experiments were performed at least three independent times.


Cells were fixed in 2% paraformaldehyde/3% sucrose/PBS in 50% conditioned medium at room temperature for 30 minutes, quenched in 10 mM glycine/PBS for 10 minutes. Alternatively, 2% paraformaldehyde/3% sucrose/0.125% glutaraldehyde/PBS was used in some experiments. The fixation conditions used do not introduce holes into the overwhelming majority of cells. Coverslips were then blocked in 5% horse serum/1% BSA/PBS ±0.05% saponin for 30 minutes. Antibodies were diluted in 1% BSA/PBS ±0.05% saponin and incubated for 1-2 hours. Coverslips were mounted in Vectashield (Vector labs) and viewed on a Zeiss Axiophot with a 40× objective lens. Images were captured with the Orca cooled CCD camera (Hamamatsu) using Openlab software (ImproVision) and processed identically in Adobe Photoshop. For surface staining, live cells were incubated for 5 minutes at room temperature in primary antibody before fixation and secondary antibody incubation.

Separate staining of surface and intracellular population of chimera proteins

To visualize the internal and surface population of NgCAM separately, a protocol from V. Lemmon's group was used with modifications (Kamiguchi et al., 1998). Briefly, cells were fixed, blocked and incubated with primary antibody as described above. The primary antibody was detected with a FITC-GaM Fab (Jackson Immunologicals). Any unbound primary antibody was subsequently blocked with unconjugated GaM-Fab (1 mg/ml) for 1 hour and fixed for 10 minutes. Cells were then permeabilized in 0.05% saponin/1% BSA/PBS for 30 minutes and incubated with a second round of primary antibody. This second round of antibody was detected with a RhodamineRed Goat-anti-Mouse secondary antibody (Jackson Immunologicals). This protocol leads to separable staining of the internal and external populations in all cells, except very highly overexpressing cells. In those cases, the surface could not be completely blocked with unconjugated Fab even at much higher concentrations.


Work in the Winckler laboratory was supported by a March of Dimes Basil O'Connor Award and NIH NS045969 from NINDS. Work in the Fölsch laboratory was supported by a grant from the NIH (GM070736). We gratefully acknowledge the generosity of Dr Benedicte Dargent (Université de la Méditerranée, Marseille, France) for providing crucial reagents. We thank Kevin Liao and Johanna Cannon for culturing neurons.


  • * These authors contributed equally to this work

  • Former address: Department of Cell Biology, Yale University Medical School, 333 Cedar Street, New Haven, CT 06520, USA

  • Accepted February 12, 2008.


View Abstract