Transcription Factor PROX1 Suppresses Notch Pathway Activation via the Nucleosome Remodeling and Deacetylase Complex in Colorectal Cancer Stem–like Cells

Activation of the Wnt pathway is the initiating event and one of the key determinants of further pathogenesis in the majority of human colorectal cancers (1). Cellular levels of free β-catenin are normally strictly controlled through a multiprotein complex that comprises the APC protein. Activation of the Wnt pathway prevents the proteasomal degradation of β-catenin, which translocates to the nucleus and binds to and activates the Tcf/Lef1 transcription factors, leading to activation of target genes stimulating cell-cycle progression and tumorigenesis (2). The homeobox transcription factor PROX1 is directly regulated by abnormally high levels of Wnt/β-catenin signaling activity in colorectal cancer (3). Altered levels of PROX1 have been demonstrated in several cancers and PROX1 has been implicated in regulating cell fate in stem and progenitor cells (reviewed in ref. 4). In the normal intestine, PROX1 is expressed in only few secretory cells (3, 5). PROX1 expression is induced in intestinal tumor progression, where it is associated with high-grade dysplasia and an invasive phenotype (3, 6). Importantly, a subpopulation of the PROX1-expressing adenoma and colorectal cancer cells displays stem cell features and deletion or silencing of Prox1 reduces tumor size and the number of LGR5⁺ stem cells (3, 5).

Cross-talk between the Notch and Wnt signaling pathways has an important function in adult intestinal homeostasis (7, 8). Genetic or chemical inhibition of the Notch pathway results in reduced number of LGR5⁺ intestinal stem cells and accumulation of differentiated secretory cells (9, 10). In intestinal adenomas, Notch activity is regulated downstream of Wnt signaling through β-catenin–mediated transcriptional activation of the Notch ligand Jagged1, which contributes to intestinal tumorigenesis in Apc^min/+ mice (11). Consistently with these results, ectopic overexpression of the active NOTCH1 intracellular domain (NICD1) leads to an elevated number of adenomas in Apc-mutant mice (7, 12). However, despite increased numbers of tumors in Apc^min/+;NICD1;villin-Cre^ERT2 mice, these tumors were predominantly low-grade adenomas, whereas control tumors from Apc^min/+ mice displayed features of high-grade adenomas, suggesting that active NOTCH1 is needed for tumor formation rather than tumor progression. Interestingly, aberrant NOTCH1 recruits the histone methyltransferase SET1 domain bifurcated 1 (SETDB1) to suppress Wnt target genes, including PROX1 (12). Although the supportive role of Notch activity in stem cells is a widely accepted concept, these data raised the possibility that Notch activity is not required in PROX1-expressing colorectal cancer stem-like cells. Here, we set out to study the relationship between PROX1 and the Notch pathway in colorectal cancer.

The National Animal Experiment Board at the Provincial State Office of Southern Finland approved all experiments performed with mice (ESAVI/6306/04.10.07/2016 and ESAVI/7945/04.10.07/2017). All mice were maintained in the C57Bl/6J background.

The Apc^flox/flox (13), Prox1^flox/flox (14) and Lgr5-EGFP-IRES-Cre^ERT2 (15) mice were crossed to generate intestinal stem cell–specific deletion of the Apc and/or Prox1 gene. To delete Apc and induce tumorigenesis, 2 mg tamoxifen (Sigma #T5648) dissolved in 100 μL corn oil (Sigma #8001-30-7) was given by oral gavage at the age of 8–10 weeks. To chemically inhibit the Notch pathway, mice were given intraperitoneal injections of the γ-secretase inhibitor dibenzazepine (Tocris #4489; 20 μmol/kg/day) diluted in 100 μL 0.5% (Hydroxylpropyl)methylcellulose (Sigma #H7509)-0.1% Tween20 (Acros organics #BP337-500) or vehicle on five consecutive days, starting at day 5 or day 21 after tamoxifen administration. The in vivo experiment was repeated twice. For each experiment, 6–10 mice were used.

To analyze the effect of Notch inhibition on Prox1 lineage tracing in intestinal adenomas, dibenzazepine (20 μmol/kg/day) was administered intraperitoneally on four consecutive days to 16-week-old Apc^min/+;Rosa26-^LSL-TdTomato (Jackson Laboratories, Stock 007914); Prox1-Cre^ERT2 (16) mice. Prox1 lineage tracing was activated by a single dose of 2 mg tamoxifen after the last dose of dibenzazepine and the mice were terminated after 24 or 72 hours. The in vivo experiment was repeated three times. For each experiment, 10–12 mice were used.

