The specific targeting of differentially expressed glycans in malignant cells has emerged as an attractive anticancer strategy. One such target is the oncodevelopmental antigen polysialic acid (polySia), a polymer of α2,8-linked sialic acid residues that is largely absent during postnatal development but is re-expressed during progression of several malignant human tumors, including small-cell and non–small cell lung carcinomas, glioma, neuroblastoma, and pancreatic carcinoma. In these cancers, expression of polySia correlates with tumor progression and poor prognosis and appears to modulate cancer cell adhesion, invasiveness, and metastasis. To evaluate the potential of PolySia as a target for anticancer therapy, we developed a chimeric human polySia-specific mAb that retained low nanomolar (nmol/L) target affinity and exhibited exquisite selectivity for polySia structures. The engineered chimeric mAb recognized several polySia-positive tumor cell lines in vitro and induced rapid endocytosis of polySia antigens. To determine whether this internalization could be exploited for delivery of conjugated cytotoxic drugs, we generated an antibody–drug conjugate (ADC) by covalently linking the chimeric human mAb to the tubulin-binding maytansinoid DM1 using a bioorthogonal chemical reaction scheme. The resulting polySia-directed ADC demonstrated potent target-dependent cytotoxicity against polySia-positive tumor cells in vitro. Collectively, these results establish polySia as a valid cell-surface, cancer-specific target for glycan-directed ADC and contribute to a growing body of evidence that the tumor glycocalyx is a promising target for synthetic immunotherapies.
These findings describe a glycan-specific antibody-drug conjugate that establishes polySia as a viable cell surface target within the tumor glycocalyx.
Glycosylation is the site-specific attachment of sugar assemblies known as glycans to a functional group of another molecule, most commonly proteins or lipids, resulting in the formation of a glycoconjugate. It is a tightly controlled cell- and microenvironment-specific mechanism that involves the coordinated expression and activity of numerous enzymes such as glycosyltransferases and glycosidases. Cellular glycosylation and its products are fundamental to a diverse range of biological processes involved in cancer progression including cell growth and proliferation, cell signaling and communication, cell–cell and cell–extracellular matrix (ECM) interactions, and immune recognition/response (1–4). Thus, it is not surprising that nearly all types of human cancers exhibit changes in glycosylation, a phenomenon that was first reported more than six decades ago (5, 6). The glycosylation changes associated with oncogenic transformation typically involve either incomplete synthesis or neosynthesis processes, both of which may arise from under- or overexpression of glycosyltransferases and glycosidases leading to the exposure of aberrant cell-surface glycans. The most common cancer-associated structural changes include N- and O-glycan branching, O-glycan truncation, increased sialylation, and increased “core” fucosylation, with these motifs occurring on all classes of glycoconjugates including glycoproteins, glycosphingolipids, and proteoglycans (7, 8).
Many of these abnormal glycan epitopes are differentially expressed on malignant cells, thereby providing novel diagnostic and even therapeutic targets that are motivating the development of affinity reagents that recognize these distinct features. However, whereas a rich and diverse collection of antibodies and antibody-derived molecules have been developed for protein antigens, reliable binders that specifically recognize carbohydrates are much less common. Indeed, the paucity of glycan-specific binding reagents was noted by the National Academy of Sciences as a key barrier for advancing glycobiology (9). This shortage was also highlighted in the recently assembled Database for Anti-Glycan Reagents, which indicates that while there are approximately 100 entries for antibodies against N- and O-linked carbohydrates, collectively these target an extremely small set of unique epitopes (10). Specifically, 55 of the 77 total antibodies to O-linked glycans target Tn, sialyl Tn, or TF antigens while 15 of the 25 total antibodies to N-linked glycans are derived from patients with HIV. There is clearly a technological deficit when one considers that glycoproteins and glycolipids are estimated to contain approximately 3,000 glycan determinants (11).
Even when anti-glycan antibodies are available, information about their specificity is often limited and, in a surprising number of cases, antibodies reported to be specific for a designated antigen were found to cross-react with other glycans (12). Moreover, for many of the glycans that differentially occur in malignant cells, it remains to be determined whether they are druggable using “synthetic” immunotherapies (13), such as mAbs, antibody–drug conjugates (ADC), bispecific antibodies, and chimeric antigen receptors (CAR), which all have the potential to initiate new immune or immune-like responses directed toward their tumor-expressed targets. One notable example along these lines is the synthetic immunotherapy dinutuximab, a first-in-class mAb that recognizes the disialoganglioside GD2 found on the surface of neuroblastic tumor cells and is administered as part of a multiagent, multimodality therapy to pediatric patients with high-risk neuroblastoma (14). There is similar potential to develop other glycan-directed antibodies and antibody derivatives; however, this will require overcoming a number of key obstacles related to (i) the current lack of antibodies against structurally diverse glycan antigens beyond the small subset discussed above and (ii) the incomplete knowledgebase surrounding known antibodies in terms of their performance characteristics such as target specificity and therapeutic function (e.g., cytotoxicity).
