The recent classification of colon cancer into molecular subtypes revealed that patients with the poorest prognosis harbor tumors with the lowest levels of Wnt signaling. This is contrary to the general understanding that overactive Wnt signaling promotes tumor progression from early initiation stages through to the later stages including invasion and metastasis. Here, we directly test this assumption by reducing the activity of ß-catenin–dependent Wnt signaling in colon cancer cell lines at either an upstream or downstream step in the pathway. We determine that Wnt-reduced cancer cells exhibit a more aggressive disease phenotype, including increased mobility in vitro and disruptive invasion into mucosa and smooth muscle in an orthotopic mouse model. RNA sequencing reveals that interference with Wnt signaling leads to an upregulation of gene programs that favor cell migration and invasion and a downregulation of inflammation signatures in the tumor microenvironment. We identify a set of upregulated genes common among the Wnt perturbations that are predictive of poor patient outcomes in early-invasive colon cancer. Our findings suggest that while targeting Wnt signaling may reduce tumor burden, an inadvertent side effect is the emergence of invasive cancer.

Implications:

Decreased Wnt signaling in colon tumors leads to a more aggressive disease phenotype due to an upregulation of gene programs favoring cell migration in the tumor and downregulation of inflammation programs in the tumor microenvironment; these impacts must be carefully considered in developing Wnt-targeting therapies.

Watch the interview with Marian L. Waterman, PhD, recipient of the 2023 MCR Michael B. Kastan Award for Research Excellence: https://vimeo.com/847435577

This article is featured in Highlights of This Issue, p. 355

Globally, colorectal cancer is one of the leading causes of cancer-related deaths. Although increased screening, changes in lifestyle habits, and improvements in treatment have decreased the colorectal cancer mortality rate, metastatic disease continues to be a significant challenge to treat. Approximately 50% to 60% of patients with colorectal cancer will develop metastatic disease, and the 5-year relative survival for these patients is 13.9% (1). Mutations in the Wnt signaling pathway are a hallmark of colorectal cancer, with 80% to 90% of patients harboring pathway-activating mutations. The most common mutations in colorectal cancer target the tumor suppressor adenomatous polyposis coli (APC), which limits the ability of cells to sequester and degrade the Wnt effector ß-catenin. Stabilized ß-catenin can translocate to the nucleus for recruitment by LEF/TCF transcription factors to activate Wnt target gene expression. Although the general consensus is that APC-inactivating mutations chronically activate ß-catenin-LEF/TCF signaling and render colorectal cancer active and independent of Wnt ligands, recent evidence has shown that Wnt ligands are still capable of modulating Wnt activity, even in APC-mutant cells (2, 3). The ability of Wnt ligands to modulate signaling in APC-mutant colorectal cancer is of particular interest, as both the tumor cells and the surrounding microenvironment express Wnt ligands throughout disease progression (4, 5).

The general model for Wnt-driven colorectal cancer is that an overactive Wnt pathway drives disease progression through all stages. However, several studies have suggested that Wnt pathway activity may actually be lower in the most invasive forms of colorectal cancer and/or in the advanced, metastatic, chemoresistant, and poor prognosis stages (6–8). One of these studies analyzed over 3,000 patient samples biopsied from the primary tumor site and associated clinical data to develop a categorization scheme of four subtypes. In their schema, tumors with the highest Wnt gene signature (labeled CMS2, or Canonical) exhibited the best overall clinical outcome, whereas the patients with the weakest Wnt signature (CMS4, or mesenchymal) exhibited invasive, poor prognosis disease. A separate effort to classify colorectal cancers analyzed 515 PDX tumors devoid of human stromal cells and identified five colorectal cancer/CRIS subtypes (9). Consistent with the CMS classification, subtypes with the strongest Wnt signaling (CRIS-C,-D,-E) had better prognoses than the subtypes with lower Wnt signaling activity (CRIS-A, -B). Using one or more of these classifications, multiple follow-up studies continue to reveal that the most aggressive subtypes of colorectal cancer are not the tumors with the highest levels of Wnt signaling. One study aligned PDX models and a large panel of colon cancer cell lines to the CMS system and revealed a clear difference in response to chemotherapy between CMS2 and CMS4 samples. Although higher levels of expression of the Wnt target gene c-MYC has been connected to chemoresistance (10), chemotherapy treatment of cell lines cultured in vitro and/or injected subcutaneously in NSG mouse models revealed that the Wnt-lower CMS4 samples were more resistant to treatment (11). Importantly, this study showed that a majority of the traditional colon cancer cell lines tested (43 lines) could be classified into the four different CMS groups, even though these cells have been maintained in an in vitro system for decades. These suggest that there is at least some level of intrinsic stability of gene program expression in cancer cell lines that preserves CMS status ex vivo and in the xenograft setting. Higher resolution studies that analyzed gene expression at various colorectal cancer stages or small budding cell clusters at the invasive edge of primary tumors also observed a correlation with lower Wnt pathway activity (8, 12).

Although these correlative studies suggest that colorectal cancers with moderate-to-low Wnt signaling are more aggressive, only a few studies have tested whether it is the weaker level of Wnt pathway activity that is linked to these phenotypic behavior differences in the primary tumor setting. One study has shown that repression of Wnt signaling through expression of dominant negative interfering LEF/TCF (dnTCF4) in colorectal cancer cell lines increased the ability of cells to seed tumors in multiple tissues when introduced into mice via tail vein injection (13, 14); another group deleted the TCF-4 gene (TCF7L2) in multiple colorectal cancer lines and observed a consistent change in morphology and invasive phenotypes (15). But these studies did not examine invasion phenotypes in the orthotopic setting of the colon where the primary tumor forms, nor were there any gene expression studies that might link these changes to colorectal cancer classification schemes and/or patient tumor data linked to outcomes.

Here we directly address these unknowns using two different approaches to genetically manipulate the level of Wnt/ß-catenin activity in colorectal cancer cells. We evaluate the tumor phenotype consequences of these manipulations via orthotopic xenografts in the mouse colon followed by histologic and RNA sequencing analysis of both the human and mouse cell populations in the tumors. We observe that interference with Wnt signaling leads to an increase in tumor invasion. Comparing the differentially expressed genes between the parental and downregulated lines, we directly connect decreases in Wnt signaling with increased expression of secreted signals for cellular invasion from the cancer cells, and gene expression changes in the tumor microenvironment that indicate reduced inflammation. Specifically, we observe changes in the localization of microenvironment components in Wnt-reduced tumors that further illustrate a diminished immune response. We demonstrate that the gene expression changes observed in our model system are enriched in CMS4 patients from The Cancer Genome Atlas. In addition, we identify signaling networks expressed in tumors that are significantly correlated with poor prognosis in multiple colorectal cancer patient datasets. We suggest that inhibition of Wnt signaling in Wnt-high colon tumors will lead to a dramatic increase in tumor invasion and aggression.

Cell lines

SW480, SW620, COLO320, and HCT116 were obtained from ATCC. Mycoplasma testing was conducted quarterly using the MycoAlert Mycoplasma Detection Kit (Lonza LT07-218). To authenticate cells, genomic DNA was submitted for short tandem repeat profiling and compared against the ATCC database annually. Cell lines were cultured in DMEM (Hyclone) or RPMI-1640 (Hyclone) supplemented with 10% FBS (Atlas Biologicals), 1% penicillin/streptomycin (Mediatech), and 2 mmol/L glutamine (Mediatech). Unless otherwise stated, all assays were conducted in the same media. dnLEF1 cell lines were prepared as described previously (16). LRP6 knockout (LRP6KO) cell lines were created by cotransfecting cell lines with 3 μg LRP6 CRISPR and 300 ng eGFP-Puro using BioT transfection reagent (Bioland Scientific, B01). After selection with puromycin, cells were sorted for LRP6 knockdown using flow cytometry (anti-human LRP6-APC, R&D Systems MAB1505, RRID:AB10889810) and successful knockdown was confirmed by Western blot analysis.

LRP5 and LRP6 CRISPR constructs

The CRISPR Kit used for constructing multiplex CRISPR/Cas9 vectors was a gift from Takashi Yamamoto (Addgene Kit #1000000054). Guide RNAs targeting LRP5 and LRP6 were designed using the Zhang Lab Optimized CRISPR Design Tool (http://crispr.mit.edu). A Cas9 nickase system was used such that two closely aligned guides per gene were developed.

For LRP5, the primer pairs used were:

  • LRP5-Cas9n-1-Fwd: CACCGCTCGGTCCAGTAGATGTAGC,

  • LRP5-Cas9n-1-Rev: AAACGCTACATCTACTGGACCGAGC,

  • LRP5-Cas9n-2-Fwd: CACCGCGGCAAGCCGAGGATCGTGC,

  • LRP6-Cas9n-2-Rev: AAACGCACGATCCTCGGCTTGCCGC.

For LRP6, the primer pairs used were:

  • LRP6-Cas9n-1-Fwd: CACCGTTGGCCAAATGGTTTAGCCT,

  • LRP6-Cas9n-1-Rev: AAACAGGCTAAACCATTTGGCCAAC,

  • LRP6-Cas9n-2-Fwd: CACCGAAGTGTTAACCAATACTACA,

  • LRP6-Cas9n-2-Rev: AAACTGTAGTATTGGTTAACACTTC.

