Galectin-3 Guides Intracellular Trafficking of Some Human Serotransferrin Glycoforms*

Background: Transferrin has two N-glycans but their biological function remains unknown. Results: About 5% of human serotransferrin glycoforms bind galectin-3 and are targeted to a different endocytic pathway. Conclusion: Transferrin glycosylation may mediate a previously unrecognized level of physiological regulation. Significance: This is the first evidence that different molecular forms of transferrin, besides iron loading, may have different cellular function. Transferrin internalization via clathrin-mediated endocytosis and subsequent recycling after iron delivery has been extensively studied. Here we demonstrate a previously unrecognized parameter regulating this recycling, the binding of galectin-3 to particular glycoforms of transferrin. Two fractions of transferrin, separated by affinity chromatography based on their binding or not to galectin-3, are targeted to kinetically different endocytic pathways in HFL-1 cells expressing galectin-3 but not in SKBR3 cells lacking galectin-3; the SKBR3 cells, however, can acquire the ability to target these transferrin glycoforms differently after preloading with exogenously added galectin-3. In all, this study provides the first evidence of a functional role for transferrin glycans, in intracellular trafficking after uptake. Moreover, the galectin-3-bound glycoform increased in cancer, suggesting a pathophysiological regulation. These are novel aspects of transferrin cell biology, which has previously considered only a degree of iron loading, but not other forms of heterogeneity.

Human serum transferrin is a glycoprotein synthesized mainly by hepatocytes and is involved in iron transport between sites of absorption and delivery. Transferrin endocytosis is considered to be a classic example of clathrin-mediated internalization, entry into early endosomes, and subsequent recycling back to the cell membrane. During this cycle, iron is released into the endosomal compartment (1). Transferrin appears to be recycled by two main kinetically distinct pathways: one rapid recycling via early endosomes, and one slower with additional passage through late and recycling endosomes before returning to the cell membrane (2)(3)(4). The regulation of these pathways by proteins acting at the cytosolic side of the plasma membrane and vesicles has been studied extensively. However, except for iron loading (5,6) and interaction with cellular transferrin receptors (7), there has been no study on the possible regulatory effect of properties of transferrin itself, such as its different glycoforms.
Transferrin has two potential N-glycosylation sites, usually with two bi-and/or triantennary N-glycans. Transferrin glycoforms have been analyzed extensively as surrogate markers for rare genetic disorders of glycosylation, and also for alcoholinduced liver injury (8). However, the biological significance of the two glycans in respect to transferrin function is not fully understood. We recently found that a fraction (ϳ5%) of normal transferrin is bound by galectin-3 (9), a member of a protein family defined by affinity for ␤-galactoside containing glycans, such as found in transferrin. Altered glycosylations are a hallmark of oncogenic transformation (10), and have been shown to alter galectin binding of glycoproteins (11)(12)(13). Galectin-3, like other galectins, has diverse roles in cancer, inflammation, and development (14), and one cellular mechanism for this is that it regulates the intracellular traffic of certain cellular glycoproteins by binding their glycans (11,13,15,16). Therefore, we asked whether it could also direct the trafficking after endocytosis of an exogenous glycoprotein such as transferrin.

EXPERIMENTAL PROCEDURES
Expression and Purification of Galectins-Galectin-3 and galectin-3C were produced in Escherichia coli BL21 Star (Invitrogen) and purified by affinity chromatography on lactosyl-Sepharose as described earlier (17). The galectin-3 R186S mutant was previously constructed as described in Ref. 18.
Galectin Affinity Column-Galectin-3 was coupled to a 1-or 5-ml NHS-activated Hi-Trap affinity column (GE Healthcare) as described by the manufacturer, using 10 (1 ml column) or 40 mg (5 ml column) of galectin-3 in 2 or 10 ml of coupling buffer (0.2 M NaHCO 3 , 0.5 M NaCl, pH 8.3). The total amount of galectin coupled to the column was obtained by subtracting the measured amount in the wash and circulating fractions from the initial amount added. 0.1 ml of serum in 1.9 ml of PBS or human transferrin (Sigma), dissolved in PBS to a final concentration of 3-4 mg/ml (total volume 5 ml), was kept on ice in a test tube and circulated on the column for 30 min at 1 (1 ml column) or 4 ml/min (5 ml column). The column was sealed for 30 min to enhance binding. The column was washed with 5 column volumes of PBS and 1-ml fractions of the unbound fraction (flow-through) were saved. Galectin-3-binding proteins were eluted with 150 mM lactose in PBS and collected in 24 0.2-ml (1 ml column) or 1-ml (5 ml column) fractions. Protein concentrations of the collected fractions were determined with the Bio-Rad protein assay (Bio-Rad) and plotted in Prism software (GraphPad software) as a chromatogram.
