Colorectal cancer (CRC) is the fourth most commonly diagnosed cancer type and the third cause of cancer-related deaths in both sexes worldwide [1
]. Classification of CRC patients into consensus molecular subtypes (CMS) based on their transcriptional profiles is of great biological and clinical significance, predicting the survival and response to therapy [2
]. The current treatment options consist of surgical resection alone (when diagnosed at an early stage) or surgery in combination with systemic administration of cytotoxic chemotherapy for locally advanced and metastatic tumors [3
]. So far, the most effective chemotherapy regimen applied to patients with advanced CRC (stages IIB–IV) includes 5-fluorouracil (5-FU), oxaliplatin and irinotecan [4
]. Moreover, targeted therapy with immune checkpoint inhibitors (e.g., anti-PD1 and anti-CTL4) [6
] or monoclonal antibodies (cetuximab and bevacizumab) [7
] is often administered in conjunction with chemotherapy and confers a survival advantage to patients with refractory disease [8
However, the response to therapy remains largely heterogeneous [9
] and therapy failure along with eventual relapse and metastatic disease has been attributed to a distinct subpopulation of tumor cells called cancer stem cells (CSCs) [10
]. These cells exhibit unique properties, such as self-renewal, infinite division, pluripotency, reduced immunogenicity, resistance to conventional chemotherapy/radiotherapy and tumor-initiating capacity under adverse microenvironmental conditions. Therefore, CSCs are well considered as the relentless engine of tumor evolution [11
]. Importantly, CSCs are characterized by high plasticity, and their functionality in CRC is predominantly dependent on stimuli derived from the surrounding tumor microenvironment [13
]. Although several therapeutic strategies either against CSCs themselves or the so-called CSC niche have recently been proposed, the obstacles that cell plasticity evokes have not been fully overridden yet [14
]. Specifically, among the remaining caveats are the lack of CSC-specific phenotypic markers, the limited efficiency of the CSC niche-signal blockade, and the regeneration of the CSC pool upon treatment. Therefore, there is an immense need for the identification of novel molecular pathways defining CSC fate and hardwiring, which can in turn be harnessed for successful therapeutic targeting.
Many of the well-known CSC markers, such as CD44, CD133 (PROM1), LGR5, CD24, EPCAM (ESA), ABCG2, CD34 and CD90/110 are heavily glycosylated [15
]. Glycosylation is a fundamental post-translational modification of proteins and lipids required for a wide range of biological and cellular functions [17
]. Fucosylation is a specific type of glycosylation that is increased in several types of cancer [18
]. Overexpression of fucosylated antigens on the surface of cancer cells is linked to increased cell survival and proliferation, epithelial to mesenchymal transition (EMT), metastasis and interaction with endothelial or immune cells [19
]. In addition, established cancer biomarkers, such as the carbohydrate antigen 125 (CA-125) [20
], carbohydrate antigen 19.9 (CA19.9 or sialyl Lewisa
] and carcinoembryonic antigen (CEA) [22
], are associated with elevated fucosylation levels during cancer progression.
(alternatively known as SSEA1 or CD15) is a tumor-associated fucosylated antigen and an established marker of glioblastoma [23
] and medulloblastoma [24
] stem cells. Among the different fucosyltransferase (FUT) enzymes that can synthesize Lewisx
, FUT9 is the most competent one regarding Lewisx
synthesis both in vitro and in vivo [25
] and it is highly conserved between human and mouse [26
]. FUT9 is expressed early during embryogenesis [27
], dictating the expression of Lewisx
in embryonic stem cells (ESCs) [28
]. Later in development, the expression of FUT9 is mainly restricted to the stomach and the brain [29
], where it is directly implicated in the synthesis of Lewisx
and the differentiation of neuronal stem cells [30
]. In CRC cells, expression of FUT9 has been associated with the induction of major metabolic changes, and for this reason FUT9 has been proposed as a metabolic driver of advanced-stage CRC [31
]. Nevertheless, the impact of FUT9 expression on stemness acquisition by colon cancer cells, along with other functional features of CSCs such as drug resistance, remains ill defined.
Here, we explored the major regulatory programs instructed in colon cancer cells upon FUT9 expression by combining computational and experimental analyses. Our results provide evidence that FUT9 dictates a stem cell-like fate in colon cancer cells both in mice and humans, suggesting a conserved role of this enzyme during malignant transformation and tumor fueling.
FUT9 is known to dictate stemness in embryonic and neuronal cells [28
] and it has been proposed to act as a metabolic driver of advanced-stage colon cancer [31
]. In the current study, we investigated the regulatory programs instructed in colon cancer cells upon FUT9 expression. Our results provide evidence that FUT9 is responsible for the hardwiring of both murine and human colon cancer cells towards a CSC-like transcriptional profile, phenotype and function, with major implications for tumor growth and resistance to chemotherapy.
