Keywords anoikis; fibronectin receptor; galectin; glycosyltransferases; pancreas tumor

Transcription

1 Tumor suppressor p16 INK4a ) modulator of glycomic profile and galectin-1 expression to increase susceptibility to carbohydrate-dependent induction of anoikis in pancreatic carcinoma cells Sabine André 1, Hugo Sanchez-Ruderisch 2, Hiroaki Nakagawa 3, Malte Buchholz 4,Jürgen Kopitz 5, Pia Forberich 6, Wolfgang Kemmner 6, Corina Böck 1, Kisaburo Deguchi 3, Katharia M. Detjen 2, Bertram Wiedenmann 2, Magnus von Knebel Doeberitz 5, Thomas M. Gress 7, Shin-Ichiro Nishimura 3, Stefan Rosewicz 2 and Hans-Joachim Gabius 1 1 Institute of Physiological Chemistry, Faculty of Veterinary Medicine, Ludwig-Maximilians-University Munich, Germany 2 Medizinische Klinik mit Schwerpunkt Hepatologie und Gastroenterologie, Charité-Universitätsmedizin Berlin, Germany 3 Graduate School of Advanced Life Science, Frontier Research Center for Post-Genome Science and Technology, Hokkaido University, Sapporo, Japan 4 Abteilung Innere Medizin I, Universität Ulm, Germany 5 Institut für Angewandte Tumorbiologie, Klinikum der Ruprecht-Karls-Universität, Heidelberg, Germany 6 Clinic of Surgery and Surgical Oncology, Robert Roessle Hospital, Charité Campus Buch, Berlin, Germany 7 Department of Gastroenterology, Endocrinology and Metabolism, University Hospital Giessen and Marburg, Germany Keywords anoikis; fibronectin receptor; galectin; glycosyltransferases; pancreas tumor Correspondence S. André, Institute of Physiological Chemistry, Faculty of Veterinary Medicine, Ludwig-Maximilians-University Munich, Veterinärstr. 13, Munich, Germany Fax: Tel: Sabine.Andre@lmu.de (Received 21 February 2007, revised 23 March 2007, accepted 27 April 2007) doi: /j x Expression of the tumor suppressor p16 INK4a after stable transfection can restore the susceptibility of epithelial tumor cells to anoikis. This property is linked to increases in the expression and cell-surface presence of the fibronectin receptor. Considering its glycan chains as pivotal signals, we assumed an effect of p16 INK4a on glycosylation. To test this hypothesis for human Capan-1 pancreatic carcinoma cells, we combined microarray for selected glycosyltransferase genes with 2D chromatographic glycan profiling and plant lectin binding. Major differences between p16-positive and control cells were detected. They concerned expression of b1,4-galactosyltransferases (down-regulation of b1,4-galactosyltransferases-i V and up-regulation of b1,4-galactosyltransferase-iv) as well as decreased a2,3-sialylation of O-glycans and a2,6-sialylation of N-glycans. The changes are compatible with increased b 1 -integrin maturation, subunit assembly and binding activity of the a 5 b 1 -integrin. Of further functional relevance in line with our hypothesis, we revealed differential reactivity towards endogenous lectins, especially galectin-1. As a result of reduced sialylation, the cells capacity to bind galectin-1 was enhanced. In parallel, the level of transcription of the galectin-1 gene increased conspicuously in p16 INK4a -positive cells, and even figured prominently in a microarray on 1996 tumor-associated genes and in proteomic analysis. The cells therefore gain optimal responsiveness. The correlation between genetically modulated galectin-1 levels and anoikis rates in engineered transfectants inferred functional significance. To connect these findings to the fibronectin receptor, galectin-1 was shown to be co-immunoprecipitated. We conclude that p16 INK4a Abbreviations b4galt, b1,4-galactosyltransferase; GalNAc, N-acetylgalactosamine; GalNAcT, N-acetylgalactosaminyltransferase; GnT-V, N-acetylglucosaminyltransferase V; LacNAc, N-acetyllactosamine; MAA, Maackia amurensis; ODS, octadecyl silane; PA, 2-aminopyridine; pi, isoelectric point; prb, retinoblastoma tumor-suppressor gene. FEBS Journal 274 (2007) ª 2007 The Authors Journal compilation ª 2007 FEBS 3233

2 New function of p16 INK4a S. André et al. orchestrates distinct aspects of glycosylation that are relevant for integrin maturation and reactivity to an endogenous effector as well as the effector s expression. This mechanism establishes a new aspect of p16 INK4a functionality. Protein glycosylation has functional potential far beyond the structural roles exerted on the protein backbone [1 3]. Endowed with a unique high-density coding capacity, oligosaccharides are versatile biochemical signals in glycoprotein maturation and routing intracellularly, as well as in diverse cell-surface activities such as adhesion or trigger mechanisms for signaling to regulate apoptosis proliferation [4 7]. This paradigm implies far-reaching functional consequences for the carefully mapped aberrations of glycosylation upon malignant transformation [8,9]. In global terms, structural studies on N-glycans from virally transformed cells (polyoma, Rous sarcoma or hamster sarcoma virus carrying the v-ras oncogene) have traced distinct, nonrandom changes in the glycomic profile (i.e. an increased degree of branching and chain extension) [10 12]. Even more important, but still mostly descriptive, oncogene presence has been shown to affect particular components of the glycosylation machinery, as documented for H- and N-ras as well as the tyrosine kinase oncogenes src and her-2 neu [13 16]. Signaling for the measured transcriptional regulation of N-acetylglucosaminyltransferase V (GnT-V) and a2,6-sialyltransferase I (ST6Gal-I) is routed through Ras Raf Ets or Ral guanine exchange factors, respectively [15,16]. However, the conclusion to invariably link the emergence of respective features, especially the increased b1,6-branching of N-glycans, with the malignant phenotype and therefore with an unfavorable prognosis in tumor patients, is not justified. The opposite correlation was reported in tumor material from nonsmall cell lung cancer, neuroblastoma and bladder cancer [17 19]. If it were known which glycoproteins are key targets, it could become possible to attribute altered glycosylation to a distinct functionality. In this respect, studies with 3T3 fibroblasts transfected with the SV40 large T antigen gene, HD3 colon epithelial cells expressing oncogenic ras and human HT1080 fibrosarcoma cells overexpressing GnT-V (mentioned above) have illustrated target selection and, of special note, prominent appearance of the b 1 -integrin or the fibronectin receptor (a 5 b 1 -integrin) within this group [20 22]. N-Glycosylation of the fibronectin receptor can be distributed over 14 potential sites in the a 5 -subunit and