Here, we addressed this latter gap by systematically characterizing the well-known mouse-derived mAb 735 (mo735; ref. 15) and a newly created chimerized human derivative (ch735), both of which target the oncodevelopmental carbohydrate antigen polysialic acid (polySia). PolySia is a unique glycan homopolymer of α2,8-linked N-acetyl neuraminic acid (NeuNAc) that occurs as a terminating structure on the N-linked glycan associated with the neural cell adhesion molecule (NCAM) and also as a capsular polysaccharide on the surface of bacterial pathogens causing meningitis (16). In vertebrates, the expression of polySia is abundant during early stages of development of the brain, heart, kidney, liver, pancreas, respiratory, and digestive tracts, but becomes significantly reduced in adults with expression largely restricted to certain regions of the brain (16). Importantly, it is aberrantly reexpressed in many cancers, appearing as part of the tumor glycocalyx in small-cell lung cancer (SCLC; ref. 17), non–small cell lung cancer (NSCLC; ref. 18), pancreatic cancer (19), Wilms' tumor (20), neuroblastoma (21), and glioma (22) among others. PolySia expression, which is catalyzed by two polysialyltransferases, ST8SiaIV (PST) and particularly ST8SiaII (STX) in cancer cells (18), is known to promote cancer cell adhesion, migration, and invasion (22–24) and is strongly correlated with aggressive and metastatic disease as well as poor prognosis in the clinic (25). For many of the aforementioned reasons, polySia was ranked as the second highest priority glycan antigen (after GD2) in a National Cancer Institute pilot project (26).
To investigate polySia targeting and its clinical potential, the mo735 and ch735 antibodies were subjected to a spectrum of biochemical and cell biological assays to characterize their polySia-binding properties. Importantly, both antibodies were observed to bind polySia with high affinity and exquisite selectivity. We also confirmed that both antibodies recognized several polySia-positive tumor cell lines in vitro and induced rapid internalization of polySia into endosomal and lysosomal compartments. In light of these findings, we hypothesized that the antibody-induced endocytosis of polySia receptors could be efficiently harnessed as part of an antitumor therapeutic strategy. To test this notion, we engineered an ADC using a bioorthogonal reaction scheme for stably linking the chimeric human ch735 mAb to the microtubule-inhibitory agent maytansinoid DM1, which has previously been developed as the cytotoxic payload in trastuzumab emtansine (T-DM1) for HER2-positive breast cancer (27). The resulting conjugate was found to exert potent target-dependent cytotoxicity against polySia-positive tumor cells in vitro, providing compelling proof-of-concept for the use of polySia receptor internalization as a carrier for delivery of cytotoxic payloads to cancer cells. Taken together, our findings add to the growing body of literature implicating aberrant glycans in the tumor glycocalyx as an attractive collection of targets for the development of glycan-directed synthetic immunotherapies.
Materials and Methods
Construction of chimeric human mAb ch735
The DNA sequences for the VH and VL domains of mAb mo735 (28) were obtained from the GenBank/EBI Data Bank (accession number AB821355) and ordered from GeneArt Gene Synthesis (Thermo Fisher Scientific). The variable regions of mAb 735 were then swapped with the existing variable regions in pVITRO1-Trastuzumab-IgG1/k (Addgene plasmid #61883) as previously described to generate the vector pVITRO-735-IgG1/k (29). Briefly, polymerase incomplete primer extension (PIPE) PCR was performed using sets of primers (Supplementary Table S3) to generate four linear fragments of the construct with 5′ PIPE overhangs. All cloned plasmids were confirmed by DNA sequencing.
Cell culture and reagents
Production of recombinant mAbs was performed using FreeStyle 293-F cells (Thermo Fisher Scientific). FreeStyle 293-F cells were maintained in FreeStyle 293 expression medium (Thermo Fisher Scientific). Cancer cell lines SH-SY5Y, H82, H69, K562, MCF7, and SKOV3 were obtained from ATCC while cell lines SW2 and A549 were kindly provided by Dr. Karen Colley (University of Illinois at Chicago, Chicago, IL). SH-SY5Y cells were maintained in high-glucose DMEM/F12 medium supplemented with 10% Hyclone FetalClone I serum (VWR), 1% MEM nonessential amino acids solution (Thermo Fisher Scientific), penicillin (100 U/mL), and streptomycin (100 μg/mL; Thermo Fisher Scientific). H82, H69, K562, and A549 cells were maintained in RPMI1640 with l-glutamine (Thermo Fisher Scientific) supplemented with 10% Hyclone FetalClone I serum, penicillin (100 U/mL), and streptomycin (100 μg/mL). MCF7 cells were maintained in high-glucose DMEM supplemented with 10% Hyclone FetalClone I serum, insulin (10 μg/mL, Sigma), penicillin (100 U/mL), and streptomycin (100 μg/mL). SKOV3 and SW2 cells were maintained in high-glucose DMEM supplemented with 10% Hyclone FetalClone I serum, penicillin (100 U/mL) and streptomycin (100 μg/mL). All cell lines were maintained at low passage numbers and routinely checked for Mycoplasma by PCR according to standard procedures.
Expression and purification of ch735 and trastuzumab
293-F cells cultured in FreeStyle 293 Expression Medium (Thermo Fisher Scientific) were transfected with pVITRO-735-IgG1/k or pVITRO1-Trastuzumab-IgG1/k using FreeStyle MAX transfection reagent (Thermo Fisher Scientific) according to the manufacturer's instructions and selected under hygromycin B as described previously (29). Purified plasmid DNA was precipitated by mixing 1/10 the volume of 3 mol/L sodium acetate pH 5.2 and 2–3 volumes of 100% ethanol and freezing at −80°C for 2 hours. The DNA was collected by centrifugation at 13,000 × g at 4°C for 30 minutes and resuspended in 100 μL of sterile tissue culture grade water (Thermo Fisher Scientific). After selection, cultures were expanded to 1 L culture volume and maintained with 50% hygromycin B (25 μg/mL). Supernatants were harvested every 48 hours, centrifuged at 1,000 × g for 15 minutes, passed over 0.2-μm filters (VWR), and stored at 4°C until use.
