Chromatin-Associated Functions of the APC Tumor Suppressor Protein

Chromatin-associated functions of the APC tumor suppressor protein

DISSERTATION

Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University

William C. Hankey

Biomedical Sciences Graduate Program

The Ohio State University

2016

Dissertation Committee:

Professor Joanna Groden, Advisor

Professor Albert de la Chapelle

Professor Kay Huebner

Professor Mark Parthun

Professor Jeffrey Parvin

William C. Hankey

2016

Abstract

Biallelic mutation of the APC tumor suppressor gene occurs in a high percentage of colorectal tumors and is considered the critical event driving tumor initiation in the large intestine. The APC protein performs multiple functions, including negative regulation of the canonical WNT signaling pathway by both cytoplasmic and nuclear mechanisms. As a result, APC suppresses proliferation, cell cycle progression, and genomic instability, while facilitating differentiation, normal directional migration, and apoptosis. The contribution of APC to these phenotypes is not mediated exclusively through its effects on canonical WNT signaling, but also through WNT-independent functions of the APC protein. Intriguing reports that APC interacts with chromatin to repress key WNT- activated targets prompted this study’s initial hypothesis that the chromatin-associated fraction of APC regulates gene transcription through multiple mechanisms, both WNT- dependent and WNT-independent.

Chromatin immunoprecipitation and next-generation sequencing identified more than

6,000 genomic peaks associated with the APC protein. Target selection was performed by comparison to whole transcriptome sequencing data from APC-deficient and APC- wild-type colon cancer cell lines and mouse tumors. 175 transcripts whose expression changes upon APC loss are linked to genomic regions physically associated with the ii

APC protein. Motif analysis of APC-associated genomic peaks found that binding sites for the TCF7L2 and AP-1 transcription factors are overrepresented and occur in many of the same peaks. Luciferase reporter assays indicate that APC antagonizes canonical

WNT signaling not only at WNT-activated genes such as PHLDB2 (Pleckstrin Homology

Like Domain Family B Member 2), but also at WNT-repressed genes such as MALL

(Mal, T-Cell Differentiation Protein-Like). A transcriptional element within the first intron of the MALL gene mediates transcriptional repression by the canonical WNT signaling pathway and transcriptional activation by AP-1. These findings demonstrate that canonical WNT signaling and the AP-1 transcription factor collaborate to fine-tune the expression of a large number of shared target genes in the colorectal epithelium.

Future therapeutic strategies for APC-deficient colorectal cancers might be expanded to include emerging agents targeting the AP-1 signaling pathway. New insights into mechanisms of transcriptional control by the canonical WNT signaling pathway indicate that known target genes should be grouped according to whether their transcription increases or decreases, as well as the presence or absence of co-regulation by AP-1.

Transcriptional changes in Apc-deficient colon tumors (from ApcMin/+ mice) and colon tumors with wild-type Apc and a degradation-resistant point mutant of β-catenin (from mice treated with azoxymethane and dextran sulfate sodium) largely overlap, with only

31-32% unique to each tumor type. Expression changes unique to Apc-deficient tumors are hypothesized to result from loss of β-catenin-independent functions of chromatin- associated APC. Further study will reveal their role in regulating apoptosis and formation of cell protrusions, as well as their relationship to colorectal cancer prognosis.

iii

Dedication

To Bill, Joyce, Nick, Robert, and Verónica

Acknowledgments

The success of this project was possible because of valuable intellectual and technical contributions from a number of sources both inside and outside of our lab. Individual lab members who contributed key ideas and struggled alongside me to troubleshoot these experiments were Dr. Michael McIlhatton, Dr. Kenechi Ebede, Dr. Alaina Martinez, Dr.

Jiang Qian, Dr. Erin Perchiniak, Max Bergman, Kevin Murnan, Max Fernandez, Jeremy

Keirsey, Dr. Patrick Grierson, Dr. Zeenia Kaul, Larissa Tangeman, Dr. April Gocha, Dr.

Michael Trimarchi, and Dr. Julia Harris. Our neighbor Dr. Samir Acharya was also a source of many good ideas.

We received valuable scientific help from members of neighboring labs, particularly Dr.

Xun Lan in the lab of Dr. Victor Jin, Dr. Sudarsharna Sharma, Dr. Madelyn Gerber in the lab of Dr. Amanda Toland, Linan Wang in the lab of Dr. Michael Freitas, and Drs. Hui- wen Liu and Mansi Arora in the lab of Dr. Jeff Parvin. We leaned heavily on the Parvin lab for materials, protocols, and expertise. Key collaborators outside the university were

Dr. Bruce Aronow, Dr. Anil Jegga, and Kendra Allton in the lab of Dr. Michelle Barton.