over 12 sites in the b 1 -chain, covering at least 35 types of oligosaccharides when analyzed for the protein from human placenta [23]. The processing of these glycan chains is an important part of integrin maturation, and the glycosylation was shown to affect integrin association and clustering, the capacity for fibronectin or lectin binding and the interaction with the regulatory gangliosides GT 1b and GD 3 [21,24 29]. These collective insights shape the hypothesis that remodeling the glycosylation of the fibronectin receptor can act as a molecular switch. If we could select a cell system in which this integrin plays a major role for the fate of the cells, then it would be feasible to put our hypothesis to the experimental test. The recent finding that the tumor suppressor p16 INK4a restores susceptibility to anoikis induction in human Capan-1 pancreatic carcinoma cells by increasing a 5 b 1 -integrin expression and surface presentation offers such a suitable test system [30]. We thus assumed that the presence of the tumor suppressor beyond the transcriptional up-regulation of the a 5 -integrin gene [30] may engender biologically significant influences on glycan synthesis and processing. Three lines of evidence support the decision to test our hypothesis in this system. First, constitutive p16 INK4a expression in human A549 lung adenocarcinoma cells reduced global b1,4- galactosyltransferase (b4galt) activity on the cell surface by 25%, mainly as a result of reduced expression of the enzyme b4galt-i [31]. Analysis of transforming growth factor b 1 -induced rapid senescence of this cell type by northern blots revealed two- to fivefold increases in transcription for the b4galt-ii, -III, -V and -VI genes and abolishment of transcription of the b4galt-iv gene [32]. Second, comparison of DNA microarray-based expression profiles between specimens of pancreatic cancer and normal tissue revealed differential gene activities, especially up-regulation for b4galt-v (factor 9.91) and b4galt-i (factor 2.54) and an inverse regulation for GnT-IVa and b (factors 3.23 versus )20.27) [33]. Third, the glycosyltransferases b4galt-v and GnT-V were shown to be strongly expressed in a panel of eight human cancer lines [34]. Transcriptional regulation of this b4galt is under the control of the transcription factor Sp1 (which can activate genes for proteins with pro-growth survival properties) and probably Ets-1; this property is shared with GnT-V [15,35,36]. Gene 3234 FEBS Journal 274 (2007) ª 2007 The Authors Journal compilation ª 2007 FEBS

3 S. André et al. New function of p16 INK4a expression of b4galt-v and also of b4galt-i [37] can be enhanced by epidermal growth factor and dominant active ras [38]. Intriguingly, both enzymes have a bearing on a 5 b 1 -integrin features. b 1 -Integrin maturation, and transcription of the a 5 -integrin gene, were enhanced in human SHG44 glioma cells upon downregulating b4galt-v expression [39], and the absence of GnT-V in murine embryonic fibroblasts led to a protein kinase C-dependent stimulation of transcription for the two integrin genes and of clustered cell-surface presentation of the fibronectin receptor [40]. In contrast, human H7721 hepatocarcinoma cells responded to treatment with GnT-V-specific antisense cdna with attenuation of gene expression for both integrins [41]. Obviously, GnT-V-dependent effects in tumors and cells thus appear to be more difficult to predict than the consequences of b4galt-v activity. To delineate any p16 INK4a effect on glycosylation at different levels, we devised a so-far unique, three-step strategy, starting with cdna microarray analysis for glycosyltransferases. Because the levels of mrnas for these enzymes may or may not directly translate into the generation of respective oligosaccharides, we performed global product analysis using 2D chromatographic profiling. The documented evidence on N-glycan alterations was the reason to focus on this type of glycosylation. In order to find out the abundance of accessible surface glycans, we mapped the glycomic pattern of native cells using 24 plant lectins, reactive with N- and or O-glycans, and then with human lectins. The p16 INK4a -positive cells were much more reactive with galectin-1 than control cells. Picking up this trail, galectin-1 production was found to be significantly up-regulated when analyzed by two separate microarrays, proteomic profiling and flow cytofluorometry, its association with a 5 b 1 -integrin was shown and a positive correlation to anoikis rates was established. The results presented thus reveal a connection between a tumor suppressor and glycosylation, at the molecular level probably between a 5 b 1 -integrin and galectin-1. This interplay is, at least in part, responsible for restoring susceptibility to anoikis in this cell system. Results Profiling of gene expression for glycosyltransferases In the first set of experiments, we addressed the question of whether p16 INK4a expression in Capan-1 pancreatic carcinoma cells will have an influence on the expression of glycosyltransferase genes. The sensitivity of detection was refined to pick up minute signals from this class of often rather low-abundant mrna species. To avoid missing important clues, we monitored enzymes involved in N- and O-glycan as well as ganglioside biosynthesis. Material from mock-treated and p16 INK4a - positive cells was processed under identical conditions, and the ratio of measured signal intensities for cdna preparations from both types of clones was calculated for each enzyme. In addition, the average signal intensity for the p16 INK4a -positive cells served as a relative measure of the expression level. When setting a threshold for a difference in ratio of ± 0.33, a total of 17 cases could be compiled; these are detailed in Table 1. The overall mrna supply for enzymatic capacity to attach N-acetylgalactosamine (GalNAc) to serine threonine residues of a target protein by a UDP- GalNAc:polypeptide N-acetylgalactosaminyltransferase (GalNAcT) of the two cell populations was measured for the initiation step and ensuing cluster building (here especially GalNAcTs-4-7). On average, it appeared to be rather similar. The same applied to three tested cases for mucin-type O-glycan extension and its a2,6-sialylation at the proximal GalNAc moiety by ST6GalNAc-IV. However, an increased expression level was measured in the cases of mrna specific for two sialyltransferases involved in the synthesis of a-series gangliosides (Table 1). A high capacity for the synthesis of