Patient Samples-De-identified human serum samples were provided by Professor Mårten Fernö and Håkan Olsson, Department of Oncology, Lund University Hospital, Lund, Sweden, under ethical permit and approval from the Ethical Review Board at Lund University (Now Regional Ethical Review Board Lund). Written informed consent was obtained from all participants. All information and data were handled confidentially, and evaluation of information linked to patients was carried out in accordance with the Swedish Personal Data Act (Personuppgiftslagen in Swedish). Serum samples from 13 female metastatic breast cancer patients and 12 age-matched healthy female volunteers, listed in Table 1, were collected and stored as described previously (9,19).
Transferrin Measurements-The amount of total serum transferrin and galectin-3-bound transferrin in pooled fractions from 13 female metastatic breast cancer patients and 12 age-matched healthy female volunteers was determined using a Transferrin Human ELISA Kit (Abcam) according to manufacturer's instructions.
Statistical Analysis-All pairwise comparisons between the cancer and control sera were done by a two-tailed unpaired t test using the GraphPad Prism software.
PHA-L 3 Affinity Chromatography-400 l of agarose-bound Phaseolus vulgaris leukoagglutinin (PHA-L) (1 mg/ml), were packed into a 1-ml syringe and equilibrated with PBS. 1 mg of human transferrin in a volume of 500 l was added to the column and flow-through was collected in 200-l fractions with continuous adding of PBS up to 15 ml. Possible retained protein was eluted using 0.1 M acetate buffer, pH 3. Protein concentrations of collected fractions were determined with the Bio-Rad protein assay (Bio-Rad) and plotted in Prism software (GraphPad software) as a chromatogram. Flowthrough fractions were pooled based on retardation time and galectin-3 binding was determined by fluorescent anisotropy inhibition assay, described below.
LC-MS, Materials-POROS TM -aldehyde self-pack medium was purchased from Applied Biosystems. Rabbit anti-human transferrin used in the immunoaffinity column was purchased from Dako. The C.128 precolumn (20 ϫ 1 mm) and fittings used for immunoaffinity was purchased from Upchurch. Apotransferrin was purchased from Sigma. The C4 cartridge Widepore C4 (butyl); 4 ϫ 2.0 mm (inner diameter) were bought from Phenomenex. All other chemicals and solvents were of the highest analytical grade available from commercial sources and were used without further purification.
Immunoaffinity Column Preparation-The method and preparation has been described before (20). In short, rabbit anti-human transferrin in phosphate-buffered saline (PBS; 0.01 mol/liter of phosphate, 0.150 M NaCl, pH 7.4) was mixed with 1.5 M Na 2 SO 4 , 100 mM phosphate, pH 7.4, followed by sodium cyanoborohydride. POROS-aldehyde self-pack medium was added and mixed. After incubation at RT the immobilized ligand was washed with PBS and capped with 0.2 M Tris, pH 7.2. After washing with PBS the resin was packed in C.128 columns.
LC-MS Analysis-The analysis was performed using an API 4000 triple quadrupole mass spectrometer (AB Sciex) equipped with the TurboIon Spray ionization probe source (operated at 5500 V). Peripherals included a CTC Autosampler, and a Shimadzu System Controller (SCL-10Avp), which controlled four Shimadzu Liquid Chromatography LC-10ADvp pumps. A 10-l aliquot of sample was applied to the immunoaffinity column in PBS with 150 mM NaCl, pH 7.4, at a flow rate of 200 l/min. Transferrin isoforms were subsequently eluted from the immunoaffinity column with 100 mM glycine, 2% acetic acid buffer (pH 2.5; flow rate, 200 ml/min) and concentrated on a C4 cartridge. The C4 column was then washed with 2 mM ammonium hydroxide, 28 mM formic acid in water:acetonitrile (98:2 by volume), pH 2.9, at a flow rate of 200 ml/min. Transferrin isoforms were eluted from the C4 column and introduced into the TurboIonSpray source using 28 mM formic acid in water: acetonitrile (2:98 by volume) at a flow rate of 200 ml/min. The mass spectrometer was operated in Q1 scan mode (m/z 2000 to 3000) with a transferrin retention time of 5 min and total instrument acquisition time of 6.5 min/sample. Data were acquired and processed using Analyst software (version 1.5.2; AB Sciex). The Bayesian protein reconstruct algorithm was used to deconvolute the charge distribution raw data to reconstructed mass data.
Neuraminidase Treatment-2 mg of human transferrin (Sigma) was treated with neuraminidase from Vibrio cholerae (Roche) (25 mol/mol of transferrin) for 1 h at 37°C. Treated or untreated samples were then separated by affinity chromatography as described above.
