Fas ligand is targeted to secretory lysosomes via a proline-rich domain in its cytoplasmic tail

Fas ligand (FasL) is a member of the tumour necrosis factor (TNF) family (Schneider et al., 1998; Tanaka et al., 1998). This 40 kDa membrane protein is expressed by activated T lymphocytes and natural killer (NK) cells (Montel et al., 1995; Suda et al., 1995), and by a small number of non-lymphoid cells. By engaging its receptor (Fas) membrane-bound FasL induces apoptosis in the target cell and, in this way, FasL plays a central role in both cell-mediated immunity and immune downregulation (Nagata and Golstein, 1995).

Tight regulation of FasL surface expression is essential to prevent non-specific killing by T cells. Several different mechanisms are involved in controlling surface expression of FasL. Firstly, T cell activation leads to transcriptional up-regulation of FasL (Suda et al., 1995). The newly synthesised FasL is targeted to specialised secretory lysosomes, present in T cells, which undergo fusion with the plasma membrane when the T cell receptor (TcR) is activated (Bossi and Griffiths, 1999). Once at the cell surface, FasL surface expression is rapidly downregulated by a metalloprotease, which sheds the extracellular domain of the protein (Schneider et al., 1998; Tanaka et al., 1998).

FasL is thought to have a relatively short half-life as it does not accumulate in the secretory lysosomes (Bossi and Griffiths, 1999). FasL-mediated killing can be inhibited by cycloheximide (el-Khatib et al., 1995), demonstrating that de novo synthesis of protein is required during killing assays. Delivery of newly synthesised FasL to the surface of T cells via the secretory lysosomes has several physiological advantages. First, as degranulation is controlled by TcR cross-linking, delivery of the granule contents occurs only under conditions of activation. Second, secretion is polarised and consequently FasL is delivered exclusively to the immunological synapse between T cell and target, ensuring that only the target cell recognised is killed, thus providing an additional level of control for the appearance of FasL on the plasma membrane.

Our previous work demonstrated that the cytoplasmic tail of FasL contains the information required for sorting to secretory lysosomes, because swapping the FasL tail onto the transmembrane and extracellular regions of the type II cell surface protein CD69 targets this chimera to the secretory lysosomes (Bossi and Griffiths, 1999). Curiously, the cytoplasmic tail of FasL does not contain any of the well-characterised di-leucine (Letourneur and Klausner, 1992; Pond et al., 1995) or tyrosine-based (Trowbridge et al., 1993) lysosomal sorting motifs.

A prominent feature of the FasL tail is a polyproline-rich region flanked by di-arginine and di-lysine residues. Such proline-rich domains (PRDs) have been shown to be important in associating with various components of the AP-2-mediated endocytic machinery, such as the interaction between amphiphysin and dynamin via their SH3 domain and PRD, respectively (Simpson et al., 1999). The functional importance of this interaction has been shown by microinjection of the SH3 domain of amphiphysin, which inhibited synaptic vesicle endocytosis at the stage of invagination of clathrin-coated pits (Shupliakov et al., 1997). The PRD of FasL might therefore mediate trafficking of the protein to secretory lysosomes by endocytosis from the plasma membrane. Alternatively, the PRD may dictate a direct pathway to the lysosomes for FasL, as has been suggested for the metalloprotease disintegrins MDC9 and MDC15 (Howard et al., 1999).

In order to determine the region of the cytoplasmic tail of FasL required for sorting to secretory lysosomes, we have expressed a series of point and deletion mutants of FasL in cells with secretory lysosomes and cells with conventional lysosomes. We find that the PRD is required for sorting to secretory lysosomes. This sorting motif is not recognised in cells with conventional lysosomes and, in these cells, FasL travels by default to the plasma membrane. We suggest that FasL is sorted directly from the Golgi to the secretory lysosomes via the interaction of the PRD with an SH3-domain-containing cytosolic protein.

The human NK cell line YT and the mouse T-cell lymphoma cell line WR19L were grown in RPMI medium (Gibco) supplemented with 10% foetal calf serum (FCS). The rat cell lines RBL (rat basophilic leukaemia), NRK (normal rat kidney) and Rat-1, and the human HeLa cell line were maintained in IMDM (Gibco) medium supplemented with 10% FCS.

