Functional characterization of wild-type and mutant human sialin

Identification of a dileucine-based sorting motif

The late endosomal and lysosomal localization of sialin has been recently shown by cell fractionation and immunofluorescence studies (Aula et al, 2002). In order to identify the signals responsible for this localization, we mutated potential sorting motifs in tagged constructs of human sialin. The protein was fused either at its amino-terminus to GFP or at its carboxy-terminus to a V5 epitope. When transiently expressed in HeLa cells, both constructs showed a punctate intracellular distribution, which overlapped with the endogenous late endosomal and lysosomal protein LAMP1 by confocal microscopy (Figure 1A and C). Consistently, when co-expressed, the V5- and GFP-tagged proteins had identical distributions (Figure 1B). These data show that the GFP or V5 tagging do not affect the intracellular localization of sialin. For practical reasons, we used GFP tags throughout this study.

Lysosomal sorting signals generally consist of short sequence motifs recognized by cytosolic membrane coats (Bonifacino and Traub, 2003). Sequences corresponding to potential dileucine-based signals are found in the amino-terminal region (DRTPLL) and in the loops connecting the fourth and fifth (ERSKLL) and the sixth and seventh (EKEYIL) transmembrane domains of sialin. These sequences are cytosolic in the predicted topological model (Figure 2A). To examine whether they act as sorting signals, their critical leucine or isoleucine residues were mutated to alanine. Mutations in the second and third sequences concentrated GFP-sialin to a perinuclear pole, but preserved its punctate, intracellular appearance (Figure 2B). In contrast, when the first sequence was mutated, GFP-sialin displayed a diffuse membrane localization corresponding to the plasma membrane. A similar effect was observed when the first pair of leucine was mutated to glycine (see Figure 5A), when the whole DRTPLL sequence was deleted (data not shown) or when the mutated constructs were expressed in other cell lines (COS-7, human embryo kidney 293 (HEK 293)).

To confirm the plasmalemmal localization of the DRTPLL mutants, we performed surface biotinylation experiments in HEK 293 cells. Biotinylated, plasmalemmal proteins were separated from other cellular components by streptavidin affinity chromatography and the relative amounts of GFP-sialin in the intracellular and plasmalemmal pools were assessed by immunoblotting with anti-GFP antibodies. A single, diffuse band centred at 80 kDa was observed in GFP-sialin transfected cells (Figure 3), in agreement with the molecular weights of human sialin (54.6 kDa) and GFP (26.9 kDa). Its heterogeneity may originate from glycosylation, as several N-glycosylation sites are predicted in the sialin sequence. As illustrated in Figure 3, lanes 4–6, most GFP-sialin was intracellular when the wild-type (WT) protein was expressed. In contrast, the amount of GFP-sialin in the plasmalemmal pool dramatically increased when leucine residues 22 and 23 were mutated to glycine, in agreement with the fluorescence microscopy analysis (compare lanes 6 and 9).

These biochemical and morphological data demonstrate that the DRTPLL motif is necessary to target sialin to its intracellular destination.

The DRTPLL motif acts as an internalization signal

Lysosomal membrane proteins reach their destination by two intracellular routes: a direct route, in which proteins are diverted to endosomes at the trans-Golgi network (TGN), and an indirect, endocytic route, in which proteins traffic to endosomes and lysosomes via the plasma membrane (Bonifacino and Traub, 2003). The relative importance of these routes for a specific protein depends on the ability of its sorting signals to be recognized at the TGN or at the plasma membrane. To examine whether the DRTPLL motif is recognized at the plasma membrane, we tested its capacity to trigger internalization of a resident plasma membrane protein.

We first constructed CD4-sialin chimera by substituting the amino-terminal region of sialin (amino acids 2–41) for the cytosolic, carboxy-terminal domain of the CD4 protein (Figure 2C), and analysed their steady-state localization in HeLa cells (Figure 2D). As reported previously (Garrido et al, 2001), the truncated CD4 lacking its cytosolic domain localized to the plasma membrane. In contrast, the chimera with an intact DRTPLL motif displayed a punctate intracellular distribution, while the chimera bearing a L22G/L23G mutation in the sialin domain remained at the plasma membrane.