Guides were inserted into vectors using T4 polynucleotide kinase (New England Biolabs), and the final combined guide RNA vector was assembled by Golden Gate Assembly using Quick Ligase (New England Biolabs). The final vector was electroporated into DH5α competent cells and purified using a Nucleospin Maxiprep Kit (Macherey Nagel). Restriction enzyme digested products were run out on an agarose gel to confirm insert size and plasmid quality. Correct assembly was verified by Sanger Sequencing.

LRP5 and LRP6 shRNA

LRP5 and LRP6 shRNA constructs were obtained from the RNAi Consortium database (Sigma-Aldrich). Lentivirus was packaged in HEK293T cells (ATCC) and purified using PEG-it (System Biosciences).

Western blotting

Whole cell lysates were prepared by resuspending a cell pellet in an appropriate amount of RIPA buffer containing protease inhibitors and phosphatase inhibitors and incubating for 30 minutes on ice. The lysates were spun down at high speed for 15 minutes at 4°C, and the supernatant was transferred to a clean tube. Lysates were quantified using a Bradford assay (Bio-Rad, 500-0205). Eighty micrograms of total cell lysates were analyzed by Western blot analysis using the following antibodies and concentrations: LRP6 (1:1,000, Cell Signaling Technology 3395, RRID:AB1950408), LRP5 (1:1,000, Cell Signaling Technology 5731, RRID:AB_10705602), ß-catenin (1:1,000, Cell Signaling Technology 8480, RRID:AB_11127855), ß-tubulin (1:2,000, Genetex GTX101279, RRID:AB_1952434). All blots were incubated overnight in primary antibody, washed, and then incubated for two hours in secondary antibody: anti-rabbit-HRP (1:5,000, GE Healthcare) or anti-mouse-HRP (1:2,000, GE Healthcare). Blots were imaged using a Syngene G-Box system. Bands were quantified using Adobe Photoshop (RRID:SCR_014199). Statistical evaluation of three or more independent biological replicates was performed using Student unpaired T test.

Luciferase assay

Cells were seeded at a density of 1 × 105 cells per well in a 12-well tissue culture plate using antibiotic-free media 24 hours prior to transfection. Cells were cotransfected with 100 ng Wnt ligand expression plasmid (17), 300 ng LRP5 or LRP6 CRISPR construct, 100 ng SuperTOPFlash or SuperFOPFlash (Gifts from Dr. R.T. Moon, Addgene plasmid 12456), and 100 ng CMV-ß-Galactosidase using BioT (Bioland Scientific B01–02) per manufacturer's instructions. Cells were harvested 24 hours post-transfection and assayed for luciferase activity and ß-galactosidase activity (as a normalization control). Statistical evaluation was performed using Student unpaired T test.

Scratch assay

Cells were seeded at a density of 2 × 106 cells per well in a 6-well plate. After 24 hours, two crosses were scratched into the cell monolayer with a P1000 tip. Each well was washed once with PBS before incubating in media. Images were taken at 0, 24, and 48 hours post-scratch. Measurements were obtained using Adobe Photoshop image analysis. Statistical evaluation was performed using Student unpaired T test.

Flow cytometry

One million cells per sample condition were collected and washed with FACS buffer (3% FBS in PBS). If needed, cells were fixed in 4% paraformaldehyde for 15 minutes at room temperature, permeabilized by 0.1% saponin in HBSS for 15 minutes on ice, and washed twice with FACS buffer prior to staining. Cells were incubated with primary antibody for 1 hour at room temperature in the dark. If secondary antibody was needed, cells were washed twice with FACS buffer prior to incubating in secondary antibody for 30 minutes at room temperature, in the dark. Cells were finally washed and resuspended in 300 to 600 μL FACS buffer per sample and analyzed on either a BD FACSAria or Acea Biosciences Novocyte. Data were analyzed using FlowJo (TreeStar, RRID:SCR_008520).

Clonogenic assay in fibrin

Briefly, 200 trypsinized cells were mixed with 100 μL of 2.5 mg/mL bovine fibrinogen (MP Biomedicals) in DMEM plus 10% FBS and 1% penicillin–streptomycin–glutamine and 1 μL of thrombin (Sigma). The fibrin gels were seeded in 96 well, flat bottom plates. After the gels solidified, 100 μL of DMEM media was layered on top. Wells were imaged regularly up to 14 days. Size measurements were taken using Adobe Photoshop. Data were analyzed using Prism (Graphpad, RRID:SCR_002798).

Orthotopic injections

Immuno-deficient NSG mice (The Jackson Laboratory) were anesthetized with 300 μL of 100 mg/kg ketamine/10 mg/kg xylazine. The surgical site was shaved and cleaned with ethanol. After an incision was made, the caecum was drawn out and washed with sterile PBS. Cells (5 × 105) were injected in the caecum wall at six to eight sites using a Hamilton syringe (Hamilton 80301), three to four injections per side. The peritoneal wall and skin were sutured separately, the former with resorbable sutures, the latter with nylon. Five mg/kg carprofen was injected as an analgesic. Mice health was monitored for 28 days prior to harvest. The University of California Irvine IACUC approved all animal experiments under protocol #AUP-17-053.

Mesentery dissociation for flow cytometry

Mesentery from orthotopically-injected mice were collected in a petri dish on ice and briefly diced with razor blades before transferring the tissue pieces into a 40 μm strainer and pushing tissue through the strainer using the plunger of a 3 mL syringe. The strainer was rinsed with 10 mL DMEM with 5% FBS, 1% P/S, and 2 mg/mL Collagenase I (Sigma). The mixture was collected in a conical tube and shaken at 37°C for 1 hour. The tube was spun down and the supernatant aspirated. The pellet was washed in 10 mL HBSS, spun down, and supernatant aspirated except for approximately 500 μL of HBSS. 50 μL DNAse I (10 U, Thermo Fisher Scientific) was added and the pellet resuspended and incubated at room temperature for 5 minutes. Following DNAse treatment, 2 mL of 0.05% trypsin was added, and the mixture was incubated at 37°C for 10 minutes. Five milliliters of DMEM with 5% FBS was then added, the mixture spun down, and the supernatant was aspirated. The pellet was resuspended in 5 mL DMEM and passed through a 40-μm cell strainer. The strainer was washed with an additional 5 mL of DMEM and the cell number and viability was measured using a Countess II (Thermo Fisher Scientific). A total of 1 × 106 cells per sample were incubated in CD298-APC (Miltenyi Biotec, RRID:AB_2657030) for 1 hour at room temperature in the dark. Samples were washed twice in FACS buffer before resuspending in FACS buffer and running on an Acea Biosciences Novocyte. Data were analyzed using FlowJo (TreeStar).

Subcutaneous injections

Immuno-deficient NSG mice (The Jackson Laboratory) were anaesthetized with 300 μL of 100 mg/kg ketamine/10 mg/kg xylazine. The surgical site was shaved and cleaned with ethanol. A total of 2 × 106 cells were resuspended in 100 μL PBS and injected under the skin on both flanks of the mouse. Mice health was monitored for 21 days prior to harvest.

IHC

Excised caecums were fixed in 10% formalin for 24 hours, cut on the latitude, and mounted on edge in paraffin. Ten μmol/L formalin‐fixed paraffin‐embedded (FFPE) sections were cut onto SuperFrost Plus (Thermo Fisher Scientific) slides. For Trichrome staining, the microwave staining protocol from the manufacturer was followed (Thermo Fisher Scientific 87019 – Masson Trichrome Stain Kit) with the following variations: maximum time for all deionized water rinses, 10-minute Biebrich Scarlet-Acid Fuchsin stain, 7-minute Aniline Blue stain followed by an addition deionized water rinse before acetic acid. Following alcohol dehydration and clearing with xylene, Trichrome stained slides were mounted with Permount mounting medium (Thermo Fisher Scientific, SP15–100). For antigen retrieval, slides were deparaffinized and rehydrated, followed by antigen retrieval in a pressure cooker using the optimized buffer for each antibody (specified below) for 5 minutes at pressure. For hematoxylin and eosin (H&E), slides were then stained by H&E, dehydrated, and mounted using Permount mounting medium. For antibody staining following antigen retrieval, slides were blocked in 3% H2O2, goat serum, and then avidin and biotin blocking reagents (Vector Labs, SP-2001). Sections were incubated in primary antibodies: cleaved caspase 9 (Thermo Fisher Scientific, MA5–32028, RRID:AB_2809322; 1:300), F4/80 (Abcam, ab111101, RRID:AB_10859466; 1:100 4°C overnight incubation; 10 mmol/L pH 6 sodium citrate retrieval buffer), Ki-67 (Thermo Fisher Scientific, MA5–14520, RRID:AB_10979488; 1:200), Ly6 g (Abcam, ab238132; 1:500 4°C overnight incubation; 10 mmol/L pH 8 tris EDTA retrieval buffer), and SMA1 (Abcam, ab5694, RRID:AB_2223021; 1:250 RT for 30 minutes; 10 mmol/L pH 6 sodium citrate retrieval buffer). A biotinylated secondary antibody was used (goat-anti-rabbit; 1:200; VectorLabs, BA-1000-1.5) and visualization using a peroxidase‐conjugated avidin‐based Vectastain protocol (ABC Elite; VectorLabs, PK-4001; DAB Quanto, Thermo Fisher Scientific, TA060QHDX). Slides were then counterstained with hematoxylin and mounted using Permount mounting medium. Whole slides scans of IHC antibody staining and Trichrome stained were performed with a Roche Ventana DP 200 slide scanner and then analyzed using QuPath (18). H&E slide images were captured with a Keyence BZ-X700 system and processed in Adobe Photoshop.

