Evaluation of Riproximin Binding Properties Reveals a Novel Mechanism for Cellular Targeting

Background: Riproximin is a cytotoxic lectin from Ximenia americana showing tumor selectivity.

Results: Riproximin selectively binds to two types of glycoconjugates present on glycoproteins, cross-linking them by its two binding sites.

Conclusion: The biologic activity of riproximin is determined by specific and dynamic interactions with multivalent, cancer-related glycan targets.

Significance: The selectivity of riproximin for cancer cells relies on its unique targeting mechanism.

Abstract

Riproximin is a cytotoxic type II ribosome-inactivating protein showing high selectivity for tumor cell lines. Its binding to cell surface glycans is crucial for subsequent internalization and cytotoxicity. In this paper, we describe a unique mechanism of interaction and discuss its implications for the cellular targeting and cytotoxicity of riproximin. On a carbohydrate microarray, riproximin specifically bound to two types of asialo-glycans, namely to bi- and triantennary complex N-glycan structures (NA2/NA3) and to repetitive N-acetyl-d-galactosamine (GalNAc), the so-called clustered Tn antigen, a cancer-specific O-glycan on mucins. Two glycoproteins showing high riproximin binding, the NA3-presenting asialofetuin and the clustered Tn-rich asialo-bovine submaxillary mucin, were subsequently chosen as model glycoproteins to mimic the binding interactions of riproximin with the two types of glycans. ELISA analyses were used to relate the two binding specificities of riproximin to its two sugar binding sites. The ability of riproximin to cross-link the two model proteins revealed that binding of the two types of glycoconjugates occurs within different binding sites. The biological implications of these binding properties were analyzed in cellular assays. The cytotoxicity of riproximin was found to depend on its specific and concomitant interaction with the two glycoconjugates as well as on dynamic avidity effects typical for lectins binding to multivalent glycoproteins. The presence of definite, cancer-related structures on the cells to be targeted determines the therapeutic potency of riproximin. Due to its cross-linking ability, riproximin is expected to show a high degree of specificity for cells exposing both NA2/NA3 and clustered Tn structures.

Introduction

Riproximin is a lectin with potent antineoplastic activity in vitro and in vivo (1, 2). It was originally identified as the active component of a powdered plant material used in African traditional medicine for treating cancer and has subsequently been isolated from the fruit kernels of Ximenia americana. Riproximin is selectively cytotoxic to cancer cell lines with IC₅₀ in the nanomolar range. For example, the breast cancer MCF7 cells showed >500-fold higher sensitivity than the non-tumorigenic breast epithelium MCF10A cells (3). This remarkable potency and selectivity prompted us to examine the molecular mechanisms of cell targeting in more detail.

Riproximin belongs to type II of the cytotoxic ribosome-inactivating proteins (RIPs)³ (4). Several members of this family were investigated as potential antineoplastic agents, most prominently ricin. The mechanism of action of the RIP family of proteins involves two key steps represented by the A- and B-chains of the protein. First, the B-chain, a lectin, binds to cell surface glycans, resulting in internalization of the RIP. Within the cell, the A-chain, an rRNA N-glycosidase, depurinates the 28S RNA, leading to transcriptional arrest and eventually cell death (5). Recently, it was shown that the mode of action also involves the induction of the unfolded protein response (6).

The cellular patterns of cytotoxicity and in vivo toxicity of a particular RIP are primarily determined by its binding and internalization efficiency (7). Several types of biomolecules, such as glycoproteins, glycosphingolipids, proteoglycans, or glycosylphosphatidylinositol-linked proteins, contribute to a cell surface glycosylation pattern (8, 9) and are potential riproximin targets.

Investigations of the glycan structures from malignant tissues showed that the tumor-associated glycosylation significantly differs from that of normal tissues. Mucins showing aberrant O-glycosylation are a typical feature of epithelia-derived cancer (8). N-Glycans are significantly altered in cancer too (10). The presence of immature (11) or highly branched core-fucosylated structures (12–15) has been related to various types of tumors.

The first attempts to use ricin as a cancer therapeutic were based on its higher cytotoxicity in transformed cells (16), a mechanism that today would be called “targeted toward cancer-related glycostructures.” Its development, however, failed due to unexpectedly high toxicity (17). Riproximin showed potent antineoplastic activity that might be derived from its glycan binding profile. The aim of this study was to identify the glycans that function as binding receptors for riproximin and are thus responsible for its tumor-specific cytotoxicity.

Using a glycan microarray, riproximin was found to specifically bind to two types of glycans, the N-glycan structures NA2/NA3 and the Tn antigen, a prominent cancer-related O-glycan. The mechanism of binding of lectins to multivalent globular and linear glycoproteins was recently elucidated (18). Both the interaction of Gal-binding galectins with multiple NA3 structures on ASF (19) and that of GalNAc-binding soybean agglutinin with Tn on Tn-rich porcine submaxillary mucin (20) were described as depending on bind-and-jump and negative cooperativity effects. Using similar model glycoproteins, the mechanism of riproximin interaction with its glycotargets and implications for its cytotoxicity were investigated.

EXPERIMENTAL PROCEDURES

Riproximin Purification and Labeling

Riproximin was purified from Ximenia americana fruit kernels as described before (3). In short, the purification procedure included an initial aqueous extraction of proteins from the crude kernel material, removal of lipids with chloroform, and subsequent chromatographic purification on a strong anion exchange resin and lactosyl-Sepharose.

For detection, riproximin was fluorescently labeled with an amine-reactive, N-hydroxysuccinimide ester-activated dye (DyLight 549, Pierce) as described by the manufacturer. Protein-containing fractions were pooled, washed, and concentrated (molecular weight cut-off, 10,000) in 20 mm Tris-HCl buffer, pH 7.5, with 200 mm NaCl.

Fluorescently labeled riproximin was additionally purified on a lactosyl-Sepharose column (Lactosyl Sepharose 4 Fast Flow, GE Healthcare) as described before (3) to exclude all conjugates with blocked binding sites. The eluate fractions were concentrated (molecular weight cut-off 10,000), and the protein content was determined by the absorbance at 280 nm.

Integrity and biological activity of the labeled riproximin were controlled by SDS-PAGE and by a cell viability assay with HeLa cells, respectively (see below). The dye payload was determined from the UV-visible spectrum.

Binding Analysis of Riproximin in a Carbohydrate Microarray

The carbohydrate microarrays were prepared as described previously, with the following modifications (21, 22). The array contained 157 components, including 97 different defined synthetic carbohydrates as bovine serum albumin (BSA) or human serum albumin conjugates, 28 synthetic glycopeptides, and 32 natural glycoproteins. For a list of the carbohydrate structures, see supplemental Table 1. Samples were printed in duplicate on SuperEpoxy 2 Protein glass slides (TeleChem International, Inc., Sunnyvale, CA) using a Biorobotics MicroGrid II microarrayer (Genomic Solutions, Ann Arbor, MI) fitted with Stealth pins (Telechem International; catalog no. SMP3, which produce ∼100-μm spots). Relative humidity within the printing chamber was maintained at >50%. Due to the small footprint of a single array (4.5 × 4.5 mm), 16 copies of the full array could be printed onto each glass slide. Printed glass slides were stored at −20 °C until use.