Our ChIP-seq experiments proved to be extremely challenging and would not have been possible without numerous contributions from Dr. Zhong Chen in the lab of Dr. Qianben

Wang. Dr. Chen was extremely generous with his time, feedback, and hands-on v technical assistance with library preparation and bioinformatic analysis. Dr. Wang shared with us his expertise in the ChIP-seq field to help plan experiments for our manuscript currently in preparation.

Core facilities at the OSU Medical Center made key experimental contributions. Dr.

Pearlly Yan successfully coordinated the sequencing of our ChIP and cDNA libraries through the Nucleic Acid Shared Resource and helped us transition into bioinformatic analysis. Dr. Baris Hancioglu and Dr. Jie Zhang from the Bioinformatics Shared

Resource processed and performed statistical analyses of our ChIP-seq and RNA-seq data and assisted us in the process of target selection.

My committee members contributed many valuable ideas, some of which facilitated major breakthroughs towards our manuscript currently in preparation. As our experiments led us into less familiar territory in the chromatin, bioinformatics, and now pre-clinical fields, we were very lucky to have access to experts who were extremely collaborative, supportive and generous with their time and their ideas. Many thanks to

Dr. Jeff Parvin, Dr. Kay Huebner, Dr. Albert de la Chapelle, and Dr. Mark Parthun.

Many of the best ideas as well as the vision that consistently pushed this project forward came from my advisor, Dr. Joanna Groden. As the project proved to be more challenging than I had anticipated and I repeatedly ran out of ideas to try next, my advisor always had a better experiment in mind to advance the project towards our bigger goal. I have been fortunate to work with a real mentor who approached my development as a student strategically, across areas ranging from scientific writing and grant applications to

vi presenting at conferences and learning how to collaborate with other labs to acquire new technical skills.

The Ohio State University Medical Center has been a great environment in which to pursue graduate study, and the Biomedical Sciences Graduate Program has been both rigorous and highly supportive. Many thanks to the current directors of the program and to Amy Lahmers, as well as to Dr. Virginia Sanders, whose positivity and enthusiasm for science were a very memorable part of the program.

Grant support was from NIH Award R01CA063507 (to Joanna Groden), and awards

8UL1TR000090-05, 8KL2TR000112-05, and 8TL1TR000091-05 from the National

Center for Advancing Translational Sciences (to Joanna Groden). Support from the

HHMI Med-Into-Grad initiative, the Pelotonia Graduate Fellowship Program, and the

National Cancer Institute (F31 CA 174260) has been critical in allowing us to bring this project to a successful conclusion.

Many, many thanks to my family (Joyce, Bill, Nick, Robert, Verónica, and Mirta) for support and encouragement in ways that are too numerous to list.

vii

Vita

2002...... B.S. Chemistry, Yale University

2005…………………….M.Phil. Molecular Biochemistry & Biophysics, Yale University

2009 to present ……………………………. Graduate Research Associate, Department of

Cancer Biology, The Ohio State University

Publications

Hankey, W., Ebede, K. Groden, J. APC, in Cancer Therapeutic Targets, J.L. Marshall,

Editor 2016, Springer New York: New York, NY. p. 1-12.

Hankey W, Goss KH, Groden J. APC (Adenomatous Polyposis Coli) Tumor Suppressor,

Reference Module in Biomedical Sciences. Elsevier. 25-Nov-2015 doi: 10.1016/B978-

0-12-801238-3.98747-6

Hankey, W. and J. Groden, The Genetics of Colorectal Cancer, in Molecular

Pathogenesis of Colorectal Cancer, P.D.K.M. Haigis, Editor 2013, Springer New York:

New York, NY. p. 1-24.

viii

Fields of Study

Major Field: Biomedical Sciences Graduate Program

Area of Emphasis: Translational Research

Table of Contents

Abstract ...... ii

Dedication ...... iv

Acknowledgments...... v

Vita ...... viii

List of Figures ...... xi

Chapter 1: Literature Review ...... 1

Chapter 2: Rationale and Research Objectives...... 18

Chapter 3: Chromatin-associated APC tumor suppressor protein antagonizes canonical

WNT repression of the Mal, T-Cell Differentiation Protein-Like (MALL) gene ...... 22