To assess cell proliferation, mice were given one intraperitoneal injection of 1 mg/mL 5-ethynyl-2′-deoxyuridine (EdU; Thermo Fisher Scientific #A10044) dissolved in 100 μL 0.9% NaCl and terminated 4 hours later. For analysis of EdU incorporation, EdU⁺ cells of 50–60 tumors from both groups were quantified and normalized to the total number of tumor cells.

The Ethical Committee of Helsinki University Central Hospital approved all the experiments involving patient samples. We obtained written informed consent from the patients and the studies were conducted in accordance with the Declaration of Helsinki. Tissue biopsies from patients were processed according to previously published method (17). Patient I harbored a KRAS^G12D mutation, but neither of the patients had clinically relevant BRAF or PI3KCA mutations (6). Intestinal crypts from Apc^flox/flox;villin-Cre^ERT2 (18), Apc^flox/flox; Lgr5-EGFP-IRES-Cre^ERT2, Apc^min/+;Rosa26^LSL-TdTomato; Prox1-Cre^ERT2 and Apc^min/+; Prox1^flox/flox;villin-Cre^ERT2 mice were isolated and cultured as described previously (5, 19). To activate gene deletion, cultures were treated with 300 nmol/L 4-hydroxytamoxifen (4-OH-Tam) for 24 hours. Organoids with endogenously active β-catenin/TCF pathway were then selected and cultured in growth factor–deficient medium. When indicated, organoids were cultured in the presence of 10 μmol/L of DAPT (Tocris #2634) or dibenzazepine (Tocris #4489) for five days, starting on day 3 for an early time period or day 12 for a later time period after Apc deletion. For determination of the replating efficiency, the organoids were extensively trypsinized to obtain clusters of 4–5 cells, which were counted and embedded into Matrigel (Corning #35623) in equal numbers. The number of organoids/well was counted under microscope 3–4 days after replating. To analyze changes in gene expression after Apc deletion, organoids were lysed at different time points after addition of 4-OH-Tam. Changes in gene expression were compared with normal organoids (d0).

The SW1222, SW620, and LS174T colorectal cancer cell lines were cultured in DMEM (Lonza #BE12-707F) + 10 % FBS (Biowest, S181B-500), 2 mmol/L l-glutamine (Corning #25-005-Cl), 100 U/mL penicillin and streptomycin (Lonza #DE-17-602E). HEK293FT cells (Thermo Fisher Scientific #R70007) were cultured in high-glucose DMEM (Gibco #11960-044) + 10% FBS, 2 mmol/L l-glutamine, and 100 U/mL penicillin and streptomycin. Cells were maintained at 37°C in a humidified incubator with 5% CO₂ and passaged every 2–4 days. The SW620 cell line was obtained from ATCC (CCL-227). The SW1222 and LS174T cell lines were a kind gift from Dr. Meenhard Herlyn, Wistar Institute (Philadelphia, PA; 2010) and Dr. Olli Kallioniemi, Institute for Molecular Medicine Finland (Helsinki, Finland; 2015), respectively. The SW620 and LS174T cells were originally authenticated by ATCC. All the cells were routinely authenticated by morphologic inspection and expression of specific markers, and tested for Mycoplasma by DAPI staining. The HEK293FT cells were used within approximately 20 passages after thawing original stocks. For experimental procedures, colorectal cancer cell lines were used between 5 and 15 passages after thawing. When indicated, cells were cultured in the presence of 1 μmol/L of entinostat (LC Laboratories #E-3866) or vorinostat (LC Laboratories #V-8477). For analysis of “megacolony” formation (20), 1,000 SW1222 cells were embedded in Matrigel and analyzed at day 10. Spheroids were imaged with EVOS FL inverted epifluorescence microscope (Thermo Fisher Scientific) using 10× or 20× LD Ph Air objectives.

Organoids were centrifuged at 1,500 rpm for 5 minutes and the Matrigel was removed. To obtain a single-cell suspension, organoids were incubated in Hank's Balanced Salt Solution (HBSS; Gibco #14175-053) containing 1 mg/mL collagenase type 1 (Worthington #LS004196), 1 mg/mL collagenase H (Roche #11074032001), 4 mg/mL dispase II (Sigma #04942078001), and 1,000 U/mL benzonase (ChemCruz #sc-202391) for 30 minutes at 37°C with gentle shaking, followed by 10-minute incubation with trypsin. After fixation with 2% PFA, the cells were blocked with mouse Fc block (1 μg/10⁶ cells, BD Biosciences #553141) and permeabilized with 1% BSA+ 0.1% Triton X-100 in HBSS. Cells were then incubated with goat anti-hPROX1 (R&D Systems #AF2727) for 20 minutes, followed by 20-minute incubation with Alexa 647 donkey anti-goat secondary antibody. HBSS washes were performed in between each step. The samples were measured with a Guava EasyCyte instrument (Millipore). The flow cytometric analysis was repeated twice, using biological replicates.