1×10⁷ cells (RBL, NRK, HeLa, Rat-1 and WR19L) were transfected by electroporation (500 μF and 250 V for RBL, NRK, Rat-1 and HeLa; 960 μF and 300 V for WR19L; BioRad, Richmond, California) with 35 μg DNA for WR19L and 20 μg for other cell lines. Transient transfectants were analysed 30-36 hours post-transfection. Stable transfectants were selected by supplementing the growth medium with 1 mg ml⁻¹ G418 (Gibco) 24 hours post-transfection.

RBL, NRK, Rat-1 and HeLa cells were grown on coverslips for 36 hours post-transfection prior to immunostaining. YT and WR19L cells were resuspended in AIM V serum-free medium (Gibco) and plated on coverslips for 20 minutes at 37°C. Cells were fixed with ice-cold methanol for 5 minutes, then stained for 40 minutes with the relevant primary antibody diluted in 1% BSA/PBS. The antibodies used were: mouse anti-human FasL Nok-1 (10 μg ml⁻¹; PharMingen, San Diego, CA), mouse anti-human lamp-1 (1:100 dilution of supernatant; clone H4A3, J. T. August, Developmental Studies Hybridoma bank, IA), rat anti-mouse lamp-1 (1:2 dilution of supernatant; clone IDB4, J. T. August), rabbit anti-human lamp (1:500 dilution of supernatant, clone AS120), mouse anti-rat Lgp-120 (sunshine; 1:100 dilution; Mark Marsh, University College, London, UK), mouse anti-human CD63 (1:2 dilution of supernatant; clone H5C6; J. T. August), mouse anti-human CD40L (10 μg ml⁻¹; PharMingen) and mouse anti-rat MHC class II (1:2 dilution of supernatant, clone OX4; N. Barclay, Dunn School of Pathology, Oxford, UK). After washing, the cells were incubated with the relevant secondary antibody diluted in 1% BSA/PBS for 35 minutes. Final concentrations of secondary antibodies used were: FITC-conjugated goat anti-mouse IgG (14 μg ml⁻¹), Texas-Red-conjugated goat anti-mouse IgG (14 μg ml⁻¹), Texas-Red-conjugated goat anti-rat IgG (14 μg ml⁻¹), FITC-conjugated goat anti-rabbit IgG (15 μg ml⁻¹) and Texas-Red-conjugated goat anti-rabbit IgG (13 μg ml⁻¹; Jackson Immunoresearch, West Grove, PA). The slides were mounted with 90% glycerol/PBS containing 2.5% DABCO (Fluka, Buchs, Switzerland). Cell staining was analysed using an MRC-1024 confocal microscope (BioRad, Hemel Hempstead, UK).