Pathology scoring

A blinded evaluation of stained slides was performed by pathologists who scored each section of caecum on the absence, low presence, or high presence of phenotypic features. Additional measurements were taken of each section by comparing the total pixels of a section to pixels of individual components (e.g., extracolonic tumors, intact epithelia, et al.) and normalizing by percentage of total section. To quantify positively stained cells in specific regions, equal-sized frames were captured from tumor, stroma, and epithelial of each orthotopic tumor sample. Positive cells were counted from each frame and frames averaged for each condition and region type. To quantify positively stained cells overall, we used QuPath to create 250 μm2 annotations over a region of interest, only considering annotations which entirely encompassed a tissue. Cell detection was done using the automated QuPath tool targeting the hematoxylin stain; positive cells marked using the object classifier tool. The same classifier settings were used for all slides stained with the same antibody. Approximately three to five sections per caecum from at least three mice per cell line were analyzed. The complete set of analyzed images is in Supplementary Fig. S1.

Total RNA isolation

RNA was extracted from cells using TRIzol (Invitrogen) and DirectZol RNA Extraction Kit following the manufacturer's instructions (Zymo Research). RNA was extracted from flash-frozen tissue samples by using a mortar and pestle to crush tissue into fine powder, homogenizing in TRIzol using a Precellys 24 (Bertin), and extracting the RNA with the DirectZol Kit. RNA quality was checked with Agilent Bioanalyzer. Samples with RNA integrity number (RIN) scores ≥9 were used for RNA-seq library construction.

qPCR

RNA was extracted from cells using TRIzol (Invitrogen) and DirectZol RNA Extraction Kit following the manufacturer's instructions (Zymo Research). RNA was extracted from flash-frozen tissue samples by using a mortar and pestle to crush tissue into fine powder, homogenizing in TRIzol using a Precellys 24 (Bertin), and extracting the RNA with the DirectZol Kit. cDNA was synthesized from 1 μg of total RNA using the High Capacity cDNA Reverse Transcription Kit (Invitrogen), as per the manufacturer's instructions. qPCR was performed in triplicate for each experimental condition using Maxima SYBR Green qPCR Master Mix (Invitrogen), according to the manufacturer's instructions. Primer pairs used are as follows:

  • AXIN: Forward: CTGGCTTTGGTGAACTGTTG Reverse: AGTTGCTCACAGCCAAGACA,

  • MYC: Forward: CTACCCTCTCAACGACAGCA Reverse: AGAGCAGAGAATCCGAGGAC,

  • SP5: Forward: AATGCTGCTGAACTGAATAGA Reverse: AACCGGTCCTAGCGAAAACC,

  • GAPDH: Forward: TCGACAGTCAGCCGCATCTTCTT Reverse: GCGCCCAATACGACCAAATCC.

Library construction

RNA-seq libraries were built using the Smart-seq2 protocol (19) with modifications according to ref. 20. Briefly, for each sample 10 ng of total RNA was converted to full-length cDNA using poly-dT primer and reverse transcriptase. The full-length cDNA was amplified using nine PCR cycles. Eighteen nanograms full-length cDNA for each sample was converted to sequencing library by tagmentation using the Illumina Nextera Kit. Eight PCR cycles were used for library amplification. The libraries were multiplexed and sequenced on an Illumina NextSeq500 sequencer as 43 bp paired-end reads.

Sequencing analysis

Adapter sequences and low-quality base pairs from the 5′ and 3′ ends of the paired-ends reads were trimmed using Trimmomatic v. 0.35 (21) using the following parameters: “PE read1.fastq read2.fastq pe_read1.fastq.gz se_read1.fastq.gz pe_read2.fastq.gz se_read2.fastq.gz ILLUMINACLIP:NexteraPE-PE.fa:2:30:8:4:true LEADING:20 TRAILING:20 SLIDINGWINDOW:4:17 MINLEN:30.″ For each sample, the reads from the host (mouse) and the graft (human) were classified and separated using Xenome (22). Xenome outputs five classes of reads: host, graft, ambiguous, both, and neither. Only the host and graft reads were kept for mapping to the host and graft transcriptomes respectively whereas the other three classes of reads (ambiguous, both, and neither) were discarded.

We used GENCODE v. 28 reference transcriptome for human and GENCODE v. M18 for mouse (23). RNA-seq reads for each sample were mapped to the reference transcriptome using Bowtie v. 1.2 (24) with the following parameters: “-X 2000 -a -m 200 -S –seedlen 25 -n 2 -v 3.” Gene expression levels and read counts were obtained using RSEM v. 1.2.31 (25).

Differential gene expression analysis

Differential gene expression analysis was performed using DESeq2 R package (26). Gene ontology was performed using ShinyGO (27), and visualized using the iGraph R package (28). Gene Set Enrichment Analysis (GSEA) was performed using the GSEA software from the Broad Institute (29).

Patient survival analyses

Differentially expressed genes of secreted factors were input into STRING (30) to identify known interactors. Interactors found to be differentially expressed in our datasets were considered for Kaplan–Meier analysis. Only genes which demonstrated a change in survival between high expression and low expression of the gene were considered for the combined gene signatures. Kaplan–Meier analyses of publicly available colon cancer datasets TCGA-COAD (https://www.cancer.gov/tcga), GSE17538 (31), GSE39582 (32), and GSE41258 (33) were performed using Prism (Graphpad). For combined gene signatures, each gene's expression was considered individually and bifurcated at the median. The resulting “high expression” and “low expression” patient sets were overlapped for all genes to define a group of patients with high or low expression for all considered genes.

Data and materials availability

The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus (34) and are accessible through GEO Series accession number GSE130236. Plasmids used for CRISPR editing will be available through Addgene.

Decreasing Wnt signaling in SW480 cells increases cell invasiveness

Our previous studies determined how decreases in Wnt signaling affect tumor progression in a subcutaneous xenograft tumor model (16). To decrease Wnt signaling, we lentivirally transduced an expression vector for dominant negative LEF1 (dnLEF1) into SW480 colon cancer cells prior to subcutaneous injection (Fig. 1A–D). LEF1 is a LEF/TCF transcription factor that binds to Wnt response elements (WRE) in enhancers and promoters and recruits ß-catenin to activate transcription of Wnt target genes. The dnLEF1 expression construct lacks coding sequences for the N-terminal ß-catenin binding domain, but retains all other sequences, including the HMG-box DNA binding domain and nuclear localization signal. This truncated transcription factor can localize to the nucleus and displace endogenous full-length LEF/TCFs from their occupancy of WREs at bona fide Wnt target genes. We used this mode of interference as it had been first developed by Van de Wetering and Clevers to effectively interfere with canonical Wnt signaling in the nucleus (35). Indeed, using a Wnt-luciferase assay, we find that dnLEF1 transduction-expression decreases Wnt signaling activity by 80% to 90% compared with Mock-transduced cells in SW480 (Fig. 1A). Expression of Wnt target genes AXIN2, MYC, and SP5 are decreased with dnLEF1 transduction, as measured by qPCR (Fig. 1B). Three weeks after subcutaneous injection of transduced cells, we harvested the tumors and examined them histologically. We observed that tumors expressing dnLEF1 exhibited a significant seven-fold decrease of in tumor weight and volume (Fig. 1C and D; ref. 16). These data are in line with the overall understanding that high levels of Wnt signaling promote tumor growth and development. However, the subcutaneous microenvironment on the back flank of a mouse is not the endogenous location for colon tumor development and further, the subcutaneous environment lacks the nutrient availability of the richly vascularized intestinal epithelium. We therefore expanded our animal studies to include an orthotopic injection model whereby tumors were developed in the mouse colon.

To study tumorigenesis in a more representative microenvironment, we injected the SW480 Mock or SW480 dnLEF1 cells into the stroma between the epithelial and muscle wall layers of the caecum (Fig. 1E and F). After 4 weeks of development, the caecums were harvested and sectioned for IHC. Unlike our findings with subcutaneous tumors, we observed no gross deficit in tumor size when comparing the SW480 Mock and SW480 dnLEF1 orthotopic tumors, implying that the high-density vasculature in the colon wall was able to compensate for the deficits in angiogenic development in the subcutaneous-dnLEF1 tumors (Fig. 1F and G). A blinded evaluation of physiological features for each tumor by pathologists revealed that SW480 dnLEF1 orthotopic tumors exhibit significantly higher rates of invasion into the submucosa and mucosa, higher prevalence of intratumor necrosis, and higher numbers of extracolonic tumors (tumors present on the exterior of the colon wall; Fig. 1H, full images with key in Supplementary Fig. S1). In vitro, the SW480 dnLEF1 cells were significantly more migratory after 24 and 48 hours, as measured by a scratch assay (Supplementary Fig. S6B). These data very clearly suggest that SW480 dnLEF1 tumors are more invasive and their cells more migratory.