Apc^flox/flox;villin-Cre^ERT2 organoids were treated with vehicle or dibenzazepine for 5 days, starting on day 3 after Apc deletion. Organoids were then dissociated to obtain a single-cell suspension. Cells in 0.04% BSA-PBS were analyzed using the Chromium Single Cell 3′RNA-sequencing system (10x Genomics) with the Reagent Kit v2 according to the manufacturer's instructions. Briefly, the cells were loaded into Chromium Single Cell Chip v2 (10x Genomics) and gel beads in emulsion (GEM) generation was performed aiming at 3,000 cell captures per sample. Subsequent cDNA purification, amplification (12 cycles), and library construction (sample index PCR 14 cycles) was performed as instructed. Sample libraries were sequenced on the Illumina NovaSeq 6000 system using S1 flow cell (Illumina) with following read lengths: 26 bp (Read 1), 8 bp (i7 Index), 0 bp (i5 Index), and 91 bp (Read 2), resulting in 115,754 and 113,705 mean reads per cell for the sample vehicle and dibenzazepine, respectively. The Cell Ranger v 2.1.1 mkfastq and count pipelines (10x Genomics) were used to demultiplex and convert Chromium single-cell 3′ RNA-sequencing barcodes and read data to FASTQ files and to generate aligned reads and gene-cell matrices. Reads were aligned to mouse reference genome mm10. We used the Seurat R package for quality control, filtering, and analysis of the data (21). Cells were filtered on the basis of UMI counts and percentage of mitochondrial genes. Cells with more than 5% of mitochondrial genes were filtered out. The expression matrix was further filtered by removing genes with expression in less than 10 cells and cells with less than 800 expressed genes. The minimum expression threshold was set at 1. The final dataset consisted of 968 cells in the Vehicle sample and 1,240 cells in the dibenzazepine sample. To be able to compare the two samples, we performed canonical correlation analysis (CCA) to identify shared correlation structures and aligned the dimensions using dynamic time warping. After this, we performed clustering using tSNE and set the resolution at 0.5. The single-cell RNA-sequencing data can be accessed from the Gene Expression Omnibus under accession number GSE118055.

The shHDAC1 constructs (TRCN0000004817, TRCN0000004818), shHDAC2 (TRCN0000196590, TRCN0000197086), shMTA1 (TRCN0000013361, TRCN0000013362), and shScramble (SH002) were acquired from the TRC library. The NOTCH1 promoter-based luciferase reporter was described previously (22). BirA-hPROX1 fusion protein was cloned into a FUW lentiviral vector. Myc-BirA (23) was amplified using primers GAAGCTTGGGCTGCAGGTCGACTCTAGAGCCACCATGGAACAAAAACTCATCTCAG and AGGGCTGTGCTGTCATGGTCAGGCATAGATCCTGAGCCCTTCTCT. hPROX1 was amplified using primers ATGCCTGACCATGACAGCACAG and TTATCGATAAGCTTGATATCGAATTCGGCGCGCCCTACTCATGAAGCAGCTCTTG. Gel-purified fragments were then assembled to FUW-Xbal-Ascl-CIP. Transfection and transduction were performed as described previously (5).

SW1222 cells were transduced with FUW-BirA-PROX1 or FUW-BirA-NLS-Cherry lentivirus, plated at 1 × 10⁷ cells per sample and incubated with 50 μmol/L biotin (Sigma Aldrich #B4501). After 24 hours, the cells were washed three times with PBS, stored at −80°C and later lysed on ice for 10 minutes in PLCLB buffer (50 mmol/L HEPES, 150 mmol/L NaCl, 10% glycerol, 1% Triton X-100, 1.5 mmol/L MgCl₂, 1 mmol/L EGTA, 10 mmol/L Na₄P₂O₇, 100 mmol/L NaF), supplemented with 1.0 mmol/L PMSF and 10 μL/mL protease inhibitor cocktail (Sigma-Aldrich #P8340), 0.1% SDS, and 80 U/mL Benzonase (ChemCruz #sc-202391), followed by three cycles of sonication and centrifugation twice at 16,000 × g at 4°C to remove insoluble material. Cleared lysates were loaded into Bio-Spin chromatography columns (Bio-Rad Laboratories #732-6008) loaded with Strep-Tactin sepharose (400 μL 50% slurry; IBA Lifesciences #2-1201-010) and washed with ice-cold PLCLB lysis buffer and ice-cold PLCLB buffer without Triton X-100 and protease inhibitors. Bound proteins were eluted twice with 300 μL of freshly prepared 0.5 mmol/L biotin in PLCLB buffer without Triton X-100 and protease inhibitors. Liquid chromatography–tandem mass spectrometry analysis was performed as described previously (24). Two biological replicates were used.