Protein A/G agarose (Thermo Fisher Scientific) was used to purify antibodies from the supernatant according to the manufacturer's recommendations. The agarose equilibrated with 10-mL PBS in a polypropylene gravity column. The supernatant was then allowed to completely pass through the column. The column was then washed with PBS until there was no signal in the flow through at an absorbance of 280 nm (Abs280). Antibodies were eluted from the column with 0.1 mol/L glycine-HCl (pH 2.0) in 1-mL fractions and neutralized with 100 μL 1 mol/L Tris (pH 8.0). Antibody purity was evaluated by SDS-PAGE under reducing and nonreducing conditions and visualized by staining with Coomassie Blue G-250. Protein A/G-purified antibodies were analyzed by size exclusion chromatography (SEC) on a 4.6 mm ID x 30 cm TSKgel SuperSW3000 SEC column with 4-μm particles. Pure antibodies were eluted from the column at 8.66 minutes in 0.1 mol/L phosphate buffer containing 0.1 mol/L Na2SO4, pH 6.7.
Costar 96-well ELISA plates (Corning) were coated overnight at 4°C with 50 μL of 1 μg/mL chicken brain NCAM (Millipore) or endoN-treated NCAM in PBS. Chicken NCAM (Millipore, AG265) was digested with 1.5 μg of endoN per 50 μg of NCAM overnight at 37°C. After blocking with 5% (w/v) milk in PBS for 1–3 hours at room temperature or overnight at 4°C, ELISA plates were washed three times with wash buffer (PBST with 0.3% BSA) and incubated with serially diluted purified ch735, mo735 (Absolute Antibody), or ab5032 (Millipore) for 1 hour at room temperature. Antibody samples were quantified with a Nanodrop. After washing three times with wash buffer, 100 μL of 1:5,000-diluted rabbit anti-human IgG (Fc) antibody–HRP conjugate (Thermo Fisher Scientific), goat anti-mouse IgG-HRP (Abcam), or goat anti-rabbit IgG-HRP (Abcam) in wash buffer was added to each well for 1 hour. Plates were washed and developed with TMB (Thermo Fisher Scientific) using standard protocols.
Specificity profiling using glycan and glycoprotein arrays
Specificity of mo735 and ch735 was determined using printed glycan arrays 5.3 and 5.4 at the CFG (http://www.functionalglycomics.org/static/consortium/resources/resourcecoreh.shtml). Both antibodies were analyzed at 10 μg/mL with 5 μg/mL of anti-mouse or anti-human Alexa-Fluor 647 (AF647)-conjugated secondary, respectively. Specificity of mo735 (1 μg/mL) and ch735 (10 μg/mL) was also assessed on a custom glycoprotein array that contained approximately 40–50 glycoproteins including chicken brain NCAM and endoN-treated chicken brain NCAM.
Surface plasmon resonance
Equilibrium binding-affinity measurements were made by surface plasmon resonance (SPR) analysis on a Biacore 3000 system. Antibodies mo735 and ch735 were bound to the surface of a Protein A sensor chip with a target level of 1,700 response units (RU). Serial dilutions of the antigen, chicken NCAM, prepared in 10 mmol/L HEPES, pH 7.4, 150 mmol/L NaCl, 3 mmol/L EDTA, 0.005% polysorbate-20 (HBS-EP buffer, GE Healthcare) at concentrations ranging from 0.25 to 250 nmol/L were injected over the chip using the same buffer at a flow rate of 20 μL/minute (10-minute injection time, 2-minute stabilization time, 20-minute dissociation). The surface of the chip was regenerated between the injections of each serial dilution with 10 mmol/L glycine, pH 1.5 (30-second injection time, 3-minute stabilization time). Kinetic parameters were determined by fitting the maximum response values for each concentration using the Hill slope nonlinear regression analysis in Prism software.
CRISPR-Cas9 genome editing
CRISPR guide RNAs targeting ST8SiaII (crRNA1: ATGCAGTGCGCACGTTGACG) and ST8Sia4 (crRNA1: ACCCGATGAGTTGCGTCTCC) were purchased from Genscript in the pLentiCRISPR v2 vector. Knockout cell lines were generated in SW2 cells using protocols described by Zhang and colleagues (30). Briefly, after lentiviral transduction, cells were maintained under selection with 1 μg/mL puromycin for at least 14 days prior to analysis by flow cytometry.
Flow cytometric analysis
Cancer cells were trypsinized and collected with media, followed by three washes in PBS. To remove polySia, cells were treated with endoN at 3 μg/mL in PBS for 1 hour at room temperature. The cells are resuspended in 4% paraformaldehyde, fixed at room temperature with constant agitation for 10 minutes, and then washed two times with PBS, and two times with 0.5% BSA in PBS. The cells were collected and resuspended to 1 × 106 cells/100 μL and pipetted into a round-bottom 96-well plate. The cells were pelleted in the 96-well plate and resuspended in 0.5% BSA in PBS containing mo735 or ch735 (5 μg/mL) and incubated for 30 minutes at room temperature with constant agitation. Cells were washed three times with 0.5% BSA in PBS and resuspended in anti-mouse IgG-Alexa-Fluor 488 (AF488) secondary or anti-human IgG-AF488 secondary (Thermo Fisher Scientific) at a 1:200 dilution for 30 minutes at room temperature in the dark with constant agitation. Cells were washed three times, resuspended in 500 μL of 0.5% BSA in PBS, and analyzed on a BD FACSCalibur flow cytometer using Cell Quest Pro software (BD Biosciences).