Construction of transcriptional regulatory networks using iRegulon [34
] has proven its value in the identification of key pathways involved in the regulation of tumor [62
] and stromal cells [63
] during colon cancer progression. Here, we followed this approach to detect the regulatory circuits of hub genes and TFs that are in play upon transcriptional activation of Fut9
in the murine colon adenocarcinoma cell line, MC38. Yy1, cMyc and Hsf1 were identified as the master TFs controlling the expression of the upregulated genes in MC38-FUT9 cells (Figure 1
H). Moreover, we observed an upregulation of target genes downstream of these TFs in FUT9+
human CRC tumor cells (Figure 5
B). This intriguing finding is supported by earlier reports showing that Yy1, cMyc and Hsf1 orchestrate early developmental processes that cancer cells hijack during disease progression and metastasis [35
]. Despite the increase in cMyc and Hsf1 expression upon FUT9 expression, no differences in the mRNA levels of Yy1 could be detected among the MC38-glyco-engineered cells (Figure 1
G). This might be explained by the fact that Yy1 function is predominantly determined by post-translational modifications, such as O
] and phosphorylation [67
Many core transcription factors (cMyc, Sox2, Klf4, Oct4 and Nanog), enzymes (ALDH) and key signaling pathways (Wnt, Notch and Hedgehog) are shared between ESCs and CSCs [68
]. Our data demonstrate that, besides cMyc, FUT9 expression is correlated with Sox2 expression both in murine (Figure 2
A–D and Figure S2B
) and human colon cancer cells (Figure 4
B and Figure 5
B). Sox2 expression in CRC is associated with several CSC features, including spheroid (3D) growth patterns, increased tumor growth and resistance to chemotherapeutic drugs, such as 5-FU [69
]. Moreover, Sox2 expression and certain tumorigenic, prosurvival signaling pathways, such as the PI3K/Akt, have been linked before [70
]. This is in agreement with all the phenotypic (Figure 2
) and functional (Figure 3
) CSC-like traits that MC38 cells displayed upon induction of FUT9 expression. Similarly, KM12 cells, highly expressing FUT9 and Sox2 (Figure 4
A,B,F), were characterized by enhanced ALDH activity (Figure 4
G), as well as increased survival upon treatment with 5-FU (Figure S4A
) and oxaliplatin (Figure S4B
), relative to HCT116 cells.
Previous studies have implicated the Wnt pathway in the resistance phenotype of human colon cancer cells against 5-FU [71
]. Here, we identified differential expression between MC38-MOCK and MC38-FUT9 cells in genes involved in the Wnt signaling pathway (Figure 2
A). Furthermore, we observed a vigorous association between FUT9 expression and the Notch (Figure 4
C and Figure 5
B) and Hedgehog (Figure 5
B) signaling pathways in human primary colon cancer cells and cell lines. Activation of these pathways usually coincides with increased ALDH activity [72
]. This could possibly explain why de novo expression of FUT9 in MC38 cells resulted in an enhanced frequency of ALDH-High cells (Figure 2
E), whereas knock out of FUT9 in KM12 and SW1116 cells diminished the percentage of ALDH-High cells relative to control cells (Figure 4
I and Figure S4D
). Based on these findings, it becomes evident that FUT9 exerts a central role in shaping molecular programs, which are linked to pluripotency and are commonly found both in ESCs and CSCs.
In general, CSCs are believed to represent a rare cell population that accounts for the 0.01–1% of the bulk tumor [73
]. This is in line with the percentage of the FUT9+
primary CRC tumor cells (0.04%, 11/272) identified from our scRNA seq reanalysis (Figure 5
A). A similar, but 50% lower, percentage (0.02%, 5/203) was detected for the healthy FUT9+
cells. Another interesting observation was that FUT9+
tumor cells did not cluster together (Figure 5
A), indicating that stem-like FUT9+
cells display differential gene expression profiles. This phenotype corroborates the heterogeneity and plasticity of CSCs that has been previously described [14
]. Additionally, the fact that FUT9+
normal cells clustered away from the FUT9+
tumor cells highlights the existence of major transcriptional programs differentiating these two cell subsets.
Currently, there is a dire need for identification, functional characterization and validation of biomarkers for cells adopting a CSC-like status. Several fucosyltransferase (FUT) enzymes and their corresponding fucosylated antigens are representative candidates of such biomarkers [19
]. For instance, the Lewisx
trisaccharide has been proposed as a stem cell selection marker in human glioblastoma cells [74
can be synthesized by different FUTs, albeit FUT9 is the most competent one [25
]. In MC38 cells, we have noted induction of a CSC-like state upon neo-expression of both FUT9 and Lewisx
A–D). However, the human colon cancer cell lines tested here, KM12 and SW1116, maintained high ALDH activity in a FUT9-specific, but Lewisx
-independent manner (Figure 4
H–I and Figure S4C–D
). This finding demonstrates that the fate of each cell type is dictated by the enzyme itself, rather than the synthesized antigen; hence, selected enzymes should be further considered as CSC-specific biomarkers in the future.
