GlycoChat Uncovers Glycan–Lectin Circuits in the Tumor Microenvironment of Pancreatic Cancer

ABSTRACT

Aberrant glycosylation is a hallmark of cancer progression, yet its functional implications within the tumor microenvironment (TME) remain poorly understood. Here, a single‐cell glycomic profiling strategy was applied to tumor tissues from patients with pancreatic ductal adenocarcinoma (PDAC), enabling the identification of cell‐type‐specific glycome signatures and their dynamic alterations during the transition from classical to basal‐like cancer subtypes. To systematically decode glycan‐mediated communication in the TME, an analytical framework termed GlycoChat was developed for mapping global glycan–lectin circuits at single‐cell resolution. GlycoChat identified CLEC10A and SIGLEC3 as lectin receptors expressed on tumor‐associated macrophages that interact with cancer cell surface glycans. Functional assays demonstrated that cancer cells promote differentiation of immunosuppressive macrophages and impair phagocytic activity through interactions with CLEC10A and SIGLEC3. This study establishes GlycoChat as a powerful tool for dissecting glycan–lectin circuits in complex TME and highlights glyco‐immune checkpoints as potential targets for therapeutic intervention in PDAC.

Aberrant glycosylation drives cancer progression, yet its role in the tumor microenvironment remains unclear. We developed GlycoChat to map glycan–lectin circuits at single‐cell resolution. We discovered that cancer cells induce immunosuppressive macrophage differentiation and impair phagocytosis through interactions with CLEC10A and SIGLEC3, highlighting novel glyco‐immune checkpoints as promising therapeutic targets.

1. Introduction

Pancreatic ductal adenocarcinoma (PDAC) is among the most lethal malignancies, with a five‐year survival rate below 8% [1]. By 2030, PDAC is expected to become the second leading cause of cancer‐related deaths in the United States [2], highlighting the urgent need for novel therapeutic strategies. Despite advances in surgical and chemotherapeutic approaches, curative resection is feasible in only 15%–20% of patients, as most present with locally advanced or metastatic disease [3]. Immunotherapies, including immune checkpoint inhibitors, have shown limited efficacy in PDAC, largely due to its profoundly immunosuppressive tumor microenvironment (TME) [4]. These challenges highlight the necessity for a deeper mechanistic understanding of PDAC biology, particularly the cellular and molecular interactions within the TME, to inform the development of more effective and targeted therapies.

A defining hallmark of PDAC is its highly complex and desmoplastic TME, composed of a complex extracellular matrix, cancer‐associated fibroblasts (CAFs), immune cells, and tumor cells [5, 6]. This architecture not only impedes drug delivery but also fosters immune evasion. Activated pancreatic stellate cells drive fibrosis and excessive matrix deposition, promoting tumor growth and metastasis. Simultaneously, the TME orchestrates the recruitment and polarization of immunosuppressive immune cells—including myeloid‐derived suppressor cells (MDSCs), tumor‐associated macrophages (TAMs), and dendritic cells (DCs)—which collectively suppress antitumor immunity and limit therapeutic efficacy [7].

Glycosylation, a major post‐translational modification occurring in the endoplasmic reticulum and Golgi apparatus, regulates the structural and functional diversity of proteins and lipids [8]. In cancer, glycosylation pathways are frequently dysregulated, resulting in the expression of aberrant glycans. These cancer‐associated glycans serve as biomarkers, such as CA19‐9 (sialyl Lewis A), widely used in PDAC diagnosis and monitoring [9, 10, 11, 12, 13]. Additional glycan markers, including STRA and BC2‐S3, have been proposed for early detection and patient stratification, which were identified by comprehensive glycomic profiling of PDAC [14, 15, 16]. However, a comprehensive understanding of glycan signatures across PDAC subtypes at single‐cell resolution remains lacking.

Emerging evidence suggests that glycans on cancer cells actively modulate immune responses by engaging glycan‐binding receptors (lectins) expressed on immune cells [17, 18]. Human lectins, including C‐type lectins and SIGLECs, mediate glycan‐dependent signaling that can promote immune tolerance and tumor progression [19]. Despite the identification of over 100 human lectins [20], the global landscape of glycan–lectin interactions within the TME remains poorly defined. Mapping these complex interactions is essential for elucidating mechanisms of immune suppression and identifying novel glyco‐immune checkpoint molecules [19, 21].

Recent advances in single‐cell sequencing technologies have transformed our understanding of PDAC biology [22, 23, 24]. The emergence of single‐cell glycan and RNA sequencing (scGR‐seq) enables simultaneous analysis of glycomic and transcriptomic landscapes in individual cells, promising unprecedented insights into glycomic heterogeneity [25, 26, 27]. Deciphering the intricate cellular communication within PDAC further necessitates technologies that are capable of capturing both the molecular signatures and functional dynamics of cell–cell interactions. Single‐cell RNA sequencing and spatial transcriptomic approaches have provided valuable insights into the cellular composition and gene expression programs of the TME; however, they are limited in their ability to directly resolve glycan‐mediated interactions, which are involved in immune regulation and tumor progression.

In this study, we applied scGR‐seq to tumor tissues from patients with PDAC to characterize glycomic and transcriptomic profiles across diverse cell types, including cancer subtypes, CAFs, and immune cells. We identified glycan alterations associated with the transition from classical to basal‐like subtypes during the epithelial–mesenchymal transition (EMT). To systematically decode glycan–lectin interactions, we developed GlycoChat, a method for mapping glyco‐immune circuits at single‐cell resolution. GlycoChat enabled the inference of glycan–lectin networks between immune cells and cancer cells by integrating human lectin reactivity profiles and human lectin gene expression data at the single‐cell level, which goes beyond the capabilities of conventional transcriptomic analysis. GlycoChat also revealed that CLEC10A and SIGLEC3—lectins expressed on TAMs—interact with glycans enriched in basal‐like cancer cells. Functional assays demonstrated that these interactions promote M2‐like macrophage differentiation and suppress phagocytic activity, contributing to immune evasion. Our findings provide a comprehensive view of glycan–lectin interactions in the PDAC TME and establish GlycoChat as a powerful tool for identifying glyco‐immune checkpoints, offering new avenues for therapeutic intervention.

2. Results

2.1. scGR‐Seq Analysis of PDAC Tumor Tissues Using Conventional Lectins

Fresh primary tumor tissues were surgically resected from five patients diagnosed with PDAC and dissociated into single‐cell suspension. The clinicopathological characteristics of these patients are summarized in Table S1. Approximately 1 × 10⁵ cells from each patient were incubated with a panel of 29 DNA‐barcoded conventional lectins refers to a group of well‐characterized lectins that have been widely used for glycan profiling, each with defined binding specificities to various glycan structures—including fucosylated, sialylated, galactosylated, GlcNAcylated, and mannosylated glycans—(Table S2), alongside a DNA‐barcoded mouse IgG₁ as a negative control. The glycan‐binding specificity of these lectins has been assessed using glycan microarrays and frontal affinity chromatography [28]. Following removal of unbound DNA‐barcoded lectins, single cells were encapsulated with 10× gel beads, to generate Gel beads‐in‐emulsion (GEMs). Within each GEM, mRNA transcripts and lectin‐derived DNA barcodes were simultaneously reverse‐transcribed and amplified. The resulting DNA products were used to construct sequencing libraries, which were subsequently analyzed using next‐generation sequencing (Figure 1A).

FIGURE 1.

Single‐cell glycan and RNA sequencing of PDAC tumor tissues. (A) Schematic overview of the experimental workflow, including sample collection, processing, and data analysis. (B) UMAP visualization of integrated single‐cell data of glycome and transcriptome, identifying 19 distinct cell types grouped into 11 annotated cell populations. (C) The heatmap illustrates lectin binding intensities across all cell types within the TME. Lectins are listed on the y‐axis. A color gradient from red to blue indicates high to low binding intensity.

After quality control and normalization, we obtained paired glycomic and transcriptomic profiles of a total of 54 634 cells across the five patient samples, with individual cell counts ranging from 10 136 to 11 529 (Figure 1A; Table S1; see Methods for details). Integrated analysis of mRNA and glycan data using Uniform Manifold Approximation and Projection (UMAP) revealed 19 distinct cell types, annotated based on established gene markers (Figure 1B; Figures S1 and S2). These included T cells, B cells, macrophages, MDSCs, DCs, CAFs, endocrine cells, endothelial cells, mast cells, and cancer cells. Notably, the composition of these cellular subpopulations varied markedly between patients, highlighting substantial interpatient heterogeneity (Figure S1).

Cancer cells were further classified into “classical,” “intermediate,” and “basal‐like” subtypes based on gene expression profiles defined by Moffitt et al. and Raghavan et al. (Figure S2B and Table S3) [29, 30]. Within the classical subtype, two major subclusters emerged: one expressing the full classical gene set, and another expressing only a subset (Figure S2B). All three subtypes were detected across patients, though their relative abundances varied significantly (Figure S3). Patients 1 and 2 predominantly harbored classical and intermediate subtypes, whereas Patients 3 and 4 exhibited a predominance of intermediate and basal‐like subtypes. Patient 5 exclusively contained the intermediate subtype (Figure S3). These subtype distributions may be closely linked to individual patient prognoses.

scGR‐seq enabled the delineation of lectin‐binding profiles—representing cell‐type‐specific glycomic landscapes—within the TME (Figure 1C; Figure S4). Heatmap revealed pronounced differences in glycan signatures not only between cancer and noncancer cells, but also among noncancer cell types, cancer subtypes, and even within the same cell type. Cancer cells exhibited restricted lectin binding, whereas noncancer cells displayed broader lectin reactivity, underscoring the aberrant glycosylation patterns characteristic of malignant cells. For example, endothelial cells showed strong binding to TJA‐II, which detects Fucα1‐2Galβ1‐3/4GlcNAc structures, consistent with previous reports [31]. Additionally, naïve CD4⁺ T cells exhibited distinct glycosylation profiles compared to effector/memory CD4⁺ T cells, suggesting functional implications of glycan diversity within T cell subsets. The top 10 lectins showing statistically significant differential binding across cell types were identified (Table S4). These cell‐type‐specific glycan signatures may serve as novel surface markers, with corresponding lectins potentially useful for cell identification and isolation. Furthermore, the unique glycan profiles of each cell type may play critical roles in cellular functions such as immune modulation and intercellular communication.

2.2. Glycome Alteration During Epithelial–Mesenchymal Transition From Classical to Basal‐Like Cancer Subtype

To investigate glycan remodeling associated with epithelial–mesenchymal transition (EMT), we analyzed lectin‐binding profiles and glycosylation‐related gene expression across classical (epithelial), intermediate, and basal‐like (mesenchymal) cancer subtypes. Distinct lectin‐binding patterns were observed among the three subtypes, indicating cancer subtype‐specific glycomic signatures (Figure S5). Notably, a progressive shift in glycan expression was evident during the transition from classical to intermediate and basal‐like states (Figure S6). To identify glycans characteristic of each subtype, we selected the top three lectins exhibiting the most significant differential binding across subtypes (Figure 2A). In parallel, we performed single‐cell analysis of glycosyltransferase gene expression to further delineate the glycosylation landscape specific to each cancer subtype.

