Cancer discrimination by on-cell N-glycan ligation

In vivo molecular imaging enables analysis of anticancer drug kinetics, accumulation, and tumor expression levels using modalities such as fluorescence imaging, MRI, and PET. RGD peptides target αβ3 integrins, which are highly expressed on tumor vasculature, and have been widely used for cancer imaging. However, because integrins are also expressed on normal endothelial tissues, conventional RGD probes often yield poor imaging contrast due to strong, non-selective binding (nM KD) and rapid receptor-mediated endocytosis. Conversely, weakly interacting probes (e.g., millimolar KD) lack sufficient binding for effective imaging even when their receptors are specific. The authors previously introduced a pre-targeting approach linking a high-affinity integrin-binding cyclic RGDyK (bearing an azide) with a weakly binding, fluorescent N-glycan (bearing DBCO) via strain-promoted azide–alkyne cycloaddition on the cell surface, thereby exploiting proximity between co-localized receptors to selectively label target cells. Building on this concept, the present study aims to discriminate αβ3 integrin-expressing cancerous and non-cancerous cells by optimizing combinations of RGD linkers and N-glycan structures and to translate these fingerprints to in vivo tumor imaging. Six cell lines (HeLaS3, A549, BxPC3, PC3, SW620, and TIG3) that all express αβ3 integrins were selected to test generalizability.

The study contextualizes prior work showing that RGD peptides are high-affinity ligands for αβ3 integrins and have been widely used in molecular imaging, though with limited selectivity due to integrin expression on normal tissues. The authors’ earlier proof-of-concept demonstrated selective imaging by in situ ligation of integrin-bound RGD and a weakly binding sialyl N-glycan. To interpret glycan-dependent cell labeling, the authors compiled a literature survey (Table 1) of lectins reported on the six examined cell lines and their glycan specificities (e.g., Siglecs for sialic acids, galectins for Gal, vimentin for GlcNAc, SP-D/DC-SIGN for Man). They note discrepancies between mRNA databases (Human Protein Atlas, RefEx) and protein-level lectin reports, emphasizing the need for biochemical validation of surface lectin expression.

Design and synthesis of ligands: Four cyclic RGDyK peptides (1a–1d) were prepared bearing an azide via PEG linkers of varying lengths (PEG3, PEG5, PEG7, PEG9) to modulate the effective distance for in situ click ligation. Five biantennary N-glycans were used as weak ligands: 2a (α2,6-sialylated), 2b (α2,3-sialylated), 2c (galactose-terminated), 2d (GlcNAc-terminated), and 2e (mannose-terminated). Fluorophores (TAMRA for in vitro imaging; Cy7.5 for in vivo imaging) and DBCO were attached to glycan reducing ends via short PEG linkers. Synthetic details, HPLC, MS, and NMR data are provided in Supplementary Methods and Figs. 1–15, 20.

Cell lines and culture: HeLaS3, A549, BxPC3, PC3, SW620, and TIG3 were sourced from ATCC/RIKEN/JCRB. HeLaS3, A549, BxPC3, PC3, and TIG3 were cultured in DMEM with 10% FBS and penicillin-streptomycin; SW620 in Leibovitz’s L-15 with 10% FBS and penicillin-streptomycin. Mycoplasma testing was routinely performed.

In vitro pre-targeting and imaging: Cells (5 × 10^4 per well) were seeded on 96-well plates. Pre-targeting: cyclic RGDyK 1a–1d (50 µM, 100 µL) incubated 15 min at room temperature; wash twice with medium. Click ligation: N-glycan ligands 2a–2e (100 µM, 25 µL) incubated 30 min at 4 °C to reduce integrin-mediated internalization and favor on-surface SPAAC. Controls included TAMRA-labeled RGDyK alone and glycans alone. After washing, cells were fixed with 4% paraformaldehyde (10 min) and nuclei labeled with Hoechst 33342. Imaging was performed on a Keyence BZ-X700 microscope. Fluorescence intensities were quantified with ImageJ, normalized to cell number. Each condition was repeated on four plates per cell line; data averaged (Supplementary Table 1). Additional optimization indicated higher internalization and reduced click efficiency at RT versus 4 °C.