a2,3 a2,6-disialyl Le a Le c epitopes is an attribute of nonmalignant epithelial cells, reducing the presence of its sialyl Le a precursor. Although this route of ganglioside synthesis could favor renormalization, the tested gene expression profile for core mucintype O-glycosylation did not reveal any impact of p16 INK4a presence. In view of the current literature on epithelial cancer or integrin glycosylation, it is reassuring to add that a marked influence on GalNAcT activity could not really be counted upon. This situation is different for the branching of N-glycans and elaboration of their chain termini. Turning therefore next to enzymes working on complex-type N-glycan structures, no major alteration was seen in the transcription of GnT-I, -III, -IVB and -V genes. Because of its importance, the result on GnT-V was deliberately ascertained by independent PCR analysis. The same picture emerged in three other groups of glycosyltransferases: b1,3-galactosyltransferases, b1,3-n-acetylglucosaminyltransferases, except for the type II/V protein (Table 1), and most a-fucosyltransferases except for a decrease to a ratio of 0.41 (signal intensity: 1249) for enzyme VIII introducing the core-fucose unit, as independently confirmed by real-time PCR (data not shown). As outlined in the Introduction, a different situation is anticipated for b4galts, and, indeed, the constitutive presence of FEBS Journal 274 (2007) ª 2007 The Authors Journal compilation ª 2007 FEBS 3235

4 New function of p16 INK4a S. André et al. Table 1. Microarray data of mrna expression for glycosyltransferases (± 0.33 from ratio of 1). Accession Symbol Ratio p16 mock Signal p16 Enzyme functionality NM_ GALNT UDP-GalNAc:polypeptide N-acetylgalactosaminyltransferase 4 (GalNAcT-4) NM_ GALNT UDP-GalNAc:polypeptide N-acetylgalactosaminyltransferase 7 (GalNAcT-7) NM_ GALNT UDP-GalNAc:polypeptide N-acetylgalactosaminyltransferase 12 (GalNAc-T-12) NM_ ST6GALNAC GD1a synthase (ST6GalNAc-V); a-series gangliosides NM_ ST6GALNAC GD1a synthase (ST6GalNAc-VI); a-series gangliosides, a3,6-disialyl Le c and Le a in a-series gangliosides NM_ B3GNT b3-n-acetylglucosaminyltransferase 2 (b3gnt-ii); initiation and elongation of poly LacNAc NM_ B3GNT b3-n-acetylglucosaminyltransferase 5 (b3gnt-v); initiation and elongation of poly LacNAc, O-linked core 3, keratan sulfate, lactotriose NM_ B4GALT b4-galactosyltransferase 1 (b4galt-i); branch extension of N- (b1,2-branch) and O-glycans (mucin, especially core 4, O-fuc, O-man), lactosylceramide NM_ B4GALT b4-galactosyltransferase 2 (b4galt-ii); branch extension of N- and O-glycans (see b4gal-t-i) NM_ B4GALT b4-galactosyltransferase 4 (b4galt-iv); poly LacNAc extension on N-glycans (b1,6-branch) and core 2 O-glycans, neolacto series glycolipids, 6 -O-sulfated LacNAc NM_ B4GALT b4-galactosyltransferase 5 (b4galt-v); branch extension of N- (b1,6-branch) and O-glycans (mucin, O-fuc, O-man), lactosylceramide NM_ ST3GAL CMP-sialic acid:galb1,3galnac a3-sialyltransferase (ST3Gal-II) NM_ ST3GAL CMP-sialic acid:galb1,3 4GlcNAcb a3-sialyltransferase (ST3Gal-III) NM_ ST3GAL CMP-sialic acid:galb1,4glc-cer a3-sialyltransferase (ST3Gal-V); GM 3 synthase NM_ ST3GAL CMP-sialic acid:galb1,4glcnacb a3-sialyltransferase (ST3Gal-VI); synthesis of 6 -O-sulfated sle x p16 INK4a made its mark on this group. Overall, the most conspicuous changes were detected in the expression levels of three b4galt proteins, explicitly downregulation of gene expression for proteins I and V and up-regulation of protein IV. Of note, transcriptional regulation of b4galts-i -V exhibits similarities noted in the Introduction. In terms of signal intensity, b4galts-i -V were the dominant species. Set into relation with the rather low signal intensities for b1,3-galactosyltransferases, a preference for type-ii termini of N-glycans is inferred. Because the activity of b4galt- IV was increased, no drastic decrease in b1,4-galactosylation of glycan chains should occur. Owing to potent galactosylation of the O-glycan core 2 structures by b4galt-iv, this synthetic route may be favored. A similar trend with up- and down-regulation within one family was observed for N-glycan-specific a2,3-sialyltransferases (ST3Gal-III versus -VI). While the opposite direction of expression levels of these two a2,3-sialyltransferase genes for enzymes of similar substrate specificity for N-glycans may act in a compensatory manner, the reduction of gene expression of GM 3 synthase (ST3Gal-V) may bear upon the capacity for ganglioside synthesis (Table 1). Looking at O-glycan a2,3-sialylation, gene expression for ST3Gal-II is reduced in the p16 INK4a -positive cells. In contrast, no modulation is seen for a2,6-sialyltransferase at the level of mrna. As with this case, it is essential to note, in general terms, that the measured extent of gene expression should not directly be extrapolated to enzyme and then product presence in a linear manner. Of course, what matters for the cellular fate is the manifestation of a detected difference at the level of glycan production. Naturally, post-transcriptional regulation and availability of the activated substrates at the appropriate site might also have their share in shifting glycan profiles. Thus, we proceeded to the analysis of actual glycan profiles via different approaches. Based on the presented results on significant differences in the display of mrna levels of enzymes acting on N-glycans and the documented relevance of N-glycans for integrin maturation, we first focused on N-glycans to spot any major differences in their profile in situ. For this purpose, we performed 2D chromatographic mapping of neutral N-glycans after labelling with 2-aminopyridine. A major difference in the branching pattern, especially affecting b1,6-branching, will hereby be readily detectable. Also, isomers can be separated and molar ratios determined. Chromatographic profiling of N-glycans The total population of N-glycans was obtained from cellular extracts, and the reproducibility of results from seven individual cell batches of the stably transfected clones was ensured in the first set of experiments. Each 3236 FEBS Journal 274 (2007) ª 2007 The Authors Journal compilation ª 2007 FEBS

