Integrated analysis of glycan and RNA in single cells

Summary Single-cell sequencing has emerged as an indispensable technology to dissect cellular heterogeneity but never been applied to the simultaneous analysis of glycan and RNA. Using oligonucleotide-labeled lectins, we first established lectin-based glycan profiling of single cells by sequencing (scGlycan-seq). We then combined the scGlycan-seq with single-cell transcriptome profiling for joint analysis of glycan and RNA in single cells (scGR-seq). Using scGR-seq, we analyzed the two modalities in human induced pluripotent stem cells (hiPSCs) before and after differentiation into neural progenitor cells at the single-cell resolution. The combination of RNA and glycan separated the two cell types clearer than either one of them. Furthermore, integrative analysis of glycan and RNA modalities in single cells found known and unknown lectins that were specific to hiPSCs and coordinated with neural differentiation. Taken together, we demonstrate that scGR-seq can reveal the cellular heterogeneity and biological roles of glycans across multicellular systems.


INTRODUCTION
Glycans are the most structurally diverse and rapidly evolving major class of molecules, which present at the surface of all living cells and play crucial roles in diverse biological processes (Varki, 2017). Glycan structures have been known to vary depending on cell types and states. Therefore, cell surface glycans are often referred to as ''cell signature'' that reflect cellular characteristics. Indeed, most of the stem cell markers (Kannagi et al., 1983;Kawabe et al., 2013;Schopperle and DeWolf, 2007;Tang et al., 2011;Wu et al., 2019) and serum tumor markers are glycoconjugates (Munkley, 2019;Pearce, 2018;Varki et al., 2015). Glycans are the secondary products of genes synthesized by the orchestration of many proteins, such as glycosyltransferases and glycosidases. Despite advances in technology, there is no established method to predict the precise glycan structures only from the gene expression profiles. Therefore, it is vital to develop methods to analyze cell surface glycans directly. In this sense, different strategies have been undertaken to analyze the glycome, including mass spectrometry (MS), high-performance liquid chromatography (HPLC), nuclear magnetic resonance, and capillary electrophoresis (Haslam et al., 2006;Nakano et al., 2011;Yamaguchi and Kato, 2010). Recently, a lectin-based glycan profiling technology called lectin microarray has played a pivotal role in surveying and mapping the informational context of complex glycans of various biological samples, indicating the applicability of lectin-based glycan profiling Narimatsu et al., 2010Narimatsu et al., , 2018Ribeiro and Mahal, 2013). However, there are limitations to the current glycan analytical methods. For instance, (1) glycans are unable to be analyzed at a single-cell level, (2) the glycan profile of each cell type in the mixed cell populations cannot be obtained without prior cell separation, and (3) the relationship between the glycome and transcriptome in single cells cannot be analyzed. Simultaneous analysis of the two modalities in single cells could lead to the understanding of cellular heterogeneity and glycan functions and the development of glycan markers of rare cells.
High-throughput single-cell sequencing has been transformative to understand the complex cell populations (Stuart and Satija, 2019). Recently, simultaneous profiling of multiple types of molecules within a single cell has been developed for building a much more comprehensive molecular view of the cell (Peterson et al., 2017;Stoeckius et al., 2017;Stuart and Satija, 2019). However, there has been no technology to jointly analyze the glycome and transcriptome in single cells since glycans cannot be amplified by polymerase chain reaction (PCR), unlike DNA and RNA. Here, we first established highly multiplexed lectin-based glycan profiling of single cells by sequencing (scGlycan-seq). We then combined the scGlycan-seq with single-cell transcriptome profiling (scRNA-seq) for joint analysis of glycan and RNA in single cells (scGR-seq). Using scGR-seq, we analyzed the relationship between the two distinct layers in human induced pluripotent To evaluate the PCR amplification bias of DNA barcodes, we prepared a mix of equal amount of 41 DNA barcodes used for the conjugation with 41 probes and performed 20 cycles of PCR followed by sequencing. The barcode counts for each DNA barcode were divided by those of all DNA barcodes and expressed as (B) Schematic representation of scGlycan-seq, scRNA-seq, and scGR-seq. Cells were incubated with DNA-barcoded lectins and separated into single cells. After DNA barcode was released by UV exposure, the released DNA barcode in the supernatants was measured by next-generation sequencing. RNA transcripts were purified from the cell pellet and analyzed by scRNA-seq. See also Figures S1-S3. iScience Article percentage (%). The average percentage was 2.43%, and the variation of the detected DNA barcodes ranged between 0.26 and 1.58-fold of the average percentage ( Figure S3), suggesting that the PCR bias is less than 2 PCR cycles. Therefore, we considered that the PCR bias is within the allowable range since the purpose of the developed method is to compare the lectin binding signals between samples prepared with the same lot of DNA-barcoded lectin library.