SW1222 cells transduced with lentivirus were trypsinized and 100,000 cells were plated into wells of a 24-well plate. Cells were transfected with 1 μg of NOTCH1 promoter–based luciferase reporter (N1PR-luc) using Fugene6 transfection agent according to manufacturer's instructions (Promega #E2692) and lysed 48 hours later. Luciferase measurement was done using the dual firefly luciferase assay kit and preformed according to manufacturer's instruction (Promega #E1910). All experiments were done in quadruplicate and repeated 2–3 times. Luciferase value was normalized to protein concentration.

Four different guide RNAs (gRNAs) targeting PROX1 start codon region were designed using http://crispr.mit.edu/ (gPROX1.1: GCTCAAGAATCCCGGGACCCTGG, gPROX1.2: ATCTTCAAAAGCTCGTCAGCTGG, gPROX1.3: CCTGAGAGCAAAGCGCGCCCGGG, gPROX1.4: CGGGTTGAGAATATAATTCGGGG). gRNA-PCR transcriptional units were assembled as described previously (25) and tested by transfection to HEK293 cells together with WT SpCas9 expressing plasmid CAG-Cas9-T2A-EGFP-ires-puro (Addgene plasmid #78311). Deletion generation was assessed by PCR of the targeted region (PROX1_Fw: GTGCCATAAATCCCAGAGCCTATG, PROX1_Rv: ACTTTCTCGGGGACTCACAGAC). gRNAs pairs (1+3 and 1+4) generating deletions most efficiently were cloned into a modified version of LentiCRISPR v2 backbone (Addgene plasmid #52961). SW1222 cells were transduced with lentivirus containing the sgPROX1-1 (gRNA pair 1+3), sgPROX1-2 (gRNA pair 1+4), or sgCTRL constructs and selected using 5 μg/mL puromycin (Merck #540222-100). The cells were then used for further experiments. PROX1 deletion was confirmed by Western blot analysis.

The quality of RNA was determined with Bioanalyzer (Agilent Technologies) and further analyzed on genome-wide Illumina Mouse WG-6 v2 Expression BeadChips (Illumina). Illumina's GenomeStudio software was used for initial data analysis and quality control and the detailed data analysis was performed with the Chipster software (www.chipster.csc.fi). The data were normalized with quantile normalization. When comparing the expression profiles of TdTomato⁺ and TdTomato⁻ organoids, the datasets derived from hybridizations at different time points were normalized separately with quantile and gene average normalizations and the data were then unified. For pathway enrichment analysis, the Enrichr software (http://amp.pharm.mssm.edu/Enrichr/) was used. The gene expression dataset was transferred to the Gene Set Enrichment Analysis software (http://www.broadinstitute.org/gsea; refs. 26, 27) and the analysis was carried out with default parameters except that the „exclude smaller sets” was set to 25 and gene permutation was applied. We used a modified list of the kegg.v3.1.symbols gene set (http://www.broadinstitute.org/gsea/msigdb) with the addition intestinal stem cell-specific genes (28), genes activated during Apc^min/+ tumorigenesis and suppressed by NOTCH1 activation, and genes upregulated in NOTCH1-activated tumors (12). FDR q < 0.1 was regarded significant. The microarray data can be accessed from the Gene Expression Omnibus under the accession number GSE117981.

Data are presented as mean + SEM unless otherwise indicated. Statistical comparison of two groups was done by two-tailed unpaired t test using the GraphPad Prism 6.0 software. P < 0.05 was considered statistically significant and the significance is marked by *, P < 0.05; **, P < 0.01; and ***, P < 0.005.

Other methods are detailed in the Supplementary Materials and Methods.