Besides FUT9, more glycosyltransferases have been interrelated with stemness induction, including ST6GAL-1, MGAT5 and B4GALT3 [16
]. ST6GAL-1 transfers α2-6-linked sialic acids to substrate proteins and its expression is enriched in CD133+
, irinotecan-resistant colon cancer cells [75
]. Additionally, in pancreatic and ovarian cancer cells, ST6GAL-1 has been shown to induce the expression of CSC-specific TFs, such as Sox9 and Slug [76
]. In colon CSCs, MGAT5 is influencing tumorigenesis and activation of Wnt signaling by synthesizing branched N
-glycans on the Wnt receptor, FZD-7 [77
]. Finally, B4GALT3 regulates stemness through modification of EGFR N
-glycosylation on the surface of colon CSCs [78
Aberrant cancer glycosylation drives drug resistance in the context of CRC via different mechanisms, such as alterations in drug absorption, drug metabolism, signaling activation and apoptosis resistance [79
]. Here, we show that FUT9 expression is associated with resistance to chemotherapeutic drugs both in murine (Figure 3
B) and human (Figure S4A,B
) colon cancer cells. Importantly, our results reveal that even a very low (6.4%) starting number of CSC-like cells expressing FUT9 was enough to repopulate the 3D tumor upon treatment with 5-FU (Figure 3
C), confirming the strong link between cancer stemness and drug resistance. Besides chemotherapy, CSCs are also resistant to radiotherapy [80
] and antibody therapy [81
]. So, going forward, the impact of FUT9 expression on resistance acquired by colon cancer cells to other types of therapy remains to be evaluated.
A lot of attention has recently been drawn to cancer immunotherapy and checkpoint inhibition in CRC patients [82
]. CSC-specific immunotherapy has been proposed as an alternative and more targeted approach [83
]. Colon CSCs have the potential to evade immune surveillance through increased PD-L1 expression [84
] and downregulation of MHC-I [85
]. In this study, we identified significant downregulation of Irf9 and Stat1, along with their target genes, upon FUT9 expression (Figure 1
G,I). Importantly, these TFs are key elements of the ISGF3 complex, which is directly implicated in cancer cell resistance against the antitumor immune response [86
]. Therefore, further investigation should focus on the immunomodulatory properties of FUT9-expressing colon cancer cells.
4. Materials and Methods
NOD.Cg-Prkdc scid Il2rg rm1Wjl/SzJ female and male mice were purchased from Jackson Laboratory (Bar Harbor, ME, USA). Mice were used at 6–8 weeks of age and maintained in a specific pathogen-free facility at the Amsterdam Animal Research Center (AARC). Mice were randomly assigned to one of the experimental groups. Each group contained 10 mice in total. Experiments were performed in accordance with national guidelines and after approval by the ‘Centrale Commissie Dierproeven’ (CCD) under number AVD1140020173844. Different absolute numbers (105, 104 and 103) of MC38-FUT9 or MC38-MOCK cells were subcutaneously injected in either the right or left flank of each mouse, respectively. Tumor growth was monitored three times per week and total tumor volume was calculated using the formula 4/3 × π × abc (a = width of the tumor/2, b = length/2 and c = the average of a and b). Mice were sacrificed when the combined tumor volume from both flanks reached the 1500 mm3. Finally, the MC38-MOCK and MC38-FUT9 tumors developed in each mouse were isolated and weighed.
