Leucine zipper-based SAIM imaging identifies therapeutic agents to disrupt the cancer cell glycocalyx for enhanced immunotherapy

Abstract

The abnormally thick glycocalyx of cancer cells can provide a physical barrier to immune cell recognition and effective immunotherapy. Here, we demonstrate an optical method based on Scanning Angle Interference Microscopy (SAIM) for the screening of therapeutic agents that can disrupt the glycocalyx layer as a strategy to improve anti-cancer immune responses. We developed a new membrane labeling strategy utilizing leucine zipper pairs to fluorescently mark the glycocalyx layer boundary for precise and robust measurement of glycocalyx thickness with SAIM. Using this platform, we evaluated the effects of glycosylation inhibitors and targeted enzymatic degraders of the glycocalyx, with particular focus on strategies for cholangiocarcinoma (CCA), a highly lethal malignancy with limited therapeutic options. We found that CCA had the highest mean expression of the cancer-associated mucin, MUC1, across all cancers represented in the cancer cell line encyclopedia. Pharmacological inhibitors of mucin-type O-glycosylation and mucin-specific proteases, such as StcE, could dramatically reduce the glycocalyx layer in the YSCCC model of intrahepatic CCA. Motivated by these findings, we engineered Natural Killer (NK) cells tethered with StcE to enhance NK cell-mediated cytotoxicity against CCA. In a CCA xenograft model, these engineered NK cells demonstrated superior anti-tumor efficacy compared to wild-type NK cells, with no observable adverse effects. Our findings not only provide a reliable imaging-based screening platform for evaluating glycocalyx-targeting pharmacological interventions but also offer mechanistic insights into how CCA may avoid immune elimination through fortification of the glycocalyx layer with mucins. Additionally, this work presents a novel therapeutic strategy for mucin-overexpressing cancers, potentially improving immunotherapy efficacy across various cancer types.

Introduction

The cellular glycocalyx is a dense meshwork of glycosylated macromolecules that covers all eukaryotic cell surfaces. As the first point of contact between two interacting cells, the glycocalyx constitutes a vital component of cell-cell interfaces, including immune synapses formed between immune effector cells and target cancer cells. Oncogenesis is commonly accompanied by the dysregulated expression of various glycosyltransferases, resulting in a remodeling of the glycan landscape presented at the cell surface. Additionally, many tumors are associated with the overexpression of bulky, hydrating glycopolymers, such as cell-surface mucins and glycosaminoglycans (GAGs), which cause a swelling and thickening of the glycocalyx¹. Cells that have undergone malignant transformation therefore exhibit marked changes in both the molecular composition and physical architecture of the glycocalyx, transforming it into a form of armor that can protect them against surveilling immune cells^2–6.

Thus far, development of immunotherapeutic strategies to overcome the cancer glycocalyx has largely been shaped by research efforts to identify tumor-specific glycan signatures that biochemically suppress cytolytic immune cell activity. Accordingly, most strategies to counter glycocalyx-mediated immune evasion focus on blocking the biochemical signaling pathways that are activated by carbohydrate-binding receptors. These strategies includes the pharmacological inhibition of the biosynthesis of glyco-immune checkpoint ligands^6–8, antibody-mediated blockade of the receptors for these ligands^9,10, and the targeted degradation of cancer-associated mucins or other glycoproteins that carry these ligands^11,12. However, evaluating the efficacy of new glycocalyx-targeting therapeutics based on biochemical activation alone is not sufficient for predicting enhancement to immune cell-mediated killing. For example, our previous work has demonstrated that the dense polymer layer formed by overexpression of the cell-surface mucin, MUC1, acts as a physical barrier to immune cell attack, offering protection from immune cells even when they are equipped with chimeric antigen receptors (CARs) to enhance their biochemical activation¹³. Notably, nanometer-scale reductions in the glycocalyx thickness can severely compromise the ability of cancer cells to evade elimination by cytotoxic immune cells¹³. These findings suggest a potential new form of anticancer immunotherapy based on therapeutic strategies that disrupt the physical structure of the glycocalyx or its biosynthesis.

Some of the most lethal solid tumor cancers express high levels of cell surface mucins and respond poorly to current immunotherapies. As one important example, cholangiocarcinoma (CCA) is a heterogeneous group of aggressive malignancies arising from different locations within the biliary tree¹⁴. MUC1 expression in CCA is closely correlated with dedifferentiation, invasion, and poor patient survival¹⁵. The malignancy often presents challenges in early detection due to its asymptomatic nature in the initial stages, leading to a grim prognosis upon diagnosis. Advanced CCA typically carries a dismal prognosis, with surgical resection being the primary curative approach, albeit feasible only in a minority of cases due in part to the complexity of the bile duct system and often diffuse nature of tumors¹⁶. Response rates of CCA to current immunotherapies remain disappointingly low¹⁷. Only limited success has been achieved with checkpoint blockade, and CCA generally lacks suitable target antigens for CAR-based cellular immunotherapies. Whether disruption of the physical structure of the glycocalyx layer could improve immune responses in CCA remains untested.

One major roadblock to the development of glycocalyx-disrupting agents is a general lack of adequate tools that enable the precise characterization of how the individual molecular constituents of the cancer glycocalyx affects its nanoscale structure. To this end, Möckl et al. has coupled metabolic labeling of glycans with single molecule localization microscopy (SMLM) to show that the height of the glycocalyx increases in correspondence with key programs in malignancy, such as the epithelial-to-mesenchymal transition and oncogenic Ras activation¹⁸. However, the utility of SMLM for drug screening and development is limited by slow acquisition times on the order of 1–30 minutes per cell, reducing sample throughout and typically restricting imaging to fixed samples¹⁹. Chemical fixation in SMLM and other high resolution imaging techniques, such as electron microscopy, can introduce artifacts that may alter the native glycocalyx architecture, complicating the ability to detect nanometer-scale changes in glycocalyx structure. A robust technique that can be used to rapidly assess how the nanoscale structure of the glycocalyx in live cells responds to drug perturbations is necessary to accelerate the discovery of new therapeutic agents that target the glycocalyx.

As one potential solution, our group has advanced an optical technology termed Scanning Angle Interference Microscopy (SAIM) for high-speed and precise measurement of the glycocalyx thickness¹³. Based on the principles of fluorescence interference contrast microscopy, SAIM can accurately measure the nanoscale space occupied by the glycocalyx between the plasma membrane and a substrate. With its sub-second acquisition rate, SAIM has strong potential for screening the effects of drug compounds on the glycocalyx structure. However, the robust application of SAIM across a broad range of cancer cell types has been limited by challenges in achieving stable and specific labeling of the plasma membrane, which is used to mark the boundary of the glycocalyx layer. For example, fluorescent proteins expressed on the cell membrane via genetic approaches often show intracellular retention in Golgi or endoplasmic reticulum (ER), disrupting the interference pattern of SAIM and compromising accuracy²⁰. Additionally, lipophilic membrane dyes frequently fail to maintain surface localization and undergo rapid endocytic internalization. Furthermore, heterogeneity in the lipid compositions of the membranes across cell lines and even within the same model can affect the labeling efficiency of lipophilic dyes, requiring painstaking optimization of labeling conditions for each cell line^21,22. Therefore, a need remains for an alternative strategy that enables efficient, long-lasting, and specific fluorescent labeling of the plasma membrane of any cell with minimal toxicity to cells over the course of a SAIM imaging experiment.