We next tested the capacity of CD4-sialin chimeras to internalize antibodies bound to their extracellular domain. Transiently transfected HeLa cells were pre-labelled with antibodies at 4°C and incubated at 37°C for increasing durations. Cells expressing the truncated CD4 or CD4-sialin L22G/L23G proteins displayed a membrane labelling that did not evolve over time and disappeared when cells were washed with an acidic medium (Supplementary Figure 1A and B). In contrast, antibodies bound to the WT CD4-sialin chimera rapidly moved to punctate structures, which resisted extracellular acid washing. The DRTPLL motif is thus an internalization signal. Interestingly, most antibodies internalized by the CD4-sialin chimera reached late endosomal compartments, since the corresponding puncta were positive for GFP-sialin when both constructs were expressed in HeLa cells (Supplementary Figure 1C). The DRTPLL motif thus seems sufficient to target sialin to its final destination.

Characterization of the transport activity of human sialin

The molecular function of sialin has been inferred from the biochemical defect of sialic acid storage diseases; however, it has not been directly studied. To develop a functional assay of sialin, we took advantage of the cell surface localization of the DRTPLL mutants and examined their capacity to translocate sialic acid across the plasma membrane. This approach allows to study lysosomal efflux as a classical cellular uptake (Kalatzis et al, 2001).

HEK 293 cells were transiently transfected with GFP-sialin plasmids and analysed for their ability to take up N-acetyl-[6-³H]neuraminic acid ([³H]Neu5Ac) from the extracellular medium. When uptake was performed at neutral pH, cells expressing GFP-sialin L22G/L23G accumulated [³H]Neu5Ac as much as mock-transfected cells (Figure 4A). However, uptake was strongly activated when the extracellular medium (topologically equivalent to the lysosomal lumen in our assay) was acidified to a pH value of 5.6. A mean increase of 5.5±0.5-fold (±standard error of the mean (s.e.m.)) over background was obtained from 19 independent experiments. In contrast, cells expressing the WT construct took up sialic acid at background (mock) level at both pH. To examine whether the L22G/L23G sialin-mediated signal represented membrane transport or binding to the cell surface, cells were perforated with digitonin for 6 min after their incubation with [³H]Neu5Ac. As shown in Figure 4A, this treatment released the radioactivity. The relocalization of GFP-sialin to the cell surface is thus associated with an ability to transport sialic acid across the plasma membrane.

The sialin-mediated process remained linear for 15 min (data not shown). We thus used uptake periods ⩽15 min throughout this study to measure transport velocities. When increasing [³H]Neu5Ac concentrations were used, uptake increased linearly up to 0.3 mM and progressively saturated for higher values (Figure 4B), with a K_M of 1.52±0.26 mM at pH 5.6 (mean±s.e.m. of four independent transfections; see Table I). The maximal velocity (V_max), which depends on the transfection efficiency, varied from 104 to 245 pmol/min per 10⁶ cells, with a mean value of 174 pmol/min per 10⁶ cells (Table I).

To characterize the dependence of sialin on protons, uptake was measured at several extracellular pH's between 7.4 and 5.0. As illustrated in Figure 4C, sialin-mediated uptake was low or undetectable at pH ⩾6.5, but increased linearly when pH decreased below 6.5. This stimulation of sialin at acidic pH might result from a higher affinity for protonated sialic acid, from an allosteric control by protonable residues exposed to the noncytosolic compartment, or from a co-transport of H⁺ with sialic acid. To discriminate between these possibilities, we first studied the dependence of transport parameters on extracellular pH (Table I). Saturation kinetics experiments performed at pH 5.0 yielded a K_M value (2.13±0.29 mM) similar to that obtained at pH 5.6, thus ruling out the hypothesis of a better affinity for protonated Neu5Ac. In contrast, the maximal velocity increased when pH decreased. To avoid variations due to differences in transfection efficiency, we performed one experiment at both pH 5.6 and 5.0 and observed a 1.6-fold increase in V_max (K_M values: 2.16 and 1.93 mM, respectively), in agreement with the pH dependence curve obtained at nonsaturating Neu5Ac concentration (Figure 4C).