Riproximin binding was evaluated using minor modifications of the previously reported protocol (21, 23). Briefly, slides were fitted with a 16-well module (Grace Bio-Labs) to physically separate the 16 printed copies of the array. Each well was blocked with 3% BSA (200 μl/well; immunoglobulin-free BSA; Sigma-Aldrich) in PBS for 1 h. DyLight-labeled riproximin was diluted 1:25 in 3% BSA in PBS and then incubated on the array (75 μl/well) for 2 h at room temperature in the dark. After washing six times with PBS, the well module was disassembled, and the slide was incubated in PBS for 5 min. The slide was centrifuged at 453 × g for 5 min and then scanned using a GenePix Scanner 4000A (Molecular Devices Corp., Union City, CA). Slides were scanned at 10 μm resolution, and image analysis was carried out with GenePix Pro 6.0 analysis software (Molecular Devices Corp.). The fluorescent spots were defined as circular regions of interest with a diameter of 100 μm. Regions of interest were allowed to be adapted to the actual feature size by ±30 μm. After local background subtraction, the median pixel intensity of regions of interest was used to obtain a single fluorescence value for each spot. The mean value of the two replicate spots was used as the final value for each array component. To investigate the competitive binding of riproximin with Tn3 structures, riproximin was incubated with 60 μg/ml Tn3-BSA (15 Tn3 molecules/BSA molecule) in 3% BSA in PBS for 1 h prior to the incubation of the mixture on the carbohydrate microarray.

Desialylation of Bovine Submaxillary Mucin

Asialo-bovine submaxillary mucin (aBSM) was prepared by incubating 9 mg/ml BSM with 28 milliunits/mg neuraminidase (Roche Applied Science) in 50 mm sodium acetate buffer, pH 5.0, for 3 h at 37 °C. As a control, the same amount of BSM was incubated in buffer without neuraminidase. The degree of desialylation was monitored by dot blot analysis using biotinylated wheat germ agglutinin (Vector Laboratories, Peterborough, UK). For cell culture experiments, aBSM and BSM control samples were filter-sterilized (0.45 μm). BSM sample concentrations were determined using the Glycoprotein Carbohydrate Estimation Kit (Thermo Scientific, Pierce) and a BSM standard curve. The BSM sample concentration was also used to estimate the aBSM concentrations because the desialylation procedure created additional reducing ends, which interfered with the measurement.

Enzymatic Deglycosylation of Asialofetuin

To deglycosylate ASF, 6 μg/μl glycoprotein was incubated for 5 min at 100 °C in denaturing buffer containing 50 mm sodium phosphate buffer, pH 7.5, 0.1% SDS, and 50 mm β-mercaptoethanol. The mixture was cooled on ice, and 0.75% (v/v) Triton X-100 as well as 1 μl of N-glycosidase F (5000 units/ml; Sigma-Aldrich) were added subsequently. The control reaction contained no N-glycosidase. Reaction mixtures were incubated overnight at 37 °C. Protein deglycosylation was monitored by SDS-PAGE.

Dot Blot Analysis

Glycoproteins were serially diluted in PBS, and 1 μl of each dilution was spotted onto a nitrocellulose transfer membrane (Whatman Protran, Dassel, Germany). The membrane was dried at room temperature and blocked with 5% BSA. The membrane was probed with riproximin (30 μg/ml in 5% BSA) followed by an anti-riproximin monoclonal mouse antibody (mRpx-Ab 62). HRP-linked, human preadsorbed goat anti-mouse antibody (Santa Cruz Biotechnology, Inc.) was used for detection.

Isolation of N-Glycans from Glycoproteins

The isolation of N-glycans from asialofetuin and fetuin was performed according to Karg et al. (24) with some modifications. 100 mg/ml glycoprotein was digested with 110 milliunits/μl pepsin (Roche Applied Science) in 10 mm HCl for 48 h at 37 °C. Pepsin was inactivated by raising the pH above 5.0 with NaOH. The samples were buffered with sodium acetate buffer, pH 5.2 (final concentration 100 mm). Peptide-N-glycosidase F (Sigma-Aldrich) was subsequently added, and the sample was incubated for 24 h at 37 °C. Free N-glycans and smaller peptides were separated by ultrafiltration on membranes with a molecular weight cut-off of 30,000.

To remove the peptides, the glycans were additionally purified on C18 resin (Waters Corp., Milford, MA) spin columns, which were prepared using 30 μl of resin per column prewashed with ethanol followed by H₂O. The N-glycan/peptide mixtures were added to the spin columns, incubated for 5 min, and centrifuged. The eluates were subsequently desalted on a cation exchange resin (AG 50W-X8, hydrogen form, 100–200 mesh; Bio-Rad). For this step, the resin was washed with H₂O before transferring 0.6 ml to an empty spin column and drying the resin by centrifugation. The N-glycan- containing eluates were loaded onto the resin, incubated at room temperature for 10 min, and eluted by centrifugation. The N-glycan samples were dried in a SpeedVac and stored at 4 °C.

Enzyme-linked Immunosorbent Assay (ELISA)

Glycoprotein binding assays were performed using Immuno 96 MicroWell solid plates (Maxisorp, NUNC, Langenselbold, Germany) coated with 1 μg/ml ASF or aBSM in PBS, respectively. After blocking with 5% BSA in PBS, a 1-h preincubated mixture of riproximin and serially diluted glycoprotein was added. Riproximin was detected using the monoclonal riproximin antibody Rpx-mAb 62, followed by an HRP-coupled, human preadsorbed anti-mouse antibody (Santa Cruz Biotechnology, Inc.). Binding was colorimetrically measured using 3,3′,5,5′-tetramethylbenzidine. When ASF binding was assessed, 5% milk in PBS was used for blocking, and an ovine anti-bovine asialofetuin IgG (AbD Serotec, Düsseldorf, Germany) followed by HRP-linked anti-sheep antibody was used for detection.

Binding signals of 2–3 replicates were averaged and plotted. Concentration-response curves were fitted using the log-logistic model. To directly refer the potential inhibition effects of the glycoproteins to the relative number of carbohydrate structures, the number of N-acetyl-d-lactosamine (LacNAc) and Tn (GalNAc) residues per ASF and aBSM was estimated as follows. For ASF, a molecular mass of 48 kDa and three NA3s resulting in nine terminal LacNAc residues were used for calculation. For aBSM, a molecular mass of 400 kDa was assumed (25, 26). 920 GalNAc residues/molecule were estimated for aBSM based on data available for porcine submaxillary mucin, which possesses a molecular mass of 10⁶ Da with ∼2300 GalNAc residues/molecule (20).