In this study, we report the development of a novel method of fluorescently labeling the plasma membranes of live cells by utilizing engineered coiled-coil dimerization pairs. By genetically engineering target cells to display a leucine zipper and recombinantly producing the cognate leucine zipper fused to a fluorescent dye, we achieve highly specific, robust labeling of the plasma membrane in a wide range of cancer cell lines. This approach demonstrated superior membrane retention and labeling efficiency compared to conventional lipophilic dyes, enabling precise quantification of glycocalyx thickness across diverse cancer cell types with SAIM. Using this platform, we systematically evaluated glycocalyx responses to pharmacological and enzymatic perturbations in a CCA culture model and identified agents that are highly effective at reducing glycocalyx thickness. From this initial screen, we further evaluated the most promising candidates for their ability to enhance immune cell-mediated killing in vitro, which informed the development of an anti-CCA cell-based immunotherapy that utilizes an enzymatic degrader. Lastly, we established a CCA xenograft model to test our cell-based therapeutic, which demonstrated pronounced efficacy in vivo.

Results

Development and validation of a leucine zipper-based membrane labelling strategy for stable SAIM imaging

Our previous work identified the importance of a single physical parameter of the cancer glycocalyx, its material thickness, as a strikingly strong predictor of tumor cells’ ability to evade immune cell attack¹³. The strong dependence of cancer cell survival on glycocalyx thickness motivated us to search for compounds that can be used to “deflate” the cancer cell glycocalyx and accordingly, enhance immune cell killing (Fig. 1a). This required an approach to accurately measure glycocalyx thickness in live cells with nanoscale precision. To this end, we have developed Scanning Angle Interference Microscopy (SAIM), which enables nanoscale axial localization of fluorescent dyes^{13,18,23–26} (Fig. 1b). However, SAIM-based measurement of the glycocalyx requires stable and specific labeling of the plasma membrane, which marks the boundary of the glycocalyx layer (Fig. 1b).

Figure 1: SAIM illustration and potential use for drug screening.

a, Schematic representation of cellular glycocalyx thickness changes induced by pharmacological inhibitors or targeted degraders to enhance immune cell recognition. b, Optical configuration for Scanning Angle Interference Microscopy (SAIM); the boundary of the glycocalyx is marked by fluorescent labeling of the plasma membrane, whose position is localized using axially varying patterns of structured illumination generated through interference of direct and reflected laser light. Silicon (Si) chips with a layer of oxide (SiO) serve as both the optical mirror and cell substrate. c, Proposed membrane labelling strategy in this work using leucine zipper pairs (Azip and Bzip). Azip is conjugated with fluorescent protein or dye. Bzip is genetically encoded and expressed as a fusion protein with the PDGFRβ transmembrane (TM) domain.

To improve the robustness of SAIM for drug screening applications, we developed a more optimal membrane-labeling strategy utilizing engineered leucine zipper dimerization motifs, previously validated for its high specificity and stability^27–29. The leucine zipper pair, which we refer to as “Azip” and “Bzip”, is comprised of several heptad repeats that favorably interact through a combination of electrostatic and hydrophobic interactions, resulting in binding with low nanomolar affinity (Fig. 1c). To specifically label the plasma membrane, we engineered target cells to display Bzip on the cell surface using the PDGFRβ transmembrane (TM) domain via lentiviral transduction, then recombinantly produced Azip fused to superfolder GFP (sfGFP) to be used as a membrane-labeling probe (Fig. 2a,b). To confirm successful lentiviral transduction and enable fluorescence-based sorting of target cells, Bzip constructs contained either an ALFA tag or mScarlet-i (Fig. 2a). This approach successfully enabled bright and specific membrane labeling with Azip-sfGFP across all tested cancer cell lines, including the pancreatic cancer cell line Capan-2 and the breast cancer cell lines KPL-1, T47D, SKBR3, and ZR-75–1 (Fig. 2c,d). Notably, expression of direct fusion of mScarlet to Bzip was not sufficient to achieve highly specific labeling of the cell surface, since mScarlet also fluoresced brightly in various intracellular organelles (Fig. 2d). We observed that the Azip-sfGFP probe was often excluded from the tight intercellular junctions within cancer cell colonies (Fig. 2d). We noted that this could limit the utility of the Azip probe for SAIM measurements, which requires a high density of fluorophores to penetrate the glycocalyx in the small cleft between the plasma membrane and the imaging substrate. To optimize the system further, we replaced the fluorescent protein markers on Azip with smaller and more efficient organic dyes through sortase-mediated conjugation (Fig. 2e). By inserting a sortag sequence between Azip and sfGFP, we enabled conjugation of small organic fluorophores (AF488 or AF647) to the purified Azip construct (Fig. 2f). This refined strategy reduced the size of the Azip probe from ~35 kDa to ~7 kDa, and demonstrated strong and specific binding to Bzip-expressing cancer cells with negligible background binding (Fig. 2g,h). We also observed that the Azip dye conjugated with AF488 or AF647 penetrated intercellular junctions in cancer cells more effectively than the sfGFP conjugate (Fig. 2h)

Figure 2: Validation of the leucine zipper-based membrane labelling strategy.

a, Schematic representation of constructs used to make Bzip-PDGFRβ TM constructs and Azip-Sortag-sfGFP construct. b, Schematic representation of cell membrane labelling using Azip-Bzip interaction. c, Fluorescence and bright-field images of Bzip-overexpressing Capan-2 pancreatic cancer cells with 1 μM Azip-sfGFP for 1 hour at 4°C. Scale bars, 10 μm. d, Fluorescence images of Bzip-mScarlet and Azip-sfGFP on indicated cancer cell lines for 1 hour at 4°C. Scale bars, 10 μm. e, Schematic of sortase-mediated reaction on Azip-sortag-sfGFP to replace the fluorescent protein with a small organic fluorophore. f, Coomassie blue staining analysis of sortase-mediated modification of Azip-sortag-sfGFP to yield Azip-Alexa Fluor 647 (AF647). g, h, Fluorescence and bright-field images of KPL-1 wild-type (g) and Bzip-overexpressing KPL-1 breast cancer cells (h) double labeled with Azip-AF488 and Azip-AF647 (1:100 dilution) for 1 hour at 4°C. Scale bars, 10 μm. i, Representative fluorescence microscopy of Bzip-expressing KPL-1 cells labeled with various membrane stains: Azip-AF488, Azip-AF647, DiO, CellBrite 650, or MemGlow 560 according to manufacturers’ recommended buffer compositions and concentrations. Cells were stained for 10 minutes at 37°C followed by two washes with culture medium. j, Quantification of normalized membrane fluorescence intensity over time (20 × 20 pixel regions, 2.98 × 2.98 μm) averaged from minimum 11 cells per condition. k, Membrane fluorescence intensity of each dye immediately after labelling (time 0). In k, statistical analysis was performed by one-way ANOVA with Tukey’s post hoc tests.

