Sialyl Lewisx-P-selectin cascade mediates tumor–mesothelial adhesion in ascitic fluid shear flow

Organ-specific colonization suggests that specific cell–cell recognition is essential. Yet, very little is known about this particular interaction. Moreover, tumor cell lodgement requires binding under shear stress, but not static, conditions. Here, we successfully isolate the metastatic populations of cancer stem/tumor-initiating cells (M-CSCs). We show that the M-CSCs tether more and roll slower than the non-metastatic (NM)-CSCs, thus resulting in the preferential binding to the peritoneal mesothelium under ascitic fluid shear stress. Mechanistically, this interaction is mediated by P-selectin expressed by the peritoneal mesothelium. Insulin-like growth factor receptor-1 carrying an uncommon non-sulfated sialyl-Lewisx (sLex) epitope serves as a distinct P-selectin binding determinant. Several glycosyltransferases, particularly α1,3-fucosyltransferase with rate-limiting activity for sLex synthesis, are highly expressed in M-CSCs. Tumor xenografts and clinical samples corroborate the relevance of these findings. These data advance our understanding on the molecular regulation of peritoneal metastasis and support the therapeutic potential of targeting the sLex-P-selectin cascade.

O varian cancer has the lowest 5-year survival rate ( < 25%) among all gynecological malignancies owing to extensive peritoneal metastatic lesions 1 . Despite not as common as blood-borne metastasis, treatment of peritoneal metastasis is notoriously challenging because of the rapid growth of metastasis in the peritoneal cavity in a positive feed-forward manner 2 . Current treatment is dissatisfying and little overall clinical benefits for patients have been achieved over the last several decades. Unraveling the underlying molecular mechanisms regulating this process will conceivably benefit the development of effective treatment strategies.
The barriers to metastasis are distinct in different organs, suggesting that specific recognition between cancer cells and peritoneal mesothelium is critical for the peritoneum-specific colonization 3 . Moreover, tumor cell lodgment in the peritoneum is consistently exposed to shear forces generated by ascitic flow 4 . However, owing to the difficulty in manipulating the dynamic flow conditions, the role of mechanical force in this adhesion has been largely neglected. Thus, the adhesion molecules and the underlying signaling operating the adhesion of cancer cells under ascitic flow remain a gap in knowledge.
Selectins (E-, P-, and L-selectin) are a family of calciumdependent glycoproteins that are the prime glycan receptors on shear-resistant interactions as described to date 5 . Although the structures of the three selectins are highly similar, they have different tissue distribution and binding kinetics, suggesting that they have different roles in various pathophysiological processes, including tumor metastasis. Thus, the selective binding between selectin and its ligand determines the metastatic destination. Tumor cells exploit the selectin-glycan binding for the initial cell-cell interaction, including tethering and rolling, under shear stress that further triggers the molecular signaling that facilitates cellular firm adhesion 6 . Such interaction has been extensively studied in blood-borne metastasis 7,8 . In contrast, selectins are expressed on peritoneal mesothelial cells, suggesting that similar receptor/ligand cascade initiated by selectin-mediated interaction may promote tumor cell peritoneal targeting 9,10 . However, the detailed underlying mechanism governing adhesion for peritoneal metastasis is poorly understood. The difference in the peritoneal shear stress (~0.1 dyne cm −2 ; at least 10-fold lower than that of the vascular shear stress) 11 suggests that the biology of peritoneal dissemination is different from that of blood-borne metastasis and probably different molecular mechanisms are involved.
In this study, we reveal a preferential binding of metastatic cancer stem/tumor-initiating cells (M-CSCs) to the peritoneal mesothelium than the non-metastatic (NM)-CSCs. We also provide evidence that P-selectin is a key molecule on the peritoneal mesothelium that mediates the binding through a unique sialyl-Lewis x (sLe x ) containing insulin-like growth factor receptor-1 (IGF-1R) ligand under ascitic flow-induced shear stress.