Desialylation of Cells with Neuraminidase

For desialylation of the cell surface, MDA-MB-231, MCF7, or HeLa cells were seeded into microplates and allowed to settle down overnight. Neuraminidase (Roche Applied Science) was diluted into medium without FCS to a concentration of 1 milliunit in 50 μl. FCS-containing medium was removed from the cells and replaced by the neuraminidase-containing FCS-free medium. Control cells received FCS free medium without neuraminidase. The plates were incubated for 1 h at 37 °C. Subsequently, neuraminidase-containing and control media were replaced by fresh FCS-containing medium, and the cells were treated with serial dilutions of riproximin. Cell viability was tested with the 3-(4,5-dimethythiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay after 72 h of incubation (see below). Three independent experiments were performed for cells in which an effect of neuraminidase exposure was observed on riproximin cytotoxicity.

For the analysis, neuraminidase-pretreated cells (Neu+/Rpx) and control cells (Neu−/Rpx) were analyzed as two separate groups. For each experiment, antiproliferative activity was measured in triplicate for each riproximin concentration as well as for cells treated with solvent control. Observed antiproliferative activity response values were normalized to mean values of solvent control cells and averaged per riproximin concentration. For each treatment group, the four-parameter log-logistic model (27) was fitted to the averaged normalized antiproliferative activity values. From each of the two fitted antiproliferative activity curves, the IC₅₀ value, defining the concentration that produces 50% of the responsive maximal cytotoxic effect of riproximin, was estimated.

The ratio of the two IC₅₀ estimates (i.e. (IC₅₀ Neu−/Rpx)/(IC₅₀ Neu+/Rpx)) and the corresponding 95% confidence interval were computed to assess statistical significance. The IC₅₀ shift was considered significant when the IC₅₀ ratio explicitly differed from a value of 1 and the respective 95% confidence interval did not include the value 1.

Competitive MTT Cell Viability Assay

The cytotoxic activity of riproximin was assessed using the MTT cell viability assay on HeLa, MCF7, and MDA-MB-231 tumor cell lines. Cells were propagated in a humid atmosphere containing 5% CO₂ at 37 °C in medium that was supplemented with 10% FCS, 2 mm l-glutamine, 100 units/ml penicillin, and 100 μg/ml streptomycin. For the assay, the cells were seeded into microplates (2500 cells/well for HeLa, 3000 cells/well for MDA-MB-231, 3500 cells/well for MCF7) and allowed to settle down overnight. To analyze the influence of single glycoproteins, serial dilutions of the glycoproteins ASF, Fet, aBSM, and BSM were added to the cells prior to the addition of riproximin in a concentration corresponding to its IC₅₀ (HeLa, 0.14 ng/ml; MCF7, 0.10 ng/ml; MDA-MB-231, 0.75 ng/ml). Plates were incubated for 72 h, and the cell growth (CG) was determined by the MTT method.

To investigate the inhibitory effect of the glycoprotein-derived N-glycans alone, the HeLa cells were treated with N-glycans obtained from ASF and Fet by peptide-N-glycosidase F treatment (see above). The glycan effect on the cytotoxicity of riproximin was correlated to the originating protein amount that had been used for deglycosylation.

For data analysis, the inhibitory effect of the single glycoproteins (GP) on riproximin (Rpx) cytotoxicity (inhibition of riproximin cytotoxicity (IRC) (%)) was calculated as (CG_GP/Rpx − CG_Rpx)/(100 − CG_Rpx) × 100. Both values, the CG_GP/Rpx (average cell growth with glycoprotein and riproximin) and CG_Rpx (average cell growth with riproximin alone), were normalized to the respective control without riproximin treatment. Computed IRC values were averaged for every glycoprotein concentration. Inverse cell viability curves were determined for each single glycoprotein by fitting the four-parameter log-logistic model (27). The IRC₅₀ value determined from the curve depicts the glycoprotein concentration that inhibits 50% of the riproximin cytotoxicity. Accordingly, IRC₂₅ and IRC₇₅ values describe the concentrations inhibiting 25 and 75% of the riproximin cytotoxicity.

Statistical two-way analysis of variance with interaction was conducted to assess significance. The effects of the independent factors “group” (i.e. native versus analogical desialylated glycoprotein) and “concentration” (i.e. the concentration of the glycoprotein) as well as the interaction effect between group and concentration on calculated average IRC values (dependent model variable) were studied by performing corresponding global F-tests. Analysis of variance was carried out on a significance level of 5% (i.e. p values of p ≤ 0.05 obtained in the F-tests were regarded as statistically significant).

Combination Cell Viability Assays

For the investigation of ASF and aBSM in combination, different concentrations of these glycoproteins were combined according to the inhibiting effect of the single proteins. Concentrations corresponding to the IRC₂₅, IRC₅₀, and IRC₇₅ of ASF were combined with aBSM concentrations corresponding to IRC₇₅, IRC₅₀, and IRC₂₅, respectively, resulting in the initial concentration for each combination. Serial dilutions of all three combinations were added to the cells prior to the treatment of the cells with riproximin, as described above.

Four MTT viability experiments were performed for each of the three combinations. Data analysis was performed as for the experiments with the single glycoproteins (see above). To circumvent potential problems due to negative or >100% IRC, the model parameters of the lower and upper curve asymptotes were constrained to be ≥0 and ≤100, respectively. Combination indices (CIs) were used to assess the effect of each combination of ASF and aBSM on riproximin cytotoxicity. CIs and the corresponding pointwise 95% confidence bounds were computed for theoretical IRC values ranging between 0.01 and 0.99 (i.e. between 1 and 99%), with a step size of Δ = 0.005 between two neighboring theoretical IRC values, as described by Lee and Kong (27). For each of the three examined combinations of ASF and aBSM, estimated CIs and their 95% confidence bounds were plotted against the theoretical IRC values. Combination effects of ASF/aBSM combinations were characterized as synergistic for CI < 1, as additive for CI = 1, and as antagonistic for CI > 1.

Statistical Software

Data analysis of the carbohydrate microarray, ELISA, and cell viability experiments was performed with Microsoft Excel. For further analysis, the open source statistical software environment R, version 2.8.0 (available from the R Project Web site) was used. Specifically, the R application package “drc” was applied for fitting cell viability curves and for estimation of IC₅₀ values from the calculated curves, as well as for computation of the corresponding IC₅₀ ratio with its 95% confidence interval. The same package was used for fitting inverse cell viability curves and for estimation of IRC₂₅, IRC₅₀, and IRC₇₅ values from the computed curves. Combination indices and according pointwise 95% confidence bounds were computed using a specific R function that was written and provided online by Lee and Kong (27).