NHS-fluorescein and NHS-sulforhodamine Labeling of Proteins-A 15-fold access of labeling dye, NHS-fluorescein or NHS-sulforhodamine (Thermo Scientific) dissolved in N,N-dimethylformamide was added to a 2 mg/ml of protein solution in 100 mM carbonate/bicarbonate buffer. The solution was mixed well and incubated for 1 h in room temperature. Labeled protein was separated from the unreacted dye by a buffer change to PBS on a PD10 column (Amersham Biosciences). Degree of labeling was calculated according to the manufacturer's instructions.
Cell Cultures-SKBR3 cells were grown in RPMI 1640 ϩ L-glutamine supplemented with 1ϫ sodium pyruvate and 1ϫ non-essential amino acids. HFL-1 and MCF-7 cells were grown in DMEM (Invitrogen) culture medium. All culture media were supplemented with 10% fetal calf serum and 1% penicillinstreptomycin (Invitrogen), Complete medium. Humidified 37°C incubators with 5% CO 2 were used for all cell incubations.
Image Acquisition-Phase-contrast or epifluorescent images were captured using a Nikon eclipse TE2000-U fluorescence microscope with a CFI Plan Apochromat ϫ100 oil objective: numeric aperture 1.40, working distance 0.13 mm, and equipped with a digital still camera DS-Qi1MC. Fluorescent images (RGB, 8 bit, 1280 ϫ 1024) were deconvolved from a three-dimensional stack (x,y,z dimensions: 129.82 ϫ 103.85 ϫ 21.6 m) with NIS-element AR software. Percent colocalization was quantitated by two different approaches. 5 images of each individual experiment were examined using NIS-element AR software colocalization application and average values of overlap coefficients according to Manders were calculated after background correction with the same threshold in each image.
In addition a minimum of 20 10-m squares were randomly placed in the cytosol of a cell. The total number of X-positive endosomes (EEA1, Rab11, and LAMP-1) and the number of structures labeled by both X and transferrin within those squares was counted. The endosomes labeled by both markers were expressed as percentage of X-positive compartments. For each experiment at least three cells were counted.
Protein Endocytosis in HFL-1 and SKBR3 Cells-HFL-1 and SKBR3 cells were grown to 50 -80% confluence in 24-well plates containing coverslips. Prior to addition of proteins cells were incubated for 30 min in serum-free medium. SKBR3 cells were in some experiments pretreated with galectin-3, galectin-3 R186S, or galectin-3C for 1 h up to 48 h. Following serumfree treatment or galectin pretreatment, cells were incubated with 1 M variously labeled proteins (NHS-fluorescein and NHS-sulforhodamine-conjugated galectin-3 binding transferrin, NHS-fluorescein and NHS-sulforhodamine-conjugated galectin-3 non-binding transferrin, NHS-fluorescein and NHSsulforhodamine-conjugated galectin-3, NHS-fluorescein-conjugated galectin-3 R186S and NHS-fluorescein-conjugated galectin-3C) in serum-free medium at 37°C. Medium containing labeled protein was replaced with serum-free medium after 5 min in HFL-1 cells to stop constant uptake, whereas it remained in SKBR3 until endocytosis was stopped. Endocytosis was stopped by removing the medium followed by addition of ice-cold PBS and fixation in 2% paraformaldehyde. Coverslips were mounted using ProLong Antifade Reagent with DAPI or Moviol and 0.01 mg/ml of Hoechst before being analyzed with a Nikon eclipse TE2000-E fluorescence microscope.
Immunocytochemistry-Fixed cells were permeabilized with 500 l of 1% Triton X-100 (Merck) in PBS for 10 min at room temperature. The cells were washed twice in ice-cold PBS before nonspecific binding was prevented by the addition of 500 l of blocking buffer (0.1% Tween ϩ 1% FBS in PBS) for 10 min at room temperature. Cells were incubated with 100 l of primary antibodies: rat anti-mouse galectin-3 (1/100), mouse antihuman transferrin receptor CD71, mouse anti-human early endosomes specific protein EEA1 (1/250) (BD Bioscience), mouse anti-human lysosome-specific protein LAMP-1 (1:250) (Santa Cruz), and mouse anti-human Rab11 (1/100) for 1 h at room temperature. After three washes with PBS for 10 min, cells were incubated with 100 l of secondary antibodies all diluted 1:320 in blocking buffer Alexa 488-and 594-conjugated anti-rat IgG (Sigma) and goat Alexa 488-and 594-conjugated anti-mouse IgG (Molecular Probes). Control staining with only the secondary antibody was performed to rule out eventual unspecific binding.