Results
Establishment of M-/NM-CSC models. To explore the limiting factors that determine the metastatic success of ovarian cancer, we reasoned that metastatic colonization is a highly inefficent process and is accomplished by only a subset of cancer stem/ tumor-initiating cells (CSCs) 3 . We have previously isolated CSCs 12 . Using in vivo selection by orthotopic implantation of CSCs into the bursa of the mouse ovary, we further isolated the highly metastatic population of CSCs (M-CSCs) in SKOV-3 ( Fig. 1a), which consistently metastasized to the peritoneum, recapitulating the clinical progression of human ovarian cancer, when implanted either by orthotopic or intraperitoneal (i.p.) injection (Fig. 1b-e). Conversely, NM-CSCs, which are equally tumorigenic (Fig. 1d, e), did not metastasize in both models ( Fig. 1b-e, Table 1). Similar metastatic abilities were observed in CSCs-enriched cultures generated from primary ovarian cancer samples (Fig. 1f, Table 1). Moreover, M-CSCs had a gene expression profile similar to the metastatic tumors in ovarian cancer patients (Fig. 1g), suggesting that the in vitro M-CSCs could closely reflect the spontaneous metastatic cellular events in patients. We have also derived M-and NM-CSCs from HEYA8 cells, and similar results were obtained ( Supplementary Fig. 1a, b).
Differential capabilities in tethering, rolling, and adhesion. To evaluate whether the distinct metastatic phenotypes in the two CSCs populations have different abilities in tumor-mesothelial interaction, we have developed a customizable microfluidic platform to overcome the technical limitations of the conventional approaches and to recapitulate peritoneal dissemination in dynamic flow 13 . CSCs spheroids were perfused to the microfluidic chip coated with primary human peritoneal mesothelial cells (HPMCs) under well-defined flow rate. As shown, the CSCs were effectively captured to HPMCs at 0.03-0.15 dynes cm −2 (a physiologically relevant range of shear stress of the ascites 14 ) with maximal activity at a shear stress of 0.05 dynes cm −2 (Fig. 2a), consistent with a catch bond that requires a clear threshold shear to initiate rolling 15 . Strikingly, both the kinetic and mechanical properties that govern tumor-mesothelial interaction were significantly different between the two CSCs populations. Compared with NM-CSC, M-CSCs had a higher tethering frequency (1.8fold at 0.05 dynes cm −2 ) (Fig. 2a, Supplementary Fig. 1c) and lower rolling velocity (Fig. 2b, Supplementary Fig. 1d) on HPMCs, resulting in a higher percentage of firm adhesion of M-CSCs to HPMCs (2.3-fold at 0.05 dynes cm −2 ) (Fig. 2c, Supplementary Fig. 1e). Although HPMCs is sufficient to induce epithelial-to-mesenchymal transition in certain contexts 16 , we showed that these cells retained their epithelial morphology with closely packed colonies of cells ( Supplementary Fig. 2a) and Ncadherin and vimentin mRNA levels were similar between the primary culture and the later subculture ( Supplementary Fig. 2b).

P-selectin mediates M-CSC-HPMC interaction under flow.
HPMCs clearly showed constitutive cell surface expression of selectins ( Fig. 3a, b). We also performed real-time PCR expression to verify the expression of E-, P-, and L-selectin. Like endothelial cells, used as a reference cell population, selectins are expressed on HPMCs, suggesting that their expression outside the vasculature is present, which are consistent with previous observations 9,10 ( Supplementary Fig. 2c). To determine which selectin on HPMCs plays a major role in capturing M-CSCs under flow, anti-selectin blocking antibodies were used. Treatment with anti-E-, P-, or L-selectin increased the rolling velocities of M-CSCs when compared with that of isotype control (24.8% increase by anti-E-selectin, 63.3% increase by anti-P-selectin, and 22.5% increase by anti-L-selectin in median rolling velocities) (Fig. 3c, Supplementary Fig. 3a). Notably, blockade of P-selectin almost completely abrogated (~88%) tumor-mesothelial adhesion at 0.05 dynes cm −2 , whereas inhibition of E-selectin or Lselectin had partial effects (48% by anti-E-selectin; 28% by anti-Lselectin) (Fig. 3d, Supplementary Fig. 6b, Supplementary Video 1, 2). Consistently, significant adherence to P-selectin-Fc (59%), but not E-selectin-or L-selectin-Fc, was