RESULTS

Carbohydrate Microarray Analysis

Fluorophore-labeled riproximin was investigated on a glycan microarray with 97 carbohydrates, 28 glycoproteins, and 32 glycopeptides. Glycan structures with riproximin binding signals of >100 relative fluorescence units (RFU) are shown in Fig. 1A. For fluorescence intensity data of all 157 glycan structures, see supplemental Table 2. From these carbohydrates, riproximin bound selectively to NA2, NA3, and Tn3 structures. The NA structures designate biantennary (NA2) and triantennary (NA3) complex N-glycans with terminal Galβ1–4GlcNAc- structures. The Tn3 glycan consists of three consecutive O-linked GalNAcα- serine residues that form a short mucin-like polypeptide backbone, a type of structure also referred to as clustered Tn. The binding signal of riproximin to NA3 was the strongest (>20,000 RFU), followed by NA2 (>10,000 RFU). The binding signal of riproximin to Tn3 was about 5-fold lower than to NA3 but still significant. Binding was strongest at the highest Tn3 density (Ac-Tn3-03 ≪ Ac-Tn3-15 ≪ Ac-Tn3-27).

FIGURE 1.

Binding profile of riproximin. A, fluorophore-labeled Rpx (27 μg/ml) was applied to the carbohydrate microarray. For the competition experiment, riproximin was preincubated with 60 μg/ml BSA carrying 15 Tn3 moieties (Rpx + Tn3). Carbohydrate structures with signals >100 RFU are shown. For carbohydrate abbreviations and the numerical data of all 157 glycan structures, see supplemental Tables 1 and 2, respectively. The number beside the carbohydrate abbreviation refers to the average number of carbohydrates per molecule of BSA (e.g. Ac-Tn3-27 has 27 Tn3s/BSA molecule). Riproximin significantly bound to the bi- and triantennary structures NA3 and NA2 as well as to Tn3 structures at high glycan density. Within the group of the glycoproteins, riproximin significantly bound to ASF, CEA, aBSM, and aOSM. B, dot blots comparing the binding of riproximin to serial dilutions of desialylated or deglycosylated glycoproteins versus their unprocessed counterparts. ASF deg, N-deglycosylated ASF. N-Deglycosylation of ASF significantly reduced its Rpx binding capacity. Error bars, S.D.

From the group of the glycoproteins, riproximin showed strong binding to ASF (>15,000 RFU), a glycoprotein with NA2 or NA3 N-glycan structures and weaker binding (<5000 RFU) to the desialylated glycoproteins bovine (aBSM) and ovine (aOSM) submaxillary mucin, both glycoproteins with massive O-glycosylation and a high proportion of Tn antigens. Riproximin also showed significant binding to the carcinoembryonic antigen (CEA) that is reported to contain N-glycan structures (Fig. 1A).

Taken together, the riproximin binding profile on the array was very narrow and confined to two types of glycans, the NA2/NA3 structures from the group of complex N-glycans and the O-glycan Tn3 structures representing the mucin-type clustered Tn.

To further investigate the relationship between the two binding activities, competitive binding experiments were performed. The preincubation of riproximin with Tn3 linked to BSA led to a decrease of riproximin binding to O-glycan-rich structures, such as Tn3 or the mucins aBSM and aOSM. In contrast, the N-glycan structures NA2 and NA3 and the glycoprotein CEA showed no decrease in the binding signal. However, the signal of ASF declined to ∼50%, suggesting the presence of a second riproximin binding site with lower affinity (Fig. 1A).

Binding Analyses

For validation of the microarray binding results, selected glycoproteins were investigated by dot blot analysis. ASF, fetuin (Fet), N-deglycosylated ASF, aBSM, and BSM were immobilized on a nitrocellulose membrane and subsequently probed with riproximin. As expected, riproximin showed distinctly stronger binding to the desialylated glycoproteins than to their native forms. N-Deglycosylation of ASF resulted in a significant loss of riproximin binding (Fig. 1B), confirming that riproximin binding is confined to the sugar component of this glycoprotein. The two glycoproteins ASF and aBSM were thus chosen as model proteins to represent the two types of glycans NA3 and clustered Tn for further characterization of the riproximin binding mechanism and its biological significance.

The binding affinity of riproximin for the two model proteins was measured using the microscale thermophoresis method. For ASF, two K_d values of 11 nm and 7 μm were determined. The second affinity constant is in line with the assumption that riproximin also binds to a non-NA3 glycoconjugate on the ASF molecule. Riproximin bound to aBSM with a K_d of >50 μm. Two similar K_d values of 48 nm and 1.1 μm were also detected in a preliminary isothermal titration calorimetry experiment, when riproximin was titrated with ASF, and the two-site binding model was applied for the analysis of the data. However, due to the high microdiversity of the binding partners, the interactions measured in this experiment turned out to be too complex to be covered by the available models, and no reasonable stoichiometry data were obtained. The complete thermophoresis and isothermal titration calorimetry data are presented in supplemental Figs. 2 and 3, respectively.

Competitive Binding Analyses of ASF and aBSM in ELISA

To further characterize the binding of riproximin to the two classes of glycans, ELISA analyses were performed with riproximin alone or under competitive conditions. On an ASF coat, the preincubation of riproximin with ASF or aBSM resulted in a decrease of the signal with increasing ASF or aBSM concentrations (Fig. 2A). On each ASF molecule, three NA3s and thus nine LacNAc structures have been reported. Mucins such as aBSM, however, show a much higher GalNAc density of >1000/molecule. The competitive effect of both glycoproteins in the ELISA was therefore also related to the absolute number of the respective carbohydrate structure, LacNAc for ASF and GalNAc for aBSM (Fig. 2B). The resulting curves indicate that the aBSM competition was only effective at very high Tn concentrations, as expected from the higher affinity of riproximin for LacNAc structures. On aBSM coat, however, preincubation with ASF did not reduce the riproximin signal, despite the higher affinity of riproximin for ASF (Fig. 2C).

FIGURE 2.

ELISA binding analysis. A, plates coated with ASF were incubated with riproximin/ASF or riproximin/aBSM mixtures. Riproximin was detected by Rpx antibody. B, binding curves from A were correlated to the numbers of glycans present on each glycoprotein (LacNAc for ASF and GalNAc for aBSM). C, plates coated with aBSM were incubated with riproximin/ASF or riproximin/aBSM mixtures. Riproximin was detected by Rpx antibody. D, plates were either coated with ASF and incubated with 5 μg/ml riproximin (ASF coat + Rpx (5)) or directly coated with 5 μg/ml riproximin (Rpx (5) coat). Serial dilutions of aBSM were subsequently applied. Riproximin was detected by Rpx antibody. E, plates coated with aBSM were incubated with serial dilutions of riproximin in the presence of 10 or 50 μg/ml ASF. Binding of ASF was detected by an anti-bovine ASF IgG. F, plates coated with decreasing ASF concentrations were incubated with 50 μg/ml ASF alone or mixed with 10 or 30 μg/ml riproximin. Binding of ASF was detected by an anti-bovine ASF IgG. Each plot shows one representative experiment. Data are represented as mean ± S.D. (error bars) from 2–3 replicates. Negative values due to background subtraction in plot D were set to zero. Concentration-response curves were fitted using the log-logistic model.