5 S. André et al. New function of p16 INK4a Fig. 1. Quantitative profiling of N-glycans. Complete representation of the N-glycan profiles of mock- and p16 INK4a -transfected pancreatic carcinoma cells and their molar ratios determined by the 2D mapping technique. The N-glycan structure is given for each case, and an inset is added for explanation. individual peak from the first column was collected separately, and all oligosaccharides with a molar ratio above 1% were further analyzed, constituting a total of 22 different types of neutral N-glycans. The elution properties of 18 N-glycan species are documented to underlie defined structures. As follows, we present the effect of p16 INK4a presence on the molar ratio and the structures of the major N-glycans. In global terms, the presence of p16 INK4a appeared to shift the balance from complex-type to oligomannosyl N-glycans, such as M6.1, M7.1 and M9.1 (Fig. 1). Bi- and tetra-antennary N-glycans surpassed the 5% threshold of molar ratio in the mock-treated cells. Invariably, the presence of b1,6-branching was reduced by p16 INK4a expression (Fig. 1). Only the biantennary complex-type N-glycan, with both a bisecting GlcNAc residue and core fucosylation, and the two type-ibranched triantennary N-glycans without core fucosylation were slightly more abundant in the p16 INK4a -expressing cells than in the control. Of note, two N-glycans of unknown structure either in the region of mannose-rich compounds or between the trimannosyl core and the biantennary structure were only found for the p16 INK4a -expressing cells, and no evidence for the emergence of an abundant N-glycan with poly N-acetyllactosamine extensions could be provided. At this stage, it should be noted that work with whole cell extracts allows us to reach a statement on total glycan presence, at the level of sensitivity of this method. Cytoplasmic N-glycans before and after maturation, as well as the cell-surface profile, irrespective of accessibility, will be simultaneously evaluated. How the pattern of glycans presented on the cell surface and accessible for binding partners looks will have to be clarified by a different method. Binding studies with glycan epitope-specific probes conducted on intact cells are suited to address this question. For this purpose, forming a panel of plant lectins is a validated approach. Although they will not monitor any glycosylation-dependent change of conformation or functional status of a glycoprotein, the glycan(s) in these cases directly acting on its (their) protein backbone, systematic application of these sensors for glycan structures is a step to define potential in situ effectors on the glycan side. In order to cover the main classes of glycan constituents we selected 24 plant agglutinins. The list of proteins and their sugar specificities are presented in Table 2. Profiling of cell-surface glycans by plant lectins In the first step of these experiments, we established the concentration and sugar dependence of lectin binding, as illustrated in the supplementary material (Fig. S1). These experiments were also instrumental in determining a common concentration to detect relative differences by comparative mapping and to avoid any toxic effects of lectins. The experimental series was systematically performed in parallel under identical conditions for the two cell populations. Hereby, any parameter change by prolonged or differential culture periods was avoided. For convenient comparison, we measured the percentage of positive cells and mean fluorescence intensity in each panel, as listed in Fig. 2. In full accordance with the results on the abundance of mrna for the GalNAcTs, the presence of GalNAc residues, measured with five different lectins, was rather similar except for VVA (Fig. 2). This lectin may react preferentially with globo- and isoglobotetraosylceramides and not mucin-type O-glycans. An apparent FEBS Journal 274 (2007) ª 2007 The Authors Journal compilation ª 2007 FEBS 3237

6 New function of p16 INK4a S. André et al. Table 2. Lectin panel for glycan profiling of cell surfaces listed in alphabetic order. Latin name (common name) Acronym Monosaccharide specificity Potent oligosaccharide a Artocarpus integrifolia (jack fruit) Jacalin (JAC) Gal GalNAc Galb3GalNAca Arachis hypogaea (peanut) PNA Gal Galb3GalNAca Canavalia ensiformis (jack bean) Con A Man Glc GlcNAcb2Mana6(GlcNAcb2Mana3)Manb4GlcNAc Datura stramonium (thorn apple) DSA GlcNAc (GlcNAc) n, Galb4GlcNAcb6(Galb4GlcNAcb2)Man (Galb4GlcNAc) 3 Dolichos biflorus (horse gram) DBA GalNAc GalNAca3GalNAca3Galb4Galb4Glc Erythrina cristagalli (coral tree) ECA Gal Galb4GlcNAcb6(Galb4GlcNAcb2)Man Galanthus nivalis (snowdrop) GNA Man Mana6(Mana3)ManaR Glycine max (soybean) SBA GalNAc GalNAca3Galb6Glc Griffonia simplicifolia I GSA I GalNAc GalNAca3Gal, GalNAca3GalNAcb3Gala4Galb4Glc Griffonia simplicifolia II GSA II GlcNAc GlcNAcb4GlcNAc, N-glycans with terminal, nonreducing-end GlcNAc Lens culinaris (lentil) LCA Man Glc N-glycan binding enhanced by core-fucosylation Lycopersicon esculentum (tomato) LEA b core and stem regions of high-mannose-type N-glycans, (GlcNAcb3Galb4GlcNAcb3Gal) n of complex-type N-glycans Maackia amurensis I (leukoagglutinin) MAA I b Neu5Aca3Galb4GlcNAc Glc Maackia amurensis II (haemagglutinin) MAA II b Neu5Aca3Galb3(a6Neu5Ac)GalNAc Phaseolus vulgaris erythroagglutinin (kidney bean) PHA-E b Bisected complex-type N-glycans: Galb4GlcNAcb2Mana6 (GlcNAcb2-Mana3)(GlcNAcb4)Manb4GlcNAc Phaseolus vulgaris leukoagglutinin PHA-L b Tetra- and triantennary N-glycans with b6-branching (kidney bean) Pisum sativum (garden pea) PSA Man Glc N-glycan binding enhanced by core-fucosylation Sambucus nigra (elderberry) SNA Gal GalNAc Neu5Aca6Gal GalNAc Solanum tuberosum (potato) STA b (GlcNAc) n with preference for high-mannose-type N-glycans Sophora japonica (pagoda tree) SJA GalNAc GalNAcb6Gal, Galb3GalNAc Triticum vulgare (wheat germ) WGA GlcNAc Neu5Ac (GlcNAc) n, Galb4GlcNAcb6Gal Ulex europaeus I (gorse) UEA I Fuc Fuca2Galb4GlcNAcb6R Vicia villosa (hairy vetch) VVA GalNAc GalNAca3(6)Gal, GalNAcb3Gal Viscum album (mistletoe) VAA Gal Galb2(3)Gal, Gala3(4)Gal, Galb3(4)GlcNAc without with a2,6-sialylation, Fuca2Gal a Based on previously compiled information [121], extended and modified; b no monosaccharide known as ligand. preference for the Thomsen-Friedenreich epitope antigen was detected for p16 INK4a -positive cells by PNA and jacalin. This result may reflect either elevated core 1 synthesis or efficient core 1 masking by a2,3-sialylation in the mock control, as suggested by the microarray data. It was therefore essential to study this issue in greater detail (please see the paragraph below). Using the standard concentration Maackia amurensis-ii (MAA-II), the percentage of positive cells was enhanced for mock-treated clones, possibly attributable to different gene expression levels for