Chapter 4: Mouse Adenoma Studies...... 74

Chapter 5: Thesis Summary and Future Directions…………………………………….89

References…………………………………….……………………………104

List of Figures

Figure 1: APC mutations are unequally distributed across different colorectal cancer pathways...... 4

Figure 2: Truncating mutations disrupt key structural features in the central and C- terminal regions of the human APC protein...... 6

Figure 3: Cytoplasmic APC protein negatively regulates the canonical WNT signaling pathway...... 8

Figure 4: Chromatin-associated APC negatively regulates canonical WNT activation of specific target genes by removing the licensing factor β-catenin...... 10

Figure 5: Anti-APC ChIP-seq exhibited high signal:background ratio and correlation between replicates...... 27

Figure 6: Anti-APC ChIP-seq exhibited genomic peaks near positive control loci...... 28

Figure 7: Motif analysis of ChIP-seq peaks detects significant enrichment of binding sites for TCF7L2 and other diverse transcription factors...... 30

Figure 8: Most transcription factor binding sites of interest are relatively APC-insensitive outside of their genomic context...... 33

Figure 9: TCF7L2 and AP-1 binding sites preferentially occur in the highest-confidence subset of peaks ...... 34 xi

Figure 10: Anti-APC peaks exhibit broad genomic distribution rather than close association with promoter elements...... 36

Figure 11: TCF7L2 and AP-1 binding sites co-occur in anti-APC ChIP-seq peaks located near genes whose transcription changes upon APC loss...... 37

Figure 12: RNA-seq from HCT-116 cells identified transcripts that change in expression upon silencing of APC...... 40

Figure 13: RNA-seq from two types of mouse colon adenoma identified transcripts that change in expression upon Apc loss of function or β-catenin gain of function...... 41

Figure 14: Colon adenomas from AOM/DSS-treated mice encode a degradation-resistant mutant form of β-catenin...... 42

Figure 15: Target selection included peaks with and without predicted TCF7L2 binding sites...... 43

Figure 16: β-catenin silencing significantly abolishes β-catenin ChIP signal at most APC targets of interest...... 45

Figure 17: β-catenin silencing does not disrupt APC localization to targets sufficiently to distinguish β-catenin-dependent vs. β-catenin-independent anti-APC peaks...... 46

Figure 18: Luciferase reporter assays identified PHLDB2 intron 1 and MALL intron 1 as transcriptional elements that change in activity following APC silencing...... 50

Figure 19: Luciferase reporter assays identified PHLDB2 intron 1 and MALL intron 1 as transcriptional elements that change in activity following silencing of β-catenin expression...... 51

xii

Figure 20: PHLDB2 intron 1 activation by canonical WNT signaling is dependent upon two TCF7L2 binding sites...... 53

Figure 21: Transcriptional activity of MALL intron 1 is dependent upon an intact AP-1 binding site...... 54

Figure 22: WNT-mediated transcriptional activation (A) and repression (B) involving

TCF7L2, AP-1, β-catenin, and APC...... 58

Figure 23: Primer Sets for Luciferase Cloning (Data in Fig. 8)...... 70

Figure 24: Primer Sets for Luciferase Cloning (Data in Figs. 18, 19)...... 71

Figure 25: Mutagenic Primers for Luciferase Constructs (Data in Figs. 20, 21)...... 72

Figure 26: ChIP qPCR primer sets (Data in Fig. 16, 17)...... 73

Figure 27: Colon adenomas from AOM/DSS mice (C57BL/6) and ApcMin/+ mice

(C57BL/6) exhibit significant overlap in their transcriptional profiles...... 77

Figure 28: Colon adenomas from ApcMin/+ mice exhibit unique transcriptional aberrations involved in the apoptosis and cell protrusion pathways...... 80

Figure 29: Colon adenomas from AOM/DSS-treated mice exhibit unique transcriptional aberrations involved in immune infiltration and inflammatory pathways...... 81

Figure 30: Transcriptional aberrations observed in colon adenomas from both ApcMin/+ and AOM/DSS-treated mice are linked to pathways associated with Apc function...... 82

Figure 31: Colon adenomas from ApcMin/+ mice exhibit activation of anti-apoptotic transcripts and repression of pro-apoptotic transcripts...... 84

Figure 32: Colon adenomas from ApcMin/+ mice exhibit aberrant activation or repression of transcripts involved in the formation of cell protrusions...... 85

xiii

Figure 33: Candidate genes repressed by canonical WNT signaling...... 92

Figure 34: Candidate genes activated by canonical WNT signaling...... 96

Figure 35: Candidate transcripts aberrantly expressed only in ApcMin/+-derived adenomas...... 101

xiv

Chapter 1: Literature Review

Driving questions in colorectal cancer research.