Glycan-seq of bulk cell populations
We assessed the ability of Glycan-seq to discriminate distinct cell populations based on cell surface glycan expression in bulk samples. Obtained data were compared with flow cytometry using fluorescence-labeled lectins as the gold standard.
We next applied Glycan-seq to Chinese hamster ovary (CHO) cells and glycosylation-defective mutants (Lec1 and Lec8) (Figure 3, Tables S2 and S3) . Wild-type (WT) cells typically express complex-type N-glycans (North et al., 2010), while Lec8 and Lec1 express agalactosylated and mannosylated N-glycans, respectively. In both flow cytometry and Glycan-seq, rLSLN (galactose binder) showed higher binding to WT than Lec8 and Lec1. Similarly, rSRL (GlcNAc binder) and rCalsepa (mannose binder) showed strong binding to Lec8 and Lec1, respectively. These results demonstrated that Glycan-seq could distinguish different bulk cell populations depending on cell surface glycan expression.

Single-cell Glycan-seq
We then tested Glycan-seq to see its applicability in single cells, which we termed glycan profiling of single cells by sequencing (scGlycan-seq) ( Figure 1B). We first applied scGlycan-seq to hiPSCs and hFibs. To assess the effect of the total barcode counts in each cell on the obtained glycan profiles, we performed principal component analysis (PCA) on the scGlycan-seq data. Single cells of hFibs with low total barcode counts showed low values in PC1 ( Figure S7A), suggesting total barcode counts were a confounding factor. Since the cells with low and high total barcode counts were separated in the histogram ( Figure S7B), we sought to remove cells with low total barcode counts from the downstream analyses and determined the cut-off value of 19,465 total barcode counts by Otsu's method. After the removal of hFibs with low total barcode counts, PCA showed no association of the total barcode counts with PC1 and PC2 ( Figure S7C), indicating that cells were no longer biased by total barcode counts. The same quality control was adapted for hiPSCs, and hiPSCs with higher than 6,126 total counts were used for the following analysis.
The signal level of rBC2LCN in hiPSCs and hFibs obtained by scGlycan-seq was shown in Figure 5A. scGlycanseq recapitulated heterogeneity of rBC2LCN observed by flow cytometry and showed statistically significant differences between hiPSCs and hFibs (p < 0.001, Brunner-Munzel test). We then performed PCA on scGlycan-seq data together with bulk Glycan-seq data of hiPSCs and hFibs ( Figure 5B). The PC1 clearly separated the two cell types, and, for each of hiPSCs and hiFibs, the PC2 showed higher variability of single cells compared to bulk samples, revealing cell-to-cell heterogeneity in glycan profiles ( Figure 5B, Tables S4 and S5).
We further applied scGlycan-seq to the hiPSCs after differentiation into NPCs (days 0, 4, and 11). The signal level of each lectin in hiPSCs and 11-day NPCs was shown in Figure S8. Relative quantitative differences in the rBC2LCN signal for hiPSCs before (day 0) and after differentiation to NPCs (day 4 and 11) observed by flow cytometry could also be captured by scGlycan-seq ( Figure 5C). Statistically significant differences iScience Article between hiPSCs (day 0) and hiPSC-derived NPCs (day 4 and 11) (p < 0.001, Brunner-Munzel test) were observed. PCA clearly separated single cells of day 0, 4, and 11, and cells were ordered as differentiation progression ( Figure 5D). Single-cell heterogeneity of hiPSCs increased after 4-day differentiation but converged after 11-day differentiation, possibly because 4-day NPCs might contain various degrees of differentiated NPCs ( Figure 5D, Tables S4 and S5). These results demonstrated that scGlycan-seq enabled glycan profiling in single cells and revealed cellular heterogeneity in glycan profiles.