Western Blot-1 ϫ 10 6 of HFL-1 or SKBR3 cells was dissolved in lysis buffer (0.9% NaCl, 2.4% Tris, 0.08% EDTA, 0.01% (v/v) Nonidet P-40) with addition of 2 g/ml of protease inhibitors: leupeptin, aprotinin, and pepstain. After cell lysis, supernatants containing equal protein amounts were separated by denaturating SDS-PAGE on 4 -20% polyacrylamide gel following the manufacturer's instructions (Pierce). Subsequently, the proteins were transferred to a PVDF methanol-activated membrane at 100 V for 1 h. To avoid unspecific binding, the membrane was incubated with blocking buffer (5% dry milk in PBS-T (1/15 M PBS, pH 7.2, ϩ 1% Tween)) for 24 h. The membrane was then incubated in 1:1000 diluted rat anti-mouse galectin-3 or goat anti-human transferrin antibody in PBS-T for 1 h at room temperature. After washing (3 ϫ 10 min), antirat or anti-goat horseradish peroxidase (HRP)-conjugated antibody in PBS-T was added at a 1:4000 dilution for 1 h at room temperature. Proteins were detected using an ECL Plus Western blotting detection system (GE Healthcare) for 5 min, and protein bands were scanned using a GelDoc imager (Bio-Rad). Band intensities from 5 parallel experiments were analyzed with ImageJ software.
Fluorescent Anisotropy Inhibition Assay-K d of galectin-3binding and non-binding transferrin and inhibitory potency of PHA-L flow-through fractions was determined using a fluorescent anisotropy inhibition assay using the different glycoforms of transferrin to inhibit the interaction of galectin-3 with a high affinity A-tetrasaccharide probe, as described in detail in Ref.

22.
Fluorescent Intensities in Cell Medium-To analyze endocytosis kinetics of galectin-3 and transferrin, fluorescence intensities in medium removed from SKBR3 cells at different time points was analyzed using POLARstar (excitation 485 nm/emission 520 nm (green) or excitation 570 nm/emission 610 nm (red)) with FLUOstar Galaxy version 4.11-0 software (BMG Lab-Technologies).
Flow Cytometry-MCF-7 cells were grown to a density of around 1 ϫ 10 6 cells/ml and harvested using an enzyme-free cell dissociation solution (Chemicon International). Cells were incubated with 1 M NHS-sulforhodamine-conjugated galectin-3-binding transferrin and allophycocyanin (APC) conjugated anti-galectin-3 (Biolegend) for 30 min. Cells were then kept untreated or treated with 150 mM lactose or 5 M galectin-3 non-binding transferrin for 1 h followed by flow cytometry analysis using a FACS Fortessa (BD Biosciences). Data were analyzed using FACS Diva software (BD Biosciences). The cells were kept on ice throughout the experiment to prevent uptake.

Galectin-3 Binds a Small Fraction of Transferrin Enriched in
Triantennary N-Glycans-Affinity chromatography of human transferrin on immobilized galectin-3 gave a bound fraction of about 5% that could be eluted as a sharp peak using lactose (Fig.  1A), clearly distinct from the unbound fraction (ϳ95%) that flowed through the column without trailing, and gave less than 0.5% additional binding when rechromatographed (Fig. 1B, dot-ted line). The potency of galectin-3-bound transferrin as an inhibitor in a fluorescence anisotropy assay corresponded to an average K d of ϳ0.5 M, whereas the unbound fraction had no measurable affinity at all. The galectin-3-bound transferrin carried a clearly different set of N-glycans compared with the unbound transferrin, as analyzed by LC-MS of intact transferrin ( Fig. 1, C and D), and previously of glycopeptides (9). The main N-glycan of total normal transferrin and galectin-3-unbound transferrin, is by far, a biantennary N-glycan, accounting for about 90%, and most of it is not expected to bind galectin-3 as it is capped by 2-6 sialic acids. The main galectin-3-bound transferrin molecules instead carried at least one triantennary N-glycan (Fig. 1D). This may bind galectin-3 because the third antenna is often capped by 2-3 sialic acids that is tolerated by galectin-3. The mere removal of 2,6-sialic acids from the main bi-antennary N-glycan was not sufficient to induce galectin-3 binding, as only another ϳ4% of the unbound fraction bound after neuraminidase treatment (Fig. 1B, solid line). Thus, the presence of at least one tri-antennary N-glycan is a feature of the main galectin-3 binding glycoforms of transferrin. At least two types of triantennary glycans exist in human transferrin depending on the actions of either N-acetylglucoseaminyltransferase (GlcNAcT) IV or V (23). GlcNAcT-IV adds a ␤4 GlcNAc to the ␣3 mannose (left in Fig. 1E), whereas GlcNAcT-V adds a ␤6 GlcNAc to the ␣6 mannose (right in Fig.  1E). The expression of the latter is significantly increased during oncogenic transformation (24) and has been suggested to increase binding to galectin-3 (13,25). To determine which type of triantennary glycan was responsible for galectin-3 binding of transferrin we used PHA-L affinity chromatography, because glycopeptides carrying the ␣6 branch but not the ␣3 branch have been shown to be retarded on a PHA-L column (26). A small fraction of transferrin was retarded on a PHA-L column, and most of the galectin-3 binding potency co-eluted with this fraction (Fig. 1F), suggesting that galectin-3 binding of transferrin to a significant extent is mediated by tri-antennary N-glycan with the ␣6 branch. Iron loading of the unfractionated, galectin-3-unbound and galectin-3-bound transferrin was similar, measured as 60, 60, and 50%, respectively. Thus, iron loading does not appear to affect galectin-3 binding of transferrin.