pFasLWT-GFP was constructed as previously described (Bossi and Griffiths, 1999). This plasmid encodes full-length human FasL in the EGFP-C1 (Clontech) vector, in which the CMV promoter was replaced with the Bos promoter (pEGFP-C1-bos) from elongation factor 1α (Mizushima and Nagata, 1990; construct from K. Campbell, Fox Chase, PA). The vector also contains the neomycin resistance gene to enable selection of transfectants with G418. Deletions of the cytoplasmic tail of FasLWT-GFP were performed by PCR or direct subcloning. For PCR, the deletion mutants of FasL (Δ54, Δ67 and Δ74) were amplified from pFasLWT-GFP using Pfu polymerase (Stratagene, La Jolla, CA) and the same 3′ primer encoding a BamHI site; 5′-CAGGTACCGGGAACCACAGCACAGG-3′ and the following 5′ primers, each of which encode a KpnI site: (Δ54) 5′-CAGGTACCCCACCTCCGCCGCCGCC-3′, (Δ67) 5′-CAGGTA-CCCTGCCACCCCTGAAGAA-3′, (Δ74) 5′-CAGGTACCGGGAA-CCACAGCACAGG-3′. The PCR products were subcloned into the pEGFP-C1-bos vector between the KpnI and BamHI sites. pFasLΔ37-GFP was constructed by digestion of pFasLWT-GFP with BglI and BamHI. The digestion product encoding FasLΔ37 was given blunt ends and subcloned in frame into pEGFP-C2 (Clontech) between the SmaI and BamHI sites. pFasLΔpro-GFP was constructed by mutagenesis using pFasLWT-GFP as a template; the restriction site BsshII was introduced either side of the R43-K71 region by two rounds of site-directed mutagenesis (QuickChange, Stratagene). The first round introduced a BsshII site at R42-R43, but also introduced a conservative amino acid change R42-A (5′ primer, 5′-AGGCCTGGTCAAGCGCGCCCACCACCACCA-3′; 3′ primer, 5′-TGGTGGTGGTGGGGCGGCTTGACCAGGCCT-3′). The second round introduced a BsshII site at K71-K72 and also a conservative amino acid change K72-A (5′ primer, 5′-CCACCCCTG-AAGGCGCGCGGGAACCACAGC-3′; 3′ primer, 5′-GCTGTGG-TTCCCGCGCGCCTTCAGGGGTGG-3′). The resulting mutant was digested with BsshII to remove the R43-K71 region and religated. The plasmids pFasL(R43R44-EE)-GFP and pFasL(K72K73R74-EEE)-GFP were also constructed by site-directed mutagenesis, using pFasLWT-GFP as a template. The primers used for RR-EE were: 5′ primer, 5′-AGGCCTGGTCAAGAGGAGCCACCACCACCA-3′; 3′ primer, 5′-TGGTGGTGGTGGCTCCTCTTGACCAGGCCT-3′. The primers used for KKR-EEE: 5′ primer, 5′-CTGCCACCCCTGG-AGGAGGAAGGGAACCACAGC-3′; 3′ primer, 5′-GCTGTGGTT-CCCTTCCTCCTCCAGGGGTGGCAG-3′. pCD63-EGFP-bos was constructed by cloning CD63 into the BamH1 site of the pEGFP-C1-bos promoter vector. This construct encodes a chimera with GFP at the N-terminus of CD63 with a 22 amino acid spacer encoded by the multiple cloning site of pEGFP-C1 before the first methionine of CD63. The sequences of all constructs were confirmed using AP Prism 377 DNA sequencer (Perkin-Elmer, Norwalk, CT).

RBL FasL-GFP transfectants were plated in 6-well dishes 12 hours prior to fluorescence-activated cell sorting (FACS) staining. 2 hours prior to FACS staining, the cell medium was supplemented with 10 μM BB3013 metalloprotease inhibitor (British Biotech, Oxford, UK), after which the dishes were placed on ice for 30 minutes. The medium was replaced with 1 μg ml⁻¹ Nok-1 antibody diluted in medium plus 10 μM BB3013 and incubated for 30 minutes on ice. The cells were washed with FACS buffer (1% FCS/PBS with 1 μM sodium azide), then fixed with 1% paraformaldehyde (Electron Microscopy Sciences, Washington, PA) in FACS buffer for 10 minutes at room temperature. The cells were washed further, and then incubated for 30 minutes with phycoerythrin-conjugated goat anti-mouse secondary antibody (final concentration 50 μg ml⁻¹ in FACS buffer; Jackson Immunoresearch). The cells were washed three times in FACS buffer and scraped from the dishes into FACS tubes and resuspended in a final volume of 300 μl FACS buffer. The samples were analysed using a FACScalibur flow cytometer (Becton Dickinson). PE fluorescence was measured at 575 nm and GFP fluorescence at 525 nm. Fluorescence data was collected for 20,000 events on a four-decade log scale and analysed using CELLQuest software (Becton Dickinson).