The SW480 cell line was derived from the primary tumor of a patient with colorectal cancer, and a metastatic lymph node from the same patient was used to derive the line SW620. It is known that SW620 cells have significantly less Wnt activity than SW480 (e.g., SW620 is closer in Wnt activity levels to SW480 dnLEF1 than SW480, Supplementary Fig. S6A), and it is closely aligned with the invasive CMS4 subtype (11). We therefore predicted that SW620 tumors would have intrinsic, invasive phenotypes when developed orthotopically. In addition, we tested whether further reduction of Wnt signaling in this cell line would enhance invasion characteristics. Lentiviral transduction of the dnLEF1 construct into SW620 cells further reduced Wnt signaling levels (Fig. 1I), and as with SW480 dnLEF1 cells, SW620 dnLEF1 showed significant or near significant decreased expression of Wnt target genes AXIN2, MYC, and SP5 (Fig. 1J). Also, consistent with our predictions in the orthotopic setting, SW620 Mock tumors are inherently invasive, and SW620 dnLEF1 tumors appeared to be even more aggressive (Fig. 1K). Blinded evaluation of SW620 orthotopic tumors showed significant increases in incidence of extracolonic SW620 dnLEF1 tumors compared with SW620 Mock (Fig. 1L). In both SW620 Mock and SW620 dnLEF1 tumors, more sections contained necrotic areas compared with SW480 Mock or SW480 dnLEF1 suggesting an innate, higher sensitivity of the SW620 cells to nutrient levels in the surrounding microenvironment. Using CD298 as a marker of human cells, we used flow cytometry analysis to detect human cancer cells that had migrated to neighboring mesenteric tissue (36). This analysis showed that a significantly higher percentage of CD298+ cells were present in the mesentery of orthotopically-injected SW620 dnLEF1 mice compared with SW620 Mock-injected mice, indicating a higher level of intravasation into the surrounding lymphatic network (Fig. 1L) even though a decrease in local lymphatic invasion was observed. These data suggest that even in cell lines with low Wnt signaling activity, the effect of further decreasing Wnt activity creates a more aggressive tumor.

LRP6 is required for Wnt ligand:receptor signaling in colon cancer cells

Several groups have reported that Wnt ligands are capable of enhancing Wnt signaling activity, even in APC-mutant colon cancers (2, 3, 37–39). We hypothesized that autocrine-acting Wnt ligands from the cancer cells and paracrine Wnt signals from the surrounding microenvironment might contribute to maintaining colon cancer cells in a noninvasive state in the orthotopic condition. The dominant negative LEF1 constructs do not directly address this hypothesis, as their effect is to interfere with ß-catenin actions in the nucleus, not the cell membrane. Therefore, to more directly probe how Wnt ligands in the tumor microenvironment affect ß-catenin-dependent signaling and colon cancer cell behavior, we used a CRISPR/Cas9 system to genetically delete the essential Wnt ligand co-receptors LRP5 (Supplementary Fig. S2A) and LRP6 (Fig. 2A). We tested the efficacy of this manipulation by transiently co-transfecting HEK293 cells with the Cas9 nickase, LRP5 or LRP6 guide RNA expression constructs, and various Wnt ligand expression plasmids for ligands known to be expressed in colon cancer. Using the SuperTOPFlash luciferase reporter, we assayed for Wnt signaling activity. Knockdown of LRP5 or LRP6 expression effectively interfered with Wnt ligand activation of the luciferase reporter. For example, we noted that the Wnt1 ligand required both LRP5 and LRP6 for reporter activation (Supplementary Fig. S2B), in line with observations from other groups (40). We tested other Wnt ligands and found that whereas a number of Wnt ligands required both LRP5 and LRP6, others required solely LRP6, whereas none of the ligands required only LRP5 (Supplementary Fig. S2C). In part, this may be due to LRP6 influencing LRP5 expression, as the knockout of LRP6 also reduced LRP5 protein (Supplementary Figs. S3A and S3D). To confirm that this was not an off-target effect of our chosen guide RNAs, we also silenced LRP6 using an shRNA construct, which yielded the same result (Supplementary Figs. S3B–S3D). Given that the LRP6 co-receptor appears to be a uniformly required co-receptor for ß-catenin–dependent signaling, we focused our remaining efforts using LRP6 knockout cells.

LRP6 modulates Wnt signaling levels in APC-mutant colon cancer

We noted that SW480 cells do not innately express much LRP6 protein and that what residual LRP6 protein is detectable, is not located on the cell membrane where Frizzled receptors reside (Fig. 2B; Supplementary Figs. S3E–S3G). In addition, we found LRP6 expression to be only modestly decreased in dnLEF1-expressing SW480 and SW620 lines (Supplementary Figs. S3H and S3I), indicating that it is not a direct Wnt target. We therefore focused our LRP6 knockout efforts on SW620 cells because LRP6 is robustly expressed on the cell surface. Using flow cytometry, we isolated a population of LRP6 knockout cells from the total transfected SW620 cell population (Fig. 2B). We validated that LRP6 expression is strongly decreased in SW620 LRP6KO cells by Western blot analysis (Fig. 2C and D). Interestingly, Wnt signaling is decreased by 10% to 20% compared with expression of Cas9 alone (Fig. 2E). However, as expected, compared with the dominant negative LEF1-expressing lines, the effect of a genetic LRP6 knockout on overall Wnt signaling in SW620 cells is not as strong (Fig. 2E and F). These results demonstrate that decreasing expression of LRP6 decreases Wnt signaling only to a moderate level, pointing to real possibilities for ligand-based Wnt pathway activation, even in the presence of a downstream APC mutation. In addition, we found that ß-catenin protein levels significantly decreased when LRP6 was deleted, but not with dnLEF1 expression (Fig. 2G and H). This decrease cannot be explained by a decrease in ß-catenin (CTNNB1) mRNA expression (Fig. 2I), suggesting that ß-catenin protein degradation is enhanced. This is in line with findings from Saito-Diaz and colleagues (41), who suggest that LRP6 associates with the ß-catenin destruction complex in the absence of a Wnt ligand, partially inhibiting its activity. Thus, the loss of LRP6 allows the destruction complex to be more active, decreasing overall ß-catenin protein. We confirmed this by decreasing LRP6 expression in colon cancer cell line HCT116, which has a mutant ß-catenin and wildtype APC. LRP6 knockout did not significantly decrease Wnt signaling activity in these cells, suggesting that the release of the destruction complex from the plasma membrane did not increase ß-catenin degradation (Supplementary Fig. S4A). Interestingly, when we decreased LRP5 or LRP6 expression in colon cancer cell line COLO320, which has a highly truncated APC, we find that Wnt signaling is reduced by 40% to 50% (Supplementary Fig. S4B). However, ß-catenin expression is not significantly reduced in LRP6-diminished cells, due to the absence of APC domains that interact with LRP6 (Supplementary Fig. S4C–S4D). We conclude that removal of LRP6 in APC-mutant colon cancer cell lines destabilizes ß-catenin and moderately decreases Wnt signaling.

Loss of LRP6 enhances aggressive tumor phenotypes

In vitro, SW620 LRP6KO cells were more migratory than SW620 Cas9, but this enhanced phenotype was only significantly evident after 48 hours (Fig. 2J). In contrast, SW620 dnLEF1 cells were significantly more migratory after 24 hours. Orthotopic tumors of SW620 Cas9 and SW620 LRP6KO were largely indistinguishable from one another by pathology scoring, except for the size of extracolonic tumors in LRP6KO sections (Fig. 2K and L; Supplementary Fig. S1). Although the number of extracolonic tumors per section was similar between the two tumor types, the size of the LRP6KO extracolonic tumors were dramatically larger than the Cas9 tumors (Fig. 2L), suggesting that the SW620 LRP6KO cells have an enhanced ability to develop into tumors outside the colon. In addition, orthotopic tumors from COLO320 LRP6KO cells showed significantly increased necrosis and lymphatic invasion, in line with our observations with the SW480 and SW620 tumors, although interestingly this line did not show any extracolonic tumors (Supplementary Fig. S4E). These suggest that there may be a graded effect to Wnt signaling inhibition of cell migration, with decreasing Wnt signaling levels yielding a graded increase in invasion. Using IHC staining approaches, we probed orthotopic tumors for phosphorylated Histone H3 as a marker of mitosis, Ki67 as a marker of active cell cycling, and cleaved caspase 9 as a marker of apoptosis. We found no measurable difference in phospho-histone H3 and caspase 9 positivity between the two tumor types, but significantly higher levels of Ki67 positivity in SW620 LRP6KO orthotopic tumors compared with the SW620 Cas9 control tumors (Supplementary Fig. S3J). Interestingly, we observed high caspase 9 positivity in the stroma of the parental SW620 tumors but very little evidence of caspase 9 positivity in the human cancer cells of either tumor type. We conclude that although the larger SW620 LRP6KO extracolonic tumors could derive from earlier migration of SW620 LRP6KO cells, their larger size is due at least in part to a greater proportion of actively cycling cells.

Suppression of ß-catenin–dependent Wnt signaling in cancer cells leads to changes in signatures of invasion and inflammation

To identify changes in gene expression that occur in the conversion to an invasive phenotype, we performed RNA-seq of bulk mRNA from the orthotopic tumors, taking advantage of the mixed mouse and human cell environment to analyze concurrent gene expression changes in the human tumor versus the mouse stroma. We also used the percentage of reads mapped to human versus mouse genomes to serve as a proxy for assessing the relative abundance of human versus mouse tissue (Fig. 3A). We found that overall, the Wnt-low, SW620 tumor cells (both control and further Wnt-diminished), comprised a higher percentage of the tissue samples than the SW480 tumors, with PBS-injected caecums serving as a negative control. Differential gene expression analysis revealed approximately 350 significantly changed genes in SW480 dnLEF1 cells compared with SW480 Mock, but twice as many significant changes in the SW620 Mock versus SW620 dnLEF1. Not surprisingly, the more moderate change in Wnt signaling induced by knocking out LRP6 led to fewer and smaller, but still-significant changes in gene expression in SW620 (Fig. 3B). We compared the expression of classical markers of epithelial-to-mesenchymal transition (EMT) in the matching tumor types and observed a trend of decreasing epithelial gene expression and increasing mesenchymal gene expression in the Wnt-reduced tumors, although few of these changes reached significance (Fig. 3C). In addition, as poorly differentiated tumors are known to be more invasive than well-differentiated tumors, we evaluated the expression of stemness and differentiation associated genes and found significant upregulation of stemness genes FOXP1 and SMOC2 in SW480 dnLEF1 cells, and significant downregulation of differentiation genes CDX2 and SHH (Fig. 3D). We observed a similar trend in the SW620 tumor types; however, the gene expression changes did not reach significance (Fig. 3D). For an in vitro test, we used clonogenic assays to assess if emergent properties of stemness were more evident in SW480 dnLEF1 or SW620 dnLEF1 cultures. We did not observe any significant differences in number of colonies or size of colonies when comparing dnLEF1-expressing lines to Mock-expressing controls, suggesting that the observed expression changes in stemness associated genes are not sufficient to impart an overt, dominant change in stemness-relevant phenotypes (Supplementary Fig. S3K and S3L).