Adherent cells were plated at 20,000 cells/cm2 on poly-l-lysine coated 35-mm glass bottom dishes and adhered overnight. To remove polySia, cells were then treated with endoN at 3 μg/mL in cell culture media overnight. Suspension cells were collected on the day of the experiment and labeled in suspension with the same protocol as the adherent cells. Cells were fixed with 4% paraformaldehyde and subsequently blocked with 5% normal goat serum PBS (NPBS) for 1 hour at room temperature. Antibodies mo735 and ch735 were diluted to 5 μg/mL in 5% NPBS and incubated overnight at 4°C. Anti-mouse and anti-human AF488-conjugated secondary antibodies A32723 and A11013 (Thermo Fisher Scientific) were diluted 1:200 in NPBS and incubated for 2 hours at room temperature. Wheatgerm agglutinin-AF647 (WGA-647) was diluted to 1 μg/mL in NPBS and incubated for 10 minutes at room temperature. Hoescht dye was used at 1 μg/mL in PBS for 5 minutes at room temperature. Samples were imaged on a Zeiss LSM inverted 880 confocal microscope using a 40× water immersion objective.
The avidin–biotin complex (ABC) immunoperoxidase method was performed essentially as described previously (31). Briefly, the sections were quenched with 0.1% H2O2 in PBS for 15 minutes, blocked with avidin and biotin reagents (Vector Laboratories) for 10 minutes each, incubated in 10% serum from which the second antibody was raised and incubated with mAb 735 at 1 μg/mL for 1 hour. This concentration was selected on the basis of strong reactivity against known positive target cells and little or no background against stroma. The sections were subsequently incubated with biotinylated secondary antibodies for 30 minutes, and then incubated in ABC reagent per manufacturer's protocols (ABC Kit, Vector Laboratories, PK-6102) for 30 minutes. Reactions were developed with liquid DAB+ Substrate Chromogen System (Dako, catalog no. K3468) for 3 minutes at room temperature. Slides were then counterstained with Mayer's Hematoxylin (Dako Cyomation, catalog no. S3309) for 1 minutes at room temperature. The immunoreactivities were graded on the basis of the percentage of positive cells and staining intensity above that seen on the negative control. Known positive and negative control slides were used in each experiment.
To calculate percent internalization, prechilled cells were incubated with 50 nmol/L ch735 on ice for 1 hour and then washed to remove unbound antibodies. For each time point, one aliquot of cells remained on ice and one was incubated at 37°C for 15, 30, or 60 minutes. Cells were fixed in 2% paraformaldehyde for 20 minutes and then stained with AF488-labeled antibody against human IgG and analyzed by flow cytometry and FlowJo software. Receptor–antibody complex internalization was calculated using the geometric mean as percent fluorescent intensity loss at 37°C relative to that on ice. For each sample, the geometric mean fluorescence intensity (MFI) of 10,000 cells was measured in triplicate.
SH-SY5Y or SW2 cells were plated at 20,000 cells/cm2 on poly-l-lysine–coated 35-mm glass bottom dishes and adhered overnight. To remove polySia, cells were treated as described above. To measure receptor–antibody internalization, cells were incubated with 150 nmol/L AF488-labeled ch735, human IgG isotype control (Thermo Fisher Scientific), mo735, or mouse IgG isotype control (anti-MBP mAb, NEB) and 100 nmol/L AF647 transferrin or AF647 anti-LAMP-3 antibody (Santa Cruz Biotechnology) for 1 hour. Cells were washed and then fixed as described above. To examine lysosomal trafficking, cells were incubated with 150 nmol/L of ch735 or isotype at 37°C for 2 hours, washed, fixed, and then permeabilized using 0.1% Triton X-100 NPBS. Cells were stained with AF488-labeled antibody against human IgG to visualize antigen–antibody complex (Thermo Fisher Scientific) and mouse anti-human LAMP-1 clone D2D11 (Cell Signaling Technology) followed by AF647-labeled anti-rabbit IgG to visualize the lysosomes (Thermo Fisher Scientific). Hoescht dye was used at 1 μg/mL in PBS for 5 minutes at room temperature. Samples were imaged on a Zeiss LSM inverted 880 confocal microscope using a 40× water immersion objective. For colocalization analysis, a 5-μm line was drawn across the apparent vesicles. The fluorescence intensity of the plot profile was analyzed using FIJI software. Fluorescence intensity was normalized to the maximum value for each channel.
Drug linker synthesis
To synthesize the drug linker, 650 μg of DM1 was incubated at 1.1 mmol/L with 3 molar equivalents of maleimide-PEG3-TCO (Click Chemistry Tools) in 3:1 DMSO:PBS overnight at room temperature. The reaction mixture was purified on a C18 analytic RP-HPLC column on a gradient of 5%–95% acetonitrile in water over 30 minutes. The product was dried, resuspended in DMSO, and quantified via Abs252 measurements (ext. coeff252nm = 26,790 mol/L cm−1). Product mass was verified via LC/MS (expected mass = 1,260.56).