4.2. Cell Culture
Wild type MC38 cells (murine colon adenocarcinoma; gift from Prof. Dr. M. van Egmond, Amsterdam UMC, Vrije Universiteit Amsterdam, Department of Surgery and Molecular Cell Biology and Immunology, Cancer Center Amsterdam), wild type HCT116 cells (human colon adenocarcinoma; gift from Prof. Dr. Remond Fijneman, Netherlands Cancer Institute, Department of Pathology) and the respective glyco-engineered cell lines were cultured in monolayers (2D culture) and maintained in DMEM (Gibco, Waltham, MA, USA, 41966-029) supplemented with 10% FBS premium (Biowest, Nuaillé, France, S182B) and 1% penicillin/streptomycin (Gibco, 15140-122). Wild type KM12 (human colon adenocarcinoma; gift from Prof. Dr. Remond Fijneman, Netherlands Cancer Institute, Department of Pathology), SW1116 cells (human colorectal adenocarcinoma; gift from Mrs. Ing. G.W van Pelt, Leiden UMC, Department of Surgery) and the corresponding glyco-engineered cell lines were cultured in monolayers (2D culture) and maintained in RPMI 1640 Medium (Gibco, 31870-025) supplemented with 10% FBS premium (Biowest, S182B), 1% L-glutamine (Gibco, 25030-024) and 1% penicillin/streptomycin (Gibco, 15140-122). Cells were detached using Trypsin-EDTA (Gibco, 15400-054). Cells were maintained at 37 °C and 5% CO2 in a humidified incubator, tested for mycoplasma infection monthly. For anchorage-independent growth and tumorsphere formation (3D culture), MC38 and KM12 cells were seeded in 96- or 24-well Corning Costar clear flat bottom ultra-low attachment plates (Merck, Kenilworth, NJ, USA, CLS7007 and CLS3473) at a concentration of 500 or 2000 cells per well, respectively, and cultured in serum-free DMEM-F12 (Thermo Fisher, Waltham, MA, USA, 11,320,033) supplemented with basic Fibroblast Growth Factor (FGF) (20 ng/mL; Thermo Fisher, 13,256,029) and 1× B27 (Thermo Fisher, 17,504,044). Medium was replaced at day 4 to replenish nutrients. At day 7, tumorspheres were dissociated with an enzyme-free cell dissociation buffer (Gibco, 13,151,014), passed through a 70 µm cell strainer (Corning, NY, USA, 352,350) and single cell suspensions were used for further experiments. During serial passages, tumorspheres were dissociated using a tumor dissociation kit (Miltenyi Biotech, Bergisch Gladbach, Germany, 130-095-929), passed through a 70 µm cell strainer (Falcon, 352,350) and the number of single cells was determined after manual counting. Cells were maintained at 37 °C and 5% CO2 in a humidified incubator.
4.3. Generation of Glyco-Engineered Colon Cancer Cell Lines
The CRISPR/dCas9-VPR constructs used for transcriptional activation of the murine Fut9
gene were designed and made as previously described [33
]. The gRNA sequences targeting the murine Fut9
promoter region were: FUT9 gRNA#1 GCATATCGGAGACGCAGCAA and FUT9 gRNA#2 GCCTCCCGACTCAACACACG. Similarly, the CRISPR/Cas9 construct used for knocking-out the human FUT9
gene was made as previously described [87
]. The following gRNA sequence targeting the human FUT9
gene was designed and used: FUT9 KO: GCATTGAAATCCATACCTAC. MC38 cells were transfected and selected with 6 μg/mL Puromycin (Invivogen, ant-pr-1) as previously reported [33
]. Human CRC cells were transfected in 6-well plates seeded with 100,000 cells per well. Of the empty pSpCas9(BB)-2A-Puro plasmid 2.5 μg (Addgene #62,988; MOCK cells) or the plasmid containing the cloned gRNA sequence (FUT9 KO cells) were delivered to the corresponding wells together with lipofectamine LTX (Thermo Fisher, 15338-100), according to the manufacturer’s instructions. Stable human cell lines were selected with 1 µg/mL puromycin (Invivogen, ant-pr-1) and selection was applied to the cells 48 h post-transfection. Manual cell separation (MACS) with the use of anti-CD15 (Lewisx
) microbeads (Miltenyi Biotec, 130-046-601) and LS-Columns (Miltenyi Biotec, 130-042-401) was performed according to the manufacturer’s instructions to enrich for the Lewisx+
cell fraction of the glyco-engineered murine and human colon cancer cell lines. All cell lines were selected in bulk in order to exclude potential clonal effects and to mimic tumor heterogeneity.
4.4. qRT-PCR Analysis
MC38 cell lysis and mRNA isolation was performed with the mRNA capture kit (Roche, Basel, Switzerland, 11,787,896,001) and cDNA synthesis was performed using the Reverse Transcription System Kit (Promega, Madison, WI, USA, A3500), according to the manufacturer’s instructions. Synthesized cDNA samples together with the KAPA SYBR® FAST qPCR Master Mix (2×) Universal (KAPA BIOSYSTEMS, KK4618) and specific qPCR primers were utilized for each qRT-PCR reaction. Precisely, the qRT-PCR primer sequences used for cDNA amplification were the following ones: Gapdh forward primer: CCTGCACCACCAACTGCTTAG; Gapdh reverse primer: CATGGACTGTGGTCATGAGCC; Fut1 forward primer: CGACACAAAGACCCCATCTT; Fut1 reverse primer: GAAGCCAAAGGTGCCAATAG; Fut4 forward primer: CAGCCTGCGCTTCAACATC; Fut4 reverse primer: CGCCTTATCCGTGCGTTCT; Fut7 forward primer: CCATCCTTATCTGGCACTGG; Fut7 reverse primer: GCTCCGGTTAGCACTCAGAC; Fut8 forward primer: GCCAAAATGCCCACAATC; Fut8 reverse primer: GTTTCCAGCCACACCAATG; Fut9 forward primer: ATCCAAGTGCCTTATGGCTTCT; Fut9 reverse primer: TGCTCAGGGTTCCAGTTACTCA; Fut10 forward primer: GGAGGGAGAGCCTAAACACC; Fut10 reverse primer: CTACCAGCATCCACCTTTGTC; Fut11 forward primer: GTCGTCGCACATGAACTGTC and Fut11 reverse primer: GCTTCCCCCTGATAGAGACC. qRT-PCR for individual genes was ran and analyzed on the CFX96 Real-Time PCR Detection System (BIORAD, Hercules, CA, USA), with all target gene expression levels normalized to Gapdh (M. Musculus).