To optimize and validate the labeling efficacy of our membrane dye, we tested its performance under various staining conditions optimized for SAIM imaging. We compared standard cold labeling (4°C, 1 hour) with rapid warm labeling (37°C, 10 minutes), finding that the Azip dye demonstrated significantly improved membrane retention without endocytosis even in warm labeling conditions (Fig. 2h and Extended Data Fig. 1). Long-term stability was assessed over 48 hours and showed that Azip-based dyes exhibited significantly higher plasma membrane retention and minimal intracellular uptake compared to popular lipophilic dyes, including DiO, CellBrite, and MemGlow³⁰ (Fig. 2i-k). While Azip conjugated with AF647 showed minor endocytosis after 48 hours, Azip-AF488 maintained remarkably stable plasma membrane localization throughout the extended imaging period, establishing it as an optimal tool for dynamic SAIM imaging applications.

Cholangiocarcinoma cell line YSCCC exhibits high MUC1 surface expression

Having established a robust method for membrane labeling, we next sought to survey a broad range of cancer cell lines that could serve as models for glycocalyx-mediated immune avoidance. Given that cell-surface MUC1 expression levels are strong determinants of the glycocalyx layer thickness¹³, we sought to identify cancer cell types and associated cell line models with high MUC1 overexpression. Analysis of publicly available datasets from the DepMap project revealed that CCA exhibits the highest mean MUC1 RNA expression levels compared to other tumor types (Fig. 3a). To investigate the correlation between MUC1 transcript levels and MUC1 protein density on the cell surface, we conducted an analysis across various human cancer cell lines using a monoclonal anti-MUC1 antibody (clone HMPV) that specifically recognizes a core peptide in the VNTR region of MUC1 in a glycosylation-independent manner. For standardization across experiments, we utilized the SKBR3 cell line as a reference control in each experimental replicate and normalized the median fluorescence intensity (MFI) of test cell lines to that of SKBR3. Linear regression analysis of normalized MFIs against MUC1 RNA transcript levels from the DepMap project demonstrated a strong correlation (R² = 0.7768) (Fig. 3b). Notably, among the panel of cell lines tested, YSCCC intrahepatic CCA cells exhibited the highest MUC1 surface expression, which strongly correlated with their elevated MUC1 transcript levels.

Figure 3: Cholangiocarcinoma cell line YSCCC exhibits high MUC1 surface expression.

a, MUC1 transcript levels as measured by the depMap project for the indicated cancer categories; center lines show the medians; means are indicated by ‘+’; box limits indicate the 25^th and 75^th percentiles; whiskers extend 1.5 times the interquartile range from the 25^th and 75^th percentiles, outliers are represented by dots. n = 8, 50, 49, 85, 35, 47, 58, 25, 44, 27, 8, 42, 30, 79, 29, 47, 74 sample points. b, Linear correlation between MUC1 transcript levels as measured by the depMap project and MUC1 surface levels as measured by flow cytometry. Points plotted as mean ± s.d. measured from at least 2 independent experiments. Data for the cholangiocarcinoma cell line, YSCCC, is shown in red. c, Representative flow cytometry histograms showing MUC1 levels in YSCCC cells following overnight treatment with 1 nM SmE and 1 nM StcE. d,e, Quantification of spreading area (d) and circularity (e) of YSCCC cells following adhesion overnight to tissue culture flasks. Results are the mean ± s.d. of n = 100 cells per condition. f, Quantification of the fraction of highly circular cells across n = 3 fields of view from phase contrast images from e. Highly circular cells were determined as those with a circularity value > 0.9. g, Representative phase-contrast images of overnight attachment of YSCCC cells in either the presence or absence of 1 nM SmE or 1 nM StcE (scale bar, 50 μm). h, Schematic of predicted behavior of cell-cell repulsion in multicellular aggregate formation assays. i, Representative brightfield images of multicellular aggregates formed by seeding 1,000 cells of YSCCC per well in round bottom, ultra-low adhesion wells in either the presence or absence of 10 nM stcE for 24 hours. Scale bar = 200 μm, inset scale bar = 50 μm. j, Quantification of overall aggregate area, normalized to untreated YSCCC. Results are the mean ± s.d. of n = 5 aggregates for untreated, n = 7 aggregates for StcE treated. Statistics were determined using a two-tailed Student’s t-test. In d,e,f, statistics were determined using a one-way ANOVA with Tukey’s multiple comparisons test.

Since there are relatively few reports on the YSCCC intrahepatic CCA model in the literature, we characterized the adhesive and morphological phenotypes of the cell line. Interestingly, we observed that YSCCC cells in culture exist as a mixed population of adherent and suspended cells, suggesting the presence of a substantially repulsive glycocalyx that potentially impairs cell-substrate adhesion^31,32. To test this hypothesis, we challenged YSCCC cells to adhere to standard tissue culture plates overnight in either the presence or absence of a low dose (1 nM) of mucin-digesting enzymes, StcE and SmE, which are known to broadly cleave mucin-domain glycoproteins from the cell surface^33–35. Following overnight treatment with the mucinases at 37°C, YSCCC cells showed significantly reduced cell-surface MUC1 levels compared to untreated controls (Fig. 3c). Morphological analysis revealed that untreated YSCCC cells exhibited partial adherence and spreading on culture plates, with a substantial proportion maintaining a highly circular morphology and poor adhesion (Fig. 3d-g). In contrast, mucinase-treated cells demonstrated increased spreading area and a significantly reduced proportion of circular cells (Fig. 3e-g). Notably, SmE treatment showed relatively modest effects, consistent with only partial MUC1 cleavage under the lower concentration protocol.

Aside from disrupting cell-substrate adhesion, we reasoned that a highly repulsive glycocalyx could also prevent cell-cell adhesion, an essential step in immune cell-mediated elimination of cancer cells. To evaluate the impact of the cellular glycocalyx on cell-cell adhesion, we assessed the capacity of YSCCC cells to form compact multicellular clusters in ultra-low attachment wells. We hypothesized that cells with high repulsive forces would resist forming close cell-cell contacts, resulting in increased intercellular spacing within aggregates as predicted previously^31,36 (Fig. 3h). YSCCC cells cultured for 24 hours in ultra-low attachment wells demonstrated minimal cell-cell association, forming diffuse aggregates at the well bottom. However, treatment with StcE mucinase significantly reduced cell-cell repulsion, enabling intercellular adhesion and resulting in more compact aggregate formation (Fig. 3i,j). These findings suggest that reduction in MUC1 levels through enzymatic treatment promotes a more adherent and spread morphology in YSCCC cells.