Purified ch735, purified trastuzumab, or human IgG1 isotype control (Thermo Fisher Scientific) was reacted with 10 molar equivalents of methyltetrazine-PEG4-NHS ester overnight at 37°C. Excess reagent was removed by centrifugation dialysis. The Tz-conjugated antibody was then incubated with 3 molar equivalents of TCO-maleimide-DM1 drug linker for 5 hours at 37°C. Excess reagent was removed by centrifugation dialysis. Average DAR was determined using absorbance spectroscopy to calculate the concentrations of antibody and drug (32). The following previously established extinction coefficients were used for each component: ϵ280 DM1 = 5,700 mol/L−1 cm−1, ϵ252 DM1 = 26,790 mol/L−1 cm−1, and ϵ280 Antibody = 218,134 mol/L−1 cm−1, ϵ252 Antibody = 76,565 mol/L−1 cm−1 (33).
Cell viability assay
SKOV3, MCF7, and SH-SY5Y cells were plated at 5,000, 2,500, and 2,500 cells/well, respectively, and allowed to rest for 24 hours. Five-fold serial dilutions of the antibodies were added starting at 150 nmol/L and incubated for 72–144 hours. The viability assays were then developed using Alamar blue according to manufacturer's protocol (Bio-Rad). Percent viability is calculated by first subtracting the value of media alone from all samples. Subsequently, the resulting values are divided by the values measured from an untreated control representing maximum viability.
Construction and characterization of a chimeric human IgG targeting polySia
To generate a mAb that is more compatible with targeting human cancers, we converted the fully mouse IgG2a mAb 735 (mo735) into a chimeric human IgG1 (ch735) by swapping the variable regions according to an antibody cloning and expression method described by Beavil and coworkers (29). Using this approach, a stable cell line was generated that was capable of producing fully assembled ch735, which could be purified to near homogeneity at yields up to 6 mg/L (Supplementary Fig. S1A–S1C). Subsequent ELISA analysis confirmed that both antibodies bound chicken brain–derived polysialylated neural cell adhesion molecule (NCAM), but not NCAM that was treated with endoneuraminidase N (endoN) that selectively removes polySia (Fig. 1A; Supplementary Fig. S2A). Probing of similarly prepared ELISA plates with an NCAM-specific antibody confirmed that both NCAM and endoN-treated NCAM were equally coated on ELISA plates (Supplementary Fig. S2B). Given the strict specificity of endoN for α2,8-linkages in sources as disparate as bacterial and neural membrane glycoconjugates, we conclude that both mo735 and ch735 specifically recognize polySia.
To investigate glycan specificity, we also analyzed both antibodies on a glycoprotein array that contained approximately 50 glycoproteins and the current glycan array (version 5.3 for mo735, version 5.4 for ch735) of the Consortium for Functional Glycomics (CFG) that contained approximately 600 natural and synthetic mammalian glycans (http://www.functionalglycomics.org). The chimeric human IgG1 ch735 showed a strong preference for polysialylated NCAM in the glycoprotein array (Fig. 1B; Chart ID #17) and the tetrasialic acid containing glycan GalNAcβ1-4(Neu5Acα2-8Neu5Acα2-8Neu5Acα2-8Neu5Acα2-3)Galβ1-4Glcβ-Sp0 in the glycan array (Fig. 1C and D; Chart ID #223 in microarray version 5.4). A lesser but still significant level of binding above background was detected for GalNAcβ1-4(Neu5Acα2-8Neu5Acα2-8Neu5Acα2-3)Galβ1-4Glcβ-Sp0 (Chart ID #224), but not for the closely related glycans #225 and 226 that differed from 224 by a single branched GalNac and Neu5Ac, respectively (Fig. 1D). In total, the glycan microarray included 145 glycans containing some form of sialic acid, often as the terminal sugar, of which 16 were α2,8-linked (Supplementary Table S1); hence, we conclude unequivocally that the specificity of ch735 is for α2,8-linked sialic acid with a degree of polymerization (DP) of three or greater. In line with this conclusion, there was no significant signal toward any other glycomolecules including endoN-treated NCAM that was spotted on the glycoprotein array. Nearly identical results were observed when the same microarrays were probed with mo735 (Supplementary Fig. S3A–S3C).
To determine affinity, we measured the equilibrium binding of both antibodies to polysialylated NCAM using SPR. Binding values were fit using the specific binding with Hill slope analysis in Prism software and the calculated Kd values for mo735 and ch735 were determined to be 10.22 and 4.79 nmol/L, respectively (Supplementary Fig. S4A and S4B). These values were in close agreement with the previously reported Kd of approximately 5 nmol/L for the mouse IgG against embryonic brain glycopeptides (34).