Colorectal cancer has emerged as a family of diseases that can develop along a number of different histological and molecular trajectories (Fig. 1). As the understanding of colorectal cancer at the molecular level becomes increasingly detailed, intense research interest surrounds the ability of these histological or molecular characteristics to predict prognosis. 65% of colorectal cancer patients survive 5-years or more after diagnosis [1], although individual outcomes are heavily dependent upon stage. Localized colorectal cancers are associated with 5-year survival of approximately 70-90% (depending on the degree of invasiveness) [1], while cancers that have spread to nearby lymph nodes are associated with only 40-50% 5-year survival. Metastatic colorectal cancers that have spread to the liver or other distant sites are associated with only 5% survival over a 5-year period, yet represent more than 20% of diagnoses [1]. These statistics confirm the fact that surgical resection remains the best tool for curing colorectal cancer, while therapeutic options against cancers that have disseminated to other sites quickly become exhausted.

The discovery and characterization of the genetic changes acquired along the malignant pathway have informed the search for novel therapeutic options. However, significant questions remain unanswered, such as whether or how colorectal tumors tend to acquire these key mutations in a particular order [2]. The field has begun to identify the patterns of mutations or aberrations in gene expression that distinguish short-term survivors from long-term survivors of advanced disease [3], or that correlate with responsiveness to certain interventions, including emerging antibody-based therapies [4]. These clinically- oriented questions may eventually be answered through a more complete understanding of how particular genetic changes translate into phenotypic changes.

The APC gene.

Biallelic mutation of the APC gene occurs in 45%-80% of colorectal cancers [5-7] and is observed in the earliest detectable lesions [2]. The APC locus was originally identified based on its link to familial adenomatous polyposis coli (FAP), an inherited syndrome of cancer predisposition [8-11]. Inherited mutations in the APC gene cause affected individuals to develop hundreds to thousands of adenomatous polyps, resulting in the onset of CRC typically before the age of 40 [12]. Individuals with FAP inherit a loss-of- function mutation in a single allele of APC, followed by an additional acquired mutation in the second allele of APC in the adenomas and adenocarcinomas that develop [13-15].

Thus, the acquisition of biallelic APC mutations represents an early and rate-limiting step in all FAP-associated and most sporadic colorectal tumors.

In studies of colorectal cancer as a whole, APC mutational status does not strongly correlate with outcome [16]. Nevertheless, APC mutations exhibit an interesting pattern of differential distribution in the recognized subtypes of colorectal cancer (Fig. 1). APC mutations correlate strongly with a large subset of colorectal cancers associated with intermediate prognosis [17]. On the other hand, APC mutations occur infrequently within a smaller subset derived from sessile serrated adenomas and associated with microsatellite instability and good prognosis [17]. This latter subset exhibits a relatively high proportion of activating mutations in the gene encoding β-catenin (CTNNB1) [18] that are mutually exclusive of APC mutations [19]. Interestingly, CTNNB1 mutations are significantly more prevalent in small adenomas than in large adenomas or adenocarcinomas, [17], whereas APC mutations are well-represented across all stages of tumorigenesis. It thus remains unclear whether APC mutational status has value as a predictive marker within any specific subgroup of colorectal cancer, as well as whether

APC loss drives clinical phenotypes relevant to prognosis.

Figure 1: APC mutations are unequally distributed across different colorectal cancer pathways.

Molecular analysis of colorectal cancers has identified at least four subsets associated with different prognoses [88, 117]. Subsets are defined by the presence or absence of the CpG island methylator phenotype (CIMP) and the microsatellite instability phenotype (MSI), two often-linked characteristics which together with the chromosomal instability phenotype (CIN) generate mutations that drive disease progression. Most colorectal cancers are characterized by neither CIMP nor MSI (B), but follow a trajectory similar to that observed in familial adenomatous polyposis (FAP), characterized by mutations in APC and TP53 [90, 118]. CIMP in the absence of MSI is associated with poor prognosis (A), while CIMP resulting in MSI [18, 119, 120] is associated with good prognosis (C), similar to the MSI-driven cancers observed in Lynch syndrome [121, 122] (D).

Functions of the APC tumor suppressor protein.