Screening glycocalyx-degrading agents using leucine zipper-based SAIM

Pharmacological inhibitors have demonstrated promising outcomes in inhibiting O-glycan biosynthesis or eliminating sialylation on cell membranes, which significantly influences cancer cell adhesion and invasion^37,38. However, a systematic evaluation of pharmacological inhibitors and glycocalyx-degrading enzymes for their specific ability to reduce glycocalyx thickness in cancer cells has not yet been performed. We investigated a peracetylated N-thioglycolyl modified N-acetylgalactosamine (GalNAc) analog (Ac₅GalNTGc) that has been shown to interfere with O-glycan extension through GalNAc-mediated core-1 synthesis inhibition^39–41. We also tested the pan-sialyltransferase inhibitor, P-3F_AX-Neu5Ac, which depletes α2,3-/α2,6-linked sialic acids without affecting cell viability or proliferation^42,43 and reduces tumor growth in vivo ⁴³. Following the validation of our membrane labeling strategy, we implemented the leucine zipper-based SAIM imaging approach to quantitatively measure glycocalyx thickness across different cells lines, including YSCCC and the breast cancer cell lines, SKBR3 and KPL-1 as additional examples (Extended Data Fig. 2a–c). Cells engineered to express the Bzip construct were cultured on fibronectin-coated silicon wafers and labeled with Azip-AF488 for 10 minutes at 37°C. As expected, Azip-AF488 effectively penetrated the space between the cell membrane and imaging substrate in all cell types, generating well-defined interference patterns that enabled precise mapping of the glycocalyx thickness with nanometer resolution (Fig. 4 and Extended Data Fig. 2).

Figure 4: Pharmacological inhibitor screening using leucine zipper-based SAIM.

a, Schematic representation of glycocalyx thickness reduction by pharmacological inhibitors as detected by leucine zipper-based SAIM. b, Representative wide-field image and glycocalyx thickness map of live YSCCC expressing Bzip construct labeled with Azip-AF488 (1:100 dilution) following treatment with the indicated inhibitors;100 μM Ac₅GalNTGc, 100 μM P-3Fax-Neu5Ac for 48 hours or 100 nM StcE mucinase for 1 hour at 37°C. Scale bars, 10 μm. c, Representative pixelwise SAIM interferograms for each condition in b. d, Quantification of glycocalyx thickness in YSCCC cells expressing Bzip treated with the indicated inhibitors. Boxes and whiskers show the first and third quartiles (boxes), median, and range of the data. Each condition includes a minimum of n = 21 cells from a representative experiment. In d, statistical analysis was performed by one-way ANOVA with Tukey’s post hoc tests.

We next investigated how the pharmacological inhibitors of glycosylation and an enzymatic degrader affect the glycocalyx structure in YSCCC cancer cells (Fig. 4a). Using the leucine zipper-based strategy to mark the membrane for SAIM, we measured changes in glycocalyx thickness in YSCCC cancer cells treated with Ac₅GalNTGc, P-3F_AX-Neu5Ac, and StcE mucinase (Fig. 4b). Clear shifts in the pixel-wise SAIM interferograms were observed following inhibition of glycosylation or enzyme-mediated degradation of mucins, indicating a measurable change in glycocalyx thickness (Fig. 4c). In YSCCC cells, the average glycocalyx thickness decreased from 96 ± 16 nm to 83 ± 16 nm after 48 hours of Ac₅GalNTGc treatment, an average reduction of 13 nm. Treatment with 100 nM StcE mucinase resulted in a glycocalyx thickness of 77 ± 10 nm, corresponding to a reduction of 19 nm (Fig. 4d). Notably, YSCCC treatment with the sialylation inhibitor, P-3F_AX-Neu5Ac, did not significantly reduce the glycocalyx thickness compared to untreated controls. These results demonstrate that leucine zipper-based membrane labeling combined with SAIM imaging provides a reliable screening platform for assessing how potential therapeutic agents alter the glycocalyx structure.

We considered the generality of our findings by testing the effects of inhibitors on an additional cell line, KPL-1. While the average glycocalyx thickness in DMSO-treated control KPL-1 cells was 67 ± 7 nm, it decreased by an average of 21 nm to 46 ± 9 nm after 48 hours of treatment with Ac₅GalNTGc (Extended Data Fig. 2d,e). Treatment with P-3F_AX-Neu5Ac reduced glycocalyx thickness to 28 ± 7 nm, representing an even larger decrease of 40 nm. This suggests that the glycocalyx structure of KPL-1 cells is markedly more sensitive to the sialylation inhibitor compared to YSCCC (Fig. 4d and Extended Data Fig. 2e). To assess glycosylation capacity in KPL-1 and YSCCC, transcriptomic data was used to reconstruct biosynthetic pathways in GlycoMaple, a glycosylation mapping tool that has been reported to estimate glycan structures in cells based on gene expression (Extended Data Fig. 3,4)⁴⁴. Although KPL-1 were predicted to have somewhat higher capacity for polysialic acid biosynthesis, sialylation capacity was largely comparable between KPL-1 and YSCCC. Thus, the greater response of the KPL-1 glycocalyx to sialylation inhibition would not have been predicted by gene expression alone. YSCCC were predicted to have a higher capacity for biosynthesis of more extended and complex core-2 O-glycans, which may in part explain their sensitivity to inhibition with Ac₅GalNTGc. Collectively, these results suggest that the response to interference strategies may be cancer cell type specific and difficult to predict, emphasizing the need for the direct measurement of the effects (Fig. 4c). Toward this end, these results highlight that leucine zipper-based SAIM imaging can serve as a powerful screening tool for evaluating the effects of inhibitors and enzymatic degraders on the glycocalyx layer.