Antibody ch735 binds surface polySia and is internalized in cancer cells
To demonstrate the relevance of this antibody in the context of human cancers, we used flow cytometric analysis to assess ch735 binding to polySia expressed on the surface of different cancer cell lines including the small-cell lung cancer (SCLC) cell lines SW2, NCI-H69, and NCI-H82, as well as neuroblastoma cell line SH-SY5Y, non–small cell lung cancer (NSCLC) cell line A549, breast cancer cell line MCF7, ovarian cancer cell line SKOV3, and chronic myeloid leukemia (CML) cell line K562. Antibody ch735 bound most avidly to SW2 cells, and also recognized NCI-H69, NCI-H82, and SH-SY5Y cancer cells although with a lower intensity (Fig. 2A), in agreement with previous cell line characterization studies (35–37). Inspection of the representative histograms revealed not only differing levels of surface expression between the different polySia-positive cancer cells, but also within each population especially for H69 cells (Supplementary Fig. S5A). The ability of ch735 to recognize polySia on the surfaces of these cells was corroborated by immunofluorescence microscopic images of the polySia-positive cell lines (Fig. 2B). When A549 and MCF7 cancer cells were similarly probed with ch735, we observed no significant binding above background, in agreement with previous studies (24, 35). Likewise, SKOV3 and K562 cells were not recognized by ch735, establishing these as polySia-negative cell lines. It should also be noted that staining of each of these cell lines with mo735 yielded virtually identical polySia-binding results (Supplementary Figs. S5B and S6A and S6B).
Importantly, surface binding was completely abrogated following endoN treatment, confirming that binding to these cell lines was specific to polySia. Further evidence of polySia-specific binding was demonstrated by a significant decrease in antibody labeling of SW2 cells in which the polysialyltransferases ST8SiaII and ST8SiaIV were knocked out by CRISPR-Cas9 gene editing (Supplementary Fig. S7A–S7C). In addition, IHC revealed strong staining of polySia in formalin-fixed, paraffin-embedded (FFPE) human tissue sections of SCLC (Fig. 2C), but little to no staining of the adjacent normal tissue except for the bronchial epithelial cells and alveolar macrophages (Fig. 2D), in close agreement with previous findings (31).
Given that clathrinmediated endocytosis is an essential pathway by which many glycoproteins are recycled or downregulated (38), we next investigated whether polySia undergoes a similar internalization process. Previous studies demonstrated that NCAM, one of the major carriers of polySia, was recycled by a clathrin-dependent endocytosis process (39, 40), whereas polySia was only detectable at the cell surface (35, 41) unless internalization was activated by the ECM (42). To further investigate this issue here, each cell line that expressed cell surface polySia was evaluated for the ability to internalize polySia. This involved first binding ch735 to the surface of tumor cells at 4°C, after which an aliquot of cells remained at 4°C while the rest were incubated at 37°C and analyzed by flow cytometry at different time points. For each of the polySia-positive cell lines, we observed that approximately 40% of the antibody was internalized after 1 hour while no internalization was observed for the MCF7 cell line (Fig. 3A), which were previously found to lack polySia at the cell surface (35). A time course of ch735 binding to SH-SY5Y cells revealed that antibody internalization occurred rapidly, with approximately 30% of the antibody internalized as early as 15 minutes and maximum internalization of 40% reached by 30 minutes (Fig. 3B). In contrast, an isotype control antibody showed no measurable internalization over the same time period. It is noteworthy that the internalization percentage and rate observed here with ch735 was on par with that reported previously with trastuzumab against HER-2–positive cancer cells (43). It should also be noted that comparable internalization of mAb mo735 into SW2 cells was observed, with intracellular fluorescence increasing as a function of polySia-specific antibody concentration and as a function of time (Supplementary Fig. S8A–S8D).
Confocal microscopy was used to investigate the compartments where the ch735 mAb accumulated after internalization using markers of early endosomes, recycling endosomes, or late endosome/lysosomes. Consistent with flow cytometry, ch735 initially labeled the plasma membrane of SH-SY5Y cells and after 1 hour at 37°C was internalized, where it clearly colocalized with early endosomal and recycling endosomal marker transferrin (Fig. 3C) and late endosomal marker LAMP-3 (Fig. 3D). Accumulation of the ch735 mAb was also observed in late endosomal/lysosomal LAMP-1–positive compartments (Fig. 3E). As expected, no detectable binding, internalization, or colocalization was observed for the isotype control (Supplementary Fig. S9A). Similar to ch735, the mo735 mAb compartmentalized in early and recycling endosomes as confirmed by colocalization with transferrin and LAMP-3 (Supplementary Fig. S9B and S9C). On the basis of these data, we conclude that mAb ch735 binds to tumor cell membranes in a target-specific manner, thereby inducing a subpopulation of bound antibodies to become rapidly internalized in endosomal/lysosomal compartments.
Glycan-directed ADC is cytotoxic against tumor cells expressing polySia
Given that ch735 induced internalization of polySia receptors in cancer cells, we next evaluated whether drug conjugation could be used to confer target-specific in vitro cytotoxicity to mAb ch735. To this end, we proposed a covalent, bioorthogonal reaction scheme between a tetrazine (Tz) and a trans-cyclooctene (TCO) as a means of linking ch735 to the cytotoxic maytansinoid DM1 that inhibits the assembly of microtubules (Fig. 4A). We chose DM1 because it has been used successfully in other ADCs including T-DM1, an FDA-approved ADC for HER2-positive breast cancer (44). DM1 is membrane permeable and travels to the cytosol, its site of action, following lysosomal degradation of the ADC. Here, a trans-cyclooctene (TCO)-maleimide-DM1 noncleavable drug linker was chemically synthesized (Supplementary Fig. S10A–S10C). Following synthesis, TCO-maleimide-DM1 was conjugated to Tz-modified ch735, forming a 1,4-dihydropyrazine (Py) linkage between the two and typically resulting in drug-to-antibody ratios (DAR) of approximately 2–3 (Supplementary Table S2). To evaluate in vitro cytotoxicity, we treated SH-SY5Y cells with ch735-Py-DM1 and then examined cell viability. The ch735-Py-DM1 conjugate, but not the isotype-Py-DM1 control, showed polySia-specific cell killing of SH-SY5Y cells, and neither showed any cytotoxicity toward MCF7 cells (Fig. 4B). For comparison, we similarly prepared a conjugate between Tz-modified trastuzumab and DM1 (T-Py-DM1) and found that it killed HER2-positive SKOV3 cells to an extent that was similar to ch735-Py-DM1 against SH-SY5Y cells (Fig. 4C). Importantly, the comparable target-specific potency that we measured for ch735-Py-DM1 relative to T-Py-DM1 (IC50 values of 17 and 23 nmol/L, respectively; Supplementary Table S2) reveals the therapeutic potential of this glycan-directed ADC against polySia-positive cancers including neuroblastoma, small-cell and non–small cell lung carcinomas, multiple myeloma, and Wilms' tumor.
