Surface glycan signatures defined by lectin microarrays in mammalian cells
To probe the accessible surface glycans or cell surface glycan signatures, we collected a panel of 94 lectins with an extensive range of specificities (supplementary Table I). Fluorescently labeled live cells selectively interacted with a subset of lectins on the microarray (supplementary Figure A). Individual cells captured on 120-micron spots could be readily distinguished by fluorescent microscopy, indicating that their interactions with lectins immobilized on solid surfaces were due to the binding of intact cells (supplementary Figure A, inset). To determine if the binding signals observed resulted from direct interactions between the lectins and glycans on the cell surfaces, we tested the specificity using the simple sugar inhibitors, α-methyl-mannoside or the disaccharide lactose (Galβ1-4Glc). Labeled HeLa and 293 cells were incubated on lectin microarrays in the presence or absence of either α-methyl-mannoside (200 mM) or lactose (200 mM). The binding of both cell lines to the lectins LCA, specific for mannose, and TKA, specific for lactose or terminal galactose, was inhibited only by their respective glycan ligands (supplementary Figure B). Cell binding to SNA, which is not specific for either mannose or lactose, was unaffected by the presence of either soluble glycans. The lectin array demonstrates glycan-specific cell binding and is useful in probing accessible cell surface glycans.
Fig. 1 Binary cell binding signature. (A) Thumbnail overview of the one-way hierarchical cluster of 24 human cell lines (rows) and 94 lectins (columns). The corresponding carbohydrate specificities were depicted to the right as cartoons on the basis of information (more ...)
A lectin binding signature for mammalian cell lines
To demonstrate the capability of the lectin microarrays to distinguish among the cell surface glycan repertoires of mammalian cells, 24 different types of human cell lines were tested (supplementary Table II). To account for significant variations in size, morphology, and labeling efficiency of cells, absolute quantification and cross-cell line normalization were not performed, but instead, a binary analysis was used to process the data (supplementary Figures and ). Each lectin was designated either bound or unbound to every cell line using a two-step method (see Materials and methods for details).
Fig. 2 Differential cell binding in lymphocytes. (A) Cell binding signature showing the binding profile of B cells (B), activated B cells (B Act), thymocytes (B6 Thy), T cells (T), PMA/Ionomycin-activated T cells (T Act), and allogeneically activated T cells (more ...)
Fig. 3 E. coli-mammalian cell binding assay. (A) Schematic diagram. The freshly cultured cells were seeded to 24-well plates and cultured until a cell monolayer formed. E. coli strains were incubated with or without d-mannose at room temperature and loaded to (more ...)
Using this analysis, a cell binding map was generated (Figure A). Significant differences in lectin binding patterns were observed among different cell lines. For example, less than 20 lectins could capture the hESC, Caco-2, D407, and U937 cells, while more than 50 lectins captured the TAg, 293, K1106, and MCF7 cells. These data indicate that the accessible cell surface glycans of mammalian cells vary significantly from one cell type to another. Accordingly, the cell lines were clustered into two major groups. Some related cell types clustered tightly, for example, the retina-derived cell lines, ARPE19, WER1, and Y79. In contrast, three breast cancer cell lines, MCF7, MDA-BA-231, and SkBr3, were not tightly clustered. This observation suggests that similar cell lines still possess variations in their cell surface glycan signatures and can be captured by the lectin microarray.
On the basis of their interactions with cell types, the 94 lectins could be grouped into four categories, i.e., zero binding (no cell line), low binding (1–9 cell lines), medium binding (10–19 cell lines), and high binding (20–24 cell lines). Three lectins recognized all of the 24 cell lines tested and 17 lectins did not show any binding activity. Therefore, the other 74 lectins were informative in terms of cell-type classification in this study. With the exception of two lectins of unknown saccharide specificities, all high binding lectins recognize simple sugars, such as mannose, galactose, and fucose, which are generally found on all mammalian cell surfaces. More complex glycan structures are recognized by lectins in the low or medium binding categories. For those lectins that showed no binding activity to the tested cells, it is possible that they either have relatively low affinity or are of low quality.
To further validate the lectin binding profile, we analyzed representative lectin-cell binding using flow cytometry. Lectins that were selected were representative of four categories: Jacalin bound to all cell types; RPA bound to most cell types; HHA and UEA-I bound to only a few cell types; and UEA-II bound to no cell types (Figure B). The flow cytometry data for each lectin correlated well with microarray binding (compare Figure B with the respective insets). High cell binding on the microarray was reflected as a large shift in the mean channel fluorescence (MCF) by flow cytometry, whereas low cell binding to the microarray was verified by negligible shifts in MCF over unstained control.