Ac₅GalNTGc enhances immune cell cytotoxicity by reducing glycocalyx thickness

To investigate whether the reduction in glycocalyx thickness by Ac₅GalNTGc corresponds with enhanced immune cell recognition and killing, we utilized a previously validated cellular model with readily tunable MUC1 expression via doxycycline-induced expression, referred to here as 1E7¹³. The green fluorescent protein, mOxGFP, was genetically inserted between the ectodomain and transmembrane domain of MUC1, enabling the visualization of cell-surface MUC1 and quantification of its expression level. Using this cellular model system, we first confirmed that the cell-surface MUC1 level of YSCCC is similar to the 1E7 clone. (Fig. 5a). We next used 1E7 cells to examine how Ac₅GalNTGc affects glycosylation of the MUC1 ectodomain displayed on the cell surface. Treatment of 1E7 cells with 100 μM Ac₅GalNTGc resulted in decreased cell-surface expression of both MUC1-GFP and core-1 glycans, as measured by GFP signal and PNA lectin labeling, respectively (Fig. 5b,c). SAIM analysis revealed that higher induction of MUC1-GFP led to a substantial reduction in glycocalyx thickness following Ac₅GalNTGc treatment compared to wild-type 1E7 cells (Fig. 5d). This effect was most pronounced at maximum doxycycline induction, where Ac₅GalNTGc treatment reduced glycocalyx thickness by 13.9 nm relative to wild-type cells (Fig. 5d). These results align with our previous findings demonstrating the impact of O-glycosylation on glycocalyx thickness¹³. We then confirmed that protection against NK-92 cell-mediated killing depended on the glycocalyx thickness of the target cells. Remarkably, we observed a significant increase in NK-92 cell-mediated cytotoxicity against Ac₅GalNTGc-treated cells compared to wild-type 1E7 cells (Fig. 5e). Consistent with the inverse correlation observed in our previous study¹³, these results further demonstrate an inverse relationship between glycocalyx thickness and NK-92 cell-mediated cytotoxicity following treatment with Ac₅GalNTGc (Fig. 5f). To extend these findings to engineered immune cells, we investigated whether Ac₅GalNTGc treatment could enhance CD19-targeted chimeric antigen receptor NK (CD19 CAR-NK) cell efficacy (Fig. 5g). Using 1E7 cells engineered to overexpress CD19, we observed a remarkable 3.02-fold increase in CD19 CAR-NK cell-mediated cytotoxicity following Ac₅GalNTGc treatment (Fig. 5g). These results demonstrate that pharmacological reduction of glycocalyx thickness can significantly enhance both natural and engineered immune cell responses, suggesting potential therapeutic applications in cancer immunotherapy.

Figure 5: Inhibition of O-glycan biosynthesis enhances NK and CAR-NK cell-based cytotoxicity in vitro.

a, Representative flow cytometry analysis comparing MUC1 surface levels between YSCCC cells and stable cell clone with titratable expression of MUC1 ectodomain (1E7 clone) induced with 1 μg/mL doxycycline for 24 hours. b, Fluorescence images of 1E7 cells induced at the indicated concentration of doxycycline with DMSO or 100 μM Ac₅GalNTGc inhibitors for 48 hours. Scale bar, 200 μm. c, Flow cytometric analysis of MUC1 mOxGFP expression and PNA binding in 1E7 cells treated with DMSO or 100 μM Ac₅GalNTGc for 48 hours at indicated doxycycline concentrations. d, Quantification of glycocalyx thickness in 1E7 cells treated with DMSO or 100 μM Ac₅GalNTGc for 48 hours at indicated doxycycline concentrations. The boxes and whiskers show the first and third quartiles (boxes), median and range of data. Each condition includes a minimum of 18 cells from a representative experiment (n = 3 independent experiments). e, NK-92-cell-mediated cytotoxicity against the wild-type and engineered 1E7 cells at the indicated doxycycline induction level. NK-cell-to-target-cell ratio is 5:1. Results are mean ± s.d. of n = 3 technical replicates for one representative of three independent experiments. f, Correlation between NK-92 cell cytotoxicity and glycocalyx thickness (R² = 0.6301); data from panels d and e; dashed line indicates linear regression; the dashed line indicates a linear fit to the data. g, NK-92 and CD19 CAR NK-92-cell-mediated cytotoxicity against 1E7, CD19 OE 1E7, or 100 μM Ac₅GalNTGc treated 1E7 cells at 1,000 ng/ml of doxycycline concentrations. Results are mean ± s.d. of n = 3 technical replicates. In d, e, and g, statistical analysis was performed by one-way ANOVA with Tukey’s post hoc tests.

Safety evaluation and anti-tumor efficacy of StcE-NK cells in YSCCC xenograft model

In addition to Ac₅GalNTGc, enzymatic mucin degraders showed potential as agents to compromise the integrity of the CCA glycocalyx (Fig. 3c-j). However, severe adverse effects have been observed in rodent models following systemic injection with StcE¹¹. To overcome this safety issue, work from our group has proposed that immune cells could be engineered to deliver enzymes locally to target cancer cells, thus, lowering the total enzyme dosage that would be required for anti-tumor functionality¹³. In one implementation of this strategy, StcE mucinases were coupled to the surface of NK cells utilizing the same leucine zipper pairs presented in this study for membrane labelling¹³ (Fig. 6a). Employing this strategy here, in vitro cytotoxicity assays demonstrated that StcE-tethered NK cells exhibited 2.9-fold enhanced killing activity against YSCCC cells compared to unmodified NK cells (Fig. 6b).

Figure 6: Safety evaluation and anti-tumor activity of StcE-NK cells in YSCCC xenograft model.

a, Cartoon showing the design of NK cells tethering with StcE cells (StcE-NK) using leucine zippers. b, Killing of YSCCC cells by NK-92, NK-92 expressing Bzip construct (NK-Zip), and NK-Zip tethered with StcE. NK-cell-to-target-cell ratio is 10:1. Results are mean ± s.d. of n = 3 technical replicates. c, Schematic of experimental setup in which 10 × 10⁶ cells of NK cell, Bzip overexpressing NK cell, or StcE-NK were treated intraperitoneally in the flank of NSG mice. d, Change in body weight over as measured (n = 3 mice per condition). e, Schematic of experimental setup in which 5 × 10⁶ YSCCC cells were subcutaneously implanted in the flank of NSG mice. f,g, Tumor volume (f) and change of body weight (g) over time as measured by caliper (n = 5 mice). h, Schematic of experimental setup in which 5 × 10⁶ YSCCC cells were subcutaneously implanted in the flank of NSG mice. Mice were treated intratumorally with PBS, 3 × 10⁶ wild-type NK-92 cells, or 3 × 10⁶ StcE-NK cells three times. i-k, Tumor volume (i,j) and change of body weight (k) over time as measured by caliper (n = 5 mice per treatment group). In b and i, statistical analysis was performed by one-way ANOVA with Tukey’s post hoc tests (b) and two-way ANOVA with correction for multiple comparisons (i).

Given these promising results, we tested the safety and functionality of the StcE-NK cells in vivo. To assess the safety profile of StcE-NK cells, we conducted toxicity studies in immunocompromised NSG mice using the maximum therapeutic dose of 10 × 10⁶ cells administered via intraperitoneal (IP) injection (Fig. 6c). Body weight monitoring over 25 days revealed no significant toxicity across all NK cell treatment groups, including the StcE-NK cohort (Fig. 6d). Comprehensive histopathological evaluation of major organs (liver, spleen, kidney, small intestine, pancreas, heart, gallbladder, lung, brain, stomach, and esophagus) by an expert pathologist revealed no histological abnormalities. These findings confirm both the suitability of the YSCCC dose for consistent tumor establishment and the favorable safety profile of StcE-NK cells in NSG mice.