scGR-seq of hiPSCs and NPCs
We combined scGlycan-seq with scRNA-seq to enable the simultaneous measurement of the glycome and transcriptome in single cells, termed scGR-seq ( Figure 1B). Specifically, we employed RamDA-seq, a iScience Article full-length single-cell total RNA-sequencing method 7 . We performed scGR-seq on the hiPSCs (n = 53) and hiPSC-derived NPCs (11-day differentiation) (n = 43) (Tables S6-S9). After quality control of scRNA-seq data of hiPSCs and NPCs (see STAR Methods and Figures S9-S12), we searched for differentially expressed genes between hiPSCs and NPCs. We found that 1,131 and 688 genes were significantly upregulated in hiPSCs and NPCs, respectively ( Figure S13, Table S10). Consistent with neural differentiation of hiPSCs, GO enrichment analysis demonstrated that gene sets annotated with neuron-associated terms were significantly enriched in NPCs (Table S11). Furthermore, transcriptome data of hiPSCs and hiPSC-derived NPCs showed cell-type-specific expression of 41 selected cell-type marker genes ( Figure S14) (Obernier and Alvarez-Buylla, 2019; Takahashi et al., 2007). For example, NPCs showed higher expression of neural progenitor marker genes such as NES (NESTIN), PAX6, and SOX1 (Obernier and Alvarez-Buylla, 2019) and lower expression of hiPSC marker genes such as NANOG and POU5F1 (Takahashi et al., 2007), which agree well with fluorescence staining ( Figure S4). These data suggest that the scRNA-seq data of GR-seq reflect transcriptome information with biological relevance.
We analyzed the correlation of lectins across cells and found lectins that fluctuated with each other based on their correlation ( Figure 6). Lectins were separated into two large clusters: one cluster containing rBC2LCN and TJAII and another cluster containing other 37 lectins. rBC2LCN and TJAII commonly recognize a1-2fucosylated glycans, which are upregulated in hiPSCs ( iScience Article Sulak et al., 2010;Tateno et al., 2011;Yamashita et al., 1992). In contrast, other lectins showed higher binding to NPCs or comparable signals between the two cell types (Figures 6 and S8).
We then addressed how the information from two modalities can be used. When we performed Uniform Manifold Approximation and Projection (UMAP), a non-linear dimensional clustering, based on only the mRNA or glycan data using the Seurat workflow, the two cell types (hiPSCs and NPCs) were partially separated ( Figures 7A and 7B). In contrast, when we performed UMAP based on both the mRNA and glycan data using the Seurat-weighted nearest neighbor workflow, the two cell types were clearly separated ( Figure 7C). We also found that the unsupervised clustering using both mRNA and glycan data completely agreed with the cell type annotation ( Figure 7F), whereas the clustering based on either mRNA or glycan data showed poorer concordance ( Figures 7D and 7E). The difference in the clustering results was quantitatively verified by the adjusted Rand index ( Figure 7G). No bias was observed by the number of genes or lectins on UMAP ( Figure S15). This tendency was also confirmed when we performed PCA based on the mRNA or lectin data and partial least squares (PLS) regression using both mRNA and glycan information, where the latter separated the two cell types clearer than the former two ( Figure S16). These results demonstrated how the combination of mRNA and glycan modalities help characterize cell identities.

Relationship between mRNA and glycan in single cells
Simultaneous transcriptome and glycome measurements could associate genes with glycans at the single-cell level. The PLS regression analysis described above found a group of mRNAs and lectins that were associated with each other differently per component (see STAR Methods, Figure 8A). For the component p1, where weights were high for rAAL, rBC2LA, rLSLN, and rBanana, the mRNAs (high p1 weight) related to brain development, cell projection morphogenesis, neural precursor cell proliferation, sensory organ development, and negative regulation of nervous system development were the most enriched gene sets ( Figures 8B and 8C, Table S12), suggesting that the glycan ligands of these lectins iScience Article might be closely associated with neural differentiation. Other components (p2-p10) also showed the relationship between lectins and the set of genes ( Figures S17-S21). This analysis allowed us to infer each glycan's potential functions and roles as a marker through the set of genes associated with the glycan. Furthermore, by summing the loadings of each component, we obtained the overall relationship between lectins and glycosylation-related genes (see STAR Methods and Figure S22). For example, ST6Gal1, which catalyzes the synthesis of a2,6Sia, showed the highest correlation with a2-6Sia-binding lectin rPSL1a (Kadirvelraj et al., 2011;Tateno et al., 2004). These exemplify how scGR-seq is useful for finding potential relevance between transcriptome and glycome layers, although further detailed analysis should be performed in future studies. ; generalized additive model, q < 0.05). rBC2LCN, the known pluripotency marker probe, was decreased along pseudotime, while other lectins such as rBanana (mannose binder), which is not known as any cell surface marker, were increased. To confirm the differential binding of rBC2LCN and rBanana, we conducted fluorescence microscopy examination. Consistently, rBC2LCN staining was diminished after differentiation into NPCs ( Figure 9C). While rBanana showed intracellular organelle staining in hiPSCs, the cell surface was stained in hiPSC-derived NPCs, suggesting a drastic change in the localization of the glycan ligands of rBanana during differentiation ( Figure 9C). Because rBanana showed high weight in component p1 ( Figure 8C) and the component p1 was associated with neuron-related gene sets (Figure 8C), the glycan ligands of rBanana may be related to neural differentiation.