FIGURE 2.

Glycan remodeling during the EMT from classical to basal‐like PDAC subtypes. (A) Dot plot showing the top three lectin‐binding profiles across cancer cell clusters. Dot size reflects the proportion of cells from which the lectin‐binding was detected; color intensity indicates relative lectin‐binding signal. (B) Differential lectin‐binding patterns among PDAC subtypes as revealed by scGR‐seq. (C) Dot plot of glycosyltransferase gene expression across cancer cell clusters, with dot size and color intensity representing expression frequency and level, respectively. (D) Spatial gene expression analysis of FUT2 and ST3GAL4 in PDAC tumor tissues. Scale bar: 200 µm. (E) Comparison of glycosyltransferase gene expression between classical and basal‐like subtypes based on spatial transcriptomics shown in (D). (F) Flow cytometry analysis of lectin binding to PDAC cell lines, visualized as a heatmap following cluster analysis. (G) Immunofluorescence staining of PDAC tissues using rBC2LCN (green) and anti‐CEA antibody (red), with DAPI nuclear counterstaining (blue). Scale bar: 50 µm. (H) Immunofluorescence staining using anti‐KRT17 antibody (green) and PVL (red), with DAPI (blue). Scale bar: 50 µm. (I) Schematic representation of glycan remodeling during the EMT from classical to basal‐like subtype. Data were analyzed by Kruskal–Wallis test followed by Dunn's multiple comparisons test in (B) and the Wilcoxon rank‐sum test and in (E) (p < 0.0001: ****, p < 0.001: ***, p < 0.01: **, p < 0.05:*, p ≥ 0.05: ns, not significant).

2.2.1. Glycomic Features of the Classical Subtype

In the classical subtype, elevated signals were observed for fucose‐binding lectins, including rBC2LCN, which specifically recognizes H type 3 structures (Fucα1–2Galβ1–3GalNAc) [15, 32], and rAOL, which exhibits broad fucose specificity (Figure 2B). Consistent with these lectin‐binding profiles, genes involved in the de novo biosynthesis of GDP‐fucose—such as GMDS, TSTA3, and SLC35C1—were upregulated, along with fucosyltransferase genes FUT2 and FUT3 (Figures S7 and S8). FUT2 encodes an α1‐2 fucosyltransferase responsible for the synthesis of H antigens recognized by rBC2LCN, while FUT3 encodes an α1‐3/4 fucosyltransferase involved in the synthesis of Lewis A (Galβ1‐3(Fucα1–4)GlcNAc), detected by rAOL (Figure 2B,C; Figures S7 and S8). These findings are consistent with previous reports indicating upregulation of α1‐2 and α1‐3 fucosylated glycans, including H antigens and (sialyl) Lewis A, in PDAC [15, 32, 33]. Correlation analysis revealed a positive association between FUT2 expression and rBC2LCN binding, as well as between FUT3 expression and rAOL binding within the classical subtype (Figure S9 and Table S5). Spatial single‐cell transcriptomic analysis further supported these observations, showing higher FUT2 expression in classical compared to basal‐like subtypes (Figure 2D,E; Figure S10). Additionally, rPSL1a, a lectin specific for α2‐6‐linked sialic acid (α2‐6Sia), exhibited stronger binding in the classical subtype. Correspondingly, genes involved in the synthesis of CMP‐Neu5Ac (the Sia donor), including GNE, NANS, CMAS, and SLC35A1, as well as α2‐6‐sialyltransferases ST6GAL1 and ST6GALNAC1, were upregulated (Figure 2B,C; Figures S7 and S8). Correlation analysis confirmed positive associations between rPSL1a binding and expression of ST6GAL1 and ST6GALNAC1 (Figure S9 and Table S5). The classical subtype also demonstrated elevated gene expression of multiple polypeptide N‐acetylgalactosaminyltransferases (GALNT3, GALNT4, GALNT5, GALNT6, GALNT7, and GALNT12), which initiates mucin‐type O‐glycosylation. This was accompanied by increased expression of heavily O‐glycosylated mucins (MUC5B, MUC13, MUC17, and MUC20) (Figure 2C), suggesting enhanced O‐glycosylation activity. Furthermore, glycosyltransferase genes involved in the synthesis of Core 2 O‐glycans (B3GNT3) and type 1 LacNAc (B3GALT5), a precursor of Lewis A, were also upregulated. Collectively, these data indicate that the classical subtype is characterized by elevated expression of H type 3 O‐glycans, α2‐6Sia, and Lewis A structures, as supported by both lectin‐binding profiles and single‐cell glycosyltransferase gene expression analyses.

2.2.2. Glycomic Features of the Intermediate Subtype

In contrast to the classical subtype, the intermediate subtype exhibited elevated binding signals for mannose/GlcNAc‐specific lectins, including NPA and rF17AG (Figure 2A,B), suggesting an increased presence of high‐mannose and agalactosylated N‐glycans. Consistent with this observation, the expression of MGAT1, an N‐acetylglucosaminyltransferase that initiates the synthesis of complex‐type N‐glycans, was significantly reduced in the intermediate subtype. Moreover, MGAT1 expression negatively correlated with the binding intensities of NPA and rF17AG (Figure 2C; Figure S9 and Table S5). Among galactosyltransferase genes, B3GALT5 and B4GALT4, which are responsible for the synthesis of type 1 and type 2 LacNAc structures, respectively, were downregulated in the intermediate subtype, further supporting the increased binding of the GlcNAc‐specific lectin rF17AG. In contrast, the expression of C1GALT1, responsible for Core 1 O‐glycan synthesis, was elevated in the intermediate subtype, in agreement with the enhanced binding of the Core 1‐specific lectin PNA. A positive correlation between C1GALT1 expression and PNA binding was also observed (Figure S9 and Table S5). Taken together, these findings indicate that the intermediate subtype is characterized by increased expression of high‐mannose‐type and agalactosylated N‐glycans, as well as enhanced synthesis of Core 1 O‐glycans.

2.2.3. Glycomic Features of the Basal‐Like Subtype

The basal‐like subtype exhibited elevated binding signals for Sia/GlcNAc‐specific lectins, including PVL and WGA, compared to other subtypes, indicating enhanced sialylation and reduced galactosylation (Figure 2A,B). Among sialyltransferase genes, ST3GAL4 was notably upregulated and showed a positive correlation with PVL and WGA binding in this subtype (Figure 2C; Figure S9 and Table S5). Spatial transcriptomic analysis further confirmed higher ST3GAL4 expression in basal‐like cells relative to classical cells (Figure 2D,E; Figure S10). Consistent with these findings, α2‐3Sia‐specific lectins (MAL, MAH, rACG) exhibited stronger binding in basal‐like cells than in classical cells, as revealed by scGR‐seq analysis (Figure S11). These results highlight a dynamic shift in sialylation patterns during EMT, transitioning from α2‐6 to α2‐3 linkages. In parallel, galactosyltransferase genes B3GALT5 and B4GALT4 were downregulated in the basal‐like subtype, supporting the increased binding of GlcNAc‐specific lectins such as PVL. A negative correlation between PVL binding and the expression of these galactosyltransferases was observed (Figure S9 and Table S5). Interestingly, although the bisecting GlcNAc‐specific lectin PHAE showed higher binding in the basal‐like subtype, the gene expression of MGAT3, the enzyme responsible for bisecting GlcNAc synthesis, was low (Figure S8). This discrepancy aligns with a previous report indicating that glycosyltransferase gene expression does not always directly reflect cell surface glycan presentation [27]. Regarding O‐glycosylation, the basal‐like subtype showed increased binding of the Tn antigen‐specific lectin HPA compared to the intermediate subtype (Figures S5 and S6). This was accompanied by reduced expression of C1GALT1, the enzyme that converts Tn antigen (GalNAc) to Core 1 O‐glycan (Galβ1‐3GalNAc), resulting in a negative correlation between HPA binding and C1GALT1 expression. These findings suggest an accumulation of Tn antigen in the basal‐like subtype.

Collectively, scGR‐seq and spatial single‐cell transcriptomic analyses revealed substantial glycomic remodeling during EMT: a shift from α2‐6 to α2‐3 sialylation, a reduction in α1‐2/4 fucosylation, and a transition in O‐glycan structures from Core 2 to Core 1 and Tn. Additionally, bisecting GlcNAc structures were more prominent in the basal‐like subtype.

To validate these findings, we performed flow cytometry using lectins that showed subtype‐specific binding patterns—rBC2LCN and rAOL for classical cells, and rF17AG, PVL, and PHAE for basal‐like cells—on classical and basal‐like PDAC cell lines, as well as nonmalignant pancreatic epithelial cells. Unsupervised hierarchical clustering of lectin‐binding profiles separated classical and basal‐like cell lines into distinct clusters (Figure 2F; Figure S12). Classical cell lines (Capan‐1, BxPC‐3) exhibited stronger binding to fucose‐specific lectins (rBC2LCN, rAOL), whereas basal‐like cell lines (AsPC‐1, MIA Paca‐2, PANC‐1) showed higher binding to GlcNAc‐ and Sia‐specific lectins (rF17AG, PVL, PHAE). These results corroborate the glycan expression patterns identified by scGR‐seq (Figure 2A,B).

To further evaluate the characteristics of glycan expression in each cancer subtype, we performed tissue staining using rBC2LCN (H‐type 3 binder) and PVL (Sia/GlcNAc binder), which showed high binding in classical and basal‐like subtypes, respectively, in scGR‐seq analysis. rBC2LCN exhibited significant colocalization with CEACAM5 (CEA), a classical subtype marker (correlation coefficient r = 0.566), while PVL showed strong colocalization with KRT17, a basal‐like marker (r = 0.886) (Figure 2G,H; Figure S13). These findings were further supported by scGR‐seq data (Figures S14 and S15). In summary, scGR‐seq revealed dynamic and cancer subtype‐specific glycan signatures during EMT, highlighting a transition from classical to basal‐like states characterized by distinct alterations in sialylation, fucosylation, and O‐glycosylation (Figure 2I).

2.3. Development of GlycoChat Predicting Human Lectins That Interact With Cancer Subtypes in the TME

Cell surface glycans on cancer cells can engage with a variety of immune cell‐expressed lectins, such as C‐type lectins and SIGLECs, potentially modulating antitumor immune responses within the TME. However, the specific lectin–glycan interactions that contribute to immune tolerance in the TME remain poorly understood. While cancer cell glycans may interact with multiple immune lectins, no existing technology has enabled comprehensive, single‐cell resolution analysis of these interactions. To address this gap, we developed GlycoChat, a platform designed to map glycan–lectin networks in the TME. GlycoChat comprises three steps (Figure 3A): (1) scGR‐seq analysis of tumor tissues using DNA‐barcoded human lectins, (2) estimation of the strength of interactions between cancer subtypes and immune cells, and (3) ranking of human lectins based on their potential to interact with cancer subtypes.