Cell lysates were prepared using the RIPA buffer (Thermo Fisher, 89,900) supplemented with protease (Thermo Fisher, 78,430) and phosphatase inhibitors (Thermo Fisher, 78,420). Protein concentrations were determined following the instructions of the Pierce BCA Protein Assay kit (Pierce, 23,225). Lysates were boiled with 2× Laemmli buffer (Bio-Rad, Hercules, CA, USA, 1,610,737) for 10 min at 95 °C. Precision Plus Protein Dual Color Standards (Bio-Rad, 1,610,374) were used as the protein ladder. Equal amounts of proteins were subjected to SDS-PAGE on 15% gels and then transferred to PVDF membranes (Bio-Rad, 162-0177). Membranes were blocked with the Odyssey Blocking Buffer in PBS (LI-COR, Lincoln, NE, USA, 927-40000) and then incubated with anti-FUT9 (1:1000; Proteintech, Rosemont, IL, USA, 60230-1-Ig) and anti-beta actin (1:10,000; Proteintech, 66009-1-Ig) in the Odyssey Blocking Buffer containing 0.1% Tween 20 for 1.5 h at room temperature. Membranes were washed 4× with PBS containing 0.05% Tween 20. For fluorescence detection, the following secondary antibodies were used: goat anti-rabbit IgG, IRDye 680 conjugated (1:15,000; LI-COR, 926-68071) for the detection of the anti-beta actin primary antibody and goat anti-mouse IgG, IRDye 800 conjugated (1:15,000; LI-COR, 926-32210) for the detection of the anti-FUT9 primary antibody. Membranes were incubated with secondary antibodies in the Odyssey Blocking Buffer containing 0.1% Tween 20 and 0.01% SDS for 1 h at room temperature. The image was recorded by an Odyssey Imaging System (LI-COR) and further analysis in the pseudo color mode was performed using the Image Studio Lite software (LI-COR).
4.6. Flow Cytometry
Cells were harvested, washed and resuspended at a concentration of 1 × 106 cells/mL in phosphate buffered saline (PBS) containing 0.5% bovine serum albumin (BSA; Fitzgerald Industries, Acton, MA, USA, 30-AB75) and 0.02% sodium azide. For extracellular staining, 50,000 cells (50 μL) were added to each well of a 96-well V-bottom plate (Merck, M9686). For intracellular staining, 100,000 cells (100 μL) were added to each well of a 96-well V-bottom plate (Merck, M9686), fixed/permeabilized in 100% methanol for 20 min at −20 °C and washed prior to staining. Cell suspensions were stained for 30 min on ice with anti-Lewisx (1:40, Merck, Kenilworth, NJ, USA, 434,631), anti-Sox2 (1:100; GeneTex, Irvine, CA, USA, GTX101507), anti-Oct4 (1:200; GeneTex, GTX100622), anti-Nanog (1:200; GeneTex, GTX627421), anti-Lewisy (1:20, GeneTex, GTX23359), anti-sialyl Lewisx (1:100, BD Pharmingen, Franklin Lakes, NJ, USA, 551344), anti-VIM2 (1:40, LabNed, Amstelveen, the Netherlands, LN1302156), isotype-APC (1:200; ImmunoTools, Friesoythe, Germany), anti-CD44-APC (1:200; ImmunoTools, 21,850,446 × 2), anti-LGR5-PE (1:100; Biolegend, 373,803), anti-CD133-BV421 (1:50; Biolegend, 372,808) and anti-CD24-AF647 (1:50; Biolegend, 311,110). 7-Aminoactinomycin D (7-AAD; 1:1000; Molecular Probes, A1310) was used for live/dead cell exclusion. Cells were washed and resuspended in 50 μL PBA (PBS, 0.5% BSA and 0.02% sodium azide) containing goat anti-mouse IgM-FITC (1:50; Jackson ImmunoResearch, Cambridge, UK, 115-096-020) for detection of the anti-Lewisx, anti-Lewisy, anti-sialyl Lewisx and anti-VIM2 primary antibodies. Cells were washed and resuspended in 50 μL PBA containing donkey anti-rabbit IgG (H+L)-AF647 (1:400, Thermo Fisher, A-31573) for detection of the anti-Sox2 and anti-Oct4 primary antibodies or goat anti-mouse IgG (H+L)-AF647 (1:400, Thermo Fisher, A-21236) for detection of the anti-Nanog primary antibody. Cell suspensions were stained for 30 min on ice with the corresponding secondary antibodies. Cells were washed with 100 μL PBA and subsequently resuspended in 100 μL PBA. Fluorescence intensities were measured using a Cyan ADP (Beckman Coulter Brea, CA, USA) or LSRII (BD Biosciences, Franklin Lakes, NJ, USA) flow cytometers. Cytometric data analysis was performed with the FlowJo V10 software (Tree Star, Ashland, OR, USA).