the enzymes responsible for the two sialylation steps. N-Glycosylation, especially with core fucosylation, appeared to be accessible on the cell surface to a higher extent in the p16 INK4a - versus mock-transfected cell populations. Glycan profiling had revealed an increase for biantennary glycans of this type containing the additional bisecting GlcNAc unit, in contrast to an otherwise decreased level of core fucosylation (Fig. 1), in full accord with array data of a-fucosyltransferase VIII. Testing two probes with similar specificities (i.e. LCA PSA) served as an internal control. Apparently, the chromatographic profiling, on average, detected more substituted N-glycan in mock controls Fig. 2. Profiling of cell-surface glycans by plant lectins. Semilogarithmic representation of fluorescent surface staining by biotinylated plant lectins (for an explanation of the acronyms and listing of oligosaccharide specificity, please see Table 2) of mock-transfected (gray line) and p16 INK4a -transfected (black line) Capan-1 pancreatic carcinoma cells determined in parallel assays. Quantitative data on the percentage of positive cells and fluorescence intensity are given in each panel (first line: mock-treated cells; second line: p16 INK4a -transfected cells). The concentration of the biotinylated lectins was 0.5 lgæml )1 except for SNA and DBA (1 lgæml )1 ), SJA and WGA (2 lgæml )1 ) and MAA-I (5 lgæml )1 ) FEBS Journal 274 (2007) ª 2007 The Authors Journal compilation ª 2007 FEBS

7 S. André et al. New function of p16 INK4a FEBS Journal 274 (2007) ª 2007 The Authors Journal compilation ª 2007 FEBS 3239

8 New function of p16 INK4a S. André et al. when compared with the relatively increased lectin reactivity on the cell surface for p16 INK4a -expressing cells, indicating disparities in the levels of accessibility and ligand preferences. The chromatographic profiling and cell-surface detection of N-glycans with bisecting GlcNAc by PHA-E could easily be reconciled, product formation substantiating efficiency of only minute quantities of detectable mrna for GnT-III, whereas no major accessibility difference could be discerned regarding lectin binding of the b1,6-branch by PHA-L (Fig. 2). Due to the potential interference by a2,6-sialylation, this result, as noted for the Thomsen-Friedenreich antigen epitope and a2,3-sialylation above, had to be further scrutinized (please see below for the effect of sialidase treatment). A clear difference in lectin binding concerned the presence of accessible GlcNAc moieties measured by applying DSA WGA but not seen to this extent with GSA-II and STA. Expression changes in the b4galt family may underlie this staining property. LEA, a marker for extensions of N-glycan branches by N-acetyllactosamine (LacNAc) units, failed to provide clear evidence for marked cell-surface differences, arguing in favour of the concept of compensatory change within the tested b1,3-n-acetylglucosaminyltransferases and b4galts, as also seen in chromatographic profiling. Because similar cell staining was measured with GNA, the contribution of the dual reactivity of LEA to high-mannose-type N-glycans will probably have no influence on this result. Similarly, the abundance of accessible poly N-acetyllactosamine chains was rather similar, prompting a final check of chain-end galactosylation. The two respective probes (VAA and ECA) revealed more intense staining of the p16 INK4a -reconstituted clones than of the mock control (Fig. 2; supplementary Fig. S1). This result does not simply reflect the microarray data when adding up signal intensities for b4galt-specific cdnas. As depicted above, sialylation will make its presence felt in this approach. Because it can mask terminal galactose residues for lectins, measurement of its status was essential. a2,3-sialylation of N-glycans monitored comparatively with MAA-I at a fairly high concentration showed a slight preference for the p16 INK4a -positive clone, indicating a compensatory balance for ST3Gal-III -IV -VI gene expression levels. By contrast, cell positivity expressed as cell percentage was lowered in these cells when measuring lectin-reactive N-glycan-specific a2,6-sialylation, despite similar extents of ST6Gal-I gene expression (Fig. 2). Ectopic ST6Gal-I expression in the p16 INK4a -positive cells did not change this parameter (data not shown). As mentioned above, this observation on the nonidentical degree of sialylation makes it mandatory to determine comparatively the impact of sialylation on the binding of plant lectins sensitive to its presence. In addition to the inhibition control with haptenic sugar, the pre-exposure of cells to neuraminidase is, at the same time, a second control for ensuring carbohydrate-dependent binding of SNA. Standard conditions for enzymatic treatment were established, minimizing the influence on the cell phenotype. In line with the inhibition studies, enzymatic pretreatment significantly reduced (mock control) or almost completely abolished (p16 INK4a -positive cells) SNA binding (Fig. 3). The same effect was observed for MAA-II used at a nearly saturating concentration of 5 lgæml )1 (Fig. 3). In addition to its control character, these results support the evidence for a quantitative difference in the a2,6-sialylation status between the two cell populations. Should the level of mucin-type O-glycan a2,3-sialylation also be lowered in the p16 INK4a -positive cells, as suggested by the microarray data, then neuraminidase activity should enhance PNA staining of the control cells, with a minor influence on the p16 INK4a -expressing cells and on DBA staining. The importance of this aspect has been pointed out above. Fittingly, PNA, but not DBA, positivity was markedly improved by the enzymatic removal of sialic acid residues from the cell surface for mock-transfected cells but not for the p16 INK4a -positive cells (Fig. 3). The minor effect on the p16 INK4a -positive cells probably indicates the presence of mucin-type core 2 tetrasaccharides, favored by an increased b4galt-iv presence and reduced O-glycan a2,3-sialylation. Thus, lectinaccessible mucin-type O-glycosylation appeared to be rather equally abundant, but the levels of its a2,3-sialylation were definitely different. The apparent preference for mock-treated cells to carry O-glycan core 1 a2,3-sialylation can be accounted for at least in part by the microarray data. Because the presence of a2,6-linked sialic acids can impede PHA-L binding, the same procedure was also carried out in this case. Using a lectin concentration of 1 lgæml )1, increased staining was seen in both cell populations, and accessibility was still at an increased level for the p16 INK4a -positive cells (Fig. 3). To avoid underestimation of the presence of b1,6-branched N-glycans in the mock control, the remarkably different levels of SNA binding after standard neuraminidase treatment must be recalled. In essence, data from chromatographic mapping square well with the lectin profiling. A major result emerging from these experiments is that the extents of sialylation of N- and mucin-type O-glycans in the two cell populations (a2,6-substitution for N-glycans, a2,3-modification for O-glycans) are different. In functional terms, these 3240 FEBS Journal 274 (2007) ª 2007 The Authors Journal compilation ª 2007 FEBS