NRK transfectants were released from flasks by incubating with 10 mM EDTA/PBS for 10 minutes at 37°C. The cells were washed once with complete media and then incubated with the relevant primary antibody (Nok-1; 10 μg ml⁻¹ or anti-CD63; 1:2 dilution) diluted in FACS buffer (see above) for 30 minutes on ice. Cells were washed three times with ice-cold complete media and one sample of each transfectant was removed at time 0. The remaining cells were divided into three aliquots and incubated in complete medium at 37°C. Cells were harvested after 15 minutes, 30 minutes and 120 minutes, and stained with PE-conjugated goat anti-mouse secondary antibody (final concentration 50 μg ml⁻¹ in FACS buffer) for 30 minutes on ice. The cells were washed three times in FACS buffer and then fixed by resuspending in 1% paraformaldehyde in FACS buffer. The samples were analysed by flow cytometry as above. To calculate internalisation, the PE fluorescence of each sample was plotted and gated to determine the mean fluorescence intensity (MFI) of each sample outside that obtained with background staining (secondary antibody alone). The resulting numbers were divided by the MFI at time 0 to obtain the percentage remaining at the cell surface.

Prior to the assay, cells (NRK transfectants or YT) were adhered onto glass coverslips as previously described. The coverslips were incubated with the relevant antibody (Nok-1; 10 μg ml⁻¹, anti-CD63; 1:2 dilution, or an isotype-matched control antibody (OX4 for YT cells and anti-CD40-L for NRK transfectants) diluted in medium. NRK transfectants were incubated with the antibody for 30 minutes at 4°C, after which they were washed and incubated with fresh medium for 2 hours at 37°C. The cells were then fixed with ice-cold methanol and incubated with the relevant Texas-Red-conjugated secondary antibody. YT cells were incubated with the primary antibody for 2 hours at 37°C, then fixed as above and stained with an anti-human Lamp antibody (clone 120). The endocytosed antibody and the anti-lamp signal were detected with the relevant FITC-conjugated and Texas-Red-conjugated secondary antibodies, respectively. All samples were analysed by confocal microscopy.

The crystal structure of the Fyn SH3 domain complexed with a synthetic peptide (PPAYPPPPPVP) solved to 2.3 Å (Musacchio et al., 1994) was taken as a template from which to construct the FasL-Fyn model. The synthetic peptide in the structure was mutated to represent the FasL peptide. The model was then subjected first to CHARMM Steepest Descent minimisation followed by soaking the model (adding water molecules) and carrying out a DISCOVER Steepest Descent minimisation from within InsightII (1300 steps) until the average absolute derivate converged below at 0.8 kcal mol⁻¹ Å⁻¹.

Our previous findings have shown that, whereas FasL is sorted to the secretory lysosomes of T, NK and mast cells, FasL is expressed directly on the plasma membrane of HeLa cells (Bossi and Griffiths, 1999). This suggests that cells with secretory lysosomes may preferentially sort FasL to this compartment, preventing surface expression, whereas FasL is expressed directly on the cell surface in other cell types. We therefore examined the expression pattern of transfected FasL in several different cell lines with conventional or secretory lysosomes. Each cell line was transfected with a GFP-tagged human FasL construct (FasLWT-GFP) and co-stained for the lysosomal marker Lamp-1 (in HeLa and WR19L) or Lgp-120 (in Rat-1 and RBL; Lewis et al., 1985). In both HeLa and Rat-1, FasLWT-GFP is expressed at the plasma membrane (Fig. 1B,E). In HeLa cells, most FasL is found at the plasma membrane, with some found intracellularly and with a small amount of co-localisation with the lysosomal compartment marker Lamp-1 (Fig. 1C). In Rat-1 cells, more FasLWT-GFP is observed intracellularly (Fig. 1E) with greater co-localisation with the lysosomal marker Lgp-120 (Fig. 1F). These results demonstrate that, in these two cell types with conventional lysosomes, FasL is expressed on the plasma membrane with variable levels of intracellular expression. Similar results are observed with NRK cells (see below). By contrast, in WR19L and RBL cells (Fig. 1G-L), which possess secretory lysosomes, expression of FasLWT-GFP is exclusively intracellular, demonstrating that FasL is retained intracellularly in these cell types. In both cell types, there is a high degree of overlap with the lysosomal markers Lamp-1 and Lgp-120.