3 × 104 MC38 cells were seeded in each well (200 µL medium/well) of an 8-well chamber slide (µ-Slide 8 well, IBIDI, 80,826) and cultured overnight at 37 °C and 5% CO2 in a humidified incubator. The next day the medium was aspirated and cells were washed 3 times with PBS. Cells were fixed with 2% paraformaldehyde (PFA) in PBS for 20 min at RT and washed 3 times with PBA prior to staining. Cells were stained for 1 h at 37 °C with anti-Lewisx (1:80, Calbiochem, 434,631) or anti-Sox2 (1:100; GeneTex, GTX101507) and then washed 3 times with PBA. Next, cells were stained for 1 h at 37 °C with goat anti-mouse IgM-FITC (1:50; Jackson ImmunoResearch, 115-096-020) for detection of the anti-Lewisx antibody and donkey anti-rabbit IgG (H+L)-AF647 (1:400, Thermo Fisher, A-31573) for detection of the anti-Sox2 antibody and then washed 3 times with PBA. Cells were incubated with DAPI (1:3000, Thermo Fisher, D1306) for 20 min at RT and washed 3× with PBA. Finally, cells were mounted with MOWIOL 4-88 (Calbiochem, 475904). Representative images of the Lewisx staining were obtained using the Axio Imager D2 (Zeiss, Germany) microscope (40× objective). For the intracellular Sox2 staining, pictures were taken with the Nikon Ti2 microscope using the 40× objective. Different Z-stack values with steps of 0.4 µm ranging from −0.4 to +1.6 µm were used for deconvolution. Images were taken from 15 different fields of view for each condition. The mean fluorescence intensity was calculated with the NIS-Elements software after deconvolution. The nuclei of the cells were delineated using the DAPI staining (blue channel) and the mean fluorescence intensity was calculated by measuring the red signal (Sox2 staining). The cytoplasm fluorescence intensity was calculated by subtracting the whole red signal from the nuclei red signal. The average and the standard deviation of the fields of view were calculated.
4.8. mRNA Library Preparation, RNA-Sequencing, Alignment and Differential Expression Analysis
The mRNA library was prepared as described previously from three independent passages (passages 6–9) of the MC38-MOCK and MC38-FUT9 harvested at three independent time points [89
]. RNA extraction, library synthesis, as well as the RNA sequencing were performed as previously described [90
]. Reads were aligned to the Ensemble M. musculus
genome (build GRCm38.90) using HiSat2 (v2.0.4) (http://daehwankimlab.github.io/hisat2/
) and subsequent processing was performed with samtools (v0.1.19) (http://www.htslib.org/
). FeatureCounts (R package Subread v1.5.0-p3, http://subread.sourceforge.net/
) was used to quantify aligned reads, excluding multioverlapping reads.
Library size adjustment, trimmed mean of M-values (TMM) normalization and differential expression analysis was done using the R package edgeR (v3.18.1) (https://www.bioconductor.org/packages//2.7/bioc/html/edgeR.html
) software. Multidimensional scaling (MDS) plots were used to visualize sample distribution among MC38 cell lines. For differential expression analysis, the negative binomial dispersion was shrunken towards the common dispersion. EdgeR’s exact test for two-group comparison was used for computing p
-values. Statistical differences in mRNA expression were identified using the following pairwise comparisons: (1) MC38-WT vs. MC38-MOCK cells and (2) MC38-FUT9 vs. MC38 MOCK cells. Here, per comparison, genes with more than 4 zeros across the 6 samples were discarded a priori. Significance was assessed using Benjamini–Hochberg false discovery rate (FDR) <0.05. Sequencing data will be made publicly available at the Sequence Read Archive (SRA) Gene Expression Omnibus thought GSO Series accession number GSE143700 upon acceptance of the manuscript.