We additionally performed RNA-seq on bulk mRNA from in vitro colorectal cancer cultures to compare gene expression to the orthotopic samples. Differential gene expression analysis showed the greatest number of significant changes when comparing SW480 dnLEF1 and SW480 Mock; fewer changes were observed in SW620 dnLEF1 and the fewest number of genes were significantly changed in SW620 LRP6KO (Supplementary Figs. S5A and S6A). Many of the significantly upregulated genes in the orthotopic tumors were also upregulated in vitro (Supplementary Figs. S5B and S6E). Using GSEA (42), we found that Wnt target genes linked to colon cancer were significantly enriched in SW480 Mock compared with SW480 dnLEF1 (Supplementary Fig. S5C). Using Gene Ontology (GO) analyses, we observed that gene programs related to extracellular matrix, cell adhesion, and cell migration were significantly changed by Wnt interference in both the in vitro and tumor data, suggesting that there are innate changes that drive the increased cell migration phenotypes (Supplementary Fig. S5D). As with the orthotopic tumors, a decrease in differentiation marker expression and a trend towards EMT was most significant in SW480 dnLEF1 (Supplementary Fig. S6D). Using immunoblotting, we found upregulation of two markers of EMT, AXL and Vimentin, supporting our transcriptomic findings (Supplementary Fig. S6F and S6G). Given the lack of overt changes in stemness phenotypes in the SW480 dnLEF1 tumors and lack of significant changes in stemness, differentiation, and EMT markers in the SW620 cell lines (Supplementary Fig. S6C), we focused our downstream analysis on the SW480 dominant negative LEF1-expressing cells and cell communication programs directed toward the tumor microenvironment.

To compare SW480 dnLEF1 and SW480 Mock orthotopic tumors, we performed GO analyses on significantly upregulated and downregulated genes and clustered the GO terms by shared function and list members (Fig. 3E; in vitro Supplementary Fig. S6H). The resulting clusters revealed an upregulation of GO terms associated with extracellular and intracellular signaling, and a downregulation of terms associated with cell differentiation and morphogenesis. We were particularly interested in the role that extracellular signaling may play in advancing tumor invasion given the literature showing APC-mutant cancer cells respond to stimulation from Wnt ligands. Thus, we looked at the expression fold change of all known genes for secreted proteins (Fig. 3F). Many of the genes that are significantly upregulated in the SW480 dnLEF1 tumors encode secreted proteins with known functions in cell migration, and many of the most significantly downregulated genes encode inflammatory mediators, suggesting that the Wnt-reduced cancer cells were eliciting a reduced immune response relative to the parental SW480 Mock tumors.

Suppression of ß-catenin–dependent Wnt signaling in cancer cells leads to a less inflamed microenvironment

To determine if the murine tumor microenvironment reacted differently to the two tumor types, we compared the expression of genes in sham-injected, SW480 Mock-injected, and SW480 dnLEF1-injected mouse stroma. As a ground truth, we performed GO analysis on significantly upregulated and downregulated genes in stromal cells comparing SW480 Mock injection to sham injection. Unsurprisingly, the GO terms that emerge in response to a tumor is clustered on programs of immune response and extracellular matrix remodeling (Fig. 4A). We then looked for tumor microenvironment gene expression differences when a SW480 Mock tumor is growing in the intestine versus a SW480 dnLEF1 tumor. We found that there is a significant downregulation in immune response GO terms (Fig. 4B; against sham Supplementary Fig. S7B), indicating that the downregulated inflammatory signals from the human SW480 dnLEF1 cancer cells appear to be met with a decreased immune response from the stroma. The expression fold change of genes for secreted proteins in the stroma confirms this as significantly downregulated genes are almost entirely associated with inflammation (Supplementary Fig. S7A). To determine if this reduction of inflammation-associated gene expression was the result of a change in the proportions of different stromal populations present in the tumor microenvironment, we looked at the expression of marker genes for endothelial cells, cancer-associated fibroblasts, tumor-associated macrophages, and neutrophils. For all cell types, few significant changes in marker gene expression were observed between SW480 Mock stroma and SW480 dnLEF1 stroma (Fig. 4C; Supplementary Fig. S7C and S7D), suggesting that the cellular composition of the overall tumor microenvironment is largely unchanged. We also used IHC approaches to determine if patterns of localization and/or accumulation in around the tumor were different. F4/80 and Ly6g were used as markers of tumor-associated macrophages and neutrophils, respectively, and CD31 was used to mark endothelial cells (Fig. 4D). Quantitation of positive staining cells from tissue sections showed that there was no significant change in the localization of tumor-associated macrophages or neutrophils (Fig. 4E and F). However, the localization of vessels revealed a significant shift from localization within the SW480 Mock tumor to the surrounding stroma in SW480 dnLEF1 tumors (Supplementary Fig. S7E).

As changes in immune cell composition do not appear to account for the decrease in immune response in SW480 dnLEF1 tumors, we examined the expression of cytokines in the stroma and found that nearly all were significantly decreased in SW480 dnLEF1 stroma (Fig. 4G; Supplementary Fig. S7H). Cytokines were significantly increased in the SW620 dnLEF1 stroma (Supplementary Fig. S8I). However, those increases derive from an extremely low baseline of expression in the SW620 Mock stroma, highlighting how even in the parental tumor these signatures of inflammation are low. Likewise, NLRP3, a key component of the innate immune response inflammasome is significantly downregulated in SW480 dnLEF1 stroma, consistent with a diminished immune response relative to the parental line (Fig. 4H). Interestingly, multiple collagen genes in the stroma are less well expressed in SW480 dnLEF1 tumors compared with SW480 Mock tumors (Fig. 4I and J; Supplementary Fig. S7G), and they are differentially localized. Trichrome staining revealed a shift in collagen localization with strongly staining bundles of collagen deposition surrounding the tumor periphery of SW480 Mock tumors versus clear patterns of collagen integration into the SW480 dnLEF1 tumors (Supplementary Fig. S7F). This finding is highly reminiscent of human colorectal cancer tumors with the low-Wnt, poor prognosis CMS4 tumor types having the highest levels of stromal infiltration (6). Similar stromal gene expression changes are observed in the SW620 dnLEF1 cells (Supplementary Fig. S6I) and orthotopic tumors (Supplementary Fig. S8A–S8J), although the stromal immune response is increased in the SW620 dnLEF1 orthotopic tumors. We hypothesize that this stromal change represents a different stage in tumor progression compared with the SW480 cell lines; as SW620 cells are inherently migratory, the diminished Wnt signal in this context pushes to advance metastasis rather than initiate it. These observations indicate that decreasing Wnt signaling in colon cancer cells leads to increased cell migration and invasion in a dosage-dependent manner relative to the magnitude of Wnt signal decrease, through both autocrine and paracrine signaling. To initiate the process of metastasis, the Wnt-decreased tumors orchestrate a weaker, diminished immune response while activating processes of tumor angiogenesis and matrix remodeling (Fig. 5A).

Decreased Wnt signaling in patients with colorectal cancer is correlated with poor prognosis

Recent attempts to categorize patients with colorectal cancer by molecular biomarkers have consistently shown that patients with the highest levels of Wnt signaling have the best prognosis, and patients with the lowest levels have the poorest (6, 9). To determine if our observations with decreased Wnt signaling in orthotopic tumors translated to the patient experience, we classified TCGA-COAD samples for CMS subtypes using the classifier described in Guinney and colleagues and performed GSEA to compare CMS2 (Wnt-high; best patient outcomes) to CMS4 (Wnt-low; poorest outcomes) samples. We found that the NLRP3 inflammasome is significantly enriched in CMS2 patients compared with CMS4, supporting our finding that Wnt-low tumors exhibit decreased expression of genes associated with an immune response (Fig. 5B). In addition, we found the genes upregulated in SW480 dnLEF1 orthotopic tumors were enriched in CMS4 patients (Fig. 5C). Finally, we combined the significantly upregulated genes for secreted proteins expressed by both the tumor and stroma and found them to be collectively enriched in CMS4 patients (Fig. 5D), suggesting that the signaling microenvironment that we modeled is most similar to the patients with the poorest outcomes.