observed with M-CSCs (Fig. 3e, Supplementary Fig. 3,c), confirming a predominant role for P-selectin. In contrast, NM-CSCs did not show similar interactions with selectin-Fcs. We further showed that CSCs in ovarian cancer patients' ascites utilized P-selectin-dependent binding as observed in M-CSCs, as tumor spheroids isolated from ascites rolled and adhered onto the P-selectin-Fc but not Fc control (Fig. 3f, Supplementary Video 3, 4). The P-selectin-Fcmediated binding was shear-resistant and Ca 2+ -dependent. Once the M-CSCs had adhered to the HPMCs, the bonding between M-CSCs and HPMCs was strong that the M-CSCs can only be completely dissociated when a maximal shear stress of 4 dynes cm −2 was applied ( Supplementary Fig. 4a). Treatment of M-CSCs with ethylenediaminetetraacetic acid (EDTA) inhibited the adhesion of M-CSCs to HPMC (Supplementary Fig. 4b) and could easily detach the adherent M-CSCs from the HMPCs and P-selectin chimera, even at shear stress~1 dynes cm −2 (Supplementary Fig. 4c).  Results are represented with mean ± SEM from two independent experiments, unpaired Student's t test. ns, not significant a P < 0.01 P-selectin is required for peritoneal metastasis in vivo. Next, we took advantage of the Selp −/− Rag2 −/− mice to study the requirement of P-selectin in intraperitoneal metastasis of M-CSCs. In mice with wild type P-selectin (Selp WT Rag2 −/− ), M-CSCs quickly disseminated throughout the peritoneal cavity after orthotopic or i.p. injection (Fig. 3g, h) with visible tumor nodules growing on the omentum, mesenteries and small bowels (Fig. 3g, h) and developed massive ascites (Supplementary Table 4). In contrast, the metastatic progression was markedly reduced in Selp −/− Rag2 −/− mice (Fig. 3g, h) with no significant difference in primary tumor growth (Fig. 3i, j). To further confirm the role of P-selectin and to explore whether therapeutic agent against P-selectin is able to eliminate tumor-mesothelium adhesion, we tested KF38789, a selective inhibitor of P-selectinmediated cell-cell adhesion in vivo. Concurrent with the above results, KF38789 significantly inhibited the extent of tumor metastases on the omentum, mesenteries, and peritoneal wall when treated 1 h prior to the injection of M-CSCs derived from both SKOV-3 and ovarian cancer patients (Fig. 3k), confirming the specific role for P-selectin in mediating M-CSCs adhesion. sLe x on an O-glycoprotein is P-selectin ligand on M-CSCs. Sialylation, fucosylation, and sulfation are the main features of Pselectin ligands 17 . As shown, the removal of sialic acid with neuraminidase reduced M-CSCs binding to HPMCs or Pselectin-Fc by~50% (Fig. 4a). Notably, almost completely abolished M-CSCs binding to HPMCs or P-selectin-Fc was observed upon the removal of fucose with fucosidase ( Fig. 4b), whereas inhibition of sulfation by sodium chlorate did not alter the binding activity (Fig. 4c), indicating that the binding between M-CSCs and HPMCs or P-selectin-Fc is highly fucosylation-and relatively weaker sialyation-, but not sulfation-dependent. We next examined the involvement of fucosylated and sialylated glycans, sLe x or sLe a , in mediating M-CSCs adhesion. As shown in Fig. 4d, e, preincubation of M-CSCs with anti-sLe x -blocking antibody reduced binding of M-CSCs to HPMCs and P-selectin-Fc, whereas anti-sLe a -blocking antibody did not affect the interaction. By flow cytometry, whereas M-CSCs, but not NM-CSCs, displayed reactivity of HECA-452 and CSLEX-1, which recognizes sLe a/x and sLe x , respectively, they showed no difference on anti-sLe a (CA19-9) expression ( Fig. 4f, Supplementary Fig. 5). We further found that sLe x -containing glycan is modified on a . Statistical analysis using chi-square test a, c and one-tailed unpaired Student's t test b. ns, not significant. *P < 0.05; **P < 0.01; ***P < 0.001 protein but not a lipid, as protease (trypsin) treatment on M-CSCs almost completely abolished the adhesion of M-CSCs on both HPMCs and P-selectin-Fc (Fig. 4g), whereas glycolipid synthesis inhibitor (PPMP) treatment had no effect (Fig. 4h). Treatment of O-sialoglycoprotein endopeptidase (OSGE) substantially reduced the adhesion of M-CSCs on both HPMCs and P-selectin-Fc ( Fig. 4i), whereas treatment of Peptide:N-glycosidase F (PNGase) for removal of N-glycans had no effect (Fig. 4j).