NOTCH1 was previously reported to suppress PROX1 expression in intestinal adenoma cells (12). To study this process in adenoma stem cells, we established organoids derived from Apc^flox/flox;Lgr5-EGFP-IRES-Cre^ERT2 (LApc) or Apc^flox/flox;villin-Cre^ERT2 (VApc) mice. For induction of tumorigenesis, 4-OH-Tam was added to the organoid cultures, leading to Apc deletion only in the LGR5+ intestinal stem cells (LApc^Δ/Δ) or in all intestinal epithelial cells (VApc^Δ/Δ). To model the effect of Notch inhibition after Apc deletion, we inhibited Notch by adding the γ-secretase inhibitor (GSI) dibenzazepine or DAPT to the cultures three days later (Fig. 1A). Notch inhibition decreased the number of viable organoids and RNAs encoding the Notch target genes hairy/enhancer of split (Hes1), Notch regulated ankyrin repeat protein (Nrarp) and Olfactomedin 4 (Olfm4), and the intestinal stem cell markers Leucine-rich repeat-containing G-protein–coupled receptor 5 (Lgr5) and Achaete-Scute family BHLH transcription factor 2 (Ascl2) in both Apc-deleted and control organoids (Supplementary Fig. S1A and S1B). Lgr5-EGFP⁺ intestinal stem cell number was also decreased, indicating that Notch is critical for LGR5⁺ stem cell activity early after Apc deletion (Fig. 1B). Furthermore, Notch inhibition increased the expression of Prox1 RNA and the number of PROX1⁺ cells (Fig. 1B; Supplementary Fig. S1A and S1B). Although the overall number of Lgr5-EGFP⁺ stem cells decreased, only a minor decrease of the Lgr5-EGFP⁺ PROX1⁺ cell population was observed (Fig. 1B). Furthermore, single-cell RNA analysis of dibenzazepine or vehicle-treated Apc^flox/flox; villin-Cre^ERT2 organoids showed that the Prox1-expressing cluster was enriched for intestinal stem cell markers Lgr5 and Tnfrsf19 (15, 29) after Notch inhibition (Fig. 1C). These data suggested that the PROX1⁺ stem-like cells had a selective growth advantage when Notch was inhibited.

Kinetic analysis of 4-OH-Tam–treated Apc^flox/flox;villin-Cre^ERT2 organoids revealed that increased expression of Prox1 and the Wnt target genes Axin2, Sox9, c-Myc correlated with decreased expression of Notch targets Olfm4 and Hes1 (Fig. 2A–D). However, Notch inhibition on day 12 after Apc deletion no longer affected the proportion of viable organoids, their colony formation efficiency, or the viability of patient-derived colorectal cancer organoids (Fig. 2E and F), suggesting that active Notch signaling was no longer required to maintain Apc-mutant intestinal epithelium. In agreement with this, the Lgr5-EGFP⁺ cells in the control intestine of Apc^flox/flox;Lgr5-EGFP-IRES-Cre^ERT2 mice were uniformly positive for NICD1 (Fig. 2G), whereas only 36% of the Lgr5-EGFP⁺ cells were positive for NICD1 21 days after Apc deletion (Fig. 2H). A majority of the Lgr5-EGFP⁺ stem cells were PROX1⁺, and PROX1⁺ cells were negative for NICD1 (Fig. 2I and J), supporting the notion that PROX1⁺ cells have a selective growth advantage after Apc deletion. Collectively, these results suggest that the Notch pathway is important for intestinal stem cell activity during a transient phase after Apc deletion, but is inactivated in later appearing PROX1-expressing stem-like cells.

To assess whether the PROX1⁺ adenoma cells retain their stem cell activity even when Notch activity is inhibited, we induced tumorigenesis by tamoxifen treatment of Apc^flox/flox; Lgr5-EGFP-IRES-Cre^ERT2 mice and applied the γ-secretase inhibitor dibenzazepine 5 days later for 5 days. As reported previously, dibenzazepine treatment converted intestinal epithelial cells into Mucin2⁺ (MUC2) goblet cells and reduced the number of Lgr5-EGFP⁺ stem cells in the normal intestine (7), leading to death of the mice within 8 days after the start of the dibenzazepine treatment. Although the number of Lgr5-EGFP⁺ adenoma stem cells was decreased at day 10 after Apc deletion when the Notch pathway was inhibited, the majority of them were PROX1⁺ (Fig. 3A and B). In contrast, Notch inhibition starting 21 days after Apc deletion, when the majority of the Lgr5-EGFP⁺ cells were PROX1⁺, did not significantly decrease the number of Lgr5-EGFP⁺ stem cells (Fig. 3C). Furthermore, the MUC2⁺ goblet cells occurred selectively in the PROX1- tumor cell population (Fig. 3D–F; Supplementary Fig. S2A). Dibenzazepine treatment also promoted increased proliferation of PROX1⁺ tumor cells and decreased proliferation of PROX1⁻ cells on days 10 and 26 after Apc deletion (Fig. 3G–I; Supplementary Fig. S2B), suggesting that Notch inhibition provides the PROX1⁺ cells an additional growth advantage.

To further analyze the PROX1⁺ stem-like cells, we induced lineage tracing in Apc^min/+;Rosa26-^LSL-TdTomato;Prox1-Cre^ERT2 mice and treated the mice with dibenzazepine or vehicle for 4 days. TdTomato fluorescence 3 days after the last dose of dibenzazepine indicated that some PROX1⁺ cells had differentiated to PROX1⁻/TdTomato⁺ Paneth cells (lysozyme⁺) and goblet cells (MUC2⁺; Supplementary Fig. S2C and S2D). Furthermore, Notch inhibition increased the number of proliferating PROX1⁺ tumor cells and the number of proliferating TdTomato-labeled PROX1⁻ progeny in more advanced tumors of the Apc^min/+ mice (Fig. 3J–L), suggesting that continued growth of the tumors originates from the PROX1⁺ tumor cells.