The remaining two hypotheses were assessed using the proton ionophore nigericin, which exchanges intracellular K⁺ for extracellular H⁺ and thus dissipates the pH gradient across the plasma membrane. As illustrated in Figure 4D, this treatment abolished Neu5Ac uptake. Therefore, a transmembrane pH gradient (rather than a mere acidic extracellular pH) is required to drive sialin, in agreement with a H⁺/sialic acid symport mechanism.

We also determined the substrate selectivity of human sialin by adding unlabelled compounds (7 mM) simultaneously to [³H]Neu5Ac in the transport assay. As shown in Table II, acidic sugars such as gluconic acid or D-glucuronic acid inhibited uptake as efficiently as, or better than, Neu5Ac. In contrast, neutral sugars such as N-acetylmannosamine or N-acetylglucosamine had no effect. We also tested aliphatic monocarboxylates and dicarboxylates, since these compounds were reported to compete with sialic acid for transport across rat liver lysosomal membranes (Havelaar et al, 1998, 1999). Monocarboxylates such as L-lactate or pyruvate inhibited uptake as efficiently as acidic sugars. However, aliphatic dicarboxylates such as maleate or succinate did not significantly inhibit sialin-mediated transport, in contrast to the aforementioned studies. L-Glutamate, which is transported by three paralogs of sialin (Reimer and Edwards, 2004), and L-aspartate had no effect.

In conclusion, sialin is a H⁺-coupled transporter able to export sialic acid and other acidic monosaccharides out of the lysosome, as anticipated from previous studies on sialic acid storage diseases.

Effect of pathogenic mutations on sialic acid transport

We then used the functional assay of sialin to characterize the molecular effect of mutations causing sialic acid storage disease (Verheijen et al, 1999; Aula et al, 2000). Only mutations that do not preclude synthesis of a full-length polypeptide were considered (Figure 2A): the R39C mutation, which is responsible for Salla disease; three missense mutations (H183R, P334R, G371V) and one short, in-frame deletion (SSLRN268–272del, referred to as ΔSSLRN hereafter) found in individuals with ISSD; and one missense mutation, K136E, found in a compound-heterozygote individual with a severe form of Salla disease (in addition to the R39C allele).

In preliminary experiments, we examined whether GFP-sialin L22G/L23G was still expressed at the plasma membrane after introducing these pathogenic mutations. As illustrated in Figure 5A, this held true for all mutations but G371V, which was retained in the endoplasmic reticulum (see Supplementary Figure 2) and thus was not investigated further. The remaining mutations were analysed for an effect on transport by expressing GFP-sialin L22G/L23G constructs in HEK 293 cells. The assay was performed at pH 5.0 to increase its sensitivity. As illustrated in Figure 5B, all mutations associated to ISSD (H183R, P334R, ΔSSLRN) abolished transport. In contrast, the R39C mutant showed positive, yet reduced, transport activity. This was also true for the K136E mutation, with a milder reduction. The residual activities observed in these conditions represented 17±2 and 55±6% of control (GFP-sialin L22G/L23G without pathogenic mutation; three independent experiments). Since fluorescence microscopy might have eluded quantitative variations in cell surface expression, we also performed surface biotinylation experiments and quantitative immunoblotting analysis to measure the relative levels of plasmalemmal sialin. As shown in Figure 5C, similar levels were observed for the control and mutant proteins. Values of 99.8±8.8 and 122±3.9% of control were obtained for the R39C and K136E mutants (four independent experiments). Therefore, the lack or decrease of activity in the mutants reflects an impaired ability to transport sialic acid.

To characterize further the R39C and K136E mutants, we determined their kinetic parameters (Table I). The R39C mutation did not alter the K_M, but induced a 3.4-fold reduction in V_max relative to control. K136E induced an identical (3.6-fold) reduction in V_max and it slightly increased the affinity for Neu5Ac, thus explaining the higher divergence from R39C at low substrate concentration (Figure 5B). Therefore, both mutants are mainly characterized by a slower turnover of the transport cycle.