9 S. André et al. New function of p16 INK4a Fig. 3. Effect of sialidase treatment on the cell-surface binding of plant lectins. Semilogarithmic representation of fluorescent surface staining of mock-transfected (Mock) and p16 INK4a -transfected (p16) Capan-1 pancreatic carcinoma cells without the incubation step using the biotinylated lectin (shaded) and after incubation with the labeled probe (0.5 lgæml )1 of PNA, 1 lgæml )1 of PHA-L, 5 lgæml )1 of SNA and MAA-II, as well as 10 lgæml )1 of DBA), without (gray line) or after (black line) sialidase treatment. Quantitative data are presented as defined in the legend to Fig. 2. processes may generate or mask sites for contact in situ with endogenous lectins. As the results with the b-galactoside-specific lectins VAA ECA revealed, the levels of accessible galactose residues were remarkably different between the two cell populations. Monitoring of cell-surface binding by plant lectins thus pinpointed disparities in accessible glycans. These observations directed our interest to the detection of endogenous lectins. They might turn these newly defined properties on the level of cell-surface glycosylation into effects, in this case on the level of susceptibility to anoikis. With focus on sialylation galactosylation, the main groups of human lectins that can read and translate such differences are the C-type lectins, siglecs and galectins [42]. The ensuing microarray monitoring of expression of 42 C-type lectins and siglecs-2, -3, -5, -6, -7 and -9 failed to provide positive data or, if positive, a difference in signal intensities. When testing the third mentioned lectin family, indications for transcription of galectin genes were collected. As further ascertained by systematic RT-PCR analysis within this lectin family, transcription of genes for galectins-1, -3, -7 and -9, respectively, was detected, the most pronounced signal (i.e. 5369) seen in the case of galectin-1 in p16 INK4a - positive cells (data not shown). Having herewith provided evidence for gene expression of members of this galactoside-specific lectin family, we next probed whether human adhesion growth-regulatory galectins can bind to the cells, as shown for the galactosidespecific lectins ECA VAA. By testing more proteins than just galectin-1, the individual binding properties of the structurally closely related members of this family can also be profiled in one cell system, a comparison so far not reported. For this purpose, we purified the lectins and then biotinylated them under activitypreserving conditions, ascertained a lack of harmful effects on sugar binding by activity assays and determined the labeling efficiency with 2 8 modified residues per carbohydrate-recognition domain. As in the binding studies with plant lectins, we routinely performed experiments to assess concentration dependence and inhibition of galectin binding by haptenic sugar. Profiling of galectin binding The measurements with the two galactoside-specific plant lectins and also with MAA-I (galectins tolerate a2,3- but not terminal a2,6-sialylation) led to the expectation that these human lectins may preferentially bind to p16 INK4a -expressing cells with their increased presence of these epitopes. Indeed, the respective studies confirmed this notion already in the first set of experiments with galectin-1, when documenting the dependence of cell staining on lectin concentration and on glycan binding (supplementary Fig. S2, first and second panels). This homodimeric family member is a potent cross-linker for glycans on the cell surface. FEBS Journal 274 (2007) ª 2007 The Authors Journal compilation ª 2007 FEBS 3241

10 New function of p16 INK4a S. André et al. Besides using haptenic sugar to relate lectin activity to binding, we tested two mutants of human galectin-1. Their carbohydrate-binding activity was impaired by a crucial substitution (W68L, E71Q) in the carbohydrate-recognition domain. The loss of binding to the p16 INK4a -transfected cells compared with the His-tagged wild-type control protein served as an independent validation of the inhibition control (supplementary Fig. S2, third panel). The concentrationand carbohydrate-dependent binding is also illustrated for the chimera-type galectin-3. In this case, it is obvious that the cells of the mock control are also rather reactive, when considering cell-percentage positivity (supplementary Fig. S2). Given this indication for intergalectin differences, we systematically assayed a series of human galectins to define staining properties with these human effector proteins. As shown in the supplementary material (Fig. S3), there is a clear trend for galectin reactivity correlating with tumor suppressor presence. The tandem-repeat-type galectin-8 reacted similarly to galectin-9 (data not shown). The most prominent change for the combination of both quantitative cell staining parameters was seen in the case of galectin-1. Glycan-dependent galectin binding to these cells is thus detectable, it is not a uniform characteristic, and, finally and even more importantly, the conspicuous difference of galectin-1 binding to the two cell populations gives further study a clear direction. As a result of the blocking effect of terminal a2,6-sialylation on galectin-1 binding, we assumed that this type of sialylation will mask galectin-1-reactive sites on the cells of the mock control. If therefore exposed to a sialidase, these cells should become reactive, as shown for SNA or PNA binding in Fig. 3. Indeed, reduction of sialylation under standard conditions increased cell binding markedly for the mock control, whereas the p16 INK4a -transfected cells showed only slightly improved binding properties (Fig. 4). Galectin-1 specificity renders it very likely that removing the