To validate these findings in vivo, we established a new YSCCC mouse xenograft model. Immunocompromised NSG mice received 5 × 10⁶ YSCCC cells via either IP or subcutaneous (SubQ) injection routes (Fig. 6e). Our experiments demonstrated robust engraftment and tumor growth following SubQ implantation (Fig. 6f,g). Engraftment as assessed by magnetic resonance imaging (MRI) was not observed in the IP model. Therefore, to evaluate therapeutic efficacy in vivo, we established SubQ YSCCC xenografts by injecting 5 × 10⁶ YSCCC cells into NSG mice. Once tumors reached an average volume of 150–250 mm³, we initiated treatment with intratumoral injections of 3 × 10⁶ wild-type or StcE-NK cells. Treatment was administered weekly for three consecutive weeks, and tumor progression was monitored via caliper measurements over a 53-day period (Fig. 6h). Consistent with our in vitro result in Fig. 6a, StcE-NK cells demonstrated significantly enhanced anti-tumor activity compared to both wild-type NK cells and PBS control groups (Fig. 6i,j). Importantly, no treatment-related toxicity was observed throughout the study period, as evidenced by stable body weights and normal behavioral patterns across all treatment groups (Fig. 6k). These results validate the therapeutic potential of StcE-NK cells against MUC1-overexpressing CCA in vivo.

Discussion

In this study, we developed a novel membrane fluorescent labeling approach by combining leucine-zipper pairs and sortase-mediated reactions to overcome critical limitations of conventional methods, such as poor membrane retention and non-specific labeling. We further combined this labeling strategy with SAIM imaging to precisely quantify cancer cell glycocalyx thickness with nanoscale resolution. We used this platform to identify the small-molecule Ac₅GalNTGc and StcE mucinase as potent agents to reduce glycocalyx thickness in multiple MUC1-overexpressing cancer cell lines, such as YSCCC and KPL-1. We further validated that StcE treatment can reverse the physical effects of the repulsive mucin barrier on YSCCC cells in cell-substrate and cell-cell adhesion assays. Based on these mechanistic insights, we validated a therapeutic approach using StcE-tethered NK cells, which effectively penetrated the mucin barrier and demonstrated significant anti-tumor activity in a xenograft model, establishing a promising strategy for targeting mucin-overexpressing cancers. These findings establish our imaging platform as a valuable tool for screening glycocalyx-modifying compounds and validate glycocalyx reduction as a promising strategy for improving immunotherapy efficacy.

Our leucine zipper-based labeling strategy, while enabling glycocalyx measurements across diverse cancer types, has certain limitations. A key technical constraint is the requirement for Bzip overexpression in target cells through genetic modification. Although our analyses revealed that Bzip expression does not significantly alter cell viability or morphology, comprehensive characterization of its potential effects on membrane properties is pending. Future studies should specifically investigate how Bzip expression might influence adhesion molecule dynamics, membrane fluctuations, and nanoscale organization of cell surface components. Additionally, while beyond the scope of the current study, the development of orthogonal leucine zipper labeling approaches could provide complementary insights into membrane protein interactions and dynamics⁴⁵. Such multi-modal analyses would further enhance our understanding of the complex interplay between glycocalyx structure and membrane organization in cancer cells.

The robustness of our system across different cancer cell types also enabled the observation of distinct patterns of glycocalyx regulation across different cancer types. KPL-1 breast cancer cells showed remarkable sensitivity to sialylation inhibition, with P-3F_AX-Neu5Ac treatment reducing glycocalyx thickness by an average of 40 nm. In contrast, YSCCC intrahepatic CCA cells demonstrated more significant response to O-GalNAc modification through Ac₅GalNTGc treatment. These differential responses suggest that cancer cells may develop specific dependencies on particular glycosylation pathways, potentially reflecting their tissue of origin or malignant adaptation⁴⁶. Understanding these cell type-specific dependencies could inform the design of more refined targeted therapeutic strategies.

While MUC1-targeted immunotherapy continues to evolve, clinically significant results remain elusive^47,48. Our approach differs fundamentally from conventional CAR-T or CAR-NK strategies by directly addressing the mucin barrier that impedes immune cell recognition. By employing the StcE enzyme to breach this barrier, we enhance immune cell recognition and targeting efficiency. This strategy holds potential for combination with various immunotherapeutic approaches, including CAR-T cells, although further investigation of StcE-NK cell mechanisms is necessary. While our current study demonstrated efficacy through intratumoral injection, future research should explore alternative administration routes, particularly intravenous delivery, to optimize therapeutic applications.

Our findings have important implications for clinical translation. The differential glycocalyx regulation patterns we observed across cancer types suggest the need for personalized approaches to glycocalyx-targeting therapies. Moreover, our platform provides a valuable tool for patient stratification and monitoring treatment response. While our study demonstrates the potential of glycocalyx-targeting approaches, several limitations should be addressed in future work. These include the need for long-term safety studies of StcE-NK cells, optimization of delivery methods, and investigation of potential resistance mechanisms. Future studies should focus on several key areas, such as optimization of StcE-NK cell delivery methods for systemic administration and investigation of potential combination strategies with existing immunotherapies.

Materials and methods

Cell culture

MCF10A cells were cultured in DMEM/F12 media (Thermo Fisher Scientific) supplemented with 5% horse serum (Thermo Fisher Scientific), 20 ng/ml EGF (Pepro Tech), 10 mg/ml insulin (Sigma), 500 ng/ml hydrocortisone (Sigma), 100 ng/ml cholera toxin (Sigma) and 1x penicillin/streptomycin (Thermo Fisher Scientific) at 37°C in 5% CO₂. HEK293T cells (gift from Valerie Weaver) were cultured in DMEM high glucose media (Thermo Fisher Scientific) supplemented with 10% fetal bovine serum (Thermo Fisher Scientific) and 1x penicillin/streptomycin at 37°C in 5% CO₂. YSCCC (RIKEN BRC; RCB1549), ZR-75–1 (ATCC; CRL-1500), T47D (ATCC; HTB-133), SKBR3 (gift from Dr. Jan Lammerding), KPL-1 (DSMZ; ACC 317), and Capan-2 (ATCC; HTB-80) cells were cultured in RPMI 1640 media (Thermo Fisher Scientific) supplemented with 10% fetal bovine serum and 1x penicillin/streptomycin at 37°C in 5% CO₂. NK-92 cells (ATCC; CRL-2407) were cultured in ɑ-MEM without ribonucleosides media (Thermo Fisher Scientific) supplemented with 12.5% fetal bovine serum, 12.5% horse serum, 0.2 mM Myo-inositol (Sigma Aldrich), 0.1 mM 2-mercaptoethanol (Thermo Fisher Scientific), 0.02 mM folic acid (Millipore Sigma), 100 U/ml recombinant human IL-2 (Pepro Tech), and 1x penicillin/streptomycin at 37°C in 5% CO_2.

Immunostaining and live cell imaging

Target cells were plated in 35-mm glass-bottom dishes and cultured for 24 hours. For standard conditions, cells were labeled with 1 μM Azip-based dyes (Azip-sfGFP, Azip-AF488, or Azip-AF647) in phenol red-free culture media for 10 minutes at 37°C. For cold conditions, cells were labeled with 1 μM Azip-based dyes in 0.5% BSA PBS for 1 hour at 4°C. For 48-hour live cell imaging experiments, cells were labeled according to manufacturers’ recommended buffer compositions and concentrations for 10 minutes at 37°C. Before imaging, cells were washed twice with either cold PBS or phenol red-free culture media. All samples were imaged at 0, 4, 24, and 48 hours at 37°C under 5% CO2.