FIGURE 3.

Global mapping of human lectin interaction across cell types in the PDAC TME. (A) Schematic illustration of GlycoChat analysis. (B) UMAP plot of integrated single‐cell data of glycome and transcriptome. (C–F) Dot plots showing binding profiles of C‐type lectins (C,E) and SIGLECs (D,F) across all cell types and cancer subtypes. Dot size and color intensity represent the proportion of cells of each cell type bound by the lectin, and the binding strength of the lectin to each cell type, respectively. (G) Dot plot of human lectin gene expression across cell types. Dot size and color intensity represent the proportion of cells expressing lectin genes in each cell type, and the relative expression levels of lectin genes in each cell type, respectively.

2.3.1. GlycoChat Step 1: ScGR‐Seq Analysis of PDAC Tumor Tissues Using Human Lectins

We recombinantly expressed 22 C‐type lectins and 12 SIGLECs—representing a broad range of immune functions including pathogen recognition, immune regulation, antigen presentation, and cell adhesion—as Fc fusion proteins. Each lectin was conjugated with a unique oligonucleotide barcode for single‐cell detection (Table S2). Single‐cell suspensions (1 × 10⁵ cells) derived from tumor tissues of three PDAC patients were incubated with the 34 DNA‐barcoded human lectins and processed using the scGR‐seq workflow to simultaneously capture lectin‐binding profiles and transcriptomic data. After quality control and normalization, we obtained paired lectin and mRNA data from 12 125 cells (Table S1). Using UMAP‐based dimensionality reduction and established gene markers, we identified 17 distinct cell types within the tumor samples (Figure 3B). Lectin‐binding profiles across these cell types were visualized using dot plots and heatmaps (Figure 3C,D; Figure S16). Notably, binding patterns of both C‐type lectins and SIGLECs varied across cell types and cancer subtypes, revealing cancer subtype‐specific glycan–lectin interactions (Figure 3E,F; Figure S17 and Tables S6 and S7). In parallel, we profiled the endogenous expression of lectin genes across cell types within the tumor tissues (Figure 3G; Figure S18). To further quantify lectin‐mediated interaction between immune and cancer cells, we computed interaction scores by multiplying the average expression of each lectin gene in immune cell types—including naïve CD4⁺ T cells, effector/memory CD4⁺ T cells, effector/memory CD8⁺ T cells, exhausted CD8⁺ T cells, M1‐like and M2‐like TAMs, MDSCs, conventional dendritic cells (cDC2), plasmacytoid dendritic cells (pDCs), and B cells with the corresponding glycan‐binding signal in cancer cell subtypes (classical, intermediate, and basal‐like). A set of interaction matrices was constructed for each lectin, and interaction matrices were visualized, allowing the identification of lectin‐specific immune–cancer communication patterns across the TME (Figure S19).

2.3.2. GlycoChat Step 2: Estimation of Interaction Strength Between Cancer Subtypes and Immune Cells

To determine which lectin plays a key role in glycan–lectin‐mediated interactions between cancer cells and immune cells within the TME, we then conducted a quantitative cell–cell interaction analysis. For each cell type, we calculated the average expression levels of all human lectin genes and corresponding lectin‐binding signals. Specifically, we assessed lectin gene expression in immune cells and their binding to cancer subtypes. Interaction scores between cancer subtypes and immune cells were computed for each lectin gene expression‐binding pair by multiplying the gene expression level of the lectin in immune cells with its binding signal in cancer cells. These scores were then aggregated across all lectin pairs to generate a composite interaction score for each immune–cancer cell‐type pair. To facilitate comparative visualization, the resulting interaction matrix was normalized to its maximum value. We then generated a directional chord diagram, illustrating the strength and directionality of predicted glycan–lectin interactions from immune cells to cancer cells. In this diagram, each arc represents a cell type, and the color intensity of the connecting chords reflects the relative interaction strength, offering a comprehensive overview of glyco‐immune communication within the TME (Figure 4A; Table S8). Strikingly, glycan–lectin interactions were most pronounced in the intermediate and basal‐like cancer subtypes, whereas interactions involving the classical subtype were comparatively weaker. Among immune cell populations, TAMs constitute one of the most abundant and functionally diverse components. In particular, M2‐like TAMs play a central immunosuppressive role by secreting IL‐10, TGF‐β, and arginase‐1, which collectively suppress cytotoxic T cell activity and promote the expansion of regulatory T cells (Tregs), thereby dampening antitumor immunity [33, 34]. Although we could see the interaction highly in the M2‐like TAMs and mast cells, current clinical and translational efforts have focused more extensively on TAMs due to their abundance, plasticity, and established targeting strategies, whereas the manipulation of mast cell activity in cancer is less developed and more unpredictable at this stage. The identification of novel human lectins functioning as immune checkpoint molecules may provide critical insights into the immunosuppressive roles of M2‐like TAMs and inform the development of TAM‐targeted therapeutic strategies. Based on the interaction analysis, M2‐like TAMs exhibited the strongest glycan‐mediated interactions with basal‐like cancer cells. We prioritized further investigation into the functional consequences and specificity of M2‐like TAM–basal‐like cancer subtype interactions.

FIGURE 4.

GlycoChat analysis of glycan–lectin interactions in the PDAC TME. (A) Chord diagram depicting cell–cell communication between cancer cells and immune cells, based on lectin gene expression and binding intensity from GlycoChat. (B,C) Dot plots ranking C‐type lectins (B) and SIGLECs (C) by their interaction scores with cancer cells. Dot size and color reflect the proportion of cells of each cancer subtype bound by lectin, and the average binding strength of lectin to each cancer subtype, respectively.

2.3.3. GlycoChat Step 3: Ranking human Lectins With Potential to Mediate Immune–Cancer Cell Interactions in the TME

To systematically prioritize human lectins that may facilitate glycan‐mediated interactions between immune and cancer cells within the TME, we developed a composite scoring framework integrating both gene expression and lectin‐binding intensity metrics. Lectin‐binding signals were quantified in cancer subtypes, while lectin gene expression was assessed across diverse immune cell types. For each lectin, we first calculated its average gene expression across immune cell types. To evaluate expression specificity, we implemented a scoring function that combines the coefficient of variation (CV) with a penalized ratio of the highest‐expressing immune cell type to the lower‐expressing ones, thereby emphasizing lectins with distinct, cell‐type‐restricted expression patterns. A similar strategy was applied to lectin‐binding signals. Lectins were ranked based on three criteria: (1) expression level of human lectin genes in immune cells, (2) binding intensity of human lectins to cancer cells, and (3) immune cell‐specific gene expression of human lectins. Each component score was normalized using min–max scaling to a range of [0, 1]. A weighted composite score was then calculated, with greater emphasis placed on specificity metrics to highlight selective interactions. This final score reflects the likelihood that a given lectin is both highly and specifically expressed in a particular immune cell type and engages in strong binding interactions with cancer cells.

Top‐ranking lectins were visualized using dot plots and bar graphs to compare total and component scores (Figure 4B,C; Figure S20 and Table S9). This integrative approach enabled the systematic identification of lectins with high potential to mediate targeted cell–cell communication in the TME. Based on this ranking, the top four C‐type lectins—L‐Selectin, CLEC2D, Prolectin, and CLEC10A—and the top four SIGLECs—SIGLEC2, SIGLEC9, SIGLEC3, and SIGLEC1—were identified as candidates with strong potential to interact with cancer cells in PDAC. Notably, L‐Selectin and SIGLEC9 are known immune checkpoint molecules, validating the effectiveness of our prioritization strategy [35, 36, 37]. Among these human lectins, CLEC10A and SIGLEC3 expressed in M2‐like TAM that interact with the basal‐like subtype were selected as targets for further analysis in this study [38, 39]. CLEC10A and SIGLEC3 genes were detected in M1 and M2‐like TAMs in each PDAC patient at a comparable level (Figure S21). In agreement with the GlycoChat results, CLEC10A and SIGLEC3 showed higher binding to basal‐like cancer cell lines than classical cancer cell lines in flow cytometry analysis (Figure S22).

2.4. Basal‐Like Cancer Subtype Interacts With CLEC10A Expressed on Macrophages

GlycoChat analysis revealed that CLEC10A expression is predominantly localized to macrophages and dendritic cells within PDAC tumor tissues (Figure 5A). To validate this observation, we performed immunostaining using an anti‐CLEC10A antibody on patient‐derived tumor sections. The results demonstrated strong spatial colocalization between CLEC10A and the macrophage marker CD163, with a Pearson correlation coefficient of r = 0.904, indicating a robust association between CLEC10A and TAMs (Figure 5B; Figure S13C). In GlycoChat analysis, CLEC10A exhibited preferential binding to basal‐like cancer cells compared to other subtypes (Figure 5C). This was further supported by double staining of KRT17—a basal‐like cancer cell marker—and CLEC10A, which showed substantial colocalization (r = 0.508), reinforcing the notion that CLEC10A selectively interacts with the basal‐like subtype (Figure 5D; Figure S13D). Spatial transcriptomics analysis also confirmed the gene expression of CLEC10A specifically within macrophage populations in the TME (Figure 5E; Figure S10).

FIGURE 5.

Interaction between basal‐like PDAC subtype and CLEC10A‐expressing macrophages. (A) CLEC10A gene expression profile from GlycoChat data. (B) Immunofluorescence staining of PDAC tissues using anti‐CD163 antibody (green) and anti‐CLEC10A antibody (red), with DAPI (blue). Scale bar: 50 µm. (C) Binding of CLEC10A across PDAC subtypes obtained from GlycoChat data. (D) Costaining of PDAC tissues with anti‐KRT17 antibody (green) and CLEC10A (red), with DAPI (blue). Scale bar: 50 µm. Statistical analysis: Kruskal–Wallis test with Dunn's post hoc test. Significance: ****p < 0.0001, ***p < 0.001, **p < 0.01, *p < 0.05, ns = not significant. (E) Spatial transcriptomic analysis of CLEC10A gene expression in macrophages within PDAC tissues. Scale bar: 200 µm. (F) Correlation between CLEC10A gene expression and M2 macrophage infiltration analyzed using TIMER 2.0. (G) Flow cytometry analysis of CLEC10A binding to AsPC‐1 cells in monoculture and coculture with macrophages.