blocking a2,6-sialylation underlies this parameter change. That the same reactivity pattern was seen for galectin-3 constitutes not only an inherent control. As galectin-3 tolerates a2,6-sialylation in poly N-acetyllactosamine chains already at the level of the dimer, in stark contrast to galectin-1 [43], these results signified no notable difference for the presence of such chain extensions between the cell types, in full agreement with LEA and microassay data. In view of an effector function, the differential status of sialylation thus appeared to influence the accessibility to ligand sites effectively. This result might become functionally relevant if a functional relationship between galectin-1 Fig. 4. Effect of sialidase treatment on the cell-surface binding of human galectins. Semilogarithmic representation of fluorescent staining of mock-transfected (Mock) and p16 INK4a -transfected (p16) Capan-1 pancreatic carcinoma cells without the incubation step using the biotinylated lectin (shaded) and after incubation with labeled galectins-1 and -3 (gal-1 and gal-3 used at 10 lgæml )1 ) without (gray line) or after (black line) sialidase treatment. Quantitative data are presented as defined in the legend to Fig. 2. and the expression of the p16 INK4a protein could be delineated. In this sense, the p16 INK4a -dependent increase of the presentation of binding sites by reduced a2,6-sialylation might even be associated with enhanced galectin-1 expression, this regulatory event accomplishing optimal sensitivity. We put this reasoning to the test in a stepwise manner by a gene array, by northern blotting nuclear run-off experiments, by proteomic profiling and by flow cytofluorometry. Identification of up-regulation of galectin-1 expression Quantitative determination of the presence of galectin- 1-specific mrna indicated a p16 INK4a -associated increase. In detail, we comparatively probed 1996 cdnas in an array designed for pancreas tissue and its cancer development and applied stringent criteria for defining up-regulation. The threshold of 50% increase was surpassed by 16 signals. Galectin-1 gene transcription was the most prominent defined case at a p16 INK4a mock ratio of 3.01 (supplementary 3242 FEBS Journal 274 (2007) ª 2007 The Authors Journal compilation ª 2007 FEBS

11 S. André et al. New function of p16 INK4a Table S1). Northern blotting and nuclear run-off experiments confirmed the array data and substantiated an increase in de novo transcription (data not shown). Extending our work from the mrna level, we next performed a proteomic analysis with the intention of establishing whether the galectin-1 protein is produced at an amount reflecting gene expression. With a total of spots per gel and a reproducibility of spot assignment between different gels of %, we detected one spot among the 48 signals that showed a consistent increase in staining by 50% in p16 INK4a -positive cells, relative to control cells, with mass isoelectric point (pi) characteristics compatible with galectin-1. As shown in Fig. 5, we confirmed this hypothesis by western blotting and mass-spectrometric fingerprinting. On the basis of staining of protein by a dye or the western blotting procedure, galectin-1 presence as protein was found to be significantly up-regulated (P ¼ , P ¼ ). The advantage of the proteomic profiling compared with western blotting after 1D electrophoresis, shown separately in Fig. 6A as a control, is the exclusion of formation of any galectin-1 variants based on different pi values. After synthesis, the protein underwent secretion, because we detected its presence in the medium (data not shown). As a consequence it may then associate with the cell surface, prompting flow-cytofluorometric analysis. Applying the antigalectin-1-specific immunoglobulin for cell-surface detection, the difference in protein production translated into increased surface presence in the p16 INK4a -positive cells (Fig. 5). We deliberately tested several cell batches and consistently measured an enhanced cell-surface presentation in the p16 INK4a -positive cells under standard culture conditions. When determining the level of inhibition of galectin-1 binding by the haptenic sugar, lactose, a notable difference became apparent. It was comparatively lower in this cell type than in the control cells, a measure for strong affinity of the endogenous lectin to a set of particular surface glycans. In fact, endogenous galectin-1 could hardly be stripped off the cell surface, even in the presence of 200 mm lactose. In comparison, lectin binding from the medium as source was much more sensitive to inhibitor presence, as observed from loading the cells with galectin-1 up to saturation (supplementary Fig. S2), intimating visualization of the gradient of decreasing affinity for binding multivalent ligands seen in a recent model study [44]. To exclude that the p16 INK4a -dependent up-regulation of galectin-1 is a singular event confined to Capan-1 cells only, we tested Dan-G pancreatic cancer cells without and with p16 INK4a expression. Of note, these two cell lines give insight into the specificity of the effect as a result of their differences in the status of the retinoblastoma tumor-suppressor gene, prb. Increased galectin-1 expression was determined by western blotting in both clones of engineered transfectants with p16 INK4a expression, despite maintained prb status (data not shown). It is thus tempting to propose a functional correlation between the detected galectin-1 up-regulation, the increased presentation of galectin-1-binding sites in p16 INK4a -expressing cells and acquisition of anoikis susceptibility associated with the fibronectin receptor. Induction of anoikis by galectin-1 In order to test the hypothesis described above, we pursued two independent approaches. First, we established stable clones with reduced galectin-1 production by transfection with a vector harboring full-length cdna for galectin-1 in the antisense orientation. After confirming the reduction of galectin-1 presence, to a level characteristic of wild-type cells, by western blotting, the cells of such a clone were subjected to monitoring levels of anoikis. In line with our hypothesis, the extent of anoikis was correlated to the level of galectin-1 present (Fig. 6A). Second, we forced an increase of galectin-1 production on wild-type cells weakly positive for galectin-1 binding, using proliferating cell nuclear antigen as an internal control. Anoikis induction was enhanced, even showing a trend for dose dependence among the