In vivo xenograft imaging: BALB/c nude mice (8–10 weeks; n=4 per condition) were implanted subcutaneously with HeLaS3 (2 × 10^6 cells) or A549 (4 × 10^6 cells) in the left shoulder and allowed to grow for 3–4 weeks. Pre-targeting: intravenous injection of RGDyK 1d (150 nmol in 100 µL saline). After 30 min, intravenous injection of Cy7.5-labeled N-glycans 3b (α2,3-sialylated) or 3d (GlcNAc-terminated) (15 nmol in 100 µL saline). Three hours later, mice were dissected and tumor regions imaged on a PerkinElmer IVIS Spectrum (excitation 640 nm; emission 710 nm). Doses were scaled from in vitro conditions assuming ~2 mL blood volume to enable on-tumor pre-targeting and bioorthogonal ligation. Statistical analysis indicated significance at p < 0.05.

• Conventional TAMRA-labeled cyclic RGDyK vividly labeled all tested αβ3-expressing cell lines but failed to discriminate among them. Glycan ligands alone did not label cells.
• Pre-targeted in situ ligation of azide-RGDyK with DBCO–glycans produced distinct, glycan-structure- and linker-length-dependent labeling fingerprints that discriminated six cell lines: • HeLaS3: strong labeling with GlcNAc- and Man-terminated glycans (2d, 2e). • A549: selective labeling with α2,3-sialylated glycan (2b). • BxPC3: labeling with Gal-, GlcNAc-, and Man-terminated glycans (2c, 2d, 2e); Man (2e) required specific RGD linker (1c). • SW620: labeling with α2,6-sialylated, α2,3-sialylated, and GlcNAc-terminated glycans (2a, 2b, 2d). • TIG3 (non-cancerous): labeling with α2,6-sialylated, α2,3-sialylated, GlcNAc-, and Man-terminated glycans (2a, 2b, 2d, 2e); GlcNAc/Man staining required RGD 1c. • PC3: no effective combinations among the tested glycans.
• Linker length between RGDyK and azide significantly influenced labeling, indicating spatial constraints between integrins and lectins on the cell surface.
• In vivo xenograft validation: Using the cell-derived fingerprints, HeLaS3 tumors were selectively labeled with GlcNAc glycan 3d and A549 tumors with α2,3-sialyl glycan 3b after pre-targeting with RGDyK 1d. Matched RGD–glycan combinations yielded >2.5-fold higher tumor fluorescence than RGD or glycan alone or non-matched combinations (n=4; p < 0.05), despite some nonspecific tumor accumulation of fluorescent glycans.

The findings demonstrate that combining a strong integrin–RGD interaction with a weak glycan–lectin interaction via in situ bioorthogonal ligation enables selective recognition of specific cell types that share a common receptor (αβ3 integrin). The approach exploits receptor proximity and dynamic spatial organization on the cell surface; the dependence on RGD linker length reflects the relative positioning of integrins and target lectins. Discrepancies between expected lectin–glycan pairings (based on literature or mRNA expression) and observed labeling likely arise from cell-surface spatial arrangement, receptor dynamics, and contributions from additional weak interactions (e.g., hydrophobic or hydrogen bonding) with membrane components. The method therefore functions both as a discriminatory imaging strategy and as a chemical probe of receptor colocalization and nanoscale organization in living systems. In vivo results corroborate the in vitro fingerprints, validating that pre-targeted on-cell ligation can enhance tumor selectivity and contrast even in the presence of background accumulation by hydrophobic fluorophores.

This work introduces a generalizable pre-targeted imaging strategy that discriminates αβ3 integrin-expressing cancerous and non-cancerous cells by on-cell ligation of high-affinity RGD peptides and low-affinity N-glycans. Rapid 96-well screening identified optimal RGD linker–glycan combinations as cell-specific fingerprints, which successfully guided selective tumor imaging in xenografted mice with significant signal enhancement over controls. Beyond glycan–lectin pairs, the concept extends to other weak interactions on cell surfaces, offering a platform to leverage low-affinity ligands for cell discrimination, imaging, and potentially modulation of receptor organization. Future work could expand ligand libraries, map spatial constraints with diverse linker architectures, integrate multiplexed ligations to profile multiple receptors simultaneously, and apply to other receptor systems or therapeutic delivery.

• Some fluorescent N-glycans exhibited nonspecific tumor accumulation due to fluorophore hydrophobicity, elevating background signals.
• PC3 cells were not effectively discriminated by any tested glycan, indicating incomplete coverage of relevant lectins or suboptimal spatial/linker parameters.
• Lack of structural data (e.g., crystallographic or high-resolution spatial maps) for receptor pairs (integrin and specific lectins) limits precise mechanistic interpretation of optimal linker lengths and spatial constraints.
• Weak glycan interactions may involve non-lectin components (membrane proteins, glycans, lipids), complicating attribution solely to lectins.
• mRNA expression datasets did not consistently align with reported surface lectin presence, highlighting uncertainties in predicting targets from transcriptomics alone.