Here, we sought to expand the knowledge base surrounding cell surface polySia and affirm its potential as a target for antibody-based cancer therapy. PolySia is a rare posttranslational modification that is found on a select group of identified carrier proteins including NCAM, SynCAM-1, Neuropilin-2, and the voltage-sensitive sodium channel α subunit (16). To create a more therapeutically relevant polySia-directed antibody, we engineered a chimerized human mAb, ch735, that was based on mouse-derived mo735, and determined that it recognized polySia with low nanomolar affinity and exquisite selectivity, binding α2,8-linked polySia structures with a DP of three sugar units or greater. It should be pointed out that previous experiments using SPR and ITC showed that mo735 prefers polySia chains of at least 8–11 sialic acid residues with increasing affinity as length increases (34), in line with our glycoprotein microarray results with polySia-NCAM that has a DP of approximately 50 (34). The binding to much shorter polySia chains that we observed with the glycan microarray could be due to differences in sensitivity and/or in how the immobilized glycans were presented to the antibody for binding (i.e., clustered). It is worth noting that a single-chain Fv (scFv) antibody derived from mo735 was observed to bind shorter α2–8-linked sialic acids (DP ∼3; ref. 28), which likely explains why ch735 and mo735 both bind to the shorter polySia structures on the glycan array. It was proposed by these authors that mo735 recognizes sialic acid trisaccharide units in a paired manner and that this lends itself to higher affinities for longer chains.
Using this chimeric human mAb, we confirmed high levels of polySia expression on several different cancer cell lines, in agreement with earlier findings that this aberrant glycan is abundantly expressed on human cancers. Importantly, polySia-positive tumor cells were observed to rapidly internalize ch735 in endosomal and lysosomal compartments. In this regard, it should be pointed out that NCAM, one of the major polySia carrier proteins, is well known to undergo internalization via the clathrin-dependent endocytic pathway in astrocytes, cortical neurons, and rat neuroblastoma cells (39, 40). However, studies of Wilms' tumor revealed that while NCAM was similarly present in intracellular compartments from the nuclear membrane to the plasma membrane, polySia was only detectable at the cell surface (41). Likewise, polySia was found exclusively on the surface of SW2 cells (35). A more recent investigation of polySia turnover in human rhabdomyosarcoma cells reported that small numbers of polySia-NCAM molecules were recurrently found colocalizing with Rab5 (early endocytic marker), but only upon activation by the ECM (42). The absence of detectable constitutive internalization of polySia in these studies leads us to believe that the rapid internalization following ch735 binding observed here is an instance of antibody-induced receptor internalization (45). Interestingly, polySia-binding Escherichia coli bacteriophages were similarly reported to induce endocytosis of polySia in human neuroblastoma cells, whereas polySia remained at the cell surface if no phage was added (46).
The ability of experimental and therapeutic antibodies to induce endocytosis of their antigens is a commonly observed phenomenon that has been leveraged as a strategy to internalize oncogenic (or survival-mediating) antigens for eliciting antitumor effects or to deliver cytotoxic payloads directly into cancer cells (45). In the case of the latter, polySia possesses a number of attributes that make it an ideal target for an ADC including: (i) it is abundantly and selectively expressed on cancer cells as discussed above; (ii) it is not detected in extracellular supernatants (42), and the NCAM ectodomains that are shed from the cell surface are devoid of polySia (47); (iii) it possesses an appropriate rate of endocytosis, comparable with that measured previously for trastuzumab (43); and (iv) it is trafficked to the endolysosomal degradation pathway and retained in a maturing endosome (rather than being recycled back to the plasma membrane; ref. 42) until finally being delivered to the lysosome, an appropriate intracellular trafficking route when using a noncleavable linker (48). To harness these traits, we synthesized an ADC in which Tz-modified ch735 was bioorthogonally conjugated to the TCO-maleimide-DM1 drug linker. The resulting ch735-Py-DM1 conjugate exhibited potent polySia-specific cytotoxicity in vitro, rivaling the potency of a similarly synthesized T-Py–DM1 conjugate. To our knowledge, this is the first ADC that targets an N-linked glycan epitope on the surface of cancer cells and one of few to leverage the Tz/TCO bioorthogonal click chemistry described here. The relative ease of component synthesis and fast reaction kinetics of this two-step method allows for rapid generation of ADCs against new targets. In addition, the aromatic stability of the pyridazine product formed could aid in stability (49). This is even more significant when one considers that unconjugated (“naked”) mo735 exhibited only limited complement-dependent cytotoxicity (CDC) against cultured neurons (50) while mAb 5A5, a polySia-specific IgM, exhibited no measurable CDC against several different SCLC cell lines (37).