To explore the immunological implications of CLEC10A expression, we analyzed its correlation with immune cell infiltration using TIMER2.0 and transcriptomic data from 179 PDAC patients in The Cancer Genome Atlas (TCGA). The gene expression of CLEC10A was significantly positively correlated with M2 macrophage infiltration (p = 1.19 × 10⁻³¹) (Figure 5F), a macrophage subset known to promote tumor progression by enhancing cancer cell proliferation and suppressing anti‐tumor immunity [40, 41]. This analysis was performed using the data of 179 patients with PDAC obtained from The Cancer Genome Atlas Program (TCGA). The results revealed a statistically significant positive correlation between the CLEC10A gene expression and M2 macrophage infiltration with a p‐value of 1.19 × 10⁻³¹ (Figure 5F). Additionally, the CLEC10A gene expression was positively associated with the infiltration of CAFs and Tregs, both of which contribute to an immunosuppressive TME [42, 43, 44]. In contrast, the CLEC10A gene expression was negatively correlated with infiltration of CD4⁺ memory T cells and Th1 cells, which are typically associated with antitumor immune responses (Figure S23A,C) [43, 44, 45]. These findings suggest that CLEC10A may play a role in shaping immune cell dynamics and promoting immune tolerance within TME.

To further investigate the functional relevance of CLEC10A–basal‐like cell interactions, we examined CLEC10A binding to AsPC‐1 cells—a basal‐like PDAC cell line—before and after coculture with phorbol 12‐myristiate‐13 acetate (PMA)‐differentiated THP‐1 macrophages. Notably, CLEC10A binding to AsPC‐1 cells was enhanced following coculture (Figure 5G), indicating that basal‐like cancer cells may upregulate CLEC10A ligands on their surface to facilitate macrophage engagement. These results suggest that the basal‐like subtype actively modulates its glycan landscape to promote interaction with CLEC10A‐expressing macrophages, potentially influencing macrophage behavior and contributing to immune evasion.

2.5. Basal‐Like Cancer Subtype Interacts With SIGLEC3 Expressed on Macrophages

Using the GlycoChat platform, SIGLEC3 was identified as a candidate lectin potentially mediating interactions with the basal‐like cancer subtype within the TME. GlycoChat analysis revealed the SIGLEC3 gene expression across multiple immune cell populations, including macrophages, myeloid‐derived suppressor cells, dendritic cells, and mast cells (Figure 6A). To validate these findings, immunostaining of PDAC tumor tissues was performed, confirming strong colocalization between SIGLEC3 and the macrophage marker CD163, with a Pearson correlation coefficient of r = 0.904 (Figure 6B; Figure S13E). Further analysis of THP‐1‐derived macrophages demonstrated SIGLEC3 expression in both M1‐ and M2‐polarized phenotypes (Figure 6C). In GlycoChat, SIGLEC3 exhibited stronger binding to intermediate and basal‐like PDAC subtypes compared to the classical subtype (Figure 6D). Tissue staining also showed SIGLEC3 binding to KRT17‐positive epithelial cells, with a moderate‐to‐strong correlation (r = 0.568), supporting its preferential interaction with the basal‐like subtype (Figure 6E; Figure S13F).

FIGURE 6.

Interaction between basal‐like PDAC subtype and SIGLEC3‐expressing macrophages. (A) SIGLEC3 gene expression profile from GlycoChat data. (B) Immunofluorescence staining of PDAC tissues using anti‐CD163 (green) and anti‐SIGLEC3 (red), and DAPI (blue). Scale bar: 50 µm. (C) Flow cytometry analysis of SIGLEC3 expression on M1 and M2 macrophages using anti‐SIGLEC3 antibody (black line). Gray indicates the isotype control antibody. (D) Binding of SIGLEC3 across PDAC subtypes from GlycoChat data. (E) Costaining of PDAC tissues with anti‐KRT17 antibody (green) and SIGLEC3 (red), and DAPI (blue). Scale bar: 50 µm. (F) Kaplan–Meier disease‐free survival curve of 177 PDAC patients stratified by SIGLEC3 gene expression using GEO RNA‐seq data. (G) Correlation between SIGLEC3 gene expression and M2 macrophage infiltration analyzed using TIMER 2.0. (H) Flow cytometry analysis of SIGLEC3 binding to AsPC‐1 cells under monoculture, coculture with macrophages, and sialidase treatment. Statistical analysis: Kruskal–Wallis test with Dunn's post hoc test (D). Significance: ****p < 0.0001, ***p < 0.001, **p < 0.01, *p < 0.05, ns = not significant.

To assess the clinical relevance of SIGLEC3, PDAC patients were stratified by SIGLEC3 expression levels using RNA‐seq data from the GEO database. Patients with low SIGLEC3 gene expression exhibited significantly improved disease‐free survival and recurrence‐free survival compared to those with high expression (Figure 6F; Figure S24). Consistent with its immunosuppressive role, analysis of TCGA data via TIMER2.0 revealed positive correlations between SIGLEC3 expression and infiltration of M2 macrophages, CAFs, and Tregs (Figure 6G; Figure S23B) [40]. In contrast, negative correlations were observed with memory CD4⁺ T cells and Th1 CD4⁺ T cells, which are associated with effective antitumor immunity (Figure S23D). Functionally, SIGLEC3 binding to AsPC‐1 basal‐like PDAC cells was disrupted by sialidase treatment, indicating Sia‐dependent interaction (Figure 6H). Moreover, the coculture of AsPC‐1 cells with THP‐1‐redived macrophage differentiated with PMA significantly enhanced SIGLEC3 binding to AsPC‐1 cells, suggesting that basal‐like cancer cells express SIGLEC3 ligands on their cell surface, thereby enhancing SIGLEC3‐mediated interactions with macrophages. Together, these findings demonstrate that basal‐like PDAC cells engage with CLEC10A and SIGLEC3—both expressed on macrophages—highlighting a glycan‐mediated mechanism by which cancer cells may modulate immune cell behavior and contribute to immune evasion within the TME.

2.6. Basal‐Like Cancer Cells Promote M2 Macrophage Differentiation via CLEC10A and SIGLEC3 Interactions

To dissect the individual and synergistic roles of CLEC10A and SIGLEC3 in modulating macrophage polarization by basal‐like PDAC cells, we engineered THP‐1 mononuclear cell mutant strains with CLEC10A‐only expression (THP1‐CLEC10A⁺/SIGLEC3⁻), SIGLEC3‐only expression (THP1‐CLEC10A⁻/SIGLEC3⁺), both lectins expressed (THP1‐CLEC10A⁺/SIGLEC3⁺), and neither lectin expressed (THP1‐CLEC10A⁻/SIGLEC3⁻) were generated (Figure 7A; see Methods for details). These THP‐1 variants were differentiated into macrophages using PMA treatment for 48 h and subsequently cocultured with AsPC‐1 cells—a basal‐like PDAC cell line—for three days in ratios 1:1, 1:2, and 2:1 of THP‐1 variants and AsPC‐1 cells. Macrophage polarization was assessed via flow cytometry by measuring the expression of classical M1 marker CD86, and M2 markers CD163 and CD206. Although no significant differences in M1 or M2 marker surface expression were observed in the 1:2 or 2:1 coculture ratios (Figure S25), the 1:1 ratio of THP‐1 variants and AsPC‐1 cells revealed that the expression of either CLEC10A or SIGLEC3 (THP1‐CLEC10A⁻/SIGLEC3⁺, THP1‐CLEC10A⁺/SIGLEC3⁻) promoted upregulation of the M2 markers relative to THP1 only used as negative control, indicating a shift toward an immunosuppressive macrophage phenotype (red and purple bars in Figure 7B). Notably, CLEC10A‐only expressed cells (THP1‐CLEC10A⁺/SIGLEC3⁻) most strongly induced CD163 expression, the M2 marker, associated with high‐grade malignancy in PDAC (red bar in Figure 7B) [46, 47]. Coexpression of CLEC10A and SIGLEC3 (THP1‐CLEC10A⁺/SIGLEC3⁺, orange bar in Figure 7B) significantly elevated CD206 levels, which have been linked to increased lymph node metastasis [48]. CLEC10A‐only expressed cells (THP1‐CLEC10A⁺/SIGLEC3⁻, red bar in Figure 7B) induced higher M2 marker expression (CD163 and CD206) than SIGLEC3‐only expressed cells (THP1‐CLEC10A⁻/SIGLEC3⁺, purple bar in Figure 7B), suggesting that CLEC10A exhibits higher M2 polarization activity than SIGLEC3. These findings underscore the potential of CLEC10A and SIGLEC3 as immune‐modulatory checkpoints that contribute to tumor progression and metastatic potential via macrophage reprogramming.

FIGURE 7.

Cancer cell‐mediated modulation of macrophage phagocytosis via CLEC10A and SIGLEC3 interactions. (A) Representative flow cytometry histograms showing CLEC10A and SIGLEC3 expression on THP‐1 cells. (B) Flow cytometry analysis of changes in MFI for M1 marker (CD86) and M2 markers (CD163, CD206) after coculture of macrophages with AsPC‐1 cells. “THP1 only” indicates THP‐1 without coculture with AsPC1 cells. (C) Flow cytometry quantification of ZsGreen1‐THP‐1 cell phagocytosis against FR‐AsPC1 cancer cells. “THP1 only” represents ZsGreen1‐THP‐1 cells without the coculture with FR‐AsPC1 cells. (D) Fluorescence microscopy images showing phagocytosis of FR‐AsPC1 cells (red) by ZsGreen1‐THP‐1 cells (green). Scale bar: 50 µm. (B,C) Data are presented as mean ± SEM from three biological replicates. Statistical analysis: Kruskal–Wallis test with Dunn's post hoc test. Significance: ****p < 0.0001, ***p < 0.001, **p < 0.01, *p < 0.05, ns = not significant.

Conversely, expression of M1 (CD86) as well as M2 markers (CD163, CD206) was markedly low in macrophages lacking both lectins (THP1‐CLEC10A⁻/SIGLEC3⁻, green bar in Figure 7B), suggesting that the absence of CLEC10A and SIGLEC3 may redirect macrophage differentiation away from M1/M2 polarization (Figure 7B; Figure S26). This phenotype resembles intermediate monocytes, which are characterized by distinct transcriptional profiles and reduced CD86 expression compared to classical and nonclassical subsets [49, 50]. Thus, dual deletion of CLEC10A and SIGLEC3 may induce an intermediate‐like macrophage state, marked by diminished expression of both M1 and M2 markers. This altered polarization profile may have functional implications for immune regulation and tumor dynamics within the TME.