These associations suggest that the signaling changes we observed in the tumor and tumor microenvironment could be used as predictors of patient outcomes. To test this, we identified the genes encoding secreted proteins that were significantly downregulated in SW480 dnLEF1 stroma and created interaction networks of known protein–protein interactors from STRING (30) that were expressed in the stroma. This process led to the identification of two stromal gene networks—one consisting of cytokines CXCL1, CXCL2, CXCL3, and matrix metalloproteases MMP9 and MMP12, and one consisting of inflammation activators IL1A, IL1B, NLRP3, and the innate immune response gene LCN2. We used publicly available datasets of patients with colon cancer from GSE17538, GSE39582, GSE41258, and TCGA-COAD to identify patients with high or low expression of each gene network and found that high expression of the cytokine-metalloprotease network significantly predicted better patient outcomes, whereas the inflammation network did not have significant predictive power (Fig. 5E).

We repeated the network analysis to identify signaling networks of genes for secreted proteins significantly upregulated in SW480 dnLEF1. From this process we identified four tumor gene networks: (i) AXL-GAS6, a well-characterized ligand–tyrosine kinase receptor interaction that activates cell migration (43, 44); (ii) CTHRC1, a migration promoting gene previously linked to Wnt signaling activation (45, 46) and two glycoproteins, TPBG and GPC4; (iii) FN1, which has been implicated in promoting cell migration (47) with matrix remodeling glycoprotein TNC; and (iv) a wound healing network consisting of plasminogen activator PLAT, coagulation factor F10, and EMT-promoting growth factor MDK (48). The CTHRC1 and FN1 networks did not significantly predict patient outcomes but high levels of expression of the AXL-GAS6 and the wound healing network (PLAT, F10, MDK) were strong predictors of poor patient outcomes (Fig. 5F). Thus, we conclude the cell signaling interactions found to be significantly changed in Wnt-decreased cell lines and tumors are predictive of poor patient outcomes for human colorectal cancer. These identified gene networks may point to therapeutic targets that may act synergistically with Wnt-targeting therapies, which have had little success in patients with colon cancer.

Here we report that interfering with ß-catenin–dependent Wnt signaling in colon cancer cells increases their cell migration and invasion activities both in vitro and in vivo. We demonstrate this in multiple ways by interfering with Wnt signal transduction at steps that lie either upstream or downstream of ß-catenin. Wnt-signaling-inhibited cell lines show increased cell migration in vitro, a phenotype that correlates well with in vivo phenotypes of increased invasiveness, including the formation of extracolonic tumors. We predict that increased invasiveness would have been eventually manifest as metastases were we able to carry out the experiments longer. However, the genetically manipulated cells are so invasive and aggressive at the primary site of injection in the caecum that the intestinal lumen was blocked and mice became moribund, therefore limiting the timeline of the study. We compared significantly upregulated genes in each of our lines to identify genes that are commonly upregulated when Wnt signaling is decreased, and discovered that a set of highly upregulated genes predicts poor outcomes in human patients with colorectal cancer. We suggest that therapies targeting one or more of these genes may find clinical application in patients with colon cancer presenting at the early stages of advanced disease.

Mutations that activate the Wnt signaling pathway have long been characterized as a hallmark of colon cancer, and multiple studies have pointed to the contribution of overactive Wnt signaling to metastatic disease. In patient samples, the leading edge of colon tumors stain strongly for nuclear ß-catenin, a marker of active Wnt signaling (49). In addition, SNAI1 and TWIST1, two genes that play key roles in EMT are direct Wnt target genes (50). Many previous studies have shown that decreasing Wnt signaling in colon cancer by targeting varying components of the pathway, can significantly reduce tumor burden in subcutaneous xenograft mouse models, the most common mouse model for preclinical testing of cancer therapeutics (34, 51–54). Indeed, studies by our group using the dnLEF1 construct have shown the same result in subcutaneous xenografts (Fig. 1A–C) (16, 55). Because of this, many efforts and deep resources have been brought to bear on the goal of bringing small molecule inhibitors of Wnt signaling into clinical practice. However, so far, none of these compounds have been able to show measurable benefit in colon cancer treatment, and the trials of Wnt inhibitors have shifted their focus to other cancer types (56). Recent evidence has shown that the genetic landscape of colon cancer is significantly more complex than previously appreciated; a meta-analysis of patient samples found that tumors with the lowest levels of Wnt signaling had the highest levels of stromal infiltration and the worst overall patient survival (6). In contrast, tumors with the highest levels of Wnt signaling had the best overall patient survival. The data that we present here support this notion, and suggest that decreasing Wnt signaling directly favors an invasive phenotype in colon cancer.

One potential caveat is that both LEF1 and family member TCF1 (encoded by TCF7), can cooperate with ATF2 and ATF4 transcription factors to activate target genes in a β-catenin–independent manner (57). We developed a lentiviral transduction system to produce moderate-to-low levels of dnLEF1 so as to avoid issues with overexpression (and only partially interfere with Wnt:β-catenin), but we cannot formally rule out that some of the gene expression changes are due to a Wnt-independent activity of dnLEF1. To test whether this possibility was a dominant feature of our gene sets, we examined the expression of ATF2 target genes (58) and performed GSEA analysis on published ATF2 target gene sets (59). We did not find any association between ATF2 and dnLEF1 regulation (Supplementary Fig. S9A and S9B). We conclude that overall, dnLEF1 is primarily and directly enforcing changes on the Wnt target gene program.

Using gene ontology, we observe that Wnt interference in colon cancer cells by dnLEF1 expression is associated with enrichment for genes associated with epithelial to mesenchymal transition, loss of cell adhesion, and invasion. We find that a high-level of expression of these genes is significantly correlated with poorer patient outcomes in multiple patient data sets (Fig. 5). The genes identified have been characterized to promote cell migration, either through changing gene expression or creating a more permissive extracellular matrix. Interestingly, RNA-seq analysis of dnLEF1 expression in SW480 and SW620 cells cultured in vitro also showed changes in gene programs for cell adhesion, mobility, cell junction, and extracellular matrix (Supplementary Fig. S8). However, the specific genes connected to these programs were different, showing that although the transcription programs and cell behaviors that link low Wnt signaling to invasive phenotypes is the same, the specific genes that change expression are different in vitro versus the orthotopic setting. Collectively, these programs highlight attempts by cancer cells to alter their microenvironment to promote invasion, as fibrotic, stiffened extracellular matrix has been shown to enhance EMT and invasion in many cancers (60).

Our findings suggest that high levels of Wnt signaling may indirectly repress invasive programs both in the tumors cells themselves and indirectly in the tumor microenvironment, fulfilling a specific role for strong Wnt signaling as an inhibitor of localized invasion. It is therefore possible that lower levels of Wnt signaling in APC-mutant colorectal cancer is linked to a colon cancer invasion-metastasis cascade and development of an invasion-promoting tumor microenvironment. Our results are concurrent with the CMS/CRIS colon cancer characterization studies and highlight how a decrease in Wnt signaling is an important contributing factor to advanced colon cancer and poor patient outcomes.

The ramifications of our findings are that while targeting Wnt signaling in colon cancer may reduce tumor burden, a potential, inadvertent side effect might be to induce surviving cancer cells to become invasive. Thus, as clinical trials continue to test the efficacy of newer, more specific Wnt inhibitors on Wnt-driven cancers, we suggest that treatment of patients with Wnt inhibitors should include concurrent treatment with drugs targeting one or more of the genes identified in this study, such as the GAS6:AXL pathway. In fact this pathway was recently reported to be increased in the invasive stage IV of colorectal cancer (61). Autocrine Gas6:Axl signaling in the cancer cells can promote migration (61) and indeed there is three-fold more Axl expression in the invasive SW620 tumors compared with the SW480 tumors. Increased paracrine Gas6:Axl signaling is also likely to be highly significant, and for both SW480 and SW620 tumors, Gas6 is very highly expressed in the stroma. Several groups have shown the Gas6:Axl signal to be immune suppressive through a reduction in inflammation-promoting cytokines—which we also observe in our invasive tumors (62–64). Small molecule inhibitors have already been developed to target AXL and these have advanced to clinical trials. However, although AXL has been observed to be significantly overexpressed in malignant cells, none of the clinical trials currently testing AXL inhibitors are focused on colon cancer (43). We suggest that AXL inhibitors may be therapeutically useful in concert with other drug therapies, particularly Wnt or VEGF. Small molecule approaches targeting extracellular matrix proteins and related protein– and cell–ECM interactions have been found to have mixed clinical outcomes (60)—given our finding that such genes are upregulated in Wnt-low colon cancer, these small molecules may be prime candidates for combination with Wnt inhibitors.

Progression to metastatic disease remains the most challenging aspect of colon cancer treatment, and the one in which patient outcomes remain poor overall. The finding that these advanced tumors harbor relatively lower levels of Wnt signaling compared with other colon cancer subtypes, points to a previously uncharacterized role of Wnt signaling. Our study demonstrates that interference of Wnt/ß-catenin signaling activities induces a more invasive phenotype, and identifies a number of contributors to this invasive state. We suggest that future studies of Wnt inhibitors for clinical use consider concurrent treatment with inhibitors of these identified targets, to both reduce tumor burden and prevent cancer invasion.

G.T. Chen reports grants from NIH/NCI during the conduct of the study. A.N. Habowski reports grants from National Science Foundation (NSF GRFP: DGE‐1321846) and NIH/NCI (NIH/NCI T32: T32CA009054) during the conduct of the study. No disclosures were reported by the other authors.