As PROX1⁺ adenoma cells retained their stem cell activity after Notch inhibition, we asked whether NOTCH1 overexpression affects the stem cell properties of PROX1⁺ cells in mouse tumor organoid cultures and human colorectal cancer cells. We transduced Apc-mutant organoids derived from Apc^min/+; Rosa26-^LSL-TdTomato; Prox1-Cre^ERT2 mice with NICD1-EGFP or control lentivirus and then subjected them to short-term genetic labeling by applying 4-OH-Tam. Sixteen hours after the addition of 4-OH-Tam, only cells positive for PROX1 displayed red TdTomato fluorescence (Fig. 4A and B), confirming the specificity of the labeling. The NICD1-overexpressing cells formed less colonies, which had fewer TdTomato⁺ cells than control lentivirus–transfected cells (Fig. 4C and D), reflecting Prox1 suppression in these cells. NICD1 overexpression in the PROX1⁺ SW620 and SW1222 human colorectal cancer cell lines that are enriched for stem-like cells (20, 30) indicated that the NICD1-overexpressing cells suppressed PROX1 RNA and protein (Fig. 4E–G), supporting previous findings (12). Furthermore, when the control and NICD1-transduced SW1222 cells were grown as three-dimensional cultures in Matrigel, the NICD1-overexpressing cells were unable to form “megacolonies” (Fig. 4H), indicating that they lack stem cell properties (31). These results suggest that NOTCH1 activation suppresses PROX1 and thus stem cell properties of PROX1⁺ colorectal cancer cells.

As increased Prox1 expression correlated with decreased Olfm4 and Hes1 expression, we next assessed whether PROX1 regulates the Notch pathway. We first analyzed the gene expression signature of FACS-isolated TdTomato⁺/PROX1⁺ cells by microarray analysis. As expected (5), we found that Prox1-expressing cells displayed an enrichment of the intestinal stem cell gene signature (Supplementary Fig. S3A; ref. 28), which had similarity to gene sets upregulated by Apc mutation and suppressed by NICD1 overexpression in intestinal adenomas (Supplementary Fig. S3B; ref. 12). Accordingly, the Prox1 cell transcriptome showed a negative enrichment of genes induced by NICD1 overexpression (Supplementary Fig. S3C), indicating that Notch signaling is suppressed in the Prox1-expressing cells.

We next induced Prox1 and Apc deletion by tamoxifen treatment of Apc^flox/flox;Prox1^flox/flox;Lgr5-EGFP-IRES-Cre^ERT2 (LApcP^Δ/Δ) mice to analyze PROX1 regulation of the Notch pathway in intestinal adenoma stem cells. Interestingly, Prox1 deletion increased the ratio of OLFM4⁺ cells in the Lgr5-EGFP⁺ cell population, suggesting Notch activation in these cells (Fig. 5A and B). Also, Prox1 deletion in organoids derived from Apc^min/+;Prox1^flox/flox;villin-Cre^ERT2 mice resulted in increased expression of the Notch targets Olfm4 and Hey2 (Fig. 5C). Furthermore, Prox1 deletion decreased the expression of Pleiotrophin (Ptn), Pyruvate dehydrogenase kinase 4 (Pdk4), and Distal-less homebox 3 (Dlx3), genes that are suppressed by NICD1 overexpression, and increased the expression of Tripartite motif-containing protein 31 (Trim31) and Carbonic anhydrase 13 (Car13), transcripts that are induced by NICD1 in intestinal adenomas (Supplementary Fig. S3D; ref. 12).

To explore PROX1 regulation of NOTCH in human colorectal cancer, we employed CRISPR/Cas9 to delete PROX1 exon 2 (Supplementary Fig. S3E and S3F). SW1222 cells were lentivirally transduced with Cas9 plus two different PROX1 gRNAs and selected with puromycin. PROX1 deletion was first confirmed by Western blotting (Supplementary Fig. S3G), and the cells were then transfected with a NOTCH1 promoter–driven luciferase reporter. When analyzed 48 hours later, the PROX1-deleted cells showed a 2-fold higher luciferase signal than the control cells (Fig. 5D). Furthermore, PROX1 deletion increased expression of the Notch targets OLFM4 and HEY2 (Fig. 5E). PROX1 overexpression in the NOTCH1-active LS174T cells (8) also decreased the expression of OLFM4 and HEY2 (Fig. 5F). These data indicate that PROX1 suppresses the Notch pathway in mouse adenomas and human colorectal cancer cells.