For live cell imaging of 1E7 cells in response to inhibitors, cells were first induced with various concentrations of doxycycline for 24 hours, then treated with either DMSO or 100 μM Ac₅GalNTGc for 48 hours. All imaging was performed using an LSM 800 confocal microscope with either a ×20 air objective (NA 0.8) or ×63 water objective (NA 1.2) (ZEISS). ImageJ was used for image analysis.

Flow cytometry

To measure cell surface Muc1 expression in cancer cell lines, cells were plated, grown for at least 48 hours, and detached using trypsin. Anti-Muc1 antibody clone HMPV (555925, BD Biosciences) was diluted 1:200 in 0.5% w/v BSA in 1x PBS and incubated with the cells at 4°C for 1 hour. Secondary labelling was with Alexa Fluor 647 conjugated goat anti-mouse, diluted 1:200 in 0.5% w/v BSA in PBS and incubated with cells at 4°C for 1 hour. In each experiment, SKBR3 cells were included as a standard for comparison. Median fluorescence intensity was calculated in FlowJo and values were normalized to the median fluorescence intensity of SKBR3 cells in each experiment. For lectin staining, detached target cancer cells were incubated with PNA-CF650R at 4°C for 1 hour. Lectin was diluted 1:200 in 0.5% BSA PBS and incubated with cells at 4°C for 1 hour. The Attune NxT flow cytometry (Thermo Fisher Scientific) was used for analysis.

Sortase-mediated labeling reaction

100 μM Azip-LPETG-sfGFP was incubated in reaction buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 10 mM CaCl2) containing 20 μM sortase and 10 mM triglycine (Gly₃)-conjugated AF488 or AF647. After 1–3 hours of incubation at room temperature, the reaction products were purified using HisPur Ni-NTA Resin to remove unreacted His6-tagged Azip dye and His6-tagged Sortase in elution buffer (20 mM HEPES, 500 mM NaCl, 10 mM imidazole, pH 7.5). The column flow-through was then buffer-exchanged to PBS to remove unbound dye.

NK-92 cell-mediated cytotoxicity assays

Target cells were detached and fluorescently labeled for 15 minutes with 10 μM CellTracker Green CMFDA Dye (Invitrogen) in growth media, followed by washing thoroughly twice with growth media. 4×10⁴ labeled target cells were mixed with varying ratios of NK-92 cells in 200 μL of growth media of the target cell line in the absence of IL-2 and co-cultured in an ultra-low attachment U-bottom 96-well plate (Corning) for 4 hours at 37°C in 5% CO₂. Following the 4-hour co-culture, the mixed cells were pelleted by centrifugation at 500 g for 5 minutes, resuspended and incubated in propidium iodide solution (PI; 20 μg/mL, Sigma) for 10 minutes. NK cell-mediated cytotoxicity was then measured by flow cytometry as previously described^7,13,49. At least 1 ×10⁴ tumor cells were analyzed after electronic gating on CellTracker Green. To calculate the percent cytotoxicity, the following formula was used: 100 × (experimental % dead – spontaneous % dead)/(100 – spontaneous % dead), where experimental % dead was the percentage of PI positive tumor cells in co-cultures and spontaneous % dead was the percentage of PI positive control tumor cells cultured in the absence of effector cells.

Scanning angle interference microscopy (SAIM)

Silicon wafers with a ~2,000 nm thermal oxide layer (Addison Engineering) were diced into 7 × 7 mm chips, and the oxide layer thickness of each chip was measured with a FilMetrics F50-EXR. Silicon chips then were functionalized using 4% (v/v) (3-mercaptopropyl)trimethoxysilane in absolute ethanol for 30 minutes at room temperature, followed by incubation with 4 mM 4-maleimidobutyric acid N-hydroxysuccinimide ester in absolute ethanol and 50 μg/ml human Alexa Fluor 647 conjugated plasma fibronectin as previously reported^13,50. Cells were seeded onto the fibronectin-coated chips at 0.5–1 × 10⁵ cells/cm² in full culture medium. After 24 hours, adhered cells were rinsed with phenol-red free RPMI and incubated with Azip-AF488 in phenol-red free RPMI for 10 minutes at 37°C. For SAIM imaging of 1E7 cells, cells were induced with various doxycycline concentrations for 24 hours and then rinsed with serum-free, phenol-red free DMEM and incubated with MemGlow dyes (MemGlow 560, MG02–2; Cytoskeleton) in serum-free, phenol red-free DMEM for 10 minutes at 37°C. Cell-seeded chips were then washed with serum-free, phenol red-free DMEM or phenol red-free RPMI again, and inverted onto a 35 mm glass-bottom imaging dish and imaged at 37°C. As previously reported¹³, SAIM was conducted on a custom circle-scanning microscope which allowed imaging at varying incidence angles, ranging from 5 to 43.75 degrees, and a total of 32 images was acquired per cell. The intensities of raw image sequences were fit pixelwise by nonlinear least-squares to an optical model:

$$I_j=A\mathrm{*}f\left({\theta }_j,H\right)+B$$

where $I_j$ is raw image intensity at each incidence angle ${\theta }_j$, $\mathrm{H}$ is the glycocalyx thickness, and $\mathrm{A}$ and $\mathrm{B}$ are additional fit parameters. The optical system maintained the s-polarization of circle-scanned excitation laser by the vortex half-wave plate. The probability of excitation is given by:

$$f\left({\theta }_j,H\right)=1+2Re\left\{r^{TE}\right\}\mathrm{c}\mathrm{o}\mathrm{s}\varphi -2Im\left\{r^{TE}\right\}\mathrm{s}\mathrm{i}\mathrm{n}\varphi +Re{\left\{r^{TE}\right\}}^2+Im{\left\{r^{TE}\right\}}^2$$