We have previously shown that the cytoplasmic tail of FasL contains the information required for secretory lysosome localisation (Bossi and Griffiths, 1999). To pinpoint the motif within the cytoplasmic tail, N-terminal deletion mutants of the FasL tail were constructed (Fig. 2a) and transiently expressed in the RBL cell line (which contains a secretory lysosomal compartment) and examined for surface expression by confocal microscopy and FACS. Membrane expression of FasL was measured by Nok-1 antibody staining (y axis; Fig. 2c) and is expressed as a percentage of total cellular FasL, as measured by GFP levels (x axis; Fig. 2c). Because the transfectants showed slight variations in their level of expression, the Nok-1 stain of transfectants with similar expression levels (10-250 GFP fluorescence units; see Fig. 2c for the gating) was plotted (Fig. 2d). This step prevented misinterpretation of membrane expression owing to overexpression of the protein and allowed the percentage of Nok-1 positive transfectants to be calculated (Fig. 2d). Wild-type (WT) FasL is sorted to the secretory lysosomes of RBL, as described previously (Bossi and Griffiths, 1999), and only low levels of surface expression can be detected by Nok-1 binding (Fig. 2d). Deletion of the first 37 amino acids gives no surface expression above WT (Fig. 2d) and FasL remains intracellular (Fig. 2b). Deletion of 54 amino acids results in the appearance of FasL at the cell surface (23.9% of total FasL at the cell surface; Fig. 2d). This percentage increases with the deletion of 67 amino acids to 31.5% of surface expression (Fig. 2d), and remains at 30.5% when a further 7 amino acids are removed (Fig. 2d; FasLΔ74-GFP mutant). In all of these mutants, FasL is still detected intracellularly (Fig. 2b), regardless of the increased surface expression. A significant proportion of the intracellular pool of FasL in these mutants co-localises with the Lgp-120-positive compartment (data not shown).

From these results, it appears that the region between amino acids 37 and 54 is an important determinant in lysosomal targeting, and that amino acids 55 onwards have an additive effect. This area contains a PRD (Fig. 2a; amino acids 45-71) flanked by basic residues. To investigate whether this proline-rich region contains the lysosomal-targeting motif, amino acids 45-71 were removed from FasLWT-GFP (Fig. 2a; FasLΔpro-GFP) and the construct was stably expressed in RBL. By both confocal and FACS analysis, FasLΔpro-GFP demonstrates substantial membrane expression (Fig. 3b; 25.2%) compared with FasLWT-GFP (1.5%), indicating that this region contains the minimal motif for lysosomal targeting in the FasL tail.

PRDs can bind SH3, WW or GYF domains (Freund et al., 1999; Ren et al., 1993; Staub and Rotin, 1996) and are important in protein-protein interactions (reviewed in Pawson, 1995). Two stretches of amino acids within the proline-rich region of FasL match the consensus sequence of an SH3 domain ligand (RRPPPPPPPP and PPLPLPPLKKR; Fig. 2a). Because peptides from the PRD region of FasL have been demonstrated to bind the Fyn SH3 domain in vitro (Hane et al., 1995), the Fyn SH3 domain was used to model the interaction with the PRD of FasL. Fig. 4 shows the peptide RRPPPPPPPP binding the SH3 domain of Fyn. The model reveals that the interaction of the eight-amino-acid proline helix positions the positively charged arginine residues near the negatively charged region of the SH3 domain. This model predicts a strong charged interaction between the Fyn SH3 domain and the RR residues at the end of the proline helix (shown in green; Fig. 4b). The other potential SH3 domain ligand (PPLPLPPLKKR) also contains positively charged residues (KKR), but the sequence is found in the opposite orientation (Fig. 2a). Because PRDs are known to be able to bind SH3 domains in either orientation (Feng et al., 1994; Finan et al., 1996), this model predicts a critical role for the charged residues in both peptides.