tested clones (Fig. 6B). Corroborating this biological effect on wild-type cells artificially overexpressing galectin-1 we could, in parallel, elicit anoikis in regular wild-type cells by adding the lectin to medium at a concentration of 125 lgæml )1 (data not shown). The presence of haptenic sugar interfered with this process, revealing carbohydrate dependence, as did the presence of the predominantly monomeric galectin- 3, revealing a requirement for cross-linking (data not shown). Finally, to connect galectin-1 with the p16 INK4a -associated increase of cell-surface presence of the fibronectin receptor, we reasoned that preparations of the integrin, when immunoprecipitated from p16 INK4a -positive cells, should contain galectin-1. Using cells grown adherent or in suspension, we tested this assumption. Western blotting revealed that galectin-1 was indeed co-immunoprecipitated with this glycoprotein, the levels of galectin-1 presence being consistently higher in p16 INK4a -positive cells than in control cells (Fig. 6C). Independently, the antibody against the a 5 -subunit was effective at markedly reducing the extent of binding of labeled galectin-1 to the cell surface (data not shown). These results imply that the a 5 b 1 -integrin is a binding partner of anoikis-inducing galectin-1. FEBS Journal 274 (2007) ª 2007 The Authors Journal compilation ª 2007 FEBS 3243

12 New function of p16 INK4a S. André et al. Fig. 5. Enhanced galectin-1 expression in p16 INK4a -reconstituted cells. Aliquots of total protein (200 lg) from mock-transfected (Mock) and p16 INK4a -transfected (p16) Capan-1 pancreatic carcinoma cells were subjected to 2D gel electrophoretic separation and silver staining. The kda-section where galectin-1 presence can be expected was marked (A, B) and the putative position of galectin-1 was labeled. Western blot (WB) analysis with equal quantities of protein and 1 lgæml )1 of galectin-1 antibody as probe revealed spots at this position after antigen visualization by enhanced chemiluminescence (top panel). Mass-spectrometric fingerprinting after digestion of the protein of this spot by trypsin ascertained identity to galectin-1, and quantification of the staining intensity in gel electrophoretic (2-DE) and WB analyses revealed statistically significant up-regulation (middle panel). Cell-surface detection of galectin-1 in flow-cytofluorometric analysis was performed using 20 lgæml )1 of polyclonal galectin-1 antibody as probe and fluorescent goat anti-rabbit IgG as the second-step reagent (bottom panel) FEBS Journal 274 (2007) ª 2007 The Authors Journal compilation ª 2007 FEBS

13 S. André et al. New function of p16 INK4a A B C Fig. 6. Role of galectin-1 in p16 INK4a -mediated anoikis induction. Quantification of p16 INK4a and galectin-1 presence in wild-type (wt), p16 INK4a -positive (p16) and p16 INK4a antisense galectin-1 (gal-1as) double-transfected cells by western blot analysis and determination of anoikis rates of cells after 24 h in suspension in at least three independent experiments (***, P < 0.01 for p16 INK4a cells versus wt; #, P < 0.05 for double transfectants versus p16 INK4a cells; the study panel includes a mock control for second transfection) (A). Quantification of galectin-1 presence in wt, mock-treated and cells transfected with vector carrying galectin-1-specific cdna showing different levels of galectin-1 positivity and determination of anoikis rates as given in panel A (***, P < 0.01, *, P < 0.05 for galectin-1 transfectants versus wt cells; sensitivity of detection was less than in panel A to avoid overexposure of the lane for clone gal-1 1) (B). Western blot detection of galectin-1 in preparations of immunoprecipitated fibronectin receptor obtained from cells grown while adherent (ad) or on polyhydroxyethylmethacrylate (PH) to keep them in suspension (C). Discussion Protein glycosylation affords a broad platform for highly versatile modulation of diverse functional aspects. The presence of distinct glycan determinants in glycoproteins can underlie quantitative aspects of intracellular routing and transport to the cell surface as well as the regulation of activities of both protein and glycan parts at the final destination. Deduced from the fundamental concept of the sugar code, cells derive much of their remarkable communication skills from presenting an array of glycan signals [3,5,6, 45,46]. In this sense, glycan remodeling is gaining a functional dimension in our understanding, and even the introduction of at first sight minor substitutions, such as a bisecting GlcNAc or core fucose residues, has remarkable consequences for the shape and ligand activity of the N-glycan [47 49]. It is thus intuitively attractive to assume that changes in the glycomic profile are nonrandom reprogramming events, acting on the protein (e.g. conformation, protection from proteolysis, aggregate formation or ligand binding) and on the glycan (e.g. conformation and affinity to lectins). Using the tumor suppressor, p16 INK4a, the fibronectin receptor and the Capan-1 cell line as study objects, we herein have delineated a new route towards re-establishing susceptibility for anoikis. The presented results lend credit to a scenario of finetuned and co-ordinated events translating into alterations of protein protein and protein carbohydrate recognition. Towards this aim, we had designed a combined approach using a cdna microarray for glycosyltransferases, 2D chromatographic profiling and binding studies with lectins on the cell surface. The array data pinpoint several changes in expression of glycosyltransferase genes and hereby afforded first evidence for functional links. To start with there is an increased potential for the synthesis of a2,3 a2,6- disialylated Le a Le c -epitopes on gangliosides, an attribute of nonmalignant epithelial cells and ligand availability for siglec-7 [50,51]. Automatically, the presence of a synthetic precursor (i.e. the sialyl Le a epitope, which is present in wild-type Capan-1 cells and can act as mediator of tumor angiogenesis and metastasis) will be diminished [51,52]. MAA-II staining at a probe concentration of 5 lgæml )1 can be interpreted to reflect the differential presence of the respective disialylated core. Next, the previously reported influence on b4galt activity [31] was confirmed at the level of transcription and extended to divergent regulation between b4galts-i -V and b4galt-iv. Of interest, the noted down-regulation of FEBS Journal 274 (2007) ª 2007 The Authors Journal compilation ª 2007 FEBS 3245