Galectin-3 Binds Increased Levels of Transferrin in Sera from Metastatic Breast Cancer Patients-13 sera from women with metastatic breast cancer (median age 69 years) and 12 sera from healthy women (median age 65 years) were analyzed by affinity chromatography on immobilized galectin-3 (subjects and respective amount of galectin-3 ligands in sera are listed in Table 1). The yield of galectin-3-binding proteins from cancer sera was significantly elevated (p Ͻ 0.0001) compared with healthy sera (average 7.4 versus 4.7 mg/ml of serum) ( Fig. 2A). Galectin-3 has previously been shown to bind a wide range of serum glycoproteins (9, 11) one being transferrin. The concentrations of total and galectin-3-bound transferrin were determined by ELISA. The level of galectin-3-bound transferrin in the cancer sera was on average about twice as high compared with the average for control sera (0.11 versus 0.05 mg/ml of sera, Fig. 2B). The serum transferrin did not differ significantly due to disease (Fig. 2C). Hence, the increase was primarily due to altered glycosylation, because there was a significant increase in percentage of galectin-3-bound transferrin in the cancer sera compared with healthy sera (4.1 versus 2.3%, Fig. 2C). Moreover, the yields from the cancer patients were compared with various aspects of the available clinical data for each patient ( Table 1) including pharmacological treatments (not shown), without finding any significant correlation. There was, however, a trend (p ϭ 0.0575) toward a correlation of increase in primary tumor size with more galectin-3-bound transferrin in serum (average: 0.09 mg/ml for Ͻ25 mm versus 0.13 mg/ml for Ͼ25 mm primary tumors).
In HFL-1 cells, significant amounts of galectin-3-binding transferrin were recycled back to the cell surface after 15-20 min, whereas the non-binding transferrin was still present in intracellular vesicles (Fig. 3A). After 45 min, none of the galectin-3-binding transferrin could be detected inside the cells, whereas some of the non-binding glycoforms could be observed along the plasma membrane and after 60 min all transferrin appeared to have left the cell (not shown). In SKBR3 cells, colocalization of binding and non-binding transferrin was primarily found in intracellular structures (10 min), showing rapid inter- nalization (Fig. 3B). After 20 min all transferrin had left the membrane and was now concentrated close to the nucleus, still with no difference between the two transferrin subpopulations. After 30 min, transferrin accumulated in close proximity to the cell membrane, suggesting an onset of recycling, supposedly followed by protein secretion because no intracellular staining could be found after 60 min (not shown). Moreover, galectin-3-binding transferrin showed almost total colocalization (90%) with endogenous galectin-3 during endocytosis in HFL-1 cells, whereas non-binding transferrin only colocalized with galectin-3 to a minor degree (20%) (Fig. 3C). To examine whether galectin-3 acts as a receptor by binding to transferrin glycoforms at the cell surface, binding of galectin-3-binding transfer-rin to MCF-7 cells known to express high levels of both the transferrin receptor (CD71) and the surface galectin-3 (27,28) were analyzed using flow cytometry (Fig. 3D). The amount of galectin-3-binding transferrin at the cell surface was significantly reduced by addition of non-binding glycoforms (ϳ50%), whereas a high concentration of (150 mM) lactose, known to completely disrupt galectin-transferrin binding (Fig. 1A), was less potent (ϳ20% reduction) (Fig. 3D, top panel), suggesting that both glycoforms compete for the same receptor. Furthermore, both galectin-3-bound and -unbound transferrin showed distinct overlap (90%) with the CD71 at all time points (not shown). The degree of labeling was similar between the binding and non-binding glycoforms (3.08 versus 3.16 mol of fluoresce-

FIGURE 2. Quantitation of total proteins and galectin-3-bound transferrin in cancer and healthy sera.