scGR-seq reveals pluripotency and differentiation glycan markers
We also examined genes correlated with hiPSC-specific rBC2LCN at the single-cell level ( Figure S23 and Table S13). The highest positive correlation coefficient (0.77) was observed with the hiPSC marker gene (POU5F1), whereas the highest negative correlation coefficient (À0.65) was observed with the NPC marker gene (VIM), supporting the previous finding that rBC2LCN was a maker probe for hiPSCs .
Furthermore, we searched for lectins correlated with hiPSC marker genes (MYC, LEFTY1, and LEFTY2) within hiPSCs ( Figure S24), all of which are known to show gene expression variability in human pluripotent stem cells. Expectedly, rBC2LCN exhibited the highest correlation with MYC. LEFTY1 and LEFTY2 exhibited the highest correlation with rGal3C (galactose binder) and rBC2LA (mannose binder), respectively, suggesting a possible link between glycan ligands of these lectins and the functions of LEFTY1 and LEFTY2 in the regulation of self-renewal and differentiation Tabibzadeh and Hemmati-Brivanlou, 2006;Takahashi et al., 2007). These results demonstrate that scGR-seq revealed the coordinated heterogeneity in pluripotency/differentiation markers across mRNA and glycans even within cells of the same cell types.  Table S12. See also Figure S15.