In order to test whether these interactions are important in the sorting motif of FasL, we mutated the basic residues flanking either side of the PRD of FasL (either R42-R43 or K72-K73-R74; Fig. 2a) to the acidic residue glutamic acid in the FasLWT-GFP construct (FasLRR-EE-GFP and FasLKKR-EEE-GFP, respectively). Both constructs were expressed in RBL and analysed for membrane expression by confocal microscopy and FACS analysis. Mutation of either the RR or the KKR residues results in a substantial increase in FasL membrane expression (14.4% and 21.6%, respectively; Fig. 5b) in RBL. These results demonstrate that sorting of FasL is disrupted by substituting negatively charged amino acids at positions 42-43 or 72-74. As WW and GYF domains preferentially interact with polyproline sequences enriched in tryptophan and glycine/histidine residues, respectively (Freund et al., 1999; Staub and Rotin, 1996), these data favour the idea that the PRD of FasL may be interacting with an SH3-domain-containing protein to sort FasL correctly to the secretory lysosomes.

Recent reports have demonstrated an important role for PRDs and SH3-domain-containing proteins in endocytosis (Shupliakov et al., 1997; Simpson et al., 1999; Slepnev et al., 2000). We therefore tested whether the PRD of FasL defines an endocytic motif. Endocytosis of FasL was compared with that of CD63-GFP, which is known to recycle from the plasma membrane to the lysosomes (Kobayashi et al., 2000). We used the non-haematopoietic cell line, NRK, as expression of each FasL mutant (WT, Δpro and Δ74) results in comparable levels at the plasma membrane. CD63-GFP and FasLWT-GFP, FasLΔpro-GFP and FasLΔ74-GFP are all expressed on the cell surface of NRK cells (Fig. 6B,E,H,K). The transfectants were incubated with the relevant antibody and then allowed to internalise any surface-bound antibody for up to 120 minutes. Samples were harvested and analysed by FACS at various time points during internalisation (Fig. 7) and the percentage of surface-bound antibody internalised was calculated (see Materials and Methods). Fig. 7 shows that CD63-GFP internalises up to 70% of the surface-bound antibody within the first 30 minutes, after which the levels reach a plateau. FasLWT-GFP and FasLΔpro-GFP demonstrate similar kinetics to CD63-GFP, although with a lower percentage of antibody internalised (82% after 30 minutes). By contrast, FasLΔ74-GFP, in which all but 6 amino acids of the cytoplasmic tail are deleted, demonstrates a gradual rate of endocytosis, reaching similar levels to the others after 120 minutes.

Fig. 6 shows confocal images of cells after 120 minutes of endocytosis. Merged images show co-localisation of internalised antibody and the GFP signal for all transfectants (Fig. 6C,F,I,L). All four proteins demonstrate internalisation to the final endocytic compartment, the lysosomes (Lgp-120-positive compartment; Fig. 8A,D,G,J), as there is significant overlap with the signals from GFP (Fig. 8B,E,H,K) and Lgp-120 (Fig. 8C,F,I,L show merged images) in NRK cells.

Taken together, these results indicate that the PRD of FasL does not affect the efficiency of endocytosis, as FasLΔpro-GFP demonstrates similar kinetics of internalisation and co-localises to the final endocytic compartment as FasLWT-GFP. Interestingly, removal of most of the cytoplasmic tail of FasL (FasLΔ74-GFP), does not prevent endocytosis (Fig. 7) and internalised protein can still reach the lysosomes (Fig. 8). However, FasLΔ74-GFP does exhibit a reduced efficiency of endocytosis compared with FasLWT-GFP and FasLΔpro-GFP. This suggests that FasLΔ74-GFP is following basal endocytic trafficking in the absence of signals, as previously observed for CD4 (Pelchen-Matthews et al., 1991) and the Fc receptor (FcR; Miettinen et al., 1992).

The NRK transfectants demonstrate the ability of FasL to undergo endocytosis and so we asked whether endogenous FasL is sorted directly to the lysosomal compartment in cells with secretory lysosomes. The human NK cell line YT constitutively expresses FasL, which, like CD63, can be detected in the secretory lysosomes (Fig. 9C,F). In order to ask whether FasL in the secretory lysosomes has transited the cell surface and entered the lysosomal compartment via the endocytic pathway, YT cells were cultured with antibody (in the presence or absence of metalloprotease inhibitor for 2 hours). This assay differs from that performed on NRK cells in that the antibody was cultured with the cells for an extended time (2 hours) to allow for constitutive internalisation. CD63 but not FasL demonstrated endocytosis of surface bound antibody (Fig. 9H,K), which co-localises with the lysosomal compartment in the case of CD63 (Fig. 9I). No internalisation with an isotype-matched control antibody was seen (results not shown). These results indicate that FasL follows a direct pathway to the secretory lysosomes rather than indirectly transiting the plasma membrane via endocytosis.