Where $\mathrm{\varphi }\left(H\right)$ is the phase shift, $\mathrm{\lambda }$ is s-polarized monochromatic excitation of wavelength, and $r^{TE}$ is the reflection coefficient for the transverse electric wave and these are given by:

$$\varphi (H)=\frac{4\pi }{\lambda }\left(n_{b}H\mathrm{c}\mathrm{o}\mathrm{s}{\theta }_b\right)$$

$$r^{TE}=\frac{\left(m_{11}^{TE}+m_{12}^{TE}{p}_{Si}\right)p_2-\left(m_{21}^{TE}+m_{22}^{TE}{p}_0\right)}{\left(m_{11}^{TE}+m_{12}^{TE}{p}_{Si}\right)p_2+\left(m_{21}^{TE}+m_{22}^{TE}{p}_0\right)}$$

$$M^{TE}=\left(\begin{array}{cc}{m}_{11}^{TE}& m_{12}^{TE}\\ m_{21}^{TE}& m_{22}^{TE}\end{array}\right)=\left(\begin{array}{cc}\mathrm{c}\mathrm{o}\mathrm{s}\left(k_{ox}{d}_{ox}cos{\theta }_{ox}\right)& -\frac{i}{p_1}\mathrm{s}\mathrm{i}\mathrm{n}\left(k_{ox}{d}_{ox}cos{\theta }_{ox}\right)\\ -i{p}_1\mathrm{s}\mathrm{i}\mathrm{n}\left(k_{ox}{d}_{ox}cos{\theta }_{ox}\right)& \mathrm{c}\mathrm{o}\mathrm{s}\left(k_{ox}{d}_{ox}cos{\theta }_{ox}\right)\end{array}\right)$$

$$p_0=n_{Si}cos{\theta }_{Si},p_1=n_{ox}cos{\theta }_{ox},p_2=n_{b}cos{\theta }_b$$

$$k_i=\frac{2\pi n_i}{\lambda },{\theta }_{ox}={sin}^{-1}\frac{nsin{\theta }_b}{n_{ox}},{\theta }_{Si}={sin}^{-1}\frac{nsin{\theta }_{ox}}{n_{Si}}$$

where $k_i$ is the wavenumber in material $\mathrm{i};n_{Si},n_{ox}$ and $n_b$ are the refractive index of the silicon, silicon oxide and sample, respectively; ${\theta }_{Si},{\theta }_{ox}$ and ${\theta }_b$ are the angles of incidence in the silicon, silicon oxide and sample, respectively; and $d_{ox}$ is the thickness of the silicon oxide layer. The angles of incidence in silicon oxide and silicon were calculated according to Snell’s Law. The average glycocalyx thickness was quantified in 200 × 200 pixel subregions of each cell by subtracting the height of their MemGlow signal or Azip signal from the height of the corresponding fluorescently labeled fibronectin on the silicon substrate.

Preparation of recombinant StcE and SmE mucinases

The cDNA for StcE-Δ35 and SmE was synthesized by custom gene synthesis (Twist Bioscience) and inserted into the pET28b expression vector³⁷. The recombinant enzymes were produced by expressing them in chemically competent NiCo21 (DE3) E. coli (NEB). Following transformation and overnight growth on Luria Broth (LB) agar plates, the cells were cultured in LB medium at 37°C until an OD600 of 0.6–0.8 was reached. At this point, the cultures were induced with 0.5 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) and grown overnight at 24°C. Cells were then harvested by centrifugation at 3,000 g for 20 minutes, resuspended in lysis buffer (20 mM HEPES, 500 mM NaCl and 10 mM imidazole, pH 7.5) with cOmplete protease inhibitor Cocktail (Roche), and lysed by a sonicator (Q125, Qsonica). Recombinant enzymes were purified by immobilized metal affinity chromatography (IMAC) on a GE ÄKTA Avant FPLC system. The lysate was applied to a HisTrap HP column (Cytiva), followed by a wash step with 20 column volumes of wash buffer (20 mM HEPES, 500 mM NaCl and 20 mM imidazole, pH 7.5). Elution was performed using a linear gradient of 20 mM to 250 mM imidazole in buffer (20 mM HEPES and 500 mM NaCl, pH 7.5). The eluted fractions containing target protein were collected and further refined by a HiPrep 26/60 Sephacryl S-200 HR (Cytiva) column equilibrated with storage buffer (20 mM HEPES and 150 mM NaCl, pH 7.5). The final protein was concentrated by using Amicon Ultra 30 kDa MWCO filters (Millipore Sigma).

Assessment of cell-substrate adhesion

YSCCC cells were plated at 80,000 cells per well in a 24-well plate (Cellstar, #662160) in RPMI 1640 media with 10% fetal bovine serum and 1% penicillin/streptomycin. Where specified, StcE or SmE mucinases were added to a final concentration of 1 nM. Cells were allowed to adhere to the plates for 18 hours in 37°C and 5% CO₂. Phase contrast images were then acquired on a BZ-X810 fluorescence box microscope (Keyence) using a 20x (NA: 0.45) air objective. Cells were manually segmented and counted in ImageJ to calculate circularity and fraction of rounded cells (cells with a Circularity > 0.9). Rounded cell fraction was calculated as (# rounded cells)/(# rounded cells + # spread cells) in each field of view.

Assessment of cell-cell adhesion

YSCCC cells were detached with 0.25% trypsin. Cells were seeded at 1,000 cells per well in 150 uL 96-well ultra-low adhesion plates (Corning #7007) in the presence of 0.05% methylcellulose (M6385, Millipore Sigma). Where specified, StcE was added to a final concentration of 10 nM. The plate was then gently centrifuged at 150×g for 3 minutes to ensure that all cells collected at the bottom of each well. Cells were then allowed to form aggregates for 24 hours at 37°C and 5% CO₂. Cell aggregates were then imaged on an LSM 800 confocal microscope using the brightfield mode with a 10x (NA: 0.3 Air) objective. Aggregates were then manually segmented in ImageJ to calculate overall aggregate area.

Mouse Xenograft Model

Male 8–10 week-old in-house bred NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ (NSG) mice (The Jackson Laboratory Catalog #005557) were housed in groups of up to five under pathogen-free conditions. The animals were kept at temperatures of 21.1–24.5°C (70–76°F), 30–70% humidity, and a 12:12 light-dark cycle. All mice were housed at the East Campus Research Facility (ECRF) at Cornell University and all care and experiments were conducted in accordance with protocols approved by the Cornell University Institutional Animal Care and Use Committee under protocol number 2017–0035. Cell line derived xenograft (CDX) mice (n = 20) were established by injecting cultured YSCCC cells at a titer of 5×10⁶ cells per 100uL and with 100uL of Corning Matrigel into the left rear flank of NSG mice while under 2.5% isoflurane anesthesia. Mice were monitored twice per week for tumor growth using digital calipers. Once tumor burden reached start criteria of 150–200mm³ mice were randomly sorted into one of three treatment groups: StcE-tethered NK-92, wild-type NK-92, or PBS (n = 5/group). Mice were intratumorally injected with treatment compounds using a 28G BD Insulin Syringe (Catalog# 329461) once per week for three weeks. Body weight and caliper measurements were taken twice per week while on treatment. At the end of the experiment mice were sacrificed using CO₂ euthanasia and tumors were excised from the animals and placed in 10% neutral buffered formalin for further histological analysis.

Statistical analysis

Unless otherwise indicated, results are presented as the mean and standard deviation (s.d.) of at least three replicates per condition using GraphPad Prism 9. Statistical differences were determined using a two-tailed unpaired t-test for two group comparisons, one-way ANOVA with multiplicity-adjusted p values from Tukey’s multiple comparisons test, and two-way ANOVA with correction for multiple comparisons. Statistical significance was determined using two-tailed unpaired t-tests, one-way ANOVA, and two-way ANOVA, with p < 0.05 considered significant across all analyses.