P-selectin ligand on M-CSCs is mainly conferred by IGF-1R.
To identify the protein presenting the sLe x on M-CSCs that mediates the P-selectin interaction on HPMCs, we examined a set of inhibitors at effective concentrations that specifically target various cell surface receptor tyrosine kinase (RTK), including c-Met, EGFR, FGFR, and IGF-1R, which have been previously implicated in the progression of ovarian cancer [18][19][20][21] . We found that only genistein (general RTK inhibitor) and AG1024 (IGF-1R inhibitor) led to a marked decrease in the binding of M-CSCs to HPMCs or P-selectin-Fc (Fig. 5a, Supplementary Fig. 6). Moreover, blockade of M-CSCs adhesion on HPMCs or P-selectin-Fc binding was observed with the anti-IGF-1R (Fig. 5b). Furthermore, endogenous IGF-1R could be readily detected in the Pselectin-Fc immunoprecipitate (Fig. 5c). Significant amount of sLe x was detected in both IGF-1R subunits (Fig. 5d). The levels of phospho-IGF-1R were elevated soon upon P-selectin-Fc binding (Fig. 5e). These experiments were all performed in the absence of IGF-1, suggesting a ligand-independent activation of IGF-1R upon P-selectin binding. The known P-selectin ligands bearing sLe x , such as CD24, was detected. However, unlike IGF-1R, there was no differential expression of CD24 between M-CSCs and NM-CSCs (Fig. 5f).
FUT5 is critical for ovarian tumor progression. To further define the molecular properties that contribute to P-selectin binding, we examined the expression profile of glycosyltransferases involved in sLe x synthesis (Fig. 6a) in M-CSCs and NM-CSCs. We found that several glycogenes, including B4GalT4, ST3Gal3, ST3Gal4, and FUT5, were significantly increased in M-CSCs when compared with NM-CSCs (Fig. 6a, Supplementary  Fig. 7a). Similar higher expression of these genes on patients' derived CSCs was also observed (Fig. 6a). The high mRNA expression of glycosyltransferases correlates with poorer progression-free survival in patients with advanced (stage III, IV) ovarian cancer ( Supplementary Fig. 7b). Among these genes, we are particularly interested in FUT5, which encodes a rate-limiting enzyme α1,3-fucosyltransferase catalyzing the addition of fucose residue for sLe x synthesis 22,23 . FUT5 knockdown abolished sLe x cell surface expression on M-CSCs (Fig. 6b) and largely reduced the adhesion of M-CSCs to HPMCs and P-selectin-Fc under shear stress (Fig. 6c). Moreover, mice inoculated with orthotopic or i.p. xenograft (Fig. 6d) of M-CSCs carrying FUT5 shRNA had significantly reduced metastatic implants and ascites formation in the peritoneal cavity when compared with mice injected with nonspecific (NS) shRNA M-CSCs (Supplementary Table 5), with no apparent difference in primary tumor growth (Fig. 6e, f). Realtime PCR of tumor samples harvested from mice at the end of the study confirmed successful target gene knockdown of FUT5 (Fig. 6g).

Discussion
Tumor microenvironment is one of the major factors controlling the metastatic progression; however, the detailed molecular  Several lines of evidence suggest that CSCs may be a key player in the metastasis of ovarian cancer. First, ovarian CSCs are enriched in cells that can undergo epithelial-mesenchymal transition, a key process in metastasis 24 . Second, CSCs have higher tumorigenicity 25 . Third, ovarian cancer cells possess a stem-like gene signature correlates with poor progression-free and overall survival in cancer patients 26 . Fourth, the existence of these CSCs at an early stage of ovarian cancer is consistent with the frequent clinical observation of early metastasis 2 . In recent years, several studies in various cancer types, including breast and colorectal cancer, showed only a subpopulation of CSCs are able to metastasize [27][28][29] . In this report, we showed that the highly metastatic subpopulation CSCs in ovarian cancer ascites; and more importantly, we provided a detailed molecular understanding on their metastatic ability through a direct interaction with the peritoneal microenvironment, which not been studied in previous metastatic CSC models.