2.7. Cancer Cells Suppress the Phagocytic Activity of Macrophages Through Interactions With CLEC10A and SIGLEC3

To investigate the functional roles of CLEC10A and SIGLEC3 in regulating macrophage phagocytic activity toward cancer cells, we conducted in vitro coculture experiments. ZsGreen1‐labeled THP‐1 cells were differentiated into macrophage‐like cells via PMA treatment for 48 h. The ZsGreen1‐labeled THP‐1‐derived macrophages were then cocultured with FR‐labeled AsPC‐1 basal‐like cancer cells for 6 h, and phagocytic activity was quantified by flow cytometry (Figure 7C; Figure S27). The scatter plot obtained from ZsGreen1‐labeled THP‐1‐derived macrophages alone served as the negative control and baseline for assessing phagocytic activity, showing fluorescence signals exclusively from ZsGreen1 (Figure S27A, left panel). Flow cytometric analysis of FR‐labeled AsPC‐1 cells revealed fluorescence signals solely originating from FR (Figure S27A, middle panel). When FR‐labeled AsPC‐1 cells were cocultured with ZsGreen1‐labeled THP‐1‐derived macrophages lacking CLEC10A and SIGLEC3 expression (THP1‐CLEC10A⁻/SIGLEC3⁻), FR‐derived fluorescence was detected within ZsGreen1‐positive THP‐1‐derived macrophages (Figure S27A, right panel). Macrophages deficient in both CLEC10A and SIGLEC3 (green bar in Figure 7C) exhibited significantly higher FR fluorescence intensity compared to macrophages coexpressing the two lectins (orange bar in Figure 7C). Conversely, macrophages expressing both CLEC10A and SIGLEC3 (THP1‐CLEC10A⁺/SIGLEC3⁺, orange bar) showed minimal phagocytic activity against AsPC‐1 cells. Among single‐lectin‐expressing cells, SIGLEC3‐only expressed cells (THP1‐SIGLEC3⁺/CLEC10A⁻, purple bar in Figure 7C) showed lower phagocytosis than CLEC10A‐only expressed cells (THP1‐CLEC10A⁺/SIGLEC3⁻, red bar in Figure 7B), indicating that SIGLEC3 exerts a stronger immune suppressive effect than CLEC10A on the phagocytosis activity. These findings indicate that CLEC10A and SIGLEC3 inhibit macrophage‐mediated phagocytosis of basal‐like PDAC cells. To corroborate this mechanistically, we performed live‐cell fluorescence imaging for direct visualization. The imaging revealed dynamic engulfment of FR‐labeled AsPC‐1 cells by ZsGreen1‐labeled THP‐1‐derived macrophages in the absence of CLEC10A and SIGLEC3 (Figure 7D; Figure S28). However, CLEC10A and SIGLEC3 exhibited no effect on the phagocytosis activity of THP‐1 cells against classical‐type PDAC cells (Figure S29), since they exhibit only low binding to classical‐type PDAC cells (Figure S22). Collectively, these results demonstrate that basal‐like cancer cells can suppress macrophage phagocytic activity through glycan–lectin interactions involving CLEC10A and SIGLEC3, highlighting a potential immune evasion mechanism within the TME.

3. Discussion

Cancer cells frequently evade immune surveillance by expressing molecules that suppress effective anti‐tumor immune responses. These include anti‐inflammatory cytokines and chemokines that recruit and activate immunosuppressive cell populations, as well as immune checkpoint ligands such as PD‐L1, which engage inhibitory receptors like PD‐1 on lymphocytes. Blocking these interactions has become a cornerstone of cancer immunotherapy [45]. Aberrant glycosylation is a hallmark of cancer, yet comprehensive analyses of glycan signatures across cancer subtypes at single‐cell resolution remain limited. Consequently, our understanding of cancer subtype‐specific glycan expression and its functional implications in the TME has been incomplete. Moreover, the complex network of interactions between cancer cell glycans and human lectins expressed on immune cells—likely involving many‐to‐many relationships—has not been fully elucidated. A deeper understanding of these glycan–lectin interactions is essential to overcome the immunosuppressive landscape of PDAC.

In this study, we performed single‐cell glycomic profiling of PDAC tumor tissues using scGR‐seq and identified distinct glycan signatures across cell types within the TME (Table S10). Cancer cells exhibited aberrant glycosylation patterns that differed markedly from nonmalignant cells. While noncancer cells showed broad lectin‐binding profiles, cancer cells displayed selective binding to specific lectins. For example, the classical subtype showed strong binding to rBC2LCN, indicating enrichment of H‐type 3 O‐glycans, consistent with previous findings [15]. During the transition to the basal‐like subtype, rBC2LCN binding decreased, while binding to Sia‐specific lectins (PVL, WGA) increased. PVL staining colocalized with KRT17‐positive basal‐like cells, suggesting its utility as a probe for identifying this subtype. α2‐3Sia‐binding lectins (MAL, MAH, rACG) also showed elevated signals, correlating with increased ST3GAL4 expression. Additionally, the basal‐like subtype exhibited strong binding to the Tn antigen‐specific lectin HPA, in association with low C1GalT1 expression, which catalyzes the conversion of Tn (GalNAc) to Core 1 O‐glycan (Galβ1‐3GalNAc). These findings demonstrate that α2‐3 sialylation and Tn antigen expression are elevated in the basal‐like subtype, both of which are known to be associated with malignancy in PDAC [46, 47].

To explore glycan–lectin interactions within the TME, we present GlycoChat as a single‐cell platform for mapping interactions between glycans and human lectins—including C‐type lectins and SIGLECs—expressed on immune cells. Spatial transcriptomics visualizes and analyzes gene expression within tissue sections, revealing cell types and spatial organization; however, this method cannot identify human lectins expressed on immune cells interacting with cancer cells at the molecular level. The previously developed single‐cell glycan and RNA sequencing (scGR‐seq) can analyze glycan and RNA expression data at the single‐cell level, but cannot analyze glycan–lectin interactions. GlycoChat enables the acquisition of comprehensive interaction data between cancer cell subtypes and human lectins expressed on immune cells within the TME. Indeed, GlycoChat identified lectins that interact with specific cell types, including cancer subtypes and CAFs. CLEC10A (Tn‐binder) and SIGLEC3 (α2‐3/6Sia‐binder) were identified as key lectins interacting with the basal‐like subtype [46]. Notably, basal‐like cancer cells upregulated ligands for CLEC10A and SIGLEC3 following coculture with THP‐1‐derived macrophages expressing these lectins, suggesting that PDAC cells dynamically remodel their glycan landscape to enhance macrophage engagement.

Although the genes of SIGLEC3 and CLEC10A are not uniquely expressed on macrophages, the immunofluorescence analysis of patient‐derived PDAC tissues demonstrated strong spatial colocalization of the two human lectins with the macrophage marker CD163, as reflected by a high Pearson correlation coefficient (r = 0.904). This pattern likely reflects the abundance of TAMs within the examined tumor regions, where these cells represent one of the dominant immune populations and express both lectins at relatively high levels. Since the colocalization analysis was performed within CD163‐positive areas enriched for TAMs, the correlation captures both molecular coexpression and spatial concentration of these lectins in macrophage‐dense niches. Together, these findings suggest that, within the PDAC microenvironment, CLEC10A and SIGLEC3 are functionally and spatially associated with macrophages.

Functionally, SIGLEC3 possesses ITIM and ITIM‐like motifs in its cytoplasmic domain, which recruit SHP1/2 phosphatases to induce inhibitory signaling [39]. While CLEC10A contains only an endocytosis motif in its cytoplasmic domain and lacks a typical ITIM motif, it may still exhibit inhibitory signaling in macrophages through binding to cancer cells via unknown signaling motifs (Figure S30) [51]. When CLEC10A and SIGLEC3 are coexpressed on macrophages, cancer cells act on macrophages via these inhibitory lectin receptors, inducing immunosuppression of the macrophages (Figure S31). This causes the macrophages to differentiate into M2‐like cells and express the surface marker CD206. When macrophages do not express both CLEC10A and SIGLEC3, cancer cells cannot influence the macrophages and cannot suppress their immune activity (such as phagocytosis). Thus, the contribution of CLEC10A and SIGLEC3 to macrophage polarization is likely mediated indirectly through glycan‐dependent recognition rather than classical cytokine stimulation. CLEC10A and SIGLEC3 expressed on macrophages acted in a synergistic and complementary manner in these effects.

The intrinsic functions of CLEC10A and SIGLEC3 are to regulate immune cell signaling, and their expression in both M1 and M2 macrophages is as expected (Figures 5A and 6A). Our results suggest that cancer cells act on CLEC10A and SIGLEC3 on M0 macrophages, promoting increased expression of these lectins on macrophages, and promoting the differentiation of M0 macrophages into M2‐like cells and inducing a tumor‐immune‐suppressive environment within the TME. Although this study did not examine the functional roles of CLEC10A and SIGLEC3 in M1/M2 macrophages in the TME, they are likely to play important roles in signaling control within these cells in the TME, representing a future research challenge.

In vivo validation of macrophage functions mediated by CLEC10A and SIGLEC3 would provide deeper insight into their pro‐ or antitumor roles; however, such experiments are complicated by substantial species‐specific differences between human and mouse lectin families. Human CLEC10A lacks a direct mouse ortholog and differs in glycan‐binding specificity and immune function from its murine paralogs Mgl1 (CD301a) and Mgl2 (CD301b) [52]. Similarly, SIGLEC3 exhibits distinct expression patterns, glycan‐binding specificities, and signaling properties between human and mouse [53]. Human SIGLEC3 is primarily expressed on monocytes and myeloid cells, exhibits binding specificity for α2‐3/6Sia, and contains both ITIM and ITIM‐like motifs [53]. In contrast, mouse SIGLEC3 is mainly expressed on mature granulocytes and myeloid progenitor cells, shows binding preference for sialyl Tn, and contains only an ITIM‐like motif [53]. Therefore, human CD33 functions as an inhibitory receptor, recruiting SHP1/2 via its ITIM motif to suppress immune signaling, whereas mouse CD33 does not strongly influence immune function. Since CD33 knockout mice exhibit normal immune responses, its inhibitory function is considered limited. This complicates direct modeling of human SIGLEC3 signaling in murine systems. Given these molecular and evolutionary differences, our study focused on single‐cell multiomics and in vitro analyses using human samples to elucidate glycan–lectin interactions and macrophage regulatory mechanisms in a human‐specific context. Future studies using humanized mouse models will be essential to further validate the in vivo relevance of CLEC10A and SIGLEC3 in macrophage polarization and tumor immunity.

In addition to their role in macrophage‐mediated immunoregulation, the glycan–lectin interactions mediated by CLEC10A and SIGLEC3 may function as innate immune checkpoints that complement the canonical adaptive immune checkpoint pathways, such as PD‐1/PD‐L1 and CTLA‐4/CD80‐CD86. These mechanisms suggest that CLEC10A and SIGLEC3 act as innate immune checkpoints, dampening antitumor immunity at an earlier stage than adaptive checkpoints. Their activity may create an immunosuppressive tumor microenvironment that reduces the efficacy of PD‐1 or CTLA‐4 blockade alone. Indeed, recent studies have demonstrated that SIGLEC‐mediated inhibitory signaling can promote resistance to checkpoint inhibitor therapy and that combined targeting of SIGLECs with PD‐1/CTLA‐4 blockade can synergistically restore immune activity [54]. Based on our findings, we propose that dual modulation of the glycan–lectin axis and adaptive checkpoint pathways could represent a promising therapeutic approach. In particular, disrupting CLEC10A‐ or SIGLEC3‐mediated suppression of macrophages may reprogram TAMs toward a proinflammatory, antitumor (M1‐like) phenotype, thereby enhancing the efficacy of PD‐1/CTLA‐4 blockade.