Effect of pathogenic mutations on localization

Since the R39C and K136E mutants are partially active, we asked whether they are present on late endosomes and lysosomes in the absence of L22G/L23G mutation. We thus introduced pathogenic mutations into GFP-sialin (with an intact DRTPLL motif) and examined the localization of the protein in HeLa cells. For comparison, ISSD-associated mutations were also investigated. Results are shown in Figure 6. In most cases (R39C, ΔSSLRN, H183R, K136E), the mutant was found on LAMP1-negative puncta and at a juxtanuclear compartment resembling the Golgi, in addition to LAMP1-positive late endosomes and lysosomes. In agreement with a previous study (Aula et al, 2002), a milder mislocalization was observed for the R39C mutant. In contrast, sialin localization was not affected by the P334R mutation since, as the WT sialin P334R fully overlapped with LAMP1.

Activity and targeting of the WT protein

Our study identified a dileucine-based sorting motif, DRTPLL, which is critical to the lysosomal and late endosomal localization of sialin. When the motif is mutated, sialin is massively redirected to the plasma membrane (Figures 2 and 3) and, conversely, its introduction into an unrelated plasmalemmal protein triggered internalization and delivery to late endosomes (Supplementary Figure 1). The motif thus plays a major role in sialin trafficking. It follows the consensus sequence [DE]XXXL[LI] and thus may bind one or several of the heterotetrameric adaptor protein complexes (AP-1–AP-4) of coat sorting machineries (Bonifacino and Traub, 2003). Its position in the amino- rather than the carboxy-terminal region of the polypeptide is unusual, but not unique, as VAMP4, the insulin-regulated aminopeptidase and the invariant chain of class II MHC proteins show similar localization.

The fact that the DRTPLL motif is recognized at the plasma membrane (Supplementary Figure 1) implies that sialin can reach lysosomes and late endosomes via an indirect, endocytic route. This route may represent the main pathway followed by sialin. However, since our experiments do not exclude DRTPLL recognition at the TGN in addition to the plasma membrane, the endocytic route may just represent a salvage pathway for a direct, intracellular route. This issue and the possible involvement of other sorting signals need further investigation.

Thus far, all knowledge on lysosomal sialic acid transport resulted from biochemical studies of rat liver lysosomal membranes (Mancini et al, 1989, 1991; Havelaar et al, 1999), of a 57-kDa protein purified from these membranes (Havelaar et al, 1998) or of human lysosomal membranes purified from cultured cells (Mancini et al, 1991). Our study now provides transport data at the molecular level. The activity of recombinant sialin is highly similar to that of lysosomal membranes, since both transports are driven by a downhill H⁺ gradient (Figure 4C and D) (Mancini et al, 1989, 1991; Havelaar et al, 1998) and both recognize acidic, but not neutral, monosaccharides as well as aliphatic monocarboxylates (Table II and references therein). The inhibition of human sialin by glucuronic acid is also consistent with its accumulation in lysosomes from Salla disease and ISSD patients (Blom et al, 1990).

Our data differ from previous biochemical studies in two minor points, however. First, the Neu5Ac affinity is six-fold lower for the recombinant human protein (K_M=1.52±0.26 mM) than for rat membranes (K_M=0.24±0.07 mM) (Mancini et al, 1989). This discrepancy should not reflect divergence between mammalian species, since D-glucuronic acid had identical K_M values in rat and human membranes (Mancini et al, 1991; Havelaar et al, 1998). One possible explanation is that substrate affinity is decreased in our model because sialin is expressed at an ectopic membrane. Alternatively, the discrepancy may result from topological differences between lysosomal vesicles and our ‘inside-out lysosome' model. Indeed, if sialin interacts asymmetrically with Neu5Ac, precisely, if its affinity is better from the cytosolic than from the luminal compartment, studies on randomly oriented vesicles would preferentially detect the process with the highest affinity (cytosol-to-lumen transport), while our assay is selective for the physiological (lumen-to-cytosol) one—for an example of kinetic asymmetry, see Reig et al (2002).