4.9. Comparative Analysis of Gene Sets
Comparative analysis was performed among the differentially expressed genes (DEGs) in the (1) MC38-WT vs. MC38-MOCK cells (864 genes, Table S1A
) and (2) MC38-FUT9 vs. MC38 MOCK cells (4118 genes; Figure S1C
) using Venny (v2.1.0, https://bioinfogp.cnb.csic.es/tools/venny/index.html
). Of the 4118 DEGs in the MC38-FUT9 cell line, 3522 genes were specifically affected in MC38-FUT9. There were 596 genes that overlapped between comparison of (1) MC38-WT vs. MC38-MOCK and comparison of (2) MC38-FUT9 vs. MC38-MOCK. Analysis of the direction of log2
fold change of these 596 overlapping DEGs, however, revealed that 61 of these DEGs showed a different direction of change between the two datasets, meaning that although these genes were differentially expressed in both datasets, differential gene expression in the FUT9 cell line was not the result of glyco-engineering and subsequent differences in cell culturing, as this would otherwise have led to a similar direction of change in gene expression. Therefore, analysis of the 596 overlapping genes yielded an additional 61 DEGs that were included for further analysis. Together, this resulted in 3583 DEGs in the MC38-FUT9 cells for further analysis, of which the expression of 1874 genes was suppressed and the expression of 1709 genes was increased in MC38-FUT9 cells compared to MC38-MOCK cells (Figure 1
E, Table S1B
4.10. Motif Enrichment Analysis
To determine whether the 3583 DEGs in the MC38-FUT9 cell line are subject to a gene regulatory network, motif enrichment analysis and transcription factor prediction analysis was performed with Cytoscape v3.6.0 (https://cytoscape.org/
), and the iRegulon plugin v1.3 [34
], using default settings. Analysis was performed separately for the upregulated and suppressed DEGs.
4.11. Gene Ontology Term Enrichment and Gene Signature Analysis
The gene ontology (GO) term enrichment analysis was performed on the 3583 DEGs by MC38-FUT9 cells using Cytoscape v3.6.0 (https://cytoscape.org/
) and the ClueGO plugin v2.5.0 (http://apps.cytoscape.org/apps/cluego
). Significantly enriched GO terms (Benjamini–Hochberg correction and false discovery rate (FDR) <0.05) were determined with the ontology source GO_BiologicalProcess-EBI-UniProt-GOA, and were subsequently visualized using view style groups, GO level 6–13 and a kappa score threshold of 0.4. For the gene signature analysis, GO term-associated genes were extracted from the Mouse Genome Informatics (MGI) database and compared to our gene set. Gene lists for the following GO terms were used; canonical Wnt signaling pathway, stem cell population maintenance and epithelial to mesenchymal transition. Normalized counts from our RNA-seq analysis of GO term-associated genes were visualized in heatmaps using Morpheus (https://software.broadinstitute.org/morpheus/
4.12. ALDH Activity
To assess ALDH enzymatic activity, the ALDEFLUOR kit (StemCell Technologies, 01700) was used. Cells were harvested and subjected to the ALDEFLUOR assay according to manufacturer’s instructions. The fluorescent ALDH-bright cells (ALDH-High) were detected in the green fluorescence channel (520–540 nm) of the Cyan ADP (Beckman Coulter, Brea, CA, USA). The gates were established using negative control cells that were stained with the ALDEFLUOR reagent and treated with the ALDH inhibitor, DEAB, provided with the ALDEFLUOR kit. Cytometric data analysis was performed with the FlowJo V10 software (Tree Star).
4.13. Tumorsphere Formation
MC38 cells were cultured in 3D as described above. At day 7, representative images of the wells were obtained using the ECLIPSE TE300 (Nikon, Minato City, Tokyo, Japan) microscope and the number of tumorspheres (>50 μm) formed was counted and plotted.
4.14. Drug Resistance Analysis
Cells were seeded in triplicates in 96-well flat-bottom plates (Merck, 0812; 5000 cells per well) or in 96-well clear flat bottom ultra-low attachment plates (Merck, CLS7007; 2000 cells per well). DMSO (vehicle control, Sigma, Saint Louis, MI, USA, 276,855), 5-FU (Selleckchem, Houston, TX, USA, S1209) and oxaliplatin (Selleckchem, S1224) were added to the corresponding wells at the indicated concentrations at day 0 (2D culture) or at day 4 (3D culture). Cells were cultured in the presence of DMSO, 5-FU or oxaliplatin for 72 h in total. Cell viability was assessed using the CellTiter-Blue® Cell Viability assay (Promega, G8080) according to the manufacturer’s instructions. Measurements were performed on the FLUOstar Galaxy (MTX Lab systems, Bradenton, FL, USA, excitation 560 nm and emission 590 nm). Fluorescence intensity was transformed to the % of viable cells by multiplying the value obtained from each concentration of the tested drug with 100 and dividing it to the fluorescent value of the corresponding DMSO concentration control.