It is worth mentioning that the IC50 value measured for ch735-Py-DM1 compared favorably to a number of previously reported ADCs against protein antigens including HER2 and NCAM, as well as a small handful of ADCs that target cell surface O-glycans including STn, Tn, and T, the blood group–related Lewis Y antigen (Supplementary Table S2). This latter group, together with our ch735-Py–DM1 conjugate, represents a new class of glycan-directed ADCs that hold promise for antitumor therapy. We anticipate that the availability of antibodies such as ch735 that recognize aberrantly expressed tumor glycans should aid the development of novel glycan-directed synthetic immunotherapies for specifically focusing immune or immune-like responses on the tumor glycocalyx. While we focused here on engineering a glycan-specific ADC, it is envisioned that molecular reformatting of antibodies or antibody domains could be used to create next-generation glycan-directed immunotherapies including bispecific antibodies or CAR-T cells.
Collectively, the results described here validate polySia as a therapeutically tractable target for ADC and pave the way for achieving selective cytotoxic effects against tumors that aberrantly express this unique oncodevelopmental antigen. The choice of polySia as a therapeutic target is supported by the fact that polySia is expressed throughout the fetus and during embryonic development, but in adults polySia expression is highly restricted (16, 31). Specifically, according to previously published IHC results, the mo735 mAb reacted with only a limited number of cells and tissues including gray matter of brain, bronchial epithelia, and pneumocytes, and capillary endothelial cells and ganglion neurons in the colon (31). Importantly, polySia is reexpressed in many types of cancer including SCLC (17), NSCLC (18), pancreatic cancer (19), Wilms' tumor (20), neuroblastoma (21), and glioma (22), and its increased expression typically correlates with later stages and increased invasive and metastatic potential (25). While recent reports indicate that polySia is also expressed on certain human immune cells (51, 52), this expression appears to be quite heterogenous and is progressively downregulated in wild-type monocytes and monocyte-derived cells during migration from bone marrow through peripheral blood to pulmonary and peritoneal sites of inflammation, with levels in peripheral blood and inflammation sites reported to be extremely low or absent relative to the levels detected in bone marrow (53). Regarding polySia's occurrence on healthy cells, it should be pointed out that Schneerson and colleagues conducted a thorough review of published data looking for evidence that anti-polySia IgG antibodies caused immunopathology in humans. From their study, they found no evidence of increased autoimmunity and urged that the use of anti-polySia immunotherapies be considered (54). We concur that further studies on the therapeutic targeting of polySia and the safety of such an approach are highly warranted, especially in light of the results presented here.
Disclosure of Potential Conflicts of Interest
M.P. DeLisa has ownership interest (including stock, patents, etc.) in Glycobia and Versatope. No potential conflicts of interest were disclosed by the other authors.
The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH, NCI, or the National Institute of General Medical Sciences.
Conception and design: E.C. Cox, J.H. Merritt, M.P. DeLisa
Development of methodology: E.C. Cox, D.N. Thornlow, M.J. Paszek, M.P. DeLisa
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E.C. Cox, D.N. Thornlow, M.A. Jones, J.L. Fuller, C.A. Alabi, M.P. DeLisa
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E.C. Cox, M.A. Jones, J.H. Merritt, M.P. DeLisa
Writing, review, and/or revision of the manuscript: E.C. Cox, M.J. Paszek, C.A. Alabi, M.P. DeLisa
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.P. DeLisa
Study supervision: M.P. DeLisa
We thank Dr. Gaurang Bhide and Dr. Karen Colley (University of Illinois at Chicago, Chicago, IL) for SW2 and A549 cell lines as well as purified endoN used in this work. We thank Carolyn Shurer (Cornell University, Ithaca, NY) for technical assistance and training in confocal microscopy. We thank Cureline for performing IHC and Dr. Andrew Miller (Cornell University, Ithaca, NY) for assistance with the IHC data analysis. We thank Dr. Cynthia Leifer (Cornell University, Ithaca, NY) for technical assistance with endocytosis experiments. We thank Dr. Brian Green and Dr. Joshua Wilson (Glycobia, Inc.) for helpful discussions of the manuscript. This work was supported by the Defense Threat Reduction Agency (GRANT11631647 to M.P. DeLisa), the National Science Foundation (CBET-1605242 to M.P. DeLisa), Cornell University Startup funds (to C.A. Alabi and M.J. Paszek), the National Cancer Institute (U54 CA210184 to M.J. Paszek), and the Nancy and Peter Meinig Investigator Fellowship (to C.A. Alabi). The work was also supported by seed project funding through the National Institutes of Health-funded Cornell Center on the Physics of Cancer Metabolism (supporting grant 1U54CA210184-01). E.C. Cox was supported by an NIH Chemical-Biology Interface (CBI) training fellowship (supporting grant T32GM008500). D.N. Thornlow was supported by a National Science Foundation Graduate Research Fellowship and a Cornell Fleming Graduate Scholarship. We acknowledge instrumentation grants NYSTEM C029155 and NIH S10OD018516 for the Zeiss LSM880 microscope. We acknowledge the participation of the Protein-Glycan Interaction Resource of the CFG (supporting grant R24 GM098791) and the National Center for Functional Glycomics (NCFG) at Beth Israel Deaconess Medical Center, Harvard Medical School (supporting grant P41 GM103694).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.