Another discrepancy concerns the effect of dicarboxylates. Havelaar et al (1998) reported that the purified rat transporter is strongly cis-inhibited by succinate and maleate, yet we did not observe any inhibition (Table II). The aforementioned topological differences might again explain this discrepancy, since dicarboxylates had no effect on membrane vesicles when added to the trans-compartment (Havelaar et al, 1998). It should be noted that the biological significance of sialin interaction with aliphatic mono- or dicarboxylates is unclear, because the lysosomal lumen is not expected to generate or import these compounds.

Molecular pathogenesis of free sialic acid storage: why two diseases?

Mutations in the sialin gene result in two diseases, which share a common biochemical defect (the accumulation of free sialic acid in lysosomes) but dramatically differ in their clinical course and symptoms: ISSD is an early fatal disorder, which displays many of the characteristic features of lysosomal disorders, whereas Salla disease is essentially a neurological disorder with near-normal lifespan (Aula and Gahl, 2001). To investigate the molecular basis of these differences, we examined how pathogenic mutations affect sialic acid transport.

We studied five missense mutations and one small in-frame deletion associated either to ISSD or to Salla disease (Verheijen et al, 1999; Aula et al, 2000). One mutation associated with ISSD, G371V, caused aggregation and degradation of the protein (Supplementary Figure 2). The remaining ISSD mutations (H183R, P334R, ΔSSLRN) did not impair expression, but abolished the capacity of the protein to translocate Neu5Ac (Figure 5). Therefore, all these mutations represent loss-of-function mutations, in agreement with the presence of frameshift mutations or large deletions in other ISSD alleles.

In contrast, the mutation responsible for Salla disease, R39C, decreased but did not abolish transport (Figure 5 and Table I), showing a correlation between the molecular and clinical phenotypes. Such a correlation has not been detected thus far, since transport studies on human lysosomes failed to detect any difference between Salla disease and ISSD (Renlund et al, 1986; Tietze et al, 1989; Mancini et al, 1991). Our whole-cell assay thus appears more sensitive than measurements on membrane vesicles. It should be noted that the residual activity of sialin R39C directly explains why Neu5Ac and glucuronic acid accumulate at a lower level in Salla disease than in ISSD (Blom et al, 1990; Aula and Gahl, 2001).

Pathogenic mutations also caused intracellular mislocalization (Figure 6), in agreement with a previous study (Aula et al, 2002). We observed sialin R39C at a perinuclear compartment identified as the Golgi (C Sagné, unpublished data), as well as in LAMP1-negative puncta, which might represent sialin molecules en route to the lysosome (Aula et al (2002), who blocked protein synthesis prior to analysis, did not observe these puncta). ΔSSLRN and H183R induced similar, but apparently stronger, mislocalization. It might be argued that mislocalized sialin, whatever its activity, has some toxic effect, which contributes to the higher severity of ISSD. However, this hypothesis is at variance with our observation that P334R, which abolishes transport and causes ISSD, does not alter intracellular localization. It is important to stress that the mutant proteins, including sialin R39C, partially localized to the lysosome (Figure 6) (Aula et al, 2002). Therefore, the residual activity of the Salla disease mutant implies positive, yet reduced, lysosomal function.

To explore the relationship between the molecular and clinical phenotypes in more detail, we also analysed a mutation (K136E) found in one Salla disease patient but not, to our knowledge, in ISSD (Aula et al, 2000). Clinical studies have distinguished two forms of Salla disease, which correlate to some extent with the genotype: whereas nearly all patients with the classical disease are homozygous for R39C, the rarer cases of ‘severe' Salla disease are generally compound heterozygotes with the R39C mutation in one allele and a large deletion, a premature stop, ΔSSLRN or K136E in the other allele (Aula et al, 2000; Biancheri et al, 2002; Kleta et al, 2003). Interestingly, K136E reduced transport as much as R39C at high Neu5Ac concentration (Table I), that is in a situation of sialic acid storage, and it mislocalized sialin apparently more strongly than R39C (Figure 6). The combination of these effects might thus explain the higher severity of Salla disease in the R39C/K136E individual. This pathogenic scenario should be considered with caution, however, because there is still a wide phenotypic variation among R39C homozygotes (Varho et al, 2002).