4.15. Cancer Stem Cell (CSC)-Enrichment Assay
Cells were cultured in 3D in the presence or absence of 5-FU as described above. For the FUT9-High Mix, 1.3 × 103 MC38-MOCK cells and 0.7 × 103 MC38-FUT9 cells were seeded in triplicates. For the FUT9-Low Mix, 1.9 × 103 MC38-MOCK cells and 0.1 × 103 MC38-FUT9 cells were seeded in triplicates. As a control, single cultures of 2 × 103 MC38-MOCK or 2 × 103 MC38-FUT9 cells were seeded in triplicates. At day 7, tumorspheres were dissociated with an enzyme-free cell dissociation buffer (Gibco, 13,151,014), passed through a 70 µm cell strainer (Falcon, 352,350) and single cell suspensions were obtained. Both at day 0 and at day 7, cells were washed and stained for Lewisx as described above. Cytometric data analysis was performed with the FlowJo V10 software (Tree Star). The % of Lewisx+ cells (indicative of MC38-FUT9 cells) was calculated for each condition and time point. Results from day 7 were compared to results from day 0.
4.16. Cell Cycle Analysis
Cells were subjected to 48 h starvation using serum-free medium, after which serum-containing medium (10% FBS) was added to the cells for 0 h, 24 h or 48 h, respectively. The cells were harvested, washed and fixed with 4% PFA (Electron Microscopy Sciences, 15,710) at 4 °C for 30 min. At RT, 1 × 106 cells per condition were stained with 1 mL staining solution containing 0.1% (v/v) Triton X-100 (Sigma, X100) and 1 µg/mL DAPI (Thermo Fisher, D1306) diluted in PBS for 10 min. The cell-cycle analysis was performed by flow cytometry using a Cyan ADP (Beckman Coulter, USA) or LSRII (BD Biosciences, USA) flow cytometers. Cytometric data analysis in the linear mode was performed with the FlowJo V10 software (Tree Star).
4.17. Cell Proliferation Assay
Cells were seeded in 96-well plates at different absolute numbers (50, 100, 200, 300, 400 or 500 cells/well) and cultured for 3 days (72 h) in 2D or 7 days in 3D as described above. The metabolic activity of the cells (indicative of the cell proliferation rate) was analyzed using the CellTiter-Blue® Cell Viability assay (Promega, G8080) according to the manufacturer’s instructions. Measurements were performed on the FLUOstar Galaxy (MTX Lab systems, USA, excitation 560 nm and emission 590 nm).
4.18. Public Databases
The MuBase Oncology Database (Crown Bioscience; https://www.crownbio.com/oncology/oncology-databases/mubase
) was used to access gene expression data (RNA-seq) of MC38-WT cells under the code: MC38-P0-3-161026. Expression levels were represented as log2
(fragments per kilobase of transcript per million mapped reads—FPKM). Values were acquired from the database and then plotted with the help of the Prism software (GraphPad V10 Software). The R2 genomics analysis and visualization platform (Academic Medical Center, Amsterdam, the Netherlands; https://hgserver1.amc.nl/cgi-bin/r2/main.cgi
) was exploited to address gene expression data (microarray) of human colon cancer cell lines available in the Cancer Cell Line Encyclopedia under the code: Cell line CCLE Cancer Cell Line Encyclopedia-Broad-MAS5.0-u133p2. Cell lines derived from the rectum were excluded from the analysis. Expression levels were represented as log2
(fragments per kilobase of transcript per million mapped reads—FPKM). Values were acquired from the database and then plotted with the help of the Prism software (GraphPad V10 Software). Single-cell RNA sequencing (sc-RNA seq) data from 11 CRC primary tumors and their matched normal samples [56
] were reanalyzed using the R package and the Seurat toolkit (https://satijalab.org/seurat/
] under default parameters. Clustering of epithelial cells based on FUT9 expression (n
> 0 classified as positive, n
= 0 classified as negative and n
= FUT9 transcript number) was followed by tSNE dimensionality reduction (https://lvdmaaten.github.io/tsne/
). For each cluster, the mean gene expression levels were obtained and pheatmap (https://cran.r-project.org/web/packages/pheatmap/index.html
) was used for the generation of the final heatmap. More information about the gene sets used can be found in Table S7
. The corresponding R script is available at https://github.com/MolecularCellBiologyImmunology/fut9-cancer
4.19. Statistical Analysis
For the differential expression analysis, significance was assessed using the Benjamini–Hochberg false discovery rate (FDR) <0.05. Unless otherwise indicated, other statistical analyses were performed with the Prism software (GraphPad V10 Software). Statistical significance was determined by an unpaired Student’s t test for comparison of two groups, by a one way ANOVA for comparison of multiple groups and by multiple Student’s t tests for comparison of two groups across a range of drug dosages: * p < 0.05, ** p < 0.01 and *** p < 0.001.