To test whether the lower signal at high aBSM concentrations was not an experimental artifact, plates were either coated with ASF followed by incubation with different riproximin concentrations or directly coated with different riproximin concentrations under identical conditions. Upon aBSM incubation, detection of riproximin decreased with rising aBSM concentrations for both ASF-bound and directly coated riproximin but more steeply for the ASF coat (Fig. 2D). Moreover, aBSM abolished the riproximin signal only for ASF bound riproximin completely.

A riproximin signal increased to more than 100% was detected on the ASF coat at low aBSM concentrations (Fig. 2, A and D). This finding indicates that riproximin may have cross-linked riproximin-loaded aBSM to the ASF coat. To further investigate cross-linking of ASF and aBSM by riproximin, an antibody against ASF was used. Coated aBSM was incubated with a serial dilution of riproximin in the presence of 10 or 50 μg/ml ASF. The results showed significant binding of ASF to aBSM that was dependent on the riproximin but not the ASF concentration (Fig. 2E). As shown in Fig. 2C, aBSM cross-linking to itself was not observed. The riproximin signal remained ≤100% even when lower aBSM concentrations were tested (data not shown). Conversely and consistent with the second binding constant observed for ASF, riproximin showed some ASF cross-linking to itself at high riproximin/ASF ratios (Fig. 2F).

Influence of Different Glycoproteins and Sugar Moieties on Riproximin Cytotoxicity

To investigate the biological significance of riproximin binding to the identified glycostructures, the chosen model glycoproteins ASF and aBSM and their sialylated counterparts Fet and BSM were tested in three different tumor cell lines (HeLa, MCF7, and MDA-MB-231) to evaluate their effect on riproximin cytotoxicity.

The neutral glycoproteins ASF and aBSM inhibited riproximin cytotoxicity significantly more strongly than their sialylated counterparts BSM and Fet in all three cell lines tested (all p < 0.01; F-test for factor “group”). In addition, the statistical analysis revealed that this effect was concentration-dependent, because with increasing concentrations, the difference between the inhibitory effect of the neutral and sialylated glycoproteins also increased (all p < 0.01; F-tests for factor “concentration” and for the interaction between “group” and “concentration”).

The relative inhibition of riproximin cytotoxicity achieved by ASF or aBSM, however, varied between the cell lines. Although ASF and aBSM were both able to completely abolish riproximin cytotoxicity in HeLa cells (Fig. 3, A and B), only 70% maximal inhibition could be achieved in MCF7 cells (Fig. 3, D and E). A particularly intriguing finding was that the inhibitory effect of aBSM decreased at high aBSM concentrations in MDA-MB-231 cells (Fig. 3G).

FIGURE 3.

Cytotoxicity experiments. ASF versus Fet (A, D, and F) and aBSM versus BSM (B, E, and G) were added in combination with riproximin to HeLa (A and B), MCF7 (D and E), and MDA-MB-231 (F and G) cells and incubated for 72 h at 37 °C. The asialo-glycoproteins showed a significantly higher degree of inhibition of riproximin cytotoxicity. C, N-glycans (N-Gly) derived from ASF or Fet were used to compete for riproximin cytotoxicity in HeLa cells as in A. The effect of the N-glycans was related to the original protein amount that had been deglycosylated. H, neuraminidase-pretreated (Neu+/Rpx) and control (Neu−/Rpx) MDA-MB-231 cells were treated with riproximin for 72 h at 37 °C. Vertical lines represent estimated IC₅₀ values. Cell viability was determined by the MTT assay. For each graph, mean values from 2–4 independent experiments were used and fitted using the four-parameter log-logistic model. For the inhibition effect of aBSM/BSM on MDA-MB-231, a five-parametric Brain-Cousens hormesis model was used for fitting (61).

Cytotoxic activity of riproximin was also reduced in HeLa cells following co-treatment with N-glycans that were obtained by deglycosylation of ASF with peptide-N-glycosidase F. This finding demonstrated that the inhibitory effects of glycoproteins on riproximin cytotoxicity were glycan- but not core protein-dependent (Fig. 3C).

Influence of the Sialyl Caps on Riproximin Cytotoxicity

To investigate the relationship between riproximin cytotoxicity and the abundance of sialyl groups on the cell surface, HeLa, MCF7, and MDA-MB-231 cells were treated with neuraminidase and subsequently exposed to riproximin in a cellular viability assay. A detectable increase in riproximin cytotoxicity was observed only in the MDA-MB-231 cells. For untreated MDA-MB-231 cells, an IC₅₀ value of 0.22 ng/ml (95% confidence interval, 0.192–0.250) was estimated for riproximin. After treatment with neuraminidase, the IC₅₀ value decreased significantly (IC₅₀ = 0.08 ng/ml, 95% confidence interval, 0.074–0.094). The ratio of the IC₅₀ values with and without neuraminidase treatment of 2.64 (95% confidence interval, 2.256–3.081) demonstrated that the enzymatic removal of the sialic acid caps from glycans on the cell surface led to significantly enhanced cytotoxicity of riproximin in MDA-MB-231 cells (Fig. 3H). In contrast, neuraminidase treatment did not influence the cytotoxic potency of riproximin in MCF7 or HeLa cells (data not shown). It is conceivable that the removal of additional sialic acids in the most sensitive, sialyl-poor MCF7 cells could not further increase its riproximin response. Furthermore, the fact that neuraminidase treatment did not influence the sensitivity of HeLa cells would indicate a low cell surface sialyl content.

Competitive Influence of ASF and aBSM on Riproximin Activity

To investigate the functional effects of the interactions between riproximin and the two different types of glycans that were found to be its binding partners, the model proteins ASF and aBSM were added in combination to compete with riproximin in cytotoxicity assays. Because of the high inhibitory effect of both glycoproteins, HeLa cells were chosen for the competitive investigation. ASF and aBSM concentrations were combined in three different proportions (ASF/aBSM of IRC₂₅/IRC₇₅, IRC₅₀/IRC₅₀, and IRC₇₅/IRC₂₅) according to the degree of riproximin cytotoxicity inhibition each of them had achieved alone (IRC values in Table 1). The three different combinations were serially diluted, resulting in three concentration ranges (supplemental Table 3) and added in combination with riproximin to HeLa cells. The respective inverse cell viability curves depicting the cytotoxicity inhibition effects of the three combinations are shown in Fig. 4, A–C. The theoretical IRC values calculated for each combination model were examined for additive, synergistic, or antagonistic effects (Fig. 4, D–F).

TABLE 1.

IRC concentration values as calculated from the fitted inverse cell viability curves and estimation of the related carbohydrate concentration

Shown is inhibition of riproximin cytotoxicity by 25% (IRC₂₅), 50% (IRC₅₀), or 75% (IRC₇₅).

	HeLa			MCF7	MDA-MB-231
	IRC25	IRC50	IRC75	IRC50	IRC50
ASF
Protein (μg/ml)	16	74	340	∼1000a	59
Protein (nm)b	333	1540	7080	19,900	1200
LacNAc (μm)c	3	14	64	179	11
aBSM
Protein (μg/ml)	0.1	0.4	1.6	6	0.3
Protein (nm)b	0.3	1	4	14	0.8
GalNAc (μm)c	0.3	0.9	3.7	13	0.7

^a The real relative IRC₅₀ value is >1000 μg/ml.