Physiological implications

In contrast to ISSD, somatic findings such as hydrops fetalis, hepatosplenomegaly and dysostosis multiplex are absent from Salla disease. Growth retardation is limited and facial dysmorphism, when present, occurs at late stages (Aula and Gahl, 2001). The residual activity of sialin R39C thus seems sufficient for most peripheral tissues. On the other hand, the CNS is severely affected, with neurological signs such as hypotonia and ataxia within the first year, mental retardation and defective cerebral myelination (Haataja et al, 1994; Sonninen et al, 1999; Aula and Gahl, 2001). Therefore, our transport measurements suggest that sialin is rate-limiting to specific processes of the brain (it may be noted that, because V_max is decreased by R39C, sialic acid accumulation in Salla lysosomes cannot compensate for the reduced activity).

An attractive possibility for such sialin-dependent processes is the interaction of gangliosides with the sialic acid-binding lectin siglec-4, also known as myelin-associated glycoprotein (MAG), because it might explain the central dysmyelination observed in sialic acid storage disorders. Indeed, MAG is located at the periaxonal membrane of myelin (Trapp et al, 1989), it binds complex gangliosides of the nerve cell membrane (Collins et al, 1999) and mice lacking either MAG (Schachner and Bartsch, 2000) or complex gangliosides (Sheikh et al, 1999) show selective CNS dysmyelination. Since MAG has a higher affinity for complex (highly sialylated) gangliosides (Collins et al, 1999), decreased sialin activity in the lysosome—or in nonlysosomal structures present in the axon (Aula et al, 2004)—could limit the availability of sialic acid in the cytosol, induce ganglioside hyposialylation and thus impair the axo-glial MAG/ganglioside interaction (see Keppler et al (1999) for a similar mechanism caused by decreased sialic acid biosynthesis). It will thus be important to analyse the sialylation level of brain glycoconjugates when an animal model of Salla disease is available.

cDNA constructs

The IMAGE cDNA clone #3847279, which encodes a full-length human sialin, was obtained from the Deutsches Ressourcenzentrum für Genomforschung (RZPD). Its coding sequence is identical to that described by Verheijen et al (1999) (DDBJ/EMBL/GenBank accession number #AJ387747), except for a silent substitution at the wobble position of codon 82 (GCA instead of GCG). This silent substitution corresponds to a single-nucleotide polymorphism (refSNP ID #rs472294).

All cDNA modifications were performed by PCR using the primers listed in Supplementary Table I. Constructs were verified by automated sequencing over the whole coding sequence. In order to fuse the carboxy-terminus of sialin to a V5 epitope (GKPIPNPLLGLDST), the coding sequence was amplified from the IMAGE cDNA using the primers SIA-7S and SIA-8A, and subcloned at the HindIII and SacII sites of the pcDNA3.1/V5-His-TOPO vector (Invitrogen). Fusion of the amino-terminus to GFP was performed by amplification with the primers SIA-24S and SIA-29A and subcloning at the EcoRI and SacII sites of the pEGFP-C2 vector (Clontech). Eleven mutations were introduced into the sialin coding sequence as described in Supplementary data.

A truncated construct of human CD4 (amino acids 1–396, designated as CD4ΔCt) was kindly provided by B Dargent (Garrido et al, 2001). The coding sequence was amplified using the CD4-1S and CD4-2A primers and the product was cloned at the KpnI and NotI sites of pEGFP-N1 (Clontech) in replacement of the GFP sequence. To construct CD4-sialin chimeras, a CD4 product amplified using the CD4-1S and CD4-4A primers was cloned at the KpnI and BamHI sites of pEGFP-N1. The sequence for the N-terminal domain of sialin, obtained using primers SIA-26S and SIA-27A, was then introduced at the BamHI and NotI sites in replacement of EGFP.

Cell culture and transfection

HeLa and HEK 293 cells were grown under 5% CO₂ in glucose-rich, Glutamax-I-containing Dulbecco's modified Eagle medium (DMEM; Invitrogen) supplemented with 7.5% foetal bovine serum, 100 U/ml penicillin and 100 μg/ml streptomycin.