The analysis also showed that there is additional information that we are currently not harvesting in the binary lectin binding profile. For example, HHA lectin had intermediate binding capacity in the microarray analysis (Figure B, inset). In our ranking system this appeared only as a positive binding event. Flow cytometry experiments confirm the intermediate nature of the interaction of K562 cells with the HHA lectin (Figure B). As microarray technology matures, one could extract additional information through quantitative analysis of lectin binding.
A lectin binding signature for differentiating lymphocytes
The lectin microarray discriminated between closely related primary murine cells and revealed glycosylation changes in differentiating immune system cells (Rudd et al. 1999
). We probed the lectin microarrays with naïve and activated B cells, thymocytes (immature T cells that are lineage committed but not fully differentiated), and naïve and activated T cells (Figure A). Freshly isolated thymocytes were captured by the largest number of lectins, indicative of the diversity of surface glycans. In contrast, naïve T cells bound to the lowest number of lectins. Two lectins BDA and PEA/PSA bound to B cells, activated B cells, and activated T cells (Figure A) but not thymocytes or naïve T cells. Interestingly, activation shifted the lectin profile of T cells from naïve toward one more similar to thymocytes. Lectin microarray analysis also showed that T cells activated with either phorbol myristyl acetate (PMA)/ionomycin or a mixed lymphocyte reaction (MLR) displayed different binding profiles. The lectins that bound MLR-activated but not PMA/ionomycin-activated T cells are specific for terminal Gal, terminal GalNAc, Man, and Fuc saccharides. This finding implies that even though both cells are “activated,” they represent phenotypically different cells. When examining B cells, we see that changes in lectin binding produced by activating B cells and by activating T cells overlap significantly with each other. Glycosylation changes in activating B cells and T cells are quite similar and may be the result of their common lymphocyte lineage.
Certain differential patterns of lectin binding to immune B and T cells were observed (Figure ). Twelve of the seventeen surface lectin binding motifs (supplementary Figure ) for lectin binding in immature, mature, and activated T lymphocytes were identical (data not shown). Despite the similarities, there are also clear differences among the lymphocyte populations. The largest segment of differentially bound lectins, such as the Gal-GalNAc specific BPL, bound all cells with the exception of naïve T cells (Figure A). Lectins which captured thymocytes and activated T cells but not naïve T cells have relatively narrow sugar specificities usually consisting of Gal, GalNAc, and GlcNAc. Differential recognition of Gal-GalNAc, the core-1 O
-glycan, reflects known changes in glycan expression when transitioning from immature thymocyte cells to naïve T cells to activated T cells (Daniels et al. 2001
; Walzel et al. 2006
). Interestingly, there were no lectins that bound exclusively to naïve T cells. The lymphocytes also provide a reminder that cell binding events are not simply binary (Figure B). After normalizing fluorescent intensities to cell labeling efficiency, as assessed by flow cytometry, there are still significant variations in fluorescence intensity between cell types represented in our reductionist binary analysis. Overall, the lectin signatures across lymphocytes show fewer differences than the 24 human cell lines. The general similarity in surface glycan expression was to be expected as the cells share a common developmental lineage.
Fig. 4 Enhanced tumorigenicity in mice injected with LEL depleted MCF7 cells. (A) Microarray image of LEL, AAL, and WGA capture of normal or sphere cultured MCF7 cells. (B) Representative tumors in two mice at the enriched cell injection sites (left) and at (more ...)