OPEN ACCESS
iScience 24, 102882, August 20, 2021 9 iScience Article DISCUSSION In this study, we first established glycan profiling by sequencing (Glycan-seq) using 41 DNA-barcoded proteins. Distinct cell populations could be discriminated based on cell surface glycan expression. Relative quantitative differences in expression levels of glycans observed by flow cytometry could be measured by Glycan-seq. No visible non-specific interaction between lectins was observed under the condition used in this study. Furthermore, Glycan-seq was proved to be applicable to single cells. Therefore, Glycan-seq is feasible for profiling cell surface glycans in both bulk and single cells. The amount of DNA barcodes conjugated with lectins affects the lectin binding signals. However, the aim of the developed  iScience Article method is to compare the lectin binding signals between samples by using the same lot of DNA-barcoded lectin library, which is similar to flow cytometry analysis, other sequencing-based analytical methods such as CITE-seq (Stoeckius et al., 2017), and lectin microarray .
We then combined scGlycan-seq with RamDA-seq (full-length total RNA sequencing in single cells), which can analyze not only mRNA but also non-poly(A) RNAs such as nascent RNAs, histone mRNAs, long noncoding RNAs (lncRNAs), circular RNAs (circRNAs), and enhancer RNAs (eRNAs) . Although we focused on the relationship between glycan and mRNA in this study, it would be interesting to analyze the relevance of glycan with non-poly(A) RNA at the single-cell resolution in future studies.
Using scGR-seq, we analyzed the two modalities in hiPSCs before and after differentiation into neural progenitor cells. The combination of RNA and glycan separated the two cell types clearer than either one of them. Furthermore, integrative analysis of glycan and RNA modalities in single cells found known and unknown lectins that were specific to hiPSCs and coordinated with neural differentiation. rBC2LCN showed the highest correlation with the well-known pluripotency marker gene, POU5F1 among all genes expressed in hiPSCs and NPCs, further supporting the previous finding that rBC2LCN signal is associated with pluripotency (Hasehira et al., 2012;Onuma et al., 2013;Tateno et al., 2011Tateno et al., , 2013Tateno et al., , 2015. rBanana with specificity to mannosylated glycans was extracted as a lectin to increase along with neural differentiation by PLS regression ( Figure 8) and pseudotime analysis ( Figure 9B). Indeed, rBanana showed cell surface staining to NPCs, while its staining was observed in cytoplasm in hiPSCs ( Figure 9C). We performed two independent experiments for the analysis of hiPSCs and NPCs in single cells and basically obtained similar results. This suggests that the plate effects (plate-to-plate variability) are visually low. These results demonstrate that mannosylated glycans might increase at the surface of NPCs, which could be used as a cell surface marker for identification and separation of NPCs and might lead to the understanding the functions of NPCs.
In terms of glycan detection probes, cost-effective and commercially available lectins, either from natural sources or recombinantly expressed, were used for Glycan-seq Ribeiro and Mahal, 2013). Any glycan-binding probes, such as anti-glycan antibody and glycan-binding peptide, can be incorporated in Glycan-seq (McKitrick et al., 2020).
scGR-seq can be widely adapted to cells, organoids, and tissues and can be applied to various scientific fields such as stem cell biology, immunology, cancer biology, and neuroscience. One of the biggest advantages of scGR-seq is that the method could be applied for the development of glycan markers of rare cells such as circulating tumor cells, fetal nucleated red blood cells, and cancer stem cells. Identification of glycan markers of rare cells could be performed as follows: (1) cell populations such as tissues, organoids, and hiPSC-derived cells/organs are analyzed by scGR-seq.
(2) Lectins with high intensity to rare cells identified by transcriptome data are selected. Such lectins might be directly used for the identification and concentration of rare cells.
(3) Glycoprotein ligands of lectins expressed in rare cells are identified by lectin pull down followed by LS-MS/MS. (4) Monoclonal antibodies recognizing both a glycan epitope and a peptide sequence could be generated as specific probes for rare cells (Watanabe et al., 2020). Such antibodies would be useful for the development of diagnosis (Egashira et al., 2019) and antibody drugs (Kato and Kaneko, 2014). scGR-seq can also be applied to cells derived from other organisms such as fungi and bacteria using the available platform, as lectins can bind glycans expressed by the cells of any organism (Hsu et al., 2006;Yasuda et al., 2011).
Recently, SUrface-protein Glycan And RNA-seq (SUGAR-seq) based on the 10x Genomics platform was reported that enables detection of a lectin-binding signal together with the analysis of extracellular epitopes and the transcriptome at the single-cell level (Kearney et al., 2021). However, SUGAR-seq detects only one lectin binding signal to single cells. In this regard, scGR-seq can acquire 39 lectin-binding signals to single cells, which can capture a whole picture of cell surface glycans (Table S14). It should be noted that the number of lectins is expandable in scGR-seq. We believe that scGR-seq has the potential to advance the understanding of cellular heterogeneity and the biological roles of glycans across diverse multicellular systems across species.

Limitations of study
There are limitations in scGlycan-seq and scGR-seq. Similar to flow cytometry and lectin microarray, absolute amounts of glycans and accurate glycan structures cannot be determined directly from the signal intensities as described above. Another limitation of the current system is the throughput. Since scGR-seq is ll OPEN ACCESS iScience 24, 102882, August 20, 2021 11 iScience Article a plate-based platform, processing of cell numbers is limited to hundreds of cells, while it can perform fulllength total RNA sequencing (Table S14) . In contrast, droplet-based methods such as 10x Genomics (CITE-seq) can sequence thousands of cells at once but target only the 3 0 ends of poly(A) transcripts (Baran-Gale et al., 2018). Due to this difference, scGR-seq will complement for studying single cells in complex biological systems. To solve this limitation, scGR-seq should be adapted to droplet (Bageritz and Raddi, 2019), nano-well (Gierahn et al., 2017), and indexing-based high-throughput single-cell technologies (Rosenberg et al., 2018) in future studies.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

ACKNOWLEDGMENTS
We greatly thank advisory board members and researchers of JST PRESTO ''Single Cell Analysis'' for thoughtful discussion and suggestion on the development of scGR-seq, Tadashi Kimura in National Institute of Advanced Industrial Science and Technology (AIST) for NGS analysis, and Sayoko Saito, Keiko Hiemori, Kayo Kiyoi, and Jinko Murakami in AIST for technical assistance on the single cell analysis. We also thank Mika Yoshimura, Tetsutaro Hayashi, and Itoshi Nikaido in the Laboratory for Bioinformatics Research, RIKEN Center for Biosystems Dynamics for thoughtful discussion on RamDA-seq data analysis.

AUTHOR CONTRIBUTIONS
H.T. designed research, analyzed the data, and wrote the paper. F.M. carried out the experiments and analyzed the data. H. Ozaki and H. Odaka carried out the experiments, performed data analysis, and wrote the paper. All authors read and approved the final manuscript.