Secretory lysosomes are unusual secretory organelles found in a small number of cell types. Most of these cells are regulated secretory cells of the haematopoietic lineage, such as T cells and NK cells. Their secretory organelles are modified lysosomes, containing both lysosomal hydrolases and secretory proteins (Stinchcombe and Griffiths, 1999). Our preliminary studies on the sorting of FasL suggested that, although FasL was sorted to the secretory lysosomes found in haematopoietic cells, it was expressed directly on the plasma membrane of other cell types containing conventional lysosomes. In order to investigate the sorting of FasL-GFP, we have expressed the protein in various cell types with either conventional or secretory lysosomes. We find that FasL is sorted to secretory lysosomes and does not appear on the plasma membrane in cells that possess this specialised organelle but is expressed on the plasma membrane of other cell types. This supports the idea that cells with secretory lysosomes may possess specialised sorting machinery that is absent in other cell types.

In this paper, we have identified the sorting motif required for FasL to reach the secretory lysosomes and asked how this affects sorting in the different cell types. We find that sorting is mediated by a PRD, which facilitates sorting of FasL directly to the secretory lysosomes from the trans-Golgi network without transiting the cell surface. Because FasL is such a potent mediator of cell death, this is an important mechanism in preventing constitutive release of FasL. We propose a model for a specialised sorting mechanism based on the interaction of the PRD of FasL with an SH3-domain-containing protein expressed selectively in cells with secretory lysosomes (Fig. 10). According to our model, FasL is sorted from the TGN by virtue of the PRD that recognises an SH3-domain-containing protein, which facilitates its sorting to the secretory lysosomes. In cells lacking this compartment, we propose that the SH3-domain-containing protein is absent and, in the absence of this component of the sorting machinery, FasL travels by default to the plasma membrane. A number of SH3-domain-containing proteins specific to cells of the haematopoietic lineage have been identified (Thomas and Brugge, 1997). Because most cells with secretory lysosomes are derived from this lineage, they are likely to express such proteins.

PRDs are known to be able to interact not only with SH3 domains but also with WW and GYF domains. In order to determine which of these domains might be involved in the interaction with the PRD of FasL, we carried out site-directed mutagenesis of the positively charged arginine and lysine residues at the end of the polyproline helices present in the FasL tail. These residues are predicted to form critical interactions with SH3, but not WW or GYF domains, based on the consensus motifs for binding to these domains (Freund et al., 1999; Staub and Rotin, 1996). Because we see significant levels of mis-sorting of FasL after site-directed mutagenesis of the positively charged residues in the motif to negative charges, this suggests that the PRDs are interacting with SH3 rather than WW or GYF domains.

Another critical aspect of our model is that FasL is sorted directly to the secretory lysosomes without transiting the plasma membrane. Our previous work supported this model, as FasL can only be bound by antibody at the plasma membrane after exocytosis in T cells in the presence of metalloprotease inhibitors, demonstrating the presence of highly effective metalloproteases in these cells. The fact that an antibody against the extracellular domain detected FasL within the secretory lysosomes (Bossi and Griffiths, 1999) indicates that FasL must reach this compartment directly. If FasL had been expressed on the cell surface prior to the lysosomal compartment then its extracellular domain would have been cleaved by a metalloprotease. In this paper, we have demonstrated that endogenous FasL cannot bind antibody and be internalised from the plasma membrane even in the presence of metalloprotease inhibitors, further supporting the model that FasL is sorted directly from the TGN to the secretory lysosomes.