The APC gene encodes a 312-kDa protein (Fig. 2) that performs diverse functions and localizes to multiple cellular compartments. Mutations in APC are often frameshifts, insertions or deletions that introduce premature stop codons and lead to the production of a truncated APC protein. Amino acids 1000 to 1600 of APC have been identified as a mutation cluster region that represents roughly 20% of the total (2,843-amino acid) protein yet contains about 60% of all identified mutation sites [7]. The following sections detail functions of the wild-type APC protein, all of which are weakened or lost through the acquisition of pathogenic mutations. Also discussed are dominant functions of truncated APC proteins that contribute to tumorigenesis.

Cytoplasmic APC negatively regulates canonical WNT signaling.

The best-known function of APC is its ability to interact with β-catenin in the cytoplasm and promote β-catenin phosphorylation, ubiquitination and subsequent proteolytic degradation (Fig. 3) [20-23]. This function occurs within the context of a cytoplasmic complex [21, 24] that includes glycogen synthase kinase 3β (GSK-3β) [25], AXIN1 [22,

26] or AXIN2 [21, 27], and other kinases and phosphatases [28, 29]. The interaction of

APC with this complex maps to a region in the center of the protein that contains three serine-alanine-methionine-proline (SAMP) repeats that mediate the binding of APC to

AXIN1/AXIN2 [21], as well as three repeats of 15 amino acids each and seven repeats of

20 amino acids each. β-catenin interacts with the 15-amino acid repeats constitutively

[30], and with the 20-amino acid repeats inducibly [31] following their phosphorylation

5 by GSK-3β [32] and casein kinase I [29, 33]. Both the 15-amino acid and 20-amino acid repeats are necessary to enable APC to promote β-catenin degradation [31]; truncating mutations to APC generally disrupt these repeats either partially or completely.

Figure 2: Truncating mutations disrupt key structural features in the central and C-terminal regions of the human APC protein.

Key structural features of the human APC protein that are disrupted by most truncating mutations include its 20-amino acid repeats (purple), SAMP repeats (red), nuclear localization sequences (green arrows), microtubule binding region (light blue), EB1 binding region (yellow), DNA-interacting S(T)PXX motifs (asterisks), and C-terminal PDZ domain-binding region (brown). Truncated N-terminal APC proteins retain some ability to interact with β-catenin due to the preservation of the 15-amino acid repeats and some of the 20-amino acid repeats, while retention of the Armadillo domain preserves the ability localize to the nucleus.

The role of APC in this cytoplasmic complex negatively regulates the canonical WNT signaling pathway [24], whose ability to mediate transcriptional changes requires the licensing factor β-catenin to bind to transcription factors of the TCF/LEF family [34].

The canonical WNT signaling pathway alters the transcription of key target genes, including activation of the genes encoding c-Myc [35], Cyclin D1 [36, 37] and Lgr5 [38,

39]. These and other changes in gene expression collectively drive proliferation, survival and maintenance of an undifferentiated state in progenitor cells of the colorectal epithelium [36, 40-42]. The pathway is critical to normal tissue homeostasis [43], as mature cells lining the colon and rectum must be frequently replaced as they become damaged, undergo apoptosis and are sloughed into the intestinal lumen [44]. Progenitor cells of the colorectal epithelium are characterized by activated canonical WNT signaling, while maturing cells are characterized by inactivation of the pathway and increasing expression of APC [45, 46]. Disruption of this balance initiates colorectal cancer, as the progeny of a colorectal progenitor cell lacking functional APC are unable to stop proliferating, differentiate or undergo apoptosis [45]. APC loss stabilizes β- catenin and constitutively activates the pathway even in the absence of a WNT signal [31,

47]. The importance of this particular APC function in the process of tumorigenesis is underscored by the observation that a significant percentage of the colorectal cancers with wild-type APC exhibit a point mutation in CTNNB1, the gene encoding β-catenin

[19] that makes the protein resistant to degradation [48]. Together, APC and β-catenin mutations form part of a larger group of molecular changes by which approximately 93% of colorectal cancers exhibit activation of the canonical WNT signaling pathway [5].

Figure 3: Cytoplasmic APC protein negatively regulates the canonical WNT signaling pathway.

The critical difference between the active (A) and inactive (B) states of the canonical WNT signaling pathway is the accumulation of the licensing factor β-catenin, especially in the nucleus, where it binds to TCF/LEF family transcription factors to promote changes in gene transcription. The active state of the pathway (A) is characteristic of colorectal cancers and colorectal progenitor cells, as it favors proliferation and survival at the expense of differentiation and sensitivity to apoptosis. The inactive state (B) is characteristic of mature, differentiated cells of the colorectal epithelium that express APC. APC (green) is a required component of the cytoplasmic complex that limits β- catenin availability by promoting its proteolytic degradation.