A, yield of galectin-3-binding proteins (sum of protein amount in bound fractions multiplied by 10 to give mg/ml of the original serum) for sera from 13 cancer patients and 12 healthy individuals. The average for each group is marked by horizontal lines, and the difference was statistically significant (p Ͻ 0.0001). B, the concentrations of transferrin in the galectin-3-bound fractions from cancer and healthy sera as determined by ELISA. The average for each group is marked by horizontal lines, and the difference was statistically significant (p Ͻ 0.0001). C, initial serum levels of transferrin (y axis) compared with % of galectin-3 bound transferrin (x axis). The averages and ranges of each parameter are shown by crossed lines, unbroken for cancer sera and dashed for healthy sera. The difference between the % galectin-3-bound transferrin was significant (p Ͻ 0.0001), whereas the difference between total transferrin was not statistically significant.
in/mol of transferrin and 2.84 and 2.62 mol of rhodamine/mol of transferrin). Moreover, comparable results were observed when exchanging fluorophores between the transferrin glycoforms (galectin-3-binding and non-binding transferrin labeled with rhodamine or fluorescein, respectively) ruling out confounding effects of the fluorophore.
For preliminary identification of the intracellular pathways taken by the two subpopulations of transferrin (galectin-3 bound and non-bound) (green in Fig. 3E) we used markers for different endosomal compartments (red in Fig. 3E) and quantitated the overlap according to Manders (Fig. 3F). After 10 min, both showed overlap with EEA1, a marker of early endosomes where transferrin is known to go initially (6); this was more prominent for the galectin-3 non-binding (ϳ45%) compared with the binding (ϳ30%) fraction. After 20 min, overlap with EEA1 had decreased and the non-binding transferrin overlapped with LAMP-1 (ϳ50%) and the galectin-3-binding transferrin with Rab11 (ϳ60%). Rab11 is a marker and regulator of recycling endosomes (29), a compartment in the main sorting route of transferrin (6), and is also known to overlap with intracellular galectin-3 in Madin-Darby canine kidney cells (30) and SKBR3 cells (22). LAMP-1 is a marker of late endosomes, which can, albeit not typically, be visited by transferrin (4). However, we found partial colocalization (ϳ30%) between LAMP-1 and EEA1, suggesting that some of the transferrin overlap with LAMP-1 could reflect a distinct population of double labeled endosomes. In further support of these different sorting routes for the different transferrin glycoforms, treatment of HFL-1 cells with the Niemann-Pick disease mimicking drug U18666A, retained galectin-3 non-binding transferrin inside the cell, whereas galectin-3-binding transferrin continued to be recycled (not shown). Together, these observations support a model with two types of recycling routes, a faster for galectin-3-bound transferrin with a return rate to the plasma membrane of 10 -20 min and a slower, U18666A inhabitable route, with a return rate of about 30 min (31).
Galectin-3 Preloaded SKBR3 Cells Acquire the Ability of Galectin-3-regulated Transferrin Trafficking-To examine the role of cellular galectin-3 in transferrin sorting, we preloaded galectin-3-deficient SKBR3 cells with exogenously added galectin-3, by a procedure shown to restore galectin-dependent functions in other galectin-deficient cells (32). Galectin-3 was rapidly taken up by the SKBR3 cells, as expected from other cell types (33,34) (Fig. 4B), and was after 24 h found in both large and small vesicular structures in the cytosol (Fig. 4A). Galectin-3 R186S mutant, deficient in binding to LacNAc and glycoproteins, was not taken up at all (Fig. 4A, middle panel). Galectin-3C, a truncated form of galectin-3 lacking the N-terminal domain and cross-linking ability, was taken up but less efficient than wild type galectin-3, and assembled only in larger structures after 24 h (Fig. 4A, right panel).
SKBR3 cells treated with galectin-3 for 24 h gained the ability to sort galectin-3-bound and -unbound transferrin differently (Fig. 4C, top panel), as they showed very little overlap (Fig. 4D,  top panel), and galectin-3-bound transferrin was recycled more rapidly and efficiently (Fig. 4E, top panel). By contrast, in galec-  (Fig. 3B) or pretreated with galectin-3, galectin-3 R186S, or galectin-3C for 24 h followed by incubation with 1 M rhodamine-labeled galectin-3binding (red) and fluorescein-labeled non-binding (green) transferrin for 5 min at 37°C to allow internalization to occur. Endocytosis was stopped after 20 min and analyzed with deconvolution fluorescence microscopy. Yellow staining in the merged images confirms eventual overlap. Scale bar represents 10 m. Nuclei were visualized with DAPI (blue) for all panels in A and C. D, percent overlap at different time points according to Manders (average of 5 images of each individual experiment of C). E, the different galectin pretreated SKBR3 cells (as described for panel C) were incubated with 1 M rhodamine-labeled galectin-3 non-binding transferrin (red) or fluorescein-labeled galectin-3-binding transferrin (green) for 5 min at 37°C to allow internalization, and then medium was changed to prevent continuous uptake. Fluorescence intensities in removed medium were analyzed using POLARstar Fluorescence polarization microplate reader. Fluorescence intensities of rhodamine-labeled galectin-3-binding (red) or fluorescein-labeled non-binding transferrin (green) remaining in the medium after different times, as a measure of uptake by the cells (decrease) and subsequent release after recycling (increase), with the first (0 -5 min) and last parts (15-65 min) on different time scales. F, fluorescence intensities in growth medium from galectin-3-pretreated SKBR3 cells as for panel E, but supplemented with U18666A.