To investigate the mechanism of how PROX1 suppresses the Notch pathway, we performed BirA-mediated proximity labeling to identify PROX1-interacting proteins in SW1222 cells (Supplementary Fig. S4A and S4B). Remarkably, we found that PROX1 interacts with several proteins that have chromatin remodeling function. In particular, several components of the NuRD corepressor complex were among the top 10 hits (Supplementary Table S1). By coimmunoprecipitation, we determined that at least HDAC1 and MTA1, key components of the complex, coprecipitated with PROX1 from lysates of SW1222 and SW620 cells (Fig. 6A). The prior report that the NuRD complex suppresses Notch pathway in Schwann cells (32), raised the possibility that PROX1 suppresses the Notch pathway in colorectal cancer cells via the NuRD complex. To assess this, we treated SW1222 cells with two HDAC inhibitors for 72 hours to inhibit HDAC1 and HDAC2. This increased the expression of the Notch target genes OLFM4 and HEY2 (Supplementary Fig. S4C). Lentiviral silencing of HDAC1, HDAC2, or MTA1 increased the NOTCH1 promoter-luciferase reporter expression and the expression of OLFM4 (Fig. 6B–E; Supplementary Fig. S4D–S4F), indicating that the NuRD complex suppresses Notch signaling in colorectal cancer. Furthermore, HDAC1, HDAC2, or MTA1 silencing in PROX1-overexpressing LS174T cells blocked the PROX1-mediated suppression of OLFM4 (Fig. 6F), suggesting that PROX1 acts via the NuRD complex to regulate the Notch signaling pathway in colorectal cancer. Chromatin immunoprecipitation using antibodies against PROX1, HDAC1, and MTA1, followed by qPCR of three regions of the NOTCH1 promoter, revealed two potential PROX1-NuRD-binding sites at −1.5 kb and −0.8 kb, whereas only MTA1 bound weakly to the NOTCH1 promoter at −3.5 kb (Fig. 6G–I; Supplementary Fig. S4G). These data indicate that PROX1 suppresses the Notch pathway by recruiting the NuRD complex to the NOTCH1 promoter (Fig. 6J).

In this study, we provide evidence that the PROX1 transcription factor regulates colorectal cancer stem-like cells via a bidirectional interaction with the Notch pathway. We show that Notch inhibition in Apc wild-type or mutant organoids decreases expression of intestinal stem cell markers while increasing PROX1⁺ cells. Chemical Notch inhibition provided PROX1⁺ cells a further growth advantage in the emerging tumors. Interestingly, Prox1 deletion increased expression of Notch target genes and NOTCH1 promoter activity. Mechanistic analysis of this finding indicated that PROX1 recruits the NuRD complex to the NOTCH1 promoter to suppress the Notch pathway in colorectal cancer.

Intestinal crypt cells are plastic and compensatory mechanisms have been reported to produce fast-cycling crypt base columnar stem cells following injury (33–35). Similarly, colorectal tumors are composed of a flexible hierarchy of stem-like cells, progenitors, and more differentiated cells (reviewed in ref. 36). In the normal intestine, PROX1 is expressed in enteroendocrine cells and some Paneth cells (3, 5). Furthermore, the PROX1⁺ enteroendocrine cells can function as injury-inducible stem cells (35). In colorectal cancer, PROX1 is directly regulated by the β-catenin/TCF4 pathway. Notch inhibition at an early phase after Apc deletion resulted in increased production of PROX1⁺ adenoma cells, but Notch inhibition had only a minor effect later during tumor development due to Notch downregulation in PROX1⁺ cells. Analysis of single-cell RNA sequencing of control and Notch-inhibited organoids showed that Prox1⁺ cell clusters express some intestinal stem cell markers, suggesting that the PROX1⁺ cells form a subpopulation of intestinal stem-like cells. Furthermore, the LGR5⁺PROX1⁻ cells were decreased by Notch inhibition, indicating that a subpopulation of the stem-like cells is Notch dependent. In line with this, Schmidt and colleagues demonstrated that high NOTCH activity marks a distinct colon cancer subpopulation with low levels of WNT and MAPK activity (37). Furthermore, NOTCH1 was shown to be coexpressed with the intestinal stem cell marker BMI1, whereas the LGR5⁺ stem-like cells expressed Wnt markers (38). Considering that PROX1 is a Wnt target, these studies suggest that PROX1 marks a high Wnt and Notch negative stem-like cell population.