Using glycan signatures to predict Escherichia coli tropism
To further explore the potential of the glycan signatures described above, an E. coli
-mammalian cell-binding system was employed. E. coli
bind to a variety of mammalian cells, such as the pathogenic E. coli
strain K1, which binds to brain microvascular endothelial cells (BMEC) with high affinity (Kim 2006
) and causes meningitis. Type 1 fimbriae-based mannose-specific E. coli
binding is one of the major contributors for E. coli
-mammalian cell binding (Teng et al. 2005
) and the FimH subunit determines the mannose-specific binding of type I fimbriae (Hung et al. 2002
). Clinically, human urinary tract and bladder infections caused by mannose-specific E. coli
-mammalian cell binding are treated with oral administration of mannose (Svanborg and Godaly 1997
The glycan signatures revealed that mannose-specific lectin binding is significantly different among different cell lines. For example, of the 16 mannose-specific lectins that bound to MCF7 only 6 bound to SkBr3. If such differences reflect a higher density of accessible mannose residues on MCF7 cells, the mannose-specific E. coli binding could be proportionally greater to those cells. To test this hypothesis, we performed E. coli-mammalian cell binding assays (Figure A) using three breast cancer cell lines MCF7, MDA-MB-231, and SkBr3, which bound to 16, 14, and 6 mannose-specific lectins, respectively. We calculated a mannose inhibition index of the bacterial binding of wildtype K12 E. coli (WT) and the FimH knockout mutant (fimHΔ) by evaluating the ratio of bacterial binding (colony counts) in the presence and absence of soluble d-mannose (Figure B). The mannose inhibition index values of the WT for MCF7, MDA-MB-231, and SkBr3 are 0.48, 0.58, and 0.76, respectively, which are statistically different among these three cell lines (P < 0.05) except MCF7 versus MDA-MB-231 (P = 0.078). These ratios are inversely proportional to the number of the mannose-specific lectins bound to them. In contrast, the mannose inhibition index values of fimHΔ cells for MCF7, MDA-MB-231, and SkBr3 are 0.96, 1, and 1.03, respectively, which are close to 1 and not significantly different (P ≥ 0.4). These results indicate that, as expected, the mannose-independent fimHΔ-mammalian cell binding is not inhibited by the addition of mannose. Taken together, the data indicate that the mannose-specific E. coli-mammalian cell binding is predicted by the mannose-specific lectin chip binding. Thus, the glycan signatures can be used to predict bacterial tropisms for mammalian cells.
Biomarker discovery based on glycan signatures
To examine the power of the microarray platform for biomarker discovery, we analyzed lectin binding in a model cancer stem-like cell system by comparing the cell surface glycan signatures of all 24 cell types including MCF7 cells under standard and sphere culture conditions. Cancer stem-like cells were chosen for analysis due to the dearth of cell surface markers to differentiate between cancer stem-like cells and cancer cell lines. Because of this issue, cancer stem-like cells are often differentiated from other cells functionally by their ability to grow as spheres when cultured, as a side population (SP) which reflects expression of drug transporters seen by flow cytometry, or by their tumorigenicity in vivo in NOD/SCID models. Ultimately, the limited ability to define unique cell surface markers has limited the phenotypic characterization of these important cells.
The model system analyzed was the breast cancer cell line, MCF7, which can be grown under both standard and sphere culture conditions. When grown as a sphere culture, MCF7 cells express the phenotypes associated with known breast cancer stem-like cells (Ponti et al. 2005
; Phillips et al. 2006
). By contrast, only a fraction of MCF7 cells grown under standard conditions express those markers or phenotypically could be seen as SP cells (Ponti et al. 2005
; Phillips et al. 2006
; Zhou et al. 2007
). Interestingly, the glycan signatures demonstrated that MCF7 cells grown under two different conditions have distinct lectin binding profiles and are not clustered together (Figure A marked by asterisks). The lectins LEL, AAL, and WGA showed the largest difference in fluorescent intensities. Each of these lectins effectively captured MCF7 cells but failed to capture MCF7 sphere cells (Figure A). The most significant difference was seen for LEL. This indicates that the LEL lectin may be used to distinguish these cancer cell subpopulations.
To demonstrate the usefulness of LEL as a potential biomarker, we looked for enhanced in vivo tumorigenicity in NOD/SCID mice injected with LEL-depleted MCF7 cells compared to those injected with regular MCF7 cells. A marked difference was seen in tumorigenicity between the two different groups (Figure B and C). The average tumor size of mice injected with the LEL-depleted cancer stem-like cell enriched cultures (8.5 × 105 cells/mouse) was 870 mm3 versus 350 mm3 for the control group injected with the similar number of regular MCF7 cells (P = 0.016), and average tumor sizes of animals injected with a 100-fold fewer LEL-depleted cancer stem-like cell enriched cultures were 420 mm3 versus 260 mm3 for the control group, respectively (P = 0.005), which was seen after two additional weeks of growth. Thus, the lectin microarray identified novel cell surface markers on cancer stem-like cells which were useful in enriching and studying cancer stem-like cells.