tin-3 R186S pretreated cells both transferrin glycoforms were endocytosed with no divergence in their sorting (ϳ80% overlap) and recycling rate (Fig. 4, C-E, middle panels). Also in galectin-3C-pretreated cells the two transferrin glycoforms mainly overlapped (Fig. 4, C-E, bottom panels), but the intracellular pattern differed (Fig. 4C, bottom panel) with accumulations of overlapping transferrin aggregates near the cell membrane as the main feature, along with smaller accumulation of non-overlapping glycoforms in the cytosol. Treatment of galectin-3-loaded SKBR3 cells with U18666A specifically blocked recycling of the non-binding glycoform (Fig. 4F), as described above for HFL-1 cells, suggesting comparable sorting machinery in these cells. Together these data show that the different intracellular targeting of the two transferrin glycoforms can be rescued by preloading the galectin-3-deficient cells with galectin-3, and that both the carbohydrate-binding activity and the N-terminal domain of galectin-3 are required for this, conceivably for cross-linking with e.g. vesicular membranes as previously proposed (35,36).

DISCUSSION
The binding of transferrin to its receptor at the cell surface does not require its N-glycans (7), but the role of N-glycans in the subsequent intracellular sorting of transferrin has not been studied before. The present results clearly demonstrate that two sets of glycoforms of transferrin can be sorted into different pathways after uptake into cells by endocytosis, and that this different sorting is dependent on galectin-3. The recycling rate (10 -20 min) and the colocalization with Rab11 of galectin-3bound transferrin most resembles the transferrin taking the route via perinuclear recycling endosomes in previous studies (2-4, 6, 31), whereas the path for the galectin-3-unbound transferrin appears slower. None of them behave as for direct fast recycling (2-4 min) from early endosomes (2)(3)(4)(5). However, the cell type and resolution of the present studies may have made this hard to see, and the pathways and rates found in one particular cell type will also depend on how much and what species of galectin-3 it expresses and perhaps other factors. For example, alterations in the sphingolipid composition of trafficking vesicles have been shown to alter transferrin sorting by redirecting it to LAMP-positive endosomes (37). In addition, it should also be noted that LAMP-1 transiently visits early endosomes after synthesis, and this may explain the unexpected distinct overlap of the unbound transferrin with LAMP-1-positive endosomes (38). Furthermore, many defining model studies have been done using human transferrin in animal model cells transfected with human transferrin receptor (2-4, 6, 31), such as CHO cells that express moderate levels of hamster galectin-3 and Madin-Darby canine kidney cells that express very high levels of dog galectin-3, but it is not known how well these galectin-3 species interact with human transferrin.
Importantly, galectin-3 endocytosis appears to be cell specific (39), consequently transferrin targeting might vary between cell type and tissue environment. Galectin-3 has recently been shown to be rapidly internalized by vascular endothelial cells (40) and macrophages (39) appearing in EEA1 positive endosomes within 10 -15 min and then partitioned into two routes, either recycling (through Rab11 positive endo-somes) or targeting to the LE/lysosomes (LAMP-1 positive endosomes). However, the dynamics of these endocytic pathways partly differed depending on cell type. Moreover, the partial targeting of galectin-3 to the LE/lysosomes may explain the minor overlap of galectin-3-bound transferrin with LAMP-1 found here. Furthermore, it has been shown that iron load can influence the kinetics of transferrin endocytosis (41). However, variance in iron loading cannot account for the different sorting of galectin-3-bound and -unbound transferrin observed here, because it was similar in the two fractions (50 and 60%, respectively). Still, because the transferrin is pooled from human blood, the small difference in iron loading may reflect a different metabolic history of the two transferrin fractions, i.e. galectin-3-binding transferrin has delivered more of its iron.