Very little is known about the selectin ligands on ovarian carcinoma cells 30 . Our present work has uncovered a sLe x -bearing P-selectin ligand in ovarian cancer cells, which is distinct from those previously defined P-selectin ligands, in particular the well-known ligand PSGL-1, which bears tyrosine sulfation. Much evidence hinted the casual relationship between sLe x and ovarian cancer metastasis. For example, antibodies against sLe x epitope have been previously shown to react with human ovarian cancer cells 31 . Glycosyltransferases that catalyze the synthesis of sLe x are significantly elevated in ovarian carcinoma tissues and cell lines 32,33 . High sLe x expression is associated with poor survival of patients with ovarian cancer 34 . The sLe x -bearing glycan present on M-CSCs binds P-selectin in a rigid fucosylation-dependent and relatively loose sialyation-dependent manner, instead of a simple sLe x structure that relies equally on fucose and sialic acid moieties 35 . Such biochemistry is of biological significance, as fucosylation has been shown with increased affinity and bind selectin more efficiently [36][37][38] . Whereas sulfation is a common feature of P-selectin ligands 39,40 , our results suggest that the sLe xbearing glycan on M-CSCs does not require sulfation for its binding to P-selectin. There is evidence that sulfated moieties of P-selectin ligands tend to be more resilient to stress for binding under high shear conditions 41 .
Although consistent with a direct role for P-selectin in mediating the tumor-mesothelial interaction, our findings also appear to differ somewhat from those of a recent work to show the sLe xbearing P-selectin ligand 10 ; however, the use of cells with different genetic backgrounds in this recent work may explain the difference. While CD24 expression is higher in some cell lines, the cell adhesion via P-selectin seems less, suggesting that there is other sLe x -containing ligand(s) of P-selectin. Using carefully controlled conditions and in comparison with a non-metastatic counterpart, while we have also shown CD24 expression in M-CSCs and NM-CSCs, there was no differential expression on CD24, unlike IGF-1R, in M-CSCs and NM-CSCs, suggesting the importance of IGF-1R in mediating the adhesion of M-CSCs. Moreover, our findings The sLe x -bearing P-selectin ligand is presented on IGF-1R, which is frequently overexpressed in ovarian cancer and other peritoneal metastasis models and confers a poor clinical prognosis 42,43 . Recently, sLe x has been identified to be decorated on various unexpected proteins, and functions of the sLe x are largely unknown 44 . IGF-1R, which to the best of our knowledge, has not been shown previously to possess sLe x . Although incompletely understood, there is growing appreciation for a role of sLe x in modulating protein activity. In gastric cancer cells, increased sLe x of c-Met was associated with increased dimerization and phosphorylation, which resulted in increased c-Metmediated signaling associated with tumor invasiveness 45 . Although we surmise that IGF-1R is functionally relevant for ovarian cancer progression and metastasis, our work suggests that blocking the sLe x mediating IGF-1R will be required to provide a promising effect.
In addition, our findings underscore a critical role for FUT5 during the development and maintenance of the metastasis. Higher incidence of sLe x in M-CSCs coincided with higher expression of glycosyltransferases. Ovarian cancer patients with FUT5-overexpressing tumors correlates with poorer survival compared with patients with low or no FUT5-expressing tumors. Importantly, blocking FUT5 expression in M-CSCs serves as an effective strategy for the treatment of peritoneal dissemination. Most of the glycosyltransferases genes involved in sLe x synthesis are constitutively expressed to produce sLe x direct precursor, whereas, FUTs encoding α1,3-fucosyltransferases catalyzing the last and rate-limiting step of sLe x synthesis by adding fucose to the precursor are normally switched off 22,23 . Moreover, peritoneal colonization of gastric cancer cells was reported to be suppressed by the downregulation of FUT5 46 . These data suggest that it is the context in which the glycan is expressed contributes to organotropism.
In summary, this study provides evidence showing P-selectin as a key molecule on HPMCs and IGF-1R carrying sLe x on M-CSCs as a ligand in mediating peritoneal dissemination under ascitic flow-induced shear stress (Fig. 7). This research is not only relevant to ovarian cancer, but also applicable to other tumor types, such as breast and colon cancers, in which peritoneal metastasis is an important pathological process, indicating that sLe x -P-selectin could become promising therapeutic targets. Antibodies. The antibodies or recombinant proteins used in the flow cytometry, western blot, immunohistochemistry, or microfluidic perfusion assays are detailed in the Supplementary Table 1.