^b Molar concentrations were calculated using molecular masses of 48 kDa for ASF and 400 kDa for aBSM.

^c The amount of the carbohydrate residues was estimated based on nine terminal LacNAc residues for an ASF and 920 GalNAc residues for an aBSM molecule.

FIGURE 4.

Analysis of the combined competitive inhibitory effect of ASF and aBSM on the cytotoxicity of riproximin. ASF and aBSM were applied in a cellular viability assay with riproximin in HeLa cells. Three glycoprotein combinations with the following ASF/aBSM active component proportions were analyzed: IRC₂₅/IRC₇₅ (A and D), IRC₅₀/IRC₅₀ (B and E), and IRC₇₅/IRC₂₅ (C and F). The cells were incubated with riproximin and a serially diluted glycoprotein combination for 72 h at 37 °C. Cell viability was determined by the MTT assay. For each combination of ASF/aBSM, average IRC values of four independent experiments and the respective inverse cell viability curves were fitted by using the four-parameter log-logistic model (A–C). The combination effect of the glycoproteins was characterized by the CIs. For all theoretical IRC values of each ASF/aBSM combination, CI values and the respective confidence bounds were calculated and plotted (D–F). Combination effects were characterized as synergistic for CI < 1, as additive for CI = 1, and as antagonistic for CI > 1.

Combinations of the IRC₂₅ of ASF with the IRC₇₅ of aBSM contained a lower ASF (i.e. NA-type active component) proportion and showed an additive effect that did not depend on the overall cytotoxicity inhibition (Fig. 4D). The other combinations (ASF with IRC₅₀ + aBSM with IRC₅₀ as well as ASF with IRC₇₅ + aBSM with IRC₂₅) showed an effect that was dependent on the overall cytotoxicity inhibition degree achieved (Fig. 4, E and F). At low riproximin cytotoxicity inhibition, <27% for ASF50/aBSM50 or <6% for ASF75/aBSM25, both combinations showed a synergistic effect. For inhibition degrees of 26–79% (ASF50/aBSM50) or 6–71% (ASF75/aBSM25), an additive effect was observed. At overall inhibition rates over 79%, the combination effect turned into an antagonistic one (Fig. 4, E and F).

In summary, the mode of inhibition of riproximin cytotoxicity was dependent on the proportion of each glycoprotein (i.e. glycan type in the combination) as well as on the cumulated glycoprotein amount. For most combinations, the effect was additive, which implies that ASF and aBSM were able to inhibit riproximin in an independent manner.

DISCUSSION

NA2/NA3 and Cancer-related, Clustered Tn Glycostructures Are Specifically Bound by Riproximin

Riproximin, a type II RIP, was purified as a galactose/lactose-specific protein (3). At first, the lactose binding and elution suggested broad galactose specificity similar to that of ricin and abrin (28, 29), both very toxic RIPs that render Ricinus communis and Abrus precatorius seeds lethal on ingestion. However, the low peroral toxicity of riproximin indicated that its binding profile would be narrower. Indeed, analysis in a carbohydrate microarray revealed that riproximin preferentially binds to two groups of glycoconjugates, the branched bi- and triantennary N-glycan structures NA2 and NA3 and the O-glycan structure known as clustered Tn, especially when present at high density. The terminal residue of NA2/NA3 structures is a Galβ that is connected to a GlcNAc of each antenna. Clustered Tn consists of a series of single GalNAcα residues (single Tn) bound to adjacent amino acids in protein regions rich in Ser/Thr repeats. They are typical for the extracellular domains of mucins. Tn and particularly clustered Tn is an established tumor-specific antigen present on many adenocarcinomas (30–32).

Riproximin specificity for the clustered Tn antigen is remarkable and could explain its tumor specificity described previously (1, 2). Prominent targets for riproximin on cancer cells could therefore be the cancer-associated mucins (MUCs). MUC 6, for example, was shown to be responsible for the high density of clustered Tn on the surface of the highly sensitive MCF7 cells (33). Cancer-specific MUC 1 and MUC 2 were also described as Tn-rich glycoproteins (34, 35). However, riproximin showed its highest binding for bi- and triantennary complex N-glycans (NA2 and NA3 structures). No direct link between the presence of NA2 or NA3 and cancer has been established so far. However, N-glycans from cancer cells show higher branching, resulting in a higher NA3 density (15, 36, 37) that would provide additional riproximin binding sites. On the other hand, the possibility cannot be excluded that these structures are also present on normal cells because 17% of the IgG molecules in human serum are NA2-glycosylated (10).

Riproximin Is Specific for Asialo-glycostructures

Riproximin showed a clear preference for desialylated glycan structures. Most cancer cells and cancer-related glycoproteins show abnormal sialylation, but its investigation revealed controversial results. For many tumors, increased sialyl-Lewis X expression was described (15, 38–40). Conversely, N-glycans of the prostate-specific antigen showed a tumor-dependent decrease of sialylation (41–43). Although cancer derived O-glycans were often described as highly sialylated, they also contain the sialyl-free, cancer-specific Tn and T antigens (44). The sialidase Neu 3, which is located in the plasma membrane and leads to sialyl-cap removal, is up-regulated during carcinogenesis (45).

An increase in riproximin cytotoxicity was observed upon neuraminidase treatment for the strongly sialylated MDA-MB-231 cells but not for the sialyl-poor MCF7 (46), the most sensitive cell line within the panel tested (2). This experiment demonstrated a direct relationship between the biological activity of riproximin and the extent of sialylation found on cell surfaces.

Riproximin Displays a Narrow Binding Profile

Riproximin demonstrated remarkable selectivity on the glycan array. Despite the presence of many other sialyl-free glycan structures with terminal Galβ or GalNAcα on the array, such as LNnT (Galβ1–4GlcNAcβ1–3Galβ-) or the tumor-related Adi (GalNAcα1–3 Galβ-) (47), Forssman antigen (GalNAcα1–3GalNAcβ-) (48), and TF antigen (Galβ1–3GalNAcα-) (49), no binding of riproximin to these structures was detected. In comparison, other commonly studied Gal- and GalNAc-binding lectins, such as R. communis agglutinin, soybean agglutinin, Helix pomatia agglutinin, jacalin, and Bauhinia purpurea lectin, recognize a wide range of glycans on the array. For example, the Gal-binding type II RIP R. communis agglutinin significantly bound to almost any structure with a terminal LacNAc and lactose (50). Thus, riproximin showed a very narrow binding profile that was dependent not only on the nature of the terminal sugars but also on their amount and constellation.