It seems most likely that both the galectin-3-binding and non-binding glycoforms of transferrin bind initially and about equally well to the transferrin receptor. The affinity of the galectin interactions is most likely too low to compete with the high affinity transferrin-CD71 interaction (42). In support of this there is a significant overlap between galectin-3-binding transferrin and the transferrin receptor, and galectin-3-non-binding transferrin competes for it (Fig. 4D). In all, it is likely that galectin-3 binding of transferrin becomes decisive after transferrin binding of its receptor and subsequent endocytosis into early endosomes. Binding of galectin-3 to the transferrin receptor itself (43), 4 might also contribute. Here galectin-3 binding of transferrin appears to divert the complex directly to Rab11positive endosomes for faster recycling kinetics. Intriguingly, Rab11 endosomes have previously been recognized as candidate compartments for galectins and glycosylated cargo proteins (44). In support of this we observe a significant overlap between galectin-3-binding transferrin and Rab11 not seen for the non-binding glycoforms.
The fact that there was an enrichment of triantennary complex N-glycans in the galectin-3-bound fraction and that neuraminidase treatment did not cause a significant increase in binding suggest that the third antenna is required for galectin-3 binding of transferrin. The increased galectin-3 binding of the PHA-L-retarded fractions imply that the mannose ␣6 branch at least to some extent is involved in this interaction. This could also explain the increased yield of galectin-3-binding glycoforms of transferrin but also total serum proteins found in sera from metastatic breast cancer patients. Amplified branching due to increased expression of GlcNAcT-V responsible for adding the ␤6 GlcNAc to the ␣6 mannose is considered a hallmark of cancer progression (45), and has been shown to facilitate tumor growth in various in vivo and in vitro models (46 -48). Moreover, Lau et al. (48) showed that the expression of GlcNAcT-V served as a regulator of the strength of the association of the EGF receptor and galectin-3 lattice and subsequently a modulator of the choice between growth and arrest. It is speculated that the preferential elongation of the ␤6 GlcNAc with poly-N-acetyllactosamine is accountable for the increased affinity, this is, however, unlikely to be essential for galectin binding of transferrin here, because of the absence of poly-N-acetyllactosamine in serum glycoproteins (11). Instead, galectin-3 binding of transferrin appears to be mediated through binding to the ␤6 GlcNAc extended with a ␤1-4-galactose, either exposed or terminated with a ␣2,3-NeuAc.
Because of their wide range of important biological activities, galectins are enigmatic in terms of identifying a central physiological function, if only one exists. The importance of galectinglycan interactions in various cellular events, especially intracellular and intercellular trafficking of proteins, has been shown by numerous studies (13,15,16,30,48). Hence, the role in intracellular guidance of glycoproteins shown here and by others emerges as a feasible explanation for these multifaceted activities. We have previously shown that the interaction of haptoglobin with galectin-1 is decisive for its intracellular trafficking (11). Moreover, this process is disturbed in cancer, because glycosylation changes associated with oncogenic transformation cause a significant increase in galectin-1-binding haptoglobin glycoforms. It should be remarked that the increase of galectin-3-binding glycoproteins and glycoforms of transferrin observed here is only based on a limited number of patients, and the clinical value of this may be of less importance. The purpose here was only to assess if there was any difference between sera from patients with severe cancer and healthy controls, and to set that finding into a biological context. Hence, even if a seemingly trivial fraction of transferrin as seen here is sorted into a different intracellular pathway after uptake this small percentage appears to go up in metastatic breast cancer. Such regulation could be critical for tumor cells, in which effective iron metabolism is crucial, because rapidly dividing cancer cells have a higher requirement for iron than their normal counterparts (49,50). Furthermore, altered cell metabolism could also influence the glycosylation machinery (15,48). The hexosamine pathway regulates the activities of the Golgi branching enzymes by production of their donor substrate UDP-GlcNAc and this metabolic control has been shown to fine-tune galectin-3 regulation of glycoprotein receptors (48). Hence it is possible that the altered cell metabolism associated with cancer accounts for the increase in galectin-3-binding glycoforms of transferrin and may serve as a regulator of transferrin trafficking. Also, other pathological conditions known to be associated with transferrin glycosylation changes, such as alcohol-induced liver injury (8) and hepatoma (51), may cause altered binding to galectin-3. Hence, under pathological circumstances, understanding the mechanisms that determine the time transferrin spends inside a cell would be of particular interest to identify design criteria for improving the drug delivery efficacy of transferrin, e.g. in anti-cancer therapy (6). Moreover, some earlier studies using transferrin as a endocytic marker will now also have to be reinterpreted in light of the fact that transferrin cannot be regarded as homogenous, and instead contains at least two different fractions with different recycling rates and paths. As a final point, the different trafficking of galectin-3-bound and -unbound transferrin points to the possibility of a previously unrecognized level of physiological regulation. Even though altered glycosylation is a feature of most cancers the functional implications of these are still enigmatic. Here we provide the first evidence that such alterations can influence the intracellular trafficking of transferrin through its binding of the endogenous lectin galectin-3.