RNA sequencing and TCGA data analysis. Total RNA extracted from three pairs of M-CSCs and NM-CSCs were sent to Beijing Genomics Institute for RNA sequencing (RNA-seq). Differential gene expression between M-CSCs and NM-CSCs was determined by a moderated t test with limma software. In total, 200 genes were identified upregulated in the M-CSCs compared with NM-CSCs (log 2 (fold-change) > 1 and P < 0.05), and 216 genes were identified downregulated (log 2 (fold-change) < −1 and P < 0.05). Hierarchical clustering was conducted on expression profile of 416 differentially expressed genes (DEG) and a high degree of internal consistency was discovered within each sample group. To discover whether M-CSCs could reflect the clinical metastatic process, we computed the fold-change of these genes in metastatic and non-metastatic tumors from The Cancer Genome Atlas (TCGA) ovarian cancer RNA-seq data. To quantify the consistence, we conduct the Gene Set Enrichment Analysis of DEG in M-CSCs samples against the ranked gene list (from up-to downregulated) in TCGA ovarian metastatic tumors.
Flow cytometry. Cultured adherent cells or tumor spheroids were detached with 2 mM EDTA PBS, suspended in PBS and counted. Cells were incubated with primary antibodies, selectin-Fc chimeras (5 μg mL −1 in 1 mM CaCl 2 , 1 mM MgCl 2 , PBS-binding buffer) or isotype control per the manufacturer's instructions. The primary antibody or recombinant proteins were stained with appropriate Alexa Fluor 488 secondary antibodies. Events were collected on FACS AriaIII (BD Biosciences). Data were analyzed using FlowJo software (Tree Star Inc.) with gating strategy shown in Supplementary Fig. 8.
Microfluidics CSCs perfusion assay. Microfluidic chips were fabricated as described 13 . To coat microfluidic channels with selectin recombinant proteins, channels were incubated with 1 μg mL −1 selectin-Fc recombinant proteins in PBS at 4°C overnight. To coat microfluidic channels with HPMCs, channels were incubated with 10 μg mL −1 human fibronectin in serum-free medium overnight at 4°C. After wash, the channels were introduced with HPMCs suspension (3.5 × 10 6 mL −1 ) and cultured overnight under CO 2 incubator at 37°C. To functionally block selectins, HPMCs were incubated with 20 μg mL −1 anti-E-selectin, anti-P-selectin, anti-L-selectin or isotype antibody at room temperature for 1 h before the assays. CSCs of 70-100 μm diameter were collected with cell strainer, fluorescently labeled with CMFDA Celltracker (2.5 µg mL −1 , C7025, Life Technologies) as per manufacturer's instructions for assays. Fluorescently labeled CSCs resuspended in binding buffer (1500 spheroids mL −1 ) were perfused into the microfluidic channel through a syringe pump (LongerPump) under desired shear stresses. The viscosity of the binding buffer (0.9 cp) is within the range of viscosity (0.95 ± 0.15 cp) of ascitic fluid in patients with various ascitic etiologies 48 . The motion of cancer spheroids was observed under fluorescent microscope (Nikon ECLIPSE Ti) and serial time lapse images (1 image sec −1 ) were captured for 30 s. More than three randomly chosen areas (× 4 objective) were captured for each treatment.
The motion of cancer spheroids under flow was analyzed offline using Image J (National Institutes of Health) and used for the classification of interactions as described 49 . In brief, cancer spheroids that tethered to the surface, then detached in the flow were defined as tethering. Cancer spheroids that moved at a velocity below the hydrodynamic velocity for more than one spheroid diameter were defined as rolling. Cancer spheroids that remained stationary for more than one frame during recording were determined as adhesion. Percentage of tethering or adhesion was calculated by dividing the number of tethering or adherent spheroids by the total number of spheroids flew through the field of observation in the 30 s duration.