Nuclear APC negatively regulates canonical WNT signaling.

The APC protein further counteracts canonical WNT signaling by participating in at least two other mechanisms that take place in the nucleus. Nuclear import of APC is dependent upon two nuclear localization sequences (NLS) as well as a separate

Armadillo domain that also promotes nuclear import [49, 50]. Once inside, nuclear APC interacts with β-catenin and facilitates its export to the cytoplasm [51-54]. This activity is dependent upon two recognized nuclear export sequences (NES) within APC [52]. The ratio of nuclear APC to cytoplasmic APC decreases as proliferation slows down and cells enter a quiescent state [55-57], and is controlled in part by phosphorylation of key residues near each APC NLS [49, 55]. The nuclear interaction between APC and β- catenin occurs in part within the chromatin fraction, in which APC promotes the removal of β-catenin from specific genomic loci (Fig. 4) [58]. Nuclear APC interacts not only with β-catenin but also with C-terminal binding protein (CtBP), a transcriptional co- repressor [59]. A detailed mechanistic study has demonstrated that APC and CtBP transiently interact with β-catenin at the WNT-activated MYC promoter and promote the removal of β-catenin from this locus, coinciding with the appearance of the more stable co-repressors TLE-1 and HDAC1 [58]. This function of chromatin-associated APC negatively regulates WNT activation of the MYC gene [58], as well as WNT activation of the AXIN2, DKK1 and SP5 genes [60]. Truncated APC proteins observed in colorectal cancer generally lack both NLS, which are located in the missing C-terminal half of the protein. However, truncated APC retains the ability to move between the nucleus and cytoplasm because of the preserved Armadillo domain [61]. It is difficult to compare the

9 relative importance of the cytoplasmic and nuclear mechanisms by which APC negatively regulates canonical WNT signaling, as both are disrupted by pathogenic APC mutations [62]. Evidence for the importance of nuclear APC includes the observations that biallelic point mutations abolishing both NLS within APC increase proliferation and expression of canonical WNT signaling targets in the mouse intestine, and that a single mutant allele increases polyp number and size in ApcMin/+ mice [63].

Figure 4: Chromatin-associated APC negatively regulates canonical WNT activation of specific target genes by removing the licensing factor β-catenin.

The nuclear fraction of APC further antagonizes canonical WNT signaling in colorectal cancer cells by interacting with chromatin-associated β-catenin at WNT-activated target genes such as MYC. This transient interaction leads to the removal of β-catenin, the appearance of co-repressors such as TLE-1 and HDAC1, and transcriptional repression [58, 60]. 10

Cytoskeletal functions link APC to adhesion, migration and cell polarity.

In addition to its roles in the cytoplasm and nucleus, APC localizes to the cell border [61] and participates in at least three mechanisms regulating epithelial organization and cell migration. APC localization to the cell periphery is dependent upon the actin cytoskeleton [64] as well as the Armadillo domain within APC [65], although truncated

APC proteins retaining this region still localize there with reduced efficiency [64]. The scaffolding protein IQGAP1 mediates the indirect interaction of APC with actin and also links APC with Rac1 and Cdc42, two Ras-family GTPases that regulate actin structures

[66]. APC also regulates the activity of these GTPases to influence actin organization

[67].

APC interacts with both β-catenin and plakoglobin (γ-catenin) at cell-cell junctions [68], which function to connect adjacent epithelial cells, organize them into layers and promote cell polarity. Β-catenin and plakoglobin physically connect cytoskeletal components including actin and intermediate filaments with transmembrane adhesion molecules such as cadherins. APC loss promotes a decrease in cell-cell and cell-matrix adhesion by altering the subcellular distribution of E-cadherin [69, 70]. This consequence of APC loss occurs at the protein level, independently of canonical WNT signaling [69]. On the other hand, the role of APC at cell-cell junctions may be reinforced by its negative regulation of canonical WNT signaling, which controls cell motility through targets including E-cadherin [71] and matrix-remodeling enzymes such as MMP-7 [72, 73].

Together, these transcriptional and non-transcriptional roles of APC at the cell periphery drive the phenotype of defective intestinal epithelial cell migration observed along crypt-

11 villus axes in both Apc-deficient [74] and wild-type Apc-overexpressing [75] mice.

Truncated APC proteins contribute in a dominant negative manner to this loss of migration directionality [76]. APC is furthermore linked to cell polarity