Inhibitors targeting the interactions between cancer cells and immune cells via CLEC10A and SIGLEC3, such as antihuman lectin blocking antibodies, glycan mimetics, and antitumor‐specific glycan antibodies, may serve as novel immune checkpoint inhibitors for treating PDAC. Simultaneously inhibiting multiple cancer cell–human lectin interactions, combined with other immune checkpoint inhibitors, may enable more effective PDAC treatment. Furthermore, by using GlycoChat data to predict the glycan‐based immune checkpoint molecules inducing immunosuppression in each patient, it may be possible to select the optimal immune checkpoint inhibitor for each patient. Modulating lectin–glycan binding may represent a promising strategy to reprogram the tumor microenvironment toward an antitumor phenotype, supporting the translational potential of the GlycoChat platform in PDAC management.

Although the integration of single‐cell datasets across multiple patients successfully reduced batch and patient‐specific effects, we recognize that interpatient heterogeneity remains an intrinsic challenge in tumor microenvironment studies. The current dataset size provides a proof‐of‐concept for the GlycoChat framework; however, validation using independent patient cohorts and additional clinical samples will be essential to confirm the generalizability of the identified glycan–lectin interaction networks. Although from a statistical analysis perspective, we are able to analyze a sufficient number of single cells (typically 100 or more cells per cluster) to reliably estimate gene expression and lectin binding profiles within each sample (Table S11). Future studies incorporating multicenter datasets and matched spatial information will further improve the robustness of patient‐level inference and strengthen the translational potential of this approach.

While our primary focus was on the interaction between TAMs and cancer cells, we also identified notable interactions between cancer cells and mast cells, suggesting that mast cells may play an underrecognized role in shaping the TME. Emerging evidence indicates that mast cells can facilitate tumor progression through the secretion of cytokines, growth factors, and proteases that promote angiogenesis, extracellular matrix remodeling, and immune suppression [55, 56]. Although the biological and therapeutic relevance of mast cells in PDAC remains less well defined than that of TAMs, their involvement observed in our dataset highlights a potentially important yet understudied component of the TME. Future investigations integrating glycan‐mediated signaling with mast cell biology may provide new insights into tumor–immune communication and uncover novel therapeutic opportunities.

Beyond CLEC10A and SIGLEC3, GlycoChat also identified other lectins with potential relevance in the PDAC TME, including L‐Selectin, CLEC2D, Prolectin, SIGLEC2, SIGLEC9, and SIGLEC1. Among these, SIGLEC9 has already been validated as a target for immune checkpoint blockade [37]. For example, Gray et al. developed an anti‐HER2–sialidase conjugate that removes sialoglycans from breast cancer cells, thereby disrupting SIGLEC7/9‐mediated immune suppression and enhancing NK cell cytotoxicity [57]. Other approaches have employed Sia mimetics to inhibit glycan‐mediated immune evasion [58].

Advantages of GlycoChat include the following: (1) It enables comprehensive single‐cell mapping of glycan–lectin interactions, capturing not only direct physical interactions but also secreted interactions, by single‐cell glycan and RNA data. (2) It identifies candidate glycan‐immune checkpoint molecules that may serve as therapeutic targets for cancer immunotherapy. (3) It is applicable to analyze glycan–lectin interactions in various normal and diseased tissues, facilitating insights into tissue homeostasis and disease mechanisms.

Limitations of GlycoChat include the following: (1) It cannot provide spatial localization data. (2) It predicts glycan–lectin interactions within tissues such as the TME but infers their functional roles. (3) It uses Fc‐fused human lectins for interaction analysis, which may differ from interactions mediated by native cell‐surface lectins.

In summary, we characterized the glycomic landscape of PDAC at single‐cell resolution and developed GlycoChat to systematically identify human lectins that interact with cancer cells. Our findings reveal that basal‐like cancer cells suppress macrophage phagocytosis and promote M2‐like polarization through interactions with CLEC10A and SIGLEC3. These lectins represent promising targets for TAM‐focused immunotherapies. The integration of GlycoChat with scGR‐seq offers a powerful framework for the rational design of next‐generation immunotherapies targeting glycan‐mediated immune checkpoints.

4. Methods

4.1. Lectins

Lectins from natural sources were purchased from J‐OIL MILLS, SEIKAGAKU CORPORATION, or EY Laboratories, Inc. Recombinant lectins were produced in bacterial (Escherichia coli) or mammalian expression systems (HEK293T). For recombinant lectins produced in E. coli, genes of carbohydrate recognition domains (CRDs) were cloned into pET27b (Merck KGaA, Darmstadt, Germany), overexpressed in E. coli BL21‐CodonPlus (DE3)‐RIL strain, and purified by affinity chromatography using appropriate sugar‐immobilized Sepharose 4B‐CL (GE Healthcare, IL, USA) [28]. The CRD or extracellular domain of C‐type lectins and SIGLECs was cloned into the pSecTag2 vector or pcDNA5/FRT containing a human IgG1 Fc region at the C‐terminus to generate Fc fusion proteins. The expression vector was transiently transfected into HEK293T cells, and the supernatant containing the Fc‐fusion protein was purified using Protein G‐Sepharose 4 Fast Flow (GE Healthcare). The purity of the lectins was analyzed using sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS‐PAGE) and western blotting. The protein concentration was calculated using the bicinchoninic acid protein assay kit (Thermo Fisher Scientific).

4.2. Cell Lines

PDAC cell lines such as AsPC‐1 (CRL‐1682, RRID: CVCL_0152), BxPC‐3 (CRL‐1687, RRID: CVCL_0186), MIA Paca‐2 (CRL‐1420, RRID: CVCL_0428), PANC‐1 (CRL‐1469, RRID: CVCL_0480), and Capan‐1 (HTB‐79, RRID: CVCL_0237), and human monocytic cells, THP‐1 (TIB‐202, RRID: CVCL_0006), were obtained from ATCC (Manassas, VA, USA) in 2015. The mycoplasma contamination test was conducted using MycoStrip (InvivoGen, New Territories, Hong Kong). No contamination was detected during the experiments. AsPC‐1 and BxPC‐3 were cultured in Roswell Park Memorial Institute RPMI‐1640 (FUJIFILM Wako Pure Chemical Co., Osaka, Japan) medium supplemented with 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin. MIAPaca‐2 and PANC‐1 were cultured in Dulbecco's Modified Eagle's Medium DMEM (FUJIFILM Wako Pure Chemical Co., Osaka, Japan) medium supplemented with 10% FBS and 1% penicillin/streptomycin. In the case of MIA PaCa‐2, horse serum was added to a final concentration of 2.5%. Capan‐1 was cultured in Iscove's Modified Dulbecco's Medium (IMDM) (FUJIFILM Wako Pure Chemical Co., Osaka, Japan) supplemented with 10% FBS and 1% penicillin/streptomycin. Human monocytic cells, THP‐1, were cultured in RPMI‐1640 (FUJIFILM Wako Pure Chemical Co., Osaka, Japan) medium supplemented with 10% FBS and 1% penicillin/streptomycin.

4.3. THP‐1 Differentiation

THP‐1 cells were differentiated into macrophages with RPMI‐1640 containing 100 ng/mL PMA (Merck KGaA, Darmstadt, Germany) over 48 h. Macrophages were polarized into M1 macrophages by incubation with 20 ng/mL of IFN‐γ (Cat No. 285‐IF, R&D Systems, Minneapolis, MN, USA) and 10 pg/mL of LPS (Cat No. 8630, Sigma, St. Louis, MO, USA) for 18 h. Macrophage M2 polarization was achieved by incubation with 20 ng/mL of interleukin 4 (Cat No. 204‐IL, R&D Systems, Minneapolis, MN, USA) and 20 ng/mL of interleukin 13 (Cat No. 213‐ILB, R&D Systems, Minneapolis, MN, USA) for 48 h.

4.4. Generation of THP‐1 Cells With and Without Expression of CLEC10A and SIGLEC3

The original THP‐1 cells express SIGLEC3, but not CLEC10A, named as “THP1‐SIGLEC3⁺/CLEC10A⁻”. A THP‐1 stable cell line with the genomic deletion on one or more copies of human SIGLEC3 (NCBI Gene ID: 945) was generated by CRISPR‐Cas9, named as “THP1‐SIGLEC3⁻/CLEC10A⁻”.

The full‐length of CLEC10A (accession No Q8IUN9, 1‐316 aa) with a FLAG tag (DYKDDDDK) at the C‐terminal end was inserted into the pLVSIN‐EF1α‐IRES‐ZsGreen1 vector via the SpeI/BamH1 sites. The CLEC10A‐pLVSIN‐EF1α‐IRES‐ZsGreen1 was transfected into Lenti‐X 293T (Takara Bio Inc., Shiga, Japan) using the TransIT‐293 transfection reagent (Takara Bio Inc., Shiga, Japan). The virus titer was evaluated by Lenti‐X GoStix Plus (Takara Bio Inc., Shiga, Japan) and concentrated with Lenti‐X Concentrator (Takara Bio Inc., Shiga, Japan). The CLEC10A lentivirus vector was infected into THP‐1 cells and THP‐1 with deletion of SIGLEC3 to generate “THP1‐SIGLEC3⁺/CLEC10A⁺” and “THP1‐SIGLEC3⁻/CLEC10A⁺” cells, respectively.

4.5. Single‐Cell Preparation of Patient Tissues

Five patients who were pathologically diagnosed with PDAC were enrolled in the University of Tsukuba. This research was conducted as per the ethical standards of the Declaration of Helsinki and was approved by the ethics committee of each institution (IRB number H28‐090). Written informed consent was obtained from all patients for the use of clinical samples for this research, and approval was obtained from the University of Tsukuba Hospital (IRB number H28‐090). The information of patients with PDAC used in this study is listed in Table S1. No organs/tissues were procured from prisoners.