Recently PRD/SH3-domain interactions have been shown to play a critical role in endocytosis. Disruption of these interactions perturbs function, as demonstrated by microinjection of the SH3 domain from amphiphysin, which inhibits endocytosis of synaptic vesicles (Shupliakov et al., 1997). In this study, we find no gross disruption of endocytosis upon removal of the PRD of FasL. In NRK cells, FasL endocytosis is unaffected by removal of the PRD (Figs 6, 7), demonstrating that the PRD is not an essential endocytic determinant.

The ability of all of the deletion mutants to undergo endocytosis explains one of the surprising results of expression of these mutants in RBL (Fig. 2b). Although we find that disruption of the PRD leads to increased cell surface expression, demonstrating that this region is required for correct intracellular sorting, some intracellular localisation is also observed. Our studies on endocytosis explain these findings. When FasL reaches the cell surface by mutation of the PRD, it is then able to undergo endocytosis even in cells with secretory lysosomes. We have shown that RBL cells endocytose CD63 efficiently, whereas WR19L cells do so poorly (data not shown). This suggests that the degree of intracellular staining of FasL may correlate with the efficiency of endocytosis in these cell types. In RBL cells, intracellular FasLΔpro-GFP and FasLΔ74-GFP co-localises with the Lgp-120-positive compartment (data not shown), consistent with it reaching this compartment via endocytosis from the plasma membrane.

From our results, we cannot determine whether the SH3 protein chaperones FasL to the secretory lysosomes directly or by other interactions with known sorting complexes, such as AP-3 (which has been shown to mediate direct sorting of proteins from the trans-Golgi network to the yeast vacuole) (Odorizzi et al., 1998) and mammalian lysosomes (Le Borgne et al., 1998). Two lines of evidence suggest that FasL itself does not interact directly with any of the adaptor complexes used during protein trafficking. First, the cytoplasmic tail of FasL does not interact with any of the μ subunits of AP-1, AP-2 or AP-3 in a yeast two-hybrid assay (D. Stephens and G. Banting, unpublished). Second, mutation of minimal tyrosine or di-leucine motifs in the tail of FasL (7Y-9Y-13Y→AAA and 29V-30L→AA) does not result in mis-sorting of FasL (G. Bossi, unpublished). We currently favour the interpretation that interaction of the PRD of FasL with an SH3-domain-containing protein allows formation of a protein complex that may allow interaction with AP-3.

Which SH3-domain-containing protein(s) are required for sorting? We have shown by computer modelling that the proline-rich region of FasL is capable of binding a Fyn-like SH3 domain. Hane et al. demonstrated an in vitro interaction between the SH3 domains of Fyn and Lck with peptides of the proline-rich region of murine FasL (Hane et al., 1995). The in vivo importance of the FasL-Fyn/Lck interaction is still unknown and there is currently no evidence to support the idea that Fyn or Lck themselves are important in the sorting of FasL. As the specificity and binding strength of an SH3 domain-ligand interaction is governed by neighbouring residues to the proline-rich core sequence (Feng et al., 1994; Ren et al., 1993), it is possible that an in vitro binding peptide partner for FasL may not reflect an in vivo partner.

We have defined a novel lysosomal-sorting motif based on a PRD found in FasL. It is likely that other membrane proteins will use the same mechanism to reach the secretory lysosomes in lymphoid cells. Although FasL is the only member of the TNF family with SH3-binding domains in the cytosolic tail, several members of the ADAMS (a disintegrin and metalloprotease) family of metalloproteases contain polyproline sequences in their tails (Rosendahl et al., 1997). The metalloprotease responsible for extracellular cleavage of FasL is thought to be a member of this family (Schneider et al., 1998; Tanaka et al., 1998), and it is tempting to speculate that the metalloprotease might use a similar sorting motif. Simultaneous sorting of FasL and its metalloprotease could enable subsequent delivery of both molecules to the cell surface provide yet another mechanism to control FasL activity precisely.

We would like to thank R. Dunbar and R. da Silva for helpful reading of the manuscript, and J. Stinchcombe, M. Marsh and D. Owen for valuable discussions. The work was supported by grants from the Wellcome Trust (040825, 059163, 059313 and 050613).