The preference of riproximin for trimeric structures, such as NA3 and Tn3, was also remarkable. The 3-fold sugar specificity might reflect the structure of the typical RIP B-chain binding domains, which are both trimers of an ancient lectin motif. However, molecular modeling of riproximin B-chain interaction with NA3 structures revealed that the terminal Gal residues on NA3 cannot span the distance between two subdomains but could interact with other aromatic residues that are frequently present in the closer neighborhood (supplemental Figs. 5 and 6). A multivalent binding of riproximin to NA2 or NA3 structures could explain the observed high specificity, which is unusual for a lectin and comparable with that of a monoclonal antibody.

On the other hand, the possibility cannot be completely ruled out that the stronger binding signals on NA3 result from stoichiometric and/or bind-and-jump effects. Binding of two or three ricin B-chains to a single NA3 structure has been described (51). However, the fact that the presence of an additional terminal Galβ in NA4 structures did not improve but instead significantly reduced the binding signal contradicts this hypothesis.

Glycoproteins as Binding Counterparts of Riproximin

Accordingly, the NA2/NA3-containing glycoproteins ASF (52) and CEA (53) as well as the Tn-rich glycoproteins aOSM and aBSM (54) on the array showed strong riproximin binding. ASF and aBSM were thus chosen as model proteins to mimic the effects of the NA2/NA3 and Tn-structures, respectively, in cellular experiments.

Both asialo-glycoproteins showed significant inhibition of riproximin cytotoxicity in HeLa, MCF7, and MDA-MB-231 cells. Moreover, deglycosylation of ASF resulted in loss of its riproximin binding, whereas the ASF-derived N-glycans significantly inhibited riproximin cytotoxicity. These findings demonstrate that specific binding of riproximin to NA2/NA3 and/or Tn glycostructures on the cellular surface is a prerequisite for cytotoxicity.

The degree of inhibition, however, strongly depended on the cell line (i.e. the cell surface glycosylation). For example, >10-fold higher glycoprotein concentrations were required in the particularly Tn-rich MCF7 cells (33) to reduce riproximin sensitivity by 50% as compared with HeLa cells, although HeLa and MCF7 are equally sensitive to riproximin. It must therefore be assumed that the abundance and surface distribution of glycostructures on a particular cell play a crucial role in its sensitivity to riproximin. Moreover, it cannot be excluded that additional riproximin glycotargets exist on the cell surface, which have not yet been identified.

Two Different Binding Sites Are Associated with Different Specificities of Riproximin

The specific binding of riproximin to both Galβ within NA2/NA3 and GalNAcα within Tn structures related to the presence of two binding domains in the B-chain of riproximin. The structures of the clustered Tn antigen and NA2/NA3 are significantly different, and it is uncommon for a lectin or monoclonal antibody to bind both. For example, Tn-specific antibodies have never shown binding to NA2/NA3 (55). The very narrow binding profile of riproximin makes the recognition of both clustered Tn and NA2/NA3 in the same binding site very unlikely. We thus hypothesized that riproximin binds the two sugar structures with different binding sites corresponding to the two sugar binding sites of the B-chain.

To evaluate this hypothesis, competitive analyses were performed. In array experiments, preincubation of riproximin with Tn3 decreased the binding of riproximin to immobilized Tn3 structures and aBSM but did not influence the NA binding. However, because the NA affinity of riproximin is significantly higher than its Tn3 affinity, it could not be excluded that the Tn3 concentrations used in the experiment were too low to affect NA binding.

The strongest argument supporting the existence of two different binding specificities is based on the finding that riproximin was able to cross-link the two proteins ASF (NA2/NA3 structures) and aBSM (Tn structures). Moreover, even at low aBSM concentrations, no cross-linking of aBSM with itself could be detected. However, at very low ASF/riproximin ratios, a significant amount of ASF cross-linking was observed, which probably results from the interaction of the riproximin GalNAc-binding site with an additional, probable non-NA3 sugar of ASF. This finding is consistent with several other observations regarding the interaction of riproximin with ASF, including the partial competition of ASF binding by Tn3 structures observed in the array, as well as the second affinity constant determined for riproximin and ASF. For fetuin, the presence of an O-linked glycoconjugate has been described (56), and it is conceivable that the desialylated structure would be present on ASF.

Affinity of Riproximin for Glycostructures Depends on Avidity Effects

In the microarray experiments, riproximin showed a distinctly stronger binding signal to NA2/NA3 than to Tn glycostructures. In microscale thermophoresis experiments, the high affinity of riproximin for ASF was >1000-fold higher than for aBSM. In the ELISA experiments, however, aBSM was able to competitively inhibit the binding of riproximin to ASF at high concentrations but not vice versa. Moreover, in cellular experiments, aBSM was the stronger cytotoxicity competitor. Even when the high number of Tn structures on aBSM was considered, the concentration needed to achieve the same inhibition was lower for the aBSM-derived Tn than for the ASF-derived LacNAc. These observations demonstrate that the interaction of riproximin with a particular glycostructure is not confined to a strict lock-and-key correlation but is strongly influenced by dynamic interactions similar to those described by Dam and Brewer (18).

Riproximin primarily interacts with terminal Gal and GalNAc, respectively, via its two binding sites, with low but significant affinity. The purification of riproximin on Gal-exposing resins was based on this primary affinity. Sequence analysis and molecular modeling data revealed that the typical type II RIP Gal-binding activity is retained for both α1 and γ2 subdomains of the riproximin B-chain (2). However, the highly specific interactions of riproximin with glycoconjugates presenting two or three moieties of the preferred terminal sugars Gal (NA2/NA3) or GalNAc (Tn3) point toward additional glycan-protein interactions.

When the interactions of riproximin with each of the proteins ASF and aBSM is considered, the main features of the mechanism of lectins binding to multivalent targets apply (19, 20). On multivalent structures, several equivalent epitopes are available for the so-called bind-and-jump effects; the lectin can be recaptured by each of the remaining free epitopes, leading to decreased dissociation. On the other hand, negative cooperativity results from decreasing functional valence when more lectin molecules occupy additional epitopes. The affinity of each consecutive binding step fractionally decreases due to the decreased number of free epitopes available for recapturing.

Based on the hypothesis of the two binding sites specific for NA2/NA3 and Tn3, respectively, riproximin would be able to interact with three NA3 structures of ASF and with >1000 Tn clusters of aBSM, respectively. There is probably a second, O-linked glycan of ASF that would be able to interact with the Tn3-binding site of riproximin. These interactions and their effect on riproximin binding and detection in ELISA experiments are schematically depicted in Fig. 5, A–D. In particular, the high epitope density on aBSM is expected to allow broad dynamic interactions with the Tn3-binding site of riproximin.

FIGURE 5.