The recovered patient tissues were suspended in the transportation medium (RPMI with 10% FBS supplemented with Protease inhibitor, 0.05 U/µL RNase inhibitor, and 10 µm Rock inhibitor) and transported on ice from the University of Tsukuba Hospital to the National Institute of Advanced Industrial Science and Technology. After decanting the media, the tissue specimens were weighed and placed on ice in petri dishes. Tissues were then washed three times with the new dissociation medium RPMI‐1640 supplemented with protease inhibitor, 0.02 µg/µL RNase inhibitor, 10 µm Rock inhibitor, 500 µg/mL Collagenase D, 20 µg/mL DNase I, 10 µg/mL Dispase II, 55 µg/mL trypsin inhibitor, and 10% bovine serum albumin (BSA). Tissues were then subjected to fine mincing in 5 mL of RPMI solution containing 10% FBS using sharp and sterile scalpel blades to make a fine slurry. After that, the medium was transferred through the MACS filter to a 15 mL Falcon tube and incubated for 30 min at 37°C under continuous shaking inside the incubator. The Falcon tubes were centrifuged at 300 × g for 7 min at 4°C and resuspended in red blood cell lysis solution (Miltenyi Biotec, Bergisch Gladbach, Germany). Following a 2‐min incubation at room temperature, samples were centrifuged (300 × g, 4 °C, 10 min) and resuspended in PBS containing 10% BSA (PBS/BSA, Merck KGaA, Darmstadt, Germany).

4.6. Generation of DNA‐Barcoded Lectins

Twenty‐nine conventional lectins, 22 C‐type lectins, and 12 SIGLECs were labeled with DNA oligonucleotides by a 5′‐Feature Barcode Antibody Conjugation Kit (Abcam, Cambridge, UK) according to the manufacturer's protocol. Briefly, each purified lectin was first buffer‐exchanged into the conjugation buffer provided in the kit. The lectin was then incubated with the 5′‐amine‐modified DNA oligonucleotide and the proprietary crosslinking reagent to enable covalent attachment between the oligonucleotide and the protein via primary amine groups on the lectin molecule. After the conjugation reaction, excess oligonucleotides and unreacted reagents were removed using the purification columns supplied in the kit to ensure high labeling specificity and purity. The concentration and the apparent molecular size of the conjugated lectins were subsequently verified using SDS‐PAGE. Protein and DNA concentrations were measured using the Bradford (BioRad, Hercules, CA, USA) and Quant‐iT OliGreen ssDNA Reagent Kit (Thermo Fisher Scientific Inc., Waltham, MA, USA), respectively, to calculate the DNA‐to‐lectin ratio. The DNA‐labeled lectins were then stored at −30°C in the provided storage buffer until further use in the experiments. A list of DNA‐barcoded lectins is shown in Table S2.

4.7. Generation of Gel Beads‐in‐Emulsion (GEMs) of scGR‐Seq

After the obtained single cells (1 × 10⁵ cells) were incubated with Human TruStain FcX (Cat. No. 422 301, BioLegend) to block Fc receptors, cells were reacted with DNA‐barcoded lectins and mouse‐IgG₁ antibody (Figures S4A and S16A,C) used as a negative control (BioLegend, CA, USA), at a final concentration of 1 mg/mL for 1 h on ice. The FlowMI 40 µm Cell Strainer (BAH136800040, Merck KGaA, Darmstadt, Germany) was used to filter cells that had attached DNA‐barcoded lectin/antibody in order to eliminate cell aggregation. The automated cell counter TC20 (Cat No. 145‐0101, BioRad, CA, USA) was used to assess the vitality and quantity of cells. To capture transcriptomic and lectin‐binding information at single‐cell resolution, cells were encapsulated using a droplet‐based microfluidic system. The GEM technology was employed to ensure that individual cells were physically isolated and barcoded, thereby preventing cross‐contamination of RNA and barcode signals between cells. This approach enables high‐throughput generation of uniquely barcoded cDNA libraries from thousands of single cells, allowing parallel profiling of RNA and glycan features [59].

For GEM generation, viable single‐cell suspensions were prepared at a final concentration of approximately 1 × 10⁶ cells/mL with > 80% viability. The cell suspension, reverse transcription (RT) reagents, and barcoded gel beads were loaded into separate channels of the Chromium Next GEM Single Cell 5′ Chip (10× Genomics, Pleasanton, CA, USA). Each gel bead contained millions of oligonucleotides consisting of a 16‐bp cell barcode, a 10‐bp unique molecular identifier (UMI), and a poly(dT) sequence for capturing polyadenylated RNA transcripts. Within the Chromium Controller, the cell suspension, reagents, and barcoded beads were coflowed through a microfluidic network, generating nanoliter‐scale oil droplets known as GEMs. Each GEM ideally contained a single cell and a single barcoded gel bead. Upon cell lysis inside the droplet, released mRNAs hybridized to the oligonucleotides on the gel bead surface. Reverse transcription was then carried out inside each GEM to generate barcoded cDNA molecules, uniquely tagging transcripts from individual cells. Following RT, GEMs were broken to release the barcoded cDNA, which was subsequently purified using silane magnetic beads. The resulting barcoded cDNA was then amplified by PCR and used for downstream library construction, including gene expression, V(D)J repertoire, and lectin barcode libraries, according to the manufacturer's protocol (Chromium Next GEM Single Cell 5′ Reagent Kit v2 User Guide, 10× Genomics). Library quality and fragment size distributions were verified by using MultiNA (Shimadzu Co., Kyoto, Japan) before sequencing.

4.8. Single‐Cell Library Preparation

The surface glycan (termed scGlycan‐seq) and gene expression (scRNA‐seq) libraries were constructed using Dual Index Plate TN Set A and Dual Index Plate TT Set A, respectively, according to the manufacturer's methodology (Chromium Next GEM Single Cell 5' Reagent Kits v2). The typical fragment sizes of PDAC tissues were 450 bp for scRNA‐seq and 222 bp for scGlycan‐seq, as verified by MultiNA (Shimadzu Co., Kyoto, Japan). The mRNA library and the lectin barcode library were combined in a 4:1 ratio for sequencing. The total read pairs for scRNA‐seq were 227 million, whereas scGlycan‐seq had 60 million.

4.9. scRNA‐Seq and scGlycan‐Seq Reading

Sequencing data were processed using Cell Ranger (10× Genomics, Pleasanton, CA, USA) following the manufacturer's standard pipeline [59]. Cell Ranger performed demultiplexing, barcode processing, and UMI counting to generate feature‐barcode matrices. Raw base call files generated by the Illumina sequencer were converted to FASTQ format using the cellranger mkfastq command. The FASTQ files were then aligned to a custom reference genome that included both human gene features (GRCh38) and the designed glycan feature barcodes corresponding to lectin‐binding probes. Alignment and UMI quantification were conducted using the cellranger count command, which produced a filtered feature‐barcode matrix containing both RNA expression and glycan‐binding signals for each single cell.

The filtered feature matrix was subsequently imported into Seurat (version 5.0.1) [60] using the Read10× function. Gene expression data were assigned to the scRNA‐seq, while antibody captures were assigned to a separate scGlycan‐seq. These two modalities were processed independently (including normalization, scaling, and variable feature selection) before integration for multi‐omic single‐cell analysis.

4.10. Seurat Analysis for scGR‐Seq

For scRNA‐seq, to filter contamination of gene expression by dead cells, cells expressing fewer than 200 genes as well as those with more than 10% mitochondrial‐derived genes, were excluded from the downstream analysis. In addition, the “LogNormalize” method in the “NormalizeData” function was used to normalize the scale factor of 10 000 scRNA‐seq data, and the “FindVariableFeatures” function was used to filter the top 3000 highly variable genes after QC by the FindVariableFeatures() function for further analysis followed by scaling of the data using the ScaleData() function. Dimension reduction was conducted using principal component analysis, and the dimensionality was calculated using the elbow plot for subsequent cell clustering. The FindNeighbour() function with the default option was used to generate k nearest neighbor cells (KNN) on the 8 components using the PCA data, which were then clustered using the FindClusters() function with a resolution of 1.0. The components from each dataset were then visualized using UMAP.

For scGlycan‐seq, the data were divided into subsets based on UMI features ranging from 10 to 31. The negative control “mouse‐IgG₁ antibody” was eliminated from both datasets prior to further analysis. The subset data is then log‐normalized with a scale factor of 10 000 using the NormalizeData() function, and finally scaled using the ScaleData() function. The FindVariableFeatures() function was not used because there are only 29, 22, and 12 lectins for conventional lectins, C‐type lectins, and siglec human lectins, respectively, as opposed to thousands of genes. Dimension reduction was conducted using principal component analysis, and the dimensionality was calculated using the elbow plot for subsequent cell clustering. The FindNeighbour() function with the default option was used to define KNN cells on the 10 PCA components, which were then clustered using the FindClusters() function with a resolution of 1.0. Then, UMAP visualization was performed on the components for each dataset.

The combined data from scRNA‐seq and scGlycan‐seq were processed with a weighted nearest neighbor (WNN) approach. To calculate the closest neighbours based on the weighted combination of RNA and lectin similarities, the FindMultiModalNeighbors() function was used with the same dimensions used in the scRNA (PC 1:20) and scGlycan analysis (PC 1:20 for C‐type lectin samples and 1:10 for SIGLEC samples) with the default parameter. UMAP was then visualized using the weighted RNA and lectin combination with default parameters, followed by WNN graph‐based clustering with the FindCluster() function with the parameter “resolution = 1.5”.

4.11. Cell Type Annotation

Single‐cell clusters were identified based on common marker genes that have been established in the literature for various cell types. Those include iCAF (C3, C7), myCAF (ACTA2, NDUFA4L2), endocrine cell (INS, SST, GCG), endothelial (VWF, PECAM1), mast cell (KIT), pDC (LILRA4, PLD4), B‐cell (CD79A), cDC2 (CD1C, FCER1A), MDSC (S100A12), M1‐like TAM, and M2‐like TAM (CD86, MRC1, CD163), naïve CD4⁺ T cell (RPL21, RPL31), exhausted CD8⁺ T cell (CD27, TIGIT, CTLA4), effector/memory CD8⁺ T cell (CD44, CD69, GZMK), and effector/memory CD4⁺ T cell (PTPRC, SELL, CCR7). The cancer cell type annotation based on the Moffitt marker genes for the “Classical” and “Basal‐like” subtypes and the Raghavan marker genes for the “Intermediate” subtype (Table S3) [29, 30].

4.12. GlycoChat

GlycoChat comprises three steps: (1) scGR‐seq analysis of tumor tissues using DNA‐barcoded human lectins, (2) estimation of the strength of interactions between cancer subtypes and immune cells, and (3) ranking of human lectins based on their potential to interact with cancer subtypes.

(1) scGR‐seq analysis of tumor tissues using DNA‐barcoded human lectins: Twenty‐two C‐type lectins and 12 SIGLECs conjugated with unique DNA barcodes were used for the scGR‐seq analysis as described above and acquired the binding signals of human lectins (22 C‐type lectins and 12 SIGLECs) and mRNA expression data in each cell.