HeLa cells were transfected by electroporation using a GHT 1287 electropulsator (Jouan). Typically, 2 × 10⁶ HeLa cells in 50 μl of ice-cold phosphate-buffered saline (PBS; pH 7.4) were mixed with 5 μg plasmid, immediately subjected to nine square pulses (200 V, 3 ms) delivered at 1 Hz by 4-mm-spaced electrodes, diluted with 6 ml of culture medium and distributed in 12 wells (15-mm diameter) of a 24-well culture plate. HEK 293 cells were plated (300 000 cells/well) into poly-D-lysine-coated 24-well plates and transfected on the following day with Lipofectamine™ 2000 (Invitrogen) according to the manufacturer's protocol. HeLa and HEK293 cells were analysed 2 or 3 days after transfection.

Immunofluorescence analysis

Cells grown on glass coverslips were washed with PBS containing 100 μM MgCl₂ and 100 μM CaCl₂ (PBS/Ca/Mg), and fixed with 4% paraformaldehyde (PFA; Sigma) for 10 min. After washing and quenching PFA with 50 mM NH₄Cl for 15 min, cells were washed with PBS and permeabilized in blocking buffer (0.05% saponin/0.2% BSA in PBS/Ca/Mg) for 20 min. Coverslips were then incubated for at least 1 h with primary antibodies in blocking buffer, washed with PBS, incubated for 1 h with secondary antibodies in the same buffer and rinsed with PBS and water. Coverslips were mounted on glass slides with Mowiol. All steps were performed at room temperature. The following antibodies were used at the indicated dilutions: mouse anti-V5-tag (Serotec) 1 μg/ml, rabbit anti-V5 (Chemicon) 0.2 μg/ml, mouse anti-LAMP1 (H4A3; Developmental Studies Hybridoma Bank, University of Iowa) 0.75 μg/ml, Cy3-conjugated donkey anti-mouse (Jackson Immunoresearch) 1.4 μg/ml, Alexa Fluor 488-conjugated goat anti-rabbit (Molecular Probes) 4 μg/ml and Alexa Fluor 568-conjugated goat anti-mouse (Molecular Probes) 2 μg/ml.

Epifluorescence pictures were acquired under a × 100 objective lens with a Leica DM RXA2 microscope equipped with a CCD camera (Coolsnap). For comparing the cellular distribution of two markers, pictures were either acquired with a spectral confocal Leica TCS/SP2 microscope or epifluorescence data were deconvoluted. Sections for deconvolution microscopy were collected throughout the cell using a z-axis focus drive with a spacing of 0.2 μm. Stacked images were deconvoluted using the PSF-based Iterative 3D Deconvolution module of the Metamorph software (Universal Imaging Corporation). All images were processed using Adobe Photoshop.

Endocytosis assay

HeLa cells transfected with CD4 constructs were briefly washed with ice-cold DMEM and incubated for 30 min at 4°C with the mouse monoclonal anti-CD4 antibodies SIM.2 or SIM.4 (NIH AIDS Research and Reference Reagent Program) diluted in DMEM at 65 and 25 ng/ml, respectively. Coverslips were then washed extensively with ice-cold PBS, transferred to 37°C-prewarmed DMEM and incubated for increasing periods of time in the air/CO₂ incubator. To remove SIM.4 antibodies bound to the cell surface (acid wash), cells were incubated for 5 min at 4°C in 0.5 M NaCl and 0.2 M acetic acid (pH 2.6). Cells were fixed with PFA and processed for immunofluorescence as described above.