NOTCH1 is known to suppress PROX1 expression in lymphatic endothelial cells (39), thyroid cancer cells (40), myoblasts (41), in the developing Drosophila intestine (42), and in mouse intestinal adenomas (12). Although the function of the Notch pathway in the WT intestinal stem cells is well established (8), how it affects colorectal cancer progression is unclear. NOTCH1 has been shown to be essential in the initiation of colorectal cancer (7, 11, 12). However, deletion of the Notch effector Rbpj did not affect tumorigenesis in Apc-mutant intestine (43). Furthermore, aberrant NOTCH1 expression decreases during tumor progression and metastasis, suggesting that the Notch pathway functions mainly in the early phase of colorectal cancer progression (44). On the other hand, PROX1 was shown to be important in colorectal cancer progression and metastatic outgrowth of Wnt high colon cancer cells (3, 45), rather than for adenoma initiation. In our models, PROX1 and the Notch pathway suppressed each other, indicating that high PROX1 expression and active Notch signaling are mutually exclusive. PROX1 has previously been shown to suppress the Notch pathway to allow differentiation of myoblasts (41) and neurons (46). We show evidence that PROX1 suppresses Notch pathway in colorectal cancer via recruiting the NuRD complex to the NOTCH1 promoter. The NuRD complex consists of seven subunits that function together in chromatin remodeling and histone deacetylation (reviewed in ref. 47). Previous studies have shown that the NuRD complex maintains the silencing of tumor suppressor genes in colorectal cancer (48). Furthermore, deficiency of Mbd2, a component of the complex, reduces tumor burden in Apc^min/+ mice and attenuates Wnt signaling (49, 50). These studies indicate that the NuRD complex performs an important function in colorectal cancer. Importantly, aberrant NOTCH1 expression suppressed the stem cell activity of the PROX1-positive stem-like cells. These observations raise the possibility that PROX1 suppresses the Notch pathway to promote malignant growth.

In conclusion, we demonstrate that PROX1 interacts with the NuRD complex to suppress Notch signaling in colorectal cancer stem-like cells. Furthermore, we show that ectopic expression of active NOTCH1 suppresses PROX1 and the stem cell activity of PROX1-positive cells, suggesting a reciprocal interaction of PROX1 and Notch pathways in regulation of colorectal cancer stem-like cells. On the basis of our results, we propose a model where the LGR5⁺ stem cells require active Notch signaling early after Apc deletion. However, some of these cells start to express PROX1, which then suppresses the Notch pathway. These PROX1⁺ LGR5⁺ cells proliferate and support tumor growth independently of the Notch pathway. Our studies thus give new insight into the complex regulation of stem-like cells and identify PROX1 as a marker of a Notch-independent stem cell population.

No potential conflicts of interest were disclosed.

Conception and design: J. Högström, Z. Wiener, K. Alitalo

Development of methodology: J. Högström, S. Heino, D. Balboa

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Högström, S. Heino, P. Kallio, M. Lähde, V.-M. Leppänen, Z. Wiener

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Högström, S. Heino, K. Alitalo

Writing, review, and/or revision of the manuscript: J. Högström, Z. Wiener, K. Alitalo

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Högström, K. Alitalo

Study supervision: K. Alitalo

This work was funded by the Academy of Finland (grants nos. 292816 and 273817, Centre of Excellence Program 2014–2019: Translational Cancer Biology grant no. 307366), Cancer Foundation in Finland, Sigrid Juselius Foundation, Hospital District of Helsinki and Uusimaa Research Grants, Helsinki Institute of Life Sciences (HiLIFE), Biocenter Finland (all to K. Alitalo), Swedish Cultural Foundation in Finland, Magnus Ehrnrooth Foundation, Biomedicum Helsinki Foundation, Ida Montini Foundation, K Albin Johansson Foundation, Orion Research Foundation, Medicinska understödsföreningen Liv&Hälsa and Maud Kuistila Memorial Foundation (all to J. Högström), János Bolyai Grant by the Hungarian Academy of Sciences (to Z. Wiener). We thank Dr Markku Varjosalo and the proteomics unit at the Institute of Biotechnology, University of Helsinki, for performing mass spectromerty, Dr. Seppo Kaijalainen for cloning the BirA-PROX1 construct, Dr. Timo Otonkoski from the Molecular Neurology Program and Biomedicum Stem Cell Center, University of Helsinki for supporting the cloning of the sgCTRL and sgPROX1 constructs, Dr. Guillermo Oliver from the Feinberg Cardiovascular Research Institute, Northwestern University, Chicago for the Prox1^flox/flox and Prox1-Cre^ERT2 mice; Dr. Tohru Kiyono from the National Cancer Center, Japan for the NOTCH1 promoter-based luciferase reporter, Dr. Riikka Kivelä for discussions, Drs. Pekka Katajisto, Saara Ollila, and Cecilia Sahlgren for reading and commenting of the manuscript, and David He, Kirsi Mattinen, Tanja Laakkonen, and Tapio Tainola for technical assistance. We also thank the Single Cell Analytics core facility at the Institute for Molecular Medicine Finland, the Biomedicum functional genomics unit and Biomedicum imaging unit, University of Helsinki for providing materials and services.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.