(2) Estimation of the strength of interactions between cancer subtypes and immune cells: To assess cell–cell interactions between immune and cancer cells, we first defined representative immune cell types and malignant epithelial subtypes. Average lectin gene expression was calculated for each immune cell type, while lectin‐binding signal refers to the normalized counts measured in the lectin‐binding signals to each cancer subtype within the scGR‐seq, using the AverageExpression() function in Seurat. In detail, raw lectin‐binding intensities were first log‐normalized and scaled using Seurat's standard preprocessing pipeline as in Seurat Analysis for scGR‐seq. For each cancer cell subtype, we then calculated the average binding signal per lectin across cells assigned to that subtype. Interaction scores were calculated by multiplying the average lectin gene expression in immune cells with the corresponding lectin‐binding signal to cancer subtype. Summing these values across all lectins yielded a composite interaction score for each immune–cancer subtype pair.

These scores reflect the potential strength of glycan–lectin interactions. The resulting interaction matrix was normalized and visualized using a directional chord diagram via the circlize R package, illustrating putative intercellular communication from immune to cancer cells.

(3) Ranking of human lectins based on their potential to interact with cancer subtypes: To prioritize human lectin candidates, we developed a composite scoring system that integrates multiple biologically relevant features. We calculated the mean‐expression of lectin‐binding signal on cancer cells and lectin gene expression in immune cells using the AverageExpression() function on single‐cell data grouped by annotated cell identities. Using these data, we calculated individual scores based on (i) expression level of each lectin gene in immune cells (Lectin gene expression level), (ii) binding intensity of lectins to cancer cells (Lectin binding level), and (iii) lectin expression specificity in immune cells (Lectin gene expression specificity). Lectin gene expression specificity was evaluated using a score that combines the CV and the expression ratio between the top two cell types. Each of these components was treated as an independent vector and scaled using min–max normalization, transforming the values to a standardized range from 0 to 1 to enable cross‐feature comparison. Following normalization, a weighted scoring scheme was applied to combine these metrics into a single interaction score, referred to as the total score. The final score was calculated as: Total Score = ʊ1 × “Lectin gene expression level” + ʊ2 × “Lectin binding level” + ʊ3 × “Lectin gene expression specificity”. Weights ʊ1​, ʊ2, and ʊ3 were user‐defined and adjusted depending on analysis priorities. For visualization, the top‐ranked glycans were plotted using dotplot and bar graphs in Seurat, showing the contribution of each score component (plot_top_glycans () function), providing interpretability into which features drive high interaction potential. The final ranked list was exported as a CSV file and used for downstream functional or experimental follow‐up.

4.13. In Vitro Assays

To analyze the macrophage polarization assay, ZsGreen1‐labeled THP‐1 cells produced by pLVSIN‐EF1α‐IRES‐ZsGreen1 were differentiated into macrophages by PMA and cocultured with AsPC‐1 at 1:1, 1:2, and 2:1 ratios for 3 days. After 3 days, the cells were collected to analyze the macrophage polarization to M1 and M2 macrophages. Cells were harvested and analyzed by flow cytometry.

To analyze the phagocytosis ability of THP‐1‐derived macrophage cells on AsPC‐1 cells, the ZsGreen1‐labeled THP‐1 cells were differentiated to macrophages by PMA and cocultured with far red (FR, CellTrace Far Red, Invitrogen, C34553)‐labeled AsPC‐1 at a 1:1 ratio for 6 h. Cells were collected to analyse the red fluorescence derived from FR on the ZsGreen1‐labeled THP‐1 cells by flow cytometry. For the control condition, ZsGreen1‐labeled macrophages were cocultured with unlabeled AsPC‐1 cells.

To evaluate the changes in glycan profiles or the binding of CLEC10A‐Fc or SIGLEC3‐Fc to AsPC‐1 pancreatic cancer cells, we performed flow cytometric analysis under different experimental conditions. AsPC‐1 cells were treated with or without sialidase to remove terminal sialic acid residues and subsequently incubated with either CLEC10A‐Fc or SIGLEC3‐Fc to assess lectin binding. For the macrophage coculture condition, THP‐1 monocytes were differentiated into macrophages by treatment with PMA for 48 h. The differentiated macrophages were labeled with FR and then cocultured with ZsGreen1‐labeled AsPC‐1 cells at a 1:1 ratio for 3 days. After coculture, the cells were harvested, stained with CLEC10A‐Fc or SIGLEC3‐Fc, and analyzed by flow cytometry.

4.14. Sialidase Treatment

To cleave terminal sialic acid residues, cells were treated with neuraminidase from Arthrobacter ureafaciens (Cat No. 10269611001, Roche, Basel, Switzerland) at a concentration of 0.04 U in PBS, and incubated at 37°C for 1 h in agitation at 50 rpm. Subsequently, the cells were washed with PBS containing 1% BSA and used for downstream analysis.

4.15. Flow Cytometry Analysis

To validate scGR‐seq and GlycoChat data, we performed flow cytometry analysis using lectins labeled with R‐phycoerythrin (PE) using the R‐phycoerythrin Labeling Kit (Dojindo Laboratories Co. Ltd., Kumamoto, Japan). Cells (1 × 10⁵) were suspended in 100 µL of PBS/BSA (PBS containing 1% bovine serum albumin) and incubated with PE‐labeled lectins (1 µg/mL) for 1 h on ice.

Cells were also incubated with Human TruStain FcX (Cat No. 422 301, BioLegend) for 10 min at room temperature to block Fc receptors and then labeled with antibodies against SIGLEC3 (Cat No. MAB9635‐100, R&D), CD86 (Cat No. 305402, Biolegend), CD163 (Cat No. 333602, Biolegend), CD206 (Cat No. 321102, Biolegend), CD301 (Cat No. 354702, Biolegend), CLEC10A‐Fc, or SIGLEC3‐Fc in PBS/BSA containing 1 mm CaCl₂ for 1 h on ice. The samples were then incubated on ice for 1 h with an Alexa Fluor 647‐conjugated goat anti‐mouse secondary antibody (Cat No. 2482945, Invitrogen). Flow cytometry analysis was performed on a CytoFLEX System (Beckman Coulter, Inc., Brea, CA, USA), and the generated data were analyzed using the FlowJo software v10.6 (BD, Franklin Lakes, NJ, USA).

4.16. Tissue Staining

Human pancreatic tissue sections were provided from the Department of Gastrointestinal and Hepato‐Biliary‐Pancreatic Surgery, Faculty of Medicine, University of Tsukuba. The tissue sections were deparaffinized by lemosol and ethanol followed by antigen retrieval in 10 mmol/L sodium citrate buffer for 2 min at 110°C. To reduce tissue autofluorescence, slides were placed in a transparent reservoir containing 3% hydrogen peroxide in methanol for 15 min at room temperature. Slides were rinsed with 1% PBS/BSA placed in a humidified stain tray, and incubated for 1 h at room temperature. The slides were then incubated with the first antibodies including anti‐KRT17 (Cat No. Sc‐393002, Santa Cruz Biotechnology), anti‐SIGLEC3 (Cat No. MAB11371‐100, R&D), anti‐CLEC10A (Cat No. 145702, Biolegend), anti‐CD163 (Cat No. 364302, Biolegend), anti‐CEA antibodies (clone No. 18‐5A), and SIGLEC3‐Fc and CLEC10A‐Fc in a humidified stain tray overnight at 4°C. After washing with PBS containing 1% Triton‐X (PBST)or PBST containing 1 mM CaCl₂ (PBSTCa) for three times, the tissue sections were incubated with Alexa Fluor 488‐conjugated goat anti‐mouse antibody (Cat No. 529465, Invitrogen) or Alexa Fluor 647‐conjugated goat anti‐mouse antibody (Cat No. 2482945, Invitrogen) in a humidified dark stain tray for 1 h at room temperature. After that, slides were washed again with PBST or PBSTCa three times and mounted in Prolong Gold Antifade with DNA Stain DAPI (Cat No. P36935, Invitrogen) and analyzed using Olympus IX51 (Olympus Corporation, Tokyo, Japan).

Colocalization analysis was performed using ImageJ with the JACoP v2.1.4 function [61]. Quantitative analysis of colocalization was carried out using Pearson's correlation coefficient (r) and Manders' overlap coefficients (M1 and M2) to evaluate the degree of signal overlap between channels. The thresholds for these coefficients were set by the plugin to reduce background noise. Interpretation of Pearson's r‐values was as follows: 0 to 0.3 indicates weak positive correlation, 0.3 to 0.5 indicates moderate positive correlation, 0.5 to 1.0 indicates strong positive correlation, and values between 0 and –1.0 suggest a negative correlation.

4.17. Spatial Analysis

Fresh frozen PDAC patient tissues were provided from the University of Tsukuba Hospital under permit IRB number H28‐090 and profiled using the in situ Xenium platform (10× Genomics, Stoneridge Mall Road, Pleasanton, CA, USA). The predesigned 300‐gene Xenium Gene Expression panel was profiled across the sections, and a total of 50 Moffitt subtype marker genes for “classical” and “basal‐like” subtypes were analyzed. Formalin‐fixed, paraffin‐embedded blocks of tumor tissue were obtained from 12 PDAC patients. Hematoxylin and eosin (H&E) staining was performed to confirm, in consultation with a pathologist, that the region (4.0 × 4.0 mm) contained sufficient and diverse cancer cells. RNA probe hybridization, ligation, and barcoding were then performed to construct a spatial transcriptome library using Xenium In Situ Reagent Kits (10× Genomics) according to the manufacturer's instructions. Autofluorescence quenching and nuclear staining were performed in the dark. Fluorescence probe hybridization and imaging were performed using a Xenium Analyzer (on‐board analysis: version 1.8.2.1, software: version 1.7.1.0, 10× Genomics) with the prepared slides. Output images and expression profiles were evaluated with Xenium Explorer (version 3.0.0, 10× Genomics). H&E staining was performed on Xenium slides after the run.

4.18. Statistical Analysis

The statistical analyses were performed using GraphPad Prism 8 or the base function in R v.4.2.1. The statistical tests used, and sample sizes are indicated in the figure legends. p‐values > 0.05 were considered not significant (n.s.), and p‐values < 0.05 were considered significant. Asterisks indicate the following: *p‐value < 0.05, **p‐value < 0.01, ***p‐value < 0.001, ****p‐value < 0.0001. All of the bars within the graphs represent mean values, and the error bars represent standard errors of the mean (SEM) or standard deviation (SD) as indicated.

Author Contributions

D.X.T.A. and H.T. designed the research. D.X.T.A., S.K., A.B., K.K., and L.O. performed experiments. D.X.T.A., S.K., A.B., and A.K. performed data analysis. D.X.T.A. and H.T. wrote the paper. O.S. and T.O. provided the patient samples.

Funding

This work was supported by Takeda COCKPI‐T Funding, JSPS KAKEN (23K26872 and 23H04796), JST A‐step (JPMJTR23U6), and AMED P‐PROMOTE (25ama221332h0002).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting File 1: advs73638‐sup‐0001‐SuppMat.pdf.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions. The custom R script used for the ranking analysis in Figures 4B,C and S20 is available at https://github.com/akikuno/GlycoChat‐Ranking/.