Sialic acid transport

HEK 293 cells were washed twice with 500 μl of uptake buffer A (5 mM D-glucose, 140 mM NaCl, 1 mM MgSO₄, 20 mM K⁺-phosphate pH 7.4) and incubated for 15 min at room temperature in 200 μl of uptake buffer A or B (identical to buffer A, but adjusted to pH 5.6) containing 0.1 μCi N-acetyl-[6-³H]neuraminic acid (20 Ci/mmol; American Radiolabeled Chemicals). The reaction was stopped by aspiration of the uptake medium, followed by two brief washes with 500 μl of ice-cold uptake buffer at the corresponding pH. Cells were lysed in 200 μl 0.1 N NaOH and the accumulated radioactivity was counted in Emulsifier-Safe cocktail (Packard) using a Tri-Carb 2100 TR liquid scintillation analyzer (Packard). Transport was studied at other pH values by replacing phosphate by 10 mM MES or MOPS. Saturation kinetics experiments performed at pH 5.0 (MES buffer) were carried out with shorter incubations (6 min) to measure transport velocities. All compounds used for inhibition studies were obtained from Sigma. Measurements were carried out in triplicate and are expressed as means±s.e.m. Each experiment was performed three or more times on independent transfections. Michaelis–Menten kinetic parameters were derived by nonlinear regression of untransformed data using the SigmaPlot 8.0 software (Systat Software, Inc.).

Western blot analysis

HEK 293 cells were chilled on ice, rinsed twice with PBS and scraped in 2 ml PBS containing protease inhibitors (5 μg/ml aprotinin, 5 μg/ml leupeptin, 5 μg/ml pepstatin and 1 mM PMSF). Cells were centrifuged at 1000 g for 10 min at 4°C and pellets were immediately frozen in liquid nitrogen and stored at −20°C. Pellets were solubilized in Laemmli's sample buffer containing Benzonase (Merck) and the equivalent of 7 × 10⁵ cells was loaded directly onto a 10% SDS–PAGE gel. The separated proteins were electrotransferred to a nitrocellulose membrane and, after blocking for 1 h in PBS containing 5% nonfat dry milk, the membrane was incubated for 1 h at room temperature with a 1:1000 dilution of mouse anti-GFP antibody (Roche Applied Science), washed three times in 0.05% Tween/PBS and incubated for 1 h at room temperature with a 1:100 000 dilution of horseradish peroxidase-conjugated antibodies against mouse whole immunoglobulins (Jackson Immunoresearch). Immune complexes were detected using the Lumi-light^PLUS Western Blotting Substrate (Roche). When specified, the membrane was stripped and re-probed with an anti-β actin monoclonal antibody (clone AC-74, Sigma).

Quantitative Western blot analysis was performed using ¹²⁵I-labelled sheep anti-mouse Ig antibody (Amersham Biosciences) as secondary antibody. The membrane was incubated for 1 h in 0.05% Tween/PBS with a 1:500 dilution (0.2 μCi/ml) of radiolabelled antibody. After washing, the membrane was exposed overnight to a Storage Phosphor Screen (Kodak). After scanning with a Phosphorimager 400E instrument (Molecular Dynamics), the signal associated to each immunoreactive band was determined using the ImageQuant software (Molecular Dynamics).

Cell surface biotinylation

At 2 days after transfection, 2 × 10⁶ HEK293 cells were washed twice with ice-cold PBS/Ca/Mg and biotinylated for 30 min at 4°C using 1 mg/ml of the cell-impermeant, cleavable reagent sulpho-NHS-SS-biotin (Pierce) in PBS/Ca/Mg. Unbound biotin was quenched for 20 min at 4°C with 100 mM glycine in PBS/Ca/Mg. After two washes, cells were lysed for 1 h in 200 μl lysis buffer (150 mM NaCl, 5 mM EDTA, 50 mM Tris–HCl (pH 7.5), 0.1% SDS, 1% Triton X-100, 1 mM PMSF, 0.1 mM leupeptin, 0.1 μM pepstatin A). The cell lysate (0.2 mg protein in 600 μl) was clarified by sedimentation at 14 000 r.p.m. for 10 min and the supernatant was incubated for 2 h at 4°C with 50 μl streptavidin–agarose beads (Sigma) under gentle agitation. The beads were sedimented at 1000 r.p.m. for 30 s. The supernatant (unbound material) was recovered and beads were washed three times with 1 ml lysis buffer, once with 500 mM NaCl and 50 mM Tris–HCl (pH 7.5), and once with 10 mM Tris/HCl (pH 7.5). The bound material was eluted in 50 μl Laemmli's sample buffer. The totality of bound proteins and an aliquot of unbound proteins or cell lysate (30 μl from the total volume of 600 μl) were resolved by SDS–PAGE and analysed by immunoblotting with anti-GFP antibodies.