Schematic of riproximin interactions with its two binding counterparts. A, schematic representation of the riproximin molecule showing its two different binding sites (BS) and its possible interactions with ASF and aBSM, respectively, as described by Dam et al. (19) for galectin and ASF and later by Dam et al. (20) for soybean agglutinin and asialo-porcine submaxillary mucin. At low riproximin concentrations, several binding epitopes are available on each glycoprotein molecule, and an increased affinity would be observed due to bind-and-jump entropic effects. At high riproximin concentrations, most binding sites on the glycoproteins have already been occupied. Due to negative cooperativity effects, the binding affinity is lower for each subsequent binding step. B–D, schematic representation of riproximin interacting with both of its counterparts as observed in ELISA experiments. On coated ASF at low aBSM concentrations, cross-linking of aBSM to the plate occurs. Due to the high riproximin concentrations in the mixture, several other riproximin molecules are bound to the cross-linked aBSM molecule, resulting in more detectable riproximin (B). At higher aBSM concentrations, competition and masking effects predominate (C). Riproximin cross-links ASF to the coated aBSM (D). E, proposed model for the selective targeting of NA3- and Tn3-presenting tumor cells. The cooperative binding to the two different glycoconjugates confers riproximin an increased affinity for cells exposing both NA2/NA3 and cancer-related clustered Tn structures, resulting in selectivity over cells presenting NA2/NA3 structures only. Moreover, cross-linking of the two structures on the cell surface would result in increased internalization efficiency and thus cytotoxicity.

The increasing affinity of riproximin for BSA bearing Tn3 at increasing density (Tn3-03 ≪ Tn3-15 ≪ Tn3-27) and its high affinity for mucins bearing up to thousands of Tn per molecule reflect the bind-and-jump effects that were described by Dam et al. (57) for the interaction of soybean agglutinin with asialo-porcine submaxillary mucin. Together with the high Tn concentration of aBSM, this dynamic increase in affinity explains the high competitive potency of aBSM in ELISA or cell experiments. On the other hand, because the NA2/NA3 structures are only present at comparatively low density, they did not provide enough binding sites for a detectable bind-and-jump-related increase in specificity.

The simultaneous interaction of riproximin with both proteins therefore strongly depends on the relative amounts of riproximin, NA2/NA3, and Tn3 epitopes. In the presence of both structures, riproximin cross-linked the NA2/NA3 and Tn3 epitopes via its two binding sites (Fig. 5D). It is very likely that cross-linking via riproximin also occurs on the cell surface. Cross-linking is known to be important for internalization of several cell surface receptors (58, 59). Because ASF alone was able to inhibit the cytotoxicity of riproximin by up to 50% in MCF7 cells that are particularly Tn-rich (33) but have <5% NA2/NA3 (60), it is very probable that the cross-linking is also important for the internalization and cytotoxicity of riproximin.

In the presence of high aBSM concentrations, however, the interaction of riproximin with the high density of Tn antigens masked its epitopes. A large cell surface mucin molecule might therefore be able to recruit riproximin and concomitantly mask its second, NA-specific binding site, thereby inhibiting its cross-linking potential (Fig. 5C).

Riproximin Cytotoxicity Was Influenced by Its Dynamic Interaction with ASF and aBSM

To investigate the interdependence of binding, cross-linking and cytotoxicity within the complex environment of a cell, viability experiments were performed in which both ASF (NA3) and aBSM (clustered Tn) were allowed to compete for riproximin binding and thus inhibit its toxicity. The results of these experiments revealed that the interaction patterns of the mixtures containing a high aBSM proportion clearly differed from those of mixtures containing equal active amounts of aBSM and ASF or low aBSM proportions. Moreover, the combinatory effects of the latter two combinations were strongly dependent on the total glycoprotein concentration. The interactions expected to be responsible for these effects are schematically depicted in supplemental Fig. 4.

Mixtures with a high proportion of aBSM exhibited a broad additivity, thus supporting the hypothesis that both binding sites of riproximin are necessary for its cytotoxic activity. In these mixtures, the riproximin/aBSM molecular ratio is particularly low. Increased internal diffusion and bind-and-jump effects lead to an enhanced aBSM affinity, resulting in riproximin sequestration and masking of its ASF binding site. Very little cross-linking occurs under these circumstances, so that the entire ASF fraction would be available to bind to free riproximin, leading to an additive effect.

Conversely, mixtures containing ≤50% of aBSM showed synergistic effects at low global concentrations that turned into additivity and eventually antagonism at high concentrations. Due to the higher riproximin/aBSM ratio, more riproximin molecules are available to bind to the same aBSM molecule. Fewer Tn3 binding sites are available for bind-and-jump effects, leading to a lower apparent affinity and less masking. At high overall glycoprotein concentrations, ASF would bind to unmasked riproximin molecules that are also bound to aBSM; cross-linking would occur. Because a significant amount of glycoproteins bind the same riproximin molecule, antagonism would be observed. At low global concentration of both glycoproteins, the chance for cross-links decreases, so that ASF and aBSM would bind different riproximin molecules. The synergistic effects observed under these circumstances strongly suggest that blocking of a single riproximin binding site significantly interferes with the activity of riproximin. A schematic of the interactions of riproximin with the two glycoproteins in each of the three different cases discussed is presented in supplemental Fig. 4. Overall, these findings indicate that cross-linking is part of the mechanism responsible for riproximin internalization and cytotoxicity (Fig. 5E).

In summary, even in the cellular context of a single cell type, the interaction pattern of riproximin with its glycotargets was very complex and depended on dynamic effects. At low riproximin concentrations, the internal diffusion along large glycoproteins predominated. At high concentrations, negative cooperativity opposed the bind-and-jump effects and favored cross-linking. Cellular experiments revealed that despite the lower affinity of aBSM to riproximin in direct binding experiments, in cellular assays it had the stronger impact on the cytotoxicity of riproximin.

Conclusions

In summary, the antineoplastic-active type II RIP riproximin was shown to specifically bind to two types of glycostructures, the N-linked NA2/NA3 and the O-linked clustered Tn tumor-specific antigen. The two specificities were related to the two binding sites present on the riproximin B-chain. The sugar interactions of riproximin were shown to combine high specificity with dynamic interactions that are typical for lectins interacting with multivalent binding targets.

Understanding the mechanism of riproximin targeting is particularly important because its therapeutic potency strongly depends on the presence of definite cancer-related structures on the cells to be targeted. Cross-linking of the two structures NA2/NA3 and Tn3 confer on riproximin an enhanced selectivity for cells exposing both structures. However, a relationship between the concomitant occurrence of these two structures and cancer has not yet been established. The ideal riproximin target cell would contain surface glycostructures with high Tn densities and sialyl-free NA3 structures. On the other hand, due to the broad range of dynamic interactions riproximin can get involved in, it is conceivable that the biological activity of riproximin might be modulated by the addition of particular glycan structures, such as glycopolymers with high riproximin binding capacity but low affinity. Such a polymer could function as a “carrier” for riproximin, preventing it from binding to low affinity structures and delivering it specifically to tumor cells, where it would find high avidity targets.

Because the field of glycobiology is rapidly expanding, the availability of synthetic complex glycan structures and the development of novel tools will support further investigations. The detailed characterization of the binding properties of riproximin reported here provides a model for functional lectin studies. Moreover, riproximin can be included in the panel of lectins with well characterized properties that currently find a broad usage in glycobiology.