DECLARATION OF INTERESTS
The authors declare no competing interests. The Seurat package (version_3.9.9.9014) was used to analyze scGR-seq data for hiPSCs and NPCs. The TPM matrix for the scRNA-seq data and the lectin expression matrix for the scGlycan-seq data was used as inputs. The RNA data was preprocessed using the 'NormalizeData' function with the default parameters, the 'FindVariableFeatures' function with the parameter 'selection.method = "vst"', and the 'ScaleData' with the default parameters. The lectin data was preprocessed using the 'NormalizeData' function with the parameters 'normalization.method = 'CLR', margin = 2', the 'FindVariableFeatures' function with the default parameters, and the 'ScaleData' with the default parameters. Principal component analysis (PCA) was then performed on the RNA data and the lectin data separately using the 'RunPCA' function with the parameter 'approx=FALSE'. Then, UMAP visualization was computed based on the PCA result of RNA (PC1 to 10) and lectin (PC1 to 20) data individually using the 'RunUMAP' function with the default parameters. In parallel, the 'FindNeighbors' function with the default parameters was used on the PCA result of RNA (PC1 to 10) and lectin (PC1 to 20) data to define k nearest neighbor cells with the parameter 'resolution = 0.5'. To search differentially expressed genes, 'FindMarkers' function with the parameter 'logfc.threshold=0.25' was performed and genes which match the criteria, Benjamini-Hochberg adjusted p-value < 0.05, were visualized by 'DoHeatmap' function. GO enrichment analysis was performed using the DAVID Bioinformatics Resources 6.8 (https://david.ncifcrf.gov/tools.jsp).
To jointly analyze the RNA and lectin modalities, the weighted nearest neighbor (WNN) workflow in Seurat was performed (Hao et al., 2020). First, to learn cell-specific modality 'weights' and construct a WNN graph that integrates the two modalities, the 'FindMultiModalNeighbors' function was used on the PCA results of mRNA (PC1 to 10) and lectin (PC1 to 20) with the parameters 'knn.range = 50, k.nn=10'. Then, UMAP visualization was computed based on a weighted combination of RNA and lectin data using the 'RunUMAP' function with the default parameters. In parallel, graph-based clustering was performed based on the WNN graph using the 'FindClusters' function with the parameter 'resolution = 0.2'. To evaluate the similarity of the clustering with the cell type annotation, the adjusted Rand index was calculated using the 'mclust' R package (version 5.4.6).

PLS regression of scRNA-seq and scGlycan-seq data
The package ropls (version 1.18.8) was used for PLS regression with scRNA-seq and scGlycan-seq data (Thevenot et al., 2015). Let X be n 3m gene expression matrix (scRNA-seq) and Y be n 3p lectin matrix (scGlycan-seq), where n is the number of cells, m is the number of genes, and p is the number of lectins. PLS regression decomposes X and Y to maximize the covariance between T and U as follows: where T and U are n3l matrices which is the projection of X and Y , respectively, and P and Q are m3 l andp3l orthogonal loading matrices, respectively, and matrices E and F are the error terms. The l components are analogous to principal components. The associations of genes and lectins were calculated as PQ u (Figures 8 and S14). The glycogene annotation was retrieved from GlycoGene Database (GGDB, https://acgg.asia/ggdb2/) (Narimatsu, 2004).

Flow cytometry
Lectins were recombinantly expressed in Escherichia coli and purified by affinity chromatography as described (Tateno, 2020;Tateno et al., 2011). Lectins were labeled with R-phycoerythrin (PE) using the R-phycoerythrin Labeling Kit (Dojindo Laboratories Co. Ltd., Kumamoto, Japan). 1x10 5 cells were suspended in 100 ml of PBS/BSA and incubated with 10 mg ml -1 of PE-labeled lectins on ice for 1 h. For intracellular staining, cells were fixed with 4% paraformaldehyde at room temperature for 10 min and permeabilized with 0.1% saponin in PBS at room temperature for 10 min. Cells were then incubated with anti-PAX6 mAb (x20, clone No. O18-1330, BD Biosciences, CA, USA) on ice for 1 h. Flow cytometry data were acquired on a CytoFLEX (Beckman Coulter, Inc., CA ) and analyzed using the FlowJo software (FlowJo, LLC., OR).

OPEN ACCESS
iScience Article