Western blot. Proteins (20 μg) were resolved on sodium dodecyl sulphate-polyacrylamide gels and transferred to the nitrocellulose membrane (Bio-Rad). Target proteins were detected using specific primary antibodies overnight at 4°C. The primary antibodies were detected with appropriate horseradish peroxidase conjugated secondary antibodies and visualized with enhanced chemiluminescence (PerkinElmer). Scans of the full blots are shown in Supplementary Fig. 9.

HPMCs
Tumor cells P-selectin ARTICLE phenylmethylsulfonyl fluoride)). In total, 1 mg of cell lysate was precleared with 40 μL of Protein A/G Agarose beads for 1 h at 4°C with agitation. Precleared lysate was incubated with 2 μg Fc or P-selectin-Fc chimera and 20 μL per tube Protein A/ G agarose beads at 4°C overnight with agitation. Precipitated beads were collected and washed with ice-cold lysis buffer. The P-selectin-Fc interacting proteins were eluted with elution buffer (50 mM Tris pH 7.4, 10 mM EDTA, 0.1% Triton, protease inhibitors cocktail) on ice for 5 min. The eluate was collected and denatured by boiling with laemmli sample buffer for 10 min. Interaction of IGF-1R with Pselectin was detected by western blot.
Immunoprecipitation. Extracellular protein was extracted with 2 mM EDTA, 150 mM NaCl, 1% NP-40, 50 mM Tris pH 7.4, and protease inhibitor cocktail. Anti-IGF-1R (sc-463, Santa Cruz Biotechnology) was immobilized on cyanogen bromide (CNBr)-activated sepharose (GE Healthcare) according to manufacturer's instruction. CNBr-activated sepharose without coupling antibody was used as a negative control. Cell lysates were precleared with CNBr-activated sepharose for 2 h followed by overnight incubation with anti-IGF-1R or isotype control conjugated beads at 4°C. The immunoprecipitates were washed with ice-cold lysis buffer and eluted with 0.1% trifluoroacetic acid and neutralized with 1 M Tris pH 8.8. The modification of sLe a/x on IGF-1R was detected via Western blot.
Histological analysis. Tissues were fixed in 4% paraformaldehyde, embedded in paraffin, and stained with hematoxylin and eosin. Immunohistochemistry detection was conducted on 5 μm formalin-fixed, paraffin-embedded human omentum tissue section collected from patients with nonmalignant disorders. Heat-induced antigen retrieval was conducted in pH 6.0 sodium citrate buffer. Monoclonal mouse anti-E-selectin, anti-P-selectin, and anti-L-selectin antibodies were used at 1:50 dilution. The streptavidin-biotin immunoperoxidase Histostain SP Bulk kit (Invitrogen) and ImmPACT AEC peroxidase substrate kit (Vector Laboratories) were applied for detection and hematoxylin was used as the nuclear counterstain.
RNA extraction, reverse transcription, and PCR. Total RNA was extracted with Trizol and 500 μg of RNA was reverse transcribed to complementary DNA (cDNA) using the first-stranded cDNA synthesis kit (Invitrogen). Specific genes were amplified and quantitated by real-time PCR using primers described in Supplementary Table 3. Real-time PCR was performed using the StepOnePlus real-time detection system and the Power SYBR Green PCR Master Mix (Applied Biosystems). Fluorescent measurements were recorded during each annealing step. The PCR quality and specificity were verified by melting curve analysis and gel electrophoresis. Relative expression was determined by normalizing to the GAPDH endogenous control. These experiments were carried out in duplicate and independently repeated three times.
FUT5 stable knockdown by shRNA. Lentiviruses carrying shRNA targeting FUT5 (sequence described in the Supplementary Table 3) or NS shRNA were generated by co-transfection of HEK293 cells with the constructs and lentiviral-packaging plasmids (Sigma-Aldrich) per the manufacturer's instructions. M-CSCs were transduced with the viral particle containing media and cells were selected with 1 μg mL −1 puromycin (Calbiochem) 24 h post transduction for 3 days. The knockdown efficiency was verified by q-PCR.
Statistical analysis. Results represent mean ± SEM. The significance of differences between categorical variables was determined using the chi-square test. Ordinal variables were assessed using the one-tailed Student's t test.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The RNA sequencing data have been deposited in the NCBI Trace Archive under the accession code PRJNA530706. All the other data supporting the findings of this study are available within the article and its supplementary information files or from the corresponding author upon reasonable request.