G.T. Chen: Conceptualization, formal analysis, funding acquisition, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. D.F. Tifrea: Resources, formal analysis, investigation, methodology. R. Murad: Resources, investigation, methodology. A.N. Habowski: Formal analysis, investigation, methodology, writing–review and editing. Y. Lyou: Formal analysis, investigation, methodology. M.R. Duong: Investigation, methodology. L. Hosohama: Investigation. A. Mortazavi: Resources, project administration. R.A. Edwards: Resources, formal analysis, supervision, project administration. M.L. Waterman: Conceptualization, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

The authors would like to thank the members of the Waterman and Donovan labs for their scientific discussion and feedback, Melanie Oakes and Jenny Wu for bioinformatics assistance. We would also like to thank Emiliana Borrelli, Christopher Hughes, Arthur Lander, John Lowengrub, Harry Mangalam, Michael McClelland, Eric Perlman, Eric Stanbridge, and Armando Villalta for providing advice and critiques. The work of G.T. Chen, Y. Lyou, A.N. Habowski, M.R. Duong, L. Hosohama, and M.L. Waterman was supported by NIH Grants CA096878, CA108697, a California CRCC award CRR-17-429379, R03CA223929, a U54CA217378 grant to the UCI Cancer Systems Biology Center (CaSB@UCI), and a P30CA062203 Cancer Center Support Grant to the Chao Family Comprehensive Cancer Center. G.T. Chen was supported by NIH Grant CA200298. L. Hosohama was supported by NSF GRFP (DGE-1839285). The work of D.F. Tifrea and R.A. Edwards were supported by P30CA062203, U54CA217378, and CRR-17-429379. This work was made possible in part, through access and support of the Genomics and High Throughput Facility and Flow Cytometry Core by the Cancer Center Support Grant (P30CA62203) and NIH shared instrumentation grants 1S10RR025496-01 and 1S10OD010794-01.

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.

1.
Davis
LE
.
The evolution of biomarkers to guide the treatment of metastatic colorectal cancer
.
Am J Manag Care
2018
;
24
:
107
17
.
2.
Voloshanenko
O
,
Erdmann
G
,
Dubash
TD
,
Augustin
I
,
Metzig
M
,
Moffa
G
, et al
.
Wnt secretion is required to maintain high levels of Wnt activity in colon cancer cells
.
Nat Commun
2013
;
4
:
2610
.
3.
Seshagiri
S
,
Stawiski
EW
,
Durinck
S
,
Modrusan
Z
,
Storm
EE
,
Conboy
CB
, et al
.
Recurrent R-spondin fusions in colon cancer
.
Nature
2012
;
488
:
660
4
.
4.
Aizawa
T
,
Karasawa
H
,
Funayama
R
,
Shirota
M
,
Suzuki
T
,
Maeda
S
, et al
.
Cancer-associated fibroblasts secrete Wnt2 to promote cancer progression in colorectal cancer
.
Cancer Med
2019
;
8
:
6370
82
.
5.
Kramer
N
,
Schmöllerl
J
,
Unger
C
,
Nivarthi
H
,
Rudisch
A
,
Unterleuthner
D
, et al
.
Autocrine WNT2 signaling in fibroblasts promotes colorectal cancer progression
.
Oncogene
2017
;
36
:
5460
72
.
6.
Guinney
J
,
Dienstmann
R
,
Wang
X
,
de Reyniès
A
,
Schlicker
A
,
Soneson
C
, et al
.
The consensus molecular subtypes of colorectal cancer
.
Nat Med
2015
;
21
:
1350
6
.
7.
De Sousa E Melo
F
,
Wang
X
,
Jansen
M
,
Fessler
E
,
Trinh
A
,
de Rooij
LP
, et al
.
Poor-prognosis colon cancer is defined by a molecularly distinct subtype and develops from serrated precursor lesions
.
Nat Med
2013
;
19
:
614
8
.
8.
de Sousa E Melo
F
,
Colak
S
,
Buikhuisen
J
,
Koster
J
,
Cameron
K
,
de Jong
JH
, et al
.
Methylation of cancer-stem-cell-associated Wnt target genes predicts poor prognosis in colorectal cancer patients
.
Cell Stem Cell
2011
;
9
:
476
85
.
9.
Isella
C
,
Brundu
F
,
Bellomo
SE
,
Galimi
F
,
Zanella
E
,
Porporato
R
, et al
.
Selective analysis of cancer-cell intrinsic transcriptional traits defines novel clinically relevant subtypes of colorectal cancer
.
Nat Commun
2017
;
8
:
1
16
.
10.
Arango
D
,
Corner
GA
,
Wadler
S
,
Catalano
PJ
,
Augenlicht
LH
.
c-myc/p53 interaction determines sensitivity of human colon carcinoma cells to 5-Fluorouracil in vitro and in vivo
.
Cancer Res
2001
;
61
:
4910 LP–4915
.
11.
Linnekamp
JF
,
Van Hooff
SR
,
Prasetyanti
PR
,
Kandimalla
R
,
Buikhuisen
JY
,
Fessler
E
, et al
.
Consensus molecular subtypes of colorectal cancer are recapitulated in in vitro and in vivo models
.
Cell Death Differ
2018
;
25
:
616
33
.
12.
De Smedt
L
,
Palmans
S
,
Govaere
O
,
Boeckx
B
,
Smeets
D
.
Expression profiling of budding cells in colorectal cancer suggests an EMT-like phenotype and molecular subtype switching
.
Eur J Cancer
2016
;
61
:
S88
.
13.
Seth
C
,
Ruiz i Altaba
A
.
Metastases and colon cancer tumor growth display divergent responses to modulation of canonical WNT signaling
.
PLoS One
2016
;
11
:
e0150697
.
14.
Varnat
F
,
Siegl-Cachedenier
I
,
Malerba
M
,
Gervaz
P
,
Ruiz i Altaba
A
.
Loss of WNT-TCF addiction and enhancement of HH-GLI1 signalling define the metastatic transition of human colon carcinomas
.
EMBO Mol Med
2010
;
2
:
440
57
.
15.
Wenzel
J
,
Rose
K
,
Haghighi
EB
,
Lamprecht
C
,
Rauen
G
,
Freihen
V
, et al
.
Loss of the nuclear Wnt pathway effector TCF7L2 promotes migration and invasion of human colorectal cancer cells
.
Oncogene
2020
;
39
:
3893
909
.
16.
Pate
KT
,
Stringari
C
,
Sprowl-Tanio
S
,
Wang
K
,
TeSlaa
T
,
Hoverter
NP
, et al
.
Wnt signaling directs a metabolic program of glycolysis and angiogenesis in colon cancer
.
EMBO J
2014
;
33
:
1454
73
.
17.
Najdi
R
,
Proffitt
K
,
Sprowl
S
,
Kaur
S
,
Yu
J
,
Covey
TM
, et al
.
A uniform human Wnt expression library reveals a shared secretory pathway and unique signaling activities
.
Differentiation
2012
;
84
:
203
13
.
18.
Bankhead
P
,
Loughrey
MB
,
Fernández
JA
,
Dombrowski
Y
,
McArt
DG
,
Dunne
PD
, et al
.
QuPath: open source software for digital pathology image analysis
.
Sci Rep
2017
;
7
:
16878
.
19.
Picelli
S
,
Faridani
OR
,
Björklund
ÅK
,
Winberg
G
,
Sagasser
S
,
Sandberg
R
.
Full-length RNA-seq from single cells using smart-seq2
.
Nat Protoc
2014
;
9
:
171
81
.
20.
Serra
L
,
Chang
D
,
Macchietto
M
,
Williams
K
,
Murad
R
,
Lu
D
, et al
.
Adapting the smart-seq2 protocol for robust single worm RNA-seq
.
Bio Protoc
2018
;
8
:
e2729
.
21.
Bolger
AM
,
Lohse
M
,
Usadel
B
.
Trimmomatic: a flexible trimmer for Illumina sequence data
.
Bioinformatics
2014
;
30
:
2114
20
.
22.
Conway
T
,
Wazny
J
,
Bromage
A
,
Tymms
M
,
Sooraj
D
,
Williams
ED
, et al
.
Xenome-a tool for classifying reads from xenograft samples
.
Bioinformatics
2012
;
28
:
i172
8
.
23.
Frankish
A
,
Diekhans
M
,
Ferreira
A-M
,
Johnson
R
,
Jungreis
I
,
Loveland
J
, et al
.
GENCODE reference annotation for the human and mouse genomes
.
Nucleic Acids Res
2018
;
47
:
766
73
.
24.
Langmead
B
,
Trapnell
C
,
Pop
M
,
Salzberg
SL
.
Ultrafast and memory-efficient alignment of short DNA sequences to the human genome
.
Genome Biol
2009
;
10
:
R25
.
25.
Li
B
,
Dewey
CN
.
RSEM: Accurate transcript quantification from RNA-Seq data with or without a reference genome
.
BMC Bioinformatics
2011
;
12
:
323
.
26.
Anders
S
,
Huber
W
.
Differential expression analysis for sequence count data
.
Genome Biol
2010
;
11
:
R106
.
27.
Ge
SX
,
Jung
D
,
Yao
R
.
ShinyGO: a graphical gene-set enrichment tool for animals and plants
.
Bioinformatics
2020
;
36
:
2628
9
.
28.
Csardi
G
,
Nepusz
T
.
The igraph software package for complex network research
.
InterJournal, Complex Syst
2006
;
1695
:
1
9
.
29.
Subramanian
A
,
Tamayo
P
,
Mootha
VK
,
Mukherjee
S
,
Ebert
BL
,
Gillette
MA
, et al
.
Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles
.
Proc Natl Acad Sci U S A
2005
;
102
:
15545
50
.
30.
Szklarczyk
D
,
Gable
AL
,
Lyon
D
,
Junge
A
,
Wyder
S
,
Huerta-Cepas
J
, et al
.
STRING v11: protein-protein association networks with increased coverage, supporting functional discovery in genome-wide experimental datasets
.
Nucleic Acids Res
2019
;
47
:
D607
13
.
31.
Smith
JJ
,
Deane
NG
,
Wu
F
,
Merchant
NB
,
Zhang
B
,
Jiang
A
, et al
.
Experimentally derived metastasis gene expression profile predicts recurrence and death in patients with colon cancer
.
Gastroenterology
2010
;
138
:
958
68
.
32.
Marisa
L
,
de Reyniès
A
,
Duval
A
,
Selves
J
,
Gaub
MP
,
Vescovo
L
, et al
.
Gene expression classification of colon cancer into molecular subtypes: characterization, validation, and prognostic value
.
PLoS Med
2013
;
10
:
e1001453
.
33.
Sheffer
M
,
Bacolod
MD
,
Zuk
O
,
Giardina
SF
,
Pincas
H
,
Barany
F
, et al
.
Association of survival and disease progression with chromosomal instability: a genomic exploration of colorectal cancer
.
Proc Natl Acad Sci U S A
2009
;
106
:
7131
6
.
34.
Edgar
R
,
Domrachev
M
,
Lash
AE
.
Gene Expression Omnibus: NCBI gene expression and hybridization array data repository
.
Nucleic Acids Res
2002
;
30
:
207
10
.
35.
Van de Wetering
M
,
Sancho
E
,
Verweij
C
,
De Lau
W
,
Oving
I
,
Hurlstone
A
, et al
.
The β-catenin/TCF-4 complex imposes a crypt progenitor phenotype on colorectal cancer cells
.
Cell
2002
;
111
:
241
50
.
36.
Lawson
DA
,
Bhakta
NR
,
Kessenbrock
K
,
Prummel
KD
,
Yu
Y
,
Takai
K
, et al
.
Single-cell analysis reveals a stem-cell program in human metastatic breast cancer cells
.
Nature
2015
;
526
:
131
5
.
37.
Giannakis
M
,
Hodis
E
,
Jasmine Mu
X
,
Yamauchi
M
,
Rosenbluh
J
,
Cibulskis
K
, et al
.
RNF43 is frequently mutated in colorectal and endometrial cancers
.
Nat Genet
2014
;
46
:
1264
6
.
38.
Jung
Y
,
Jun
S
,
Lee
SH
,
Sharma
A
,
Park
J
.
Wnt2 complements Wnt/β-catenin signaling in colorectal cancer
.
Oncotarget
2015
;
6
:
37257
68
.
39.
Nishioka
M
,
Ueno
K
,
Hazama
S
,
Okada
T
,
Sakai
K
,
Suehiro
Y
, et al
.
Possible involvement of Wnt11 in colorectal cancer progression
.
Mol Carcinog
2011
;
52
:
1
11
.
40.
Goel
S
,
Chin
EN
,
Fakhraldeen
SA
,
Berry
SM
,
Beebe
DJ
,
Alexander
CM
.
Both LRP5 and LRP6 receptors are required to respond to physiological Wnt ligands in mammary epithelial cells and fibroblasts
.
J Biol Chem
2012
;
287
:
16454
66
.
41.
Saito-diaz
K
,
Benchabane
H
,
Tiwari
A
,
Tian
A
,
Li
B
,
Joshua
J
, et al
.
APC inhibits ligand-independent Wnt signaling by the clathrin endocytic pathway
.
Dev Cell
2018
;
44
:
566
81
.
42.
Subramanian
A
,
Tamayo
P
,
Mootha
V
.
GSEA: GseaGsea: Ak-bafig-we profiles
.
PNSA
2014
;
43.
Axelrod
H
,
Pienta
KJ
.
Axl as a mediator of cellular growth and survival
.
Oncotarget
2014
;
5
:
8818
52
.
44.
Goyette
MA
,
Duhamel
S
,
Aubert
L
,
Pelletier
A
,
Savage
P
,
Thibault
MP
, et al
.
The receptor tyrosine kinase AXL is required at multiple steps of the metastatic cascade during HER2-positive breast cancer progression
.
Cell Rep
2018
;
23
:
1476
90
.
45.
Yang
XM
,
You
HY
,
Li
Q
,
Ma
H
,
Wang
YH
,
Zhang
YL
, et al
.
CTHRC1 promotes human colorectal cancer cell proliferation and invasiveness by activating Wnt/PCP signaling
.
Int J Clin Exp Pathol
2015
;
8
:
12793
801
.
46.
Kim
HC
,
Kim
YS
,
Oh
HW
,
Kim
K
,
Oh
SS
,
Kim
JT
, et al
.
Collagen triple helix repeat containing 1 (CTHRC1) acts via ERK-dependent induction of MMP9 to promote invasion of colorectal cancer cells
.
Oncotarget
2014
;
5
:
519
29
.
47.
Pupa
SM
,
Ménard
S
,
Forti
S
,
Tagliabue
E
.
New insights into the role of extracellular matrix during tumor onset and progression
.
J Cell Physiol
2002
;
192
:
259
67
.
48.
Filippou
PS
,
Karagiannis
GS
,
Constantinidou
A
.
Midkine (MDK) growth factor: a key player in cancer progression and a promising therapeutic target
.
Oncogene
2020
;
39
:
2040
54
.
49.
Jung
A
,
Schrauder
M
,
Oswald
U
,
Knoll
C
,
Sellberg
P
,
Palmqvist
R
, et al
.
The invasion front of human colorectal adenocarcinomas shows co-localization of nuclear β-catenin, cyclin D1, and p16INK4A and is a region of low proliferation
.
Am J Pathol
2001
;
159
:
1613
7
.
50.
ten Berge
D
,
Koole
W
,
Fuerer
C
,
Fish
M
,
Eroglu
E
,
Nusse
R
.
Wnt signaling mediates self-organization and axis formation in embryoid bodies
.
Cell Stem Cell
2008
;
3
:
508
18
.
51.
Morin
PJ
,
Vogelstein
B
,
Kinzlertt
KW
.
Apoptosis and APC in colorectal tumorigenesis
.
Med Sci
1996
;
93
:
7950
4
.
52.
Satoh
S
,
Daigo
Y
,
Furukawa
Y
,
Kato
T
,
Miwa
N
,
Nishiwaki
T
, et al
.
AXIN1 mutations in hepatocellular carcinomas, and growth suppression
.
Nat Genet
2000
;
24
:
245
50
.
53.
Tetsu
O
,
McCormick
F
.
Beta-catenin regulates expression of cyclin D1 in colon carcinoma cells
.
Nature
1999
;
398
:
422
6
.
54.
Polakis
P
.
Drugging Wnt signalling in cancer
.
EMBO J
2012
;
31
:
2737
46
.
55.
Lee
M
,
Chen
GT
,
Puttock
E
,
Wang
K
,
Edwards
RA
,
Waterman
ML
, et al
.
Mathematical modeling links Wnt signaling to emergent patterns of metabolism in colon cancer
.
Mol Syst Biol
2017
;
13
:
912
.
56.
Kahn
M
.
Can we safely target the WNT pathway?
Nat Rev Drug Discov
2014
;
13
:
513
32
.
57.
Grumolato
L
,
Liu
G
,
Haremaki
T
,
Mungamuri
SK
,
Mong
P
,
Akiri
G
, et al
.
β-catenin-independent activation of TCF1/LEF1 in human hematopoietic tumor cells through interaction with ATF2 transcription factors
.
PLOS Genet
2013
;
9
:
e1003603
.
58.
Watson
G
,
Ronai
ZA
,
Lau
E
.
ATF2, a paradigm of the multifaceted regulation of transcription factors in biology and disease
.
Pharmacol Res
2017
;
119
:
347
57
.
59.
Bailey
J
,
Tyson-Capper
AJ
,
Gilmore
K
,
Robson
SC
,
Europe-Finner
GN
.
Identification of human myometrial target genes of the cAMP pathway: the role of cAMP-response element binding (CREB) and modulator (CREMalpha and CREMtau2alpha) proteins
.
J Mol Endocrinol
2005
;
34
:
1
17
.
60.
Kai
FB
,
Drain
AP
,
Weaver
VM
.
The extracellular matrix modulates the metastatic journey
.
Dev Cell
2019
;
49
:
332
46
.
61.
Uribe
DJ
,
Mandell
EK
,
Watson
A
,
Martinez
JD
,
Leighton
JA
,
Ghosh
S
, et al
.
The receptor tyrosine kinase AXL promotes migration and invasion in colorectal cancer
.
PLoS One
2017
;
12
:
1
16
.
62.
Dunne
PD
,
McArt
DG
,
Blayney
JK
,
Kalimutho
M
,
Greer
S
,
Wang
T
, et al
.
AXL is a key regulator of inherent and chemotherapy-induced invasion and predicts a poor clinical outcome in early-stage colon cancer
.
Clin Cancer Res
2014
;
20
:
164 LP –175
.
63.
Rothlin
CV
,
Ghosh
S
,
Zuniga
EI
,
Oldstone
MBA
,
Lemke
G
.
TAM receptors are pleiotropic inhibitors of the innate immune response
.
Cell
2007
;
131
:
1124
36
.
64.
Tanaka
M
,
Siemann
DW
.
Gas6/Axl signaling pathway in the tumor immune microenvironment
.
Cancers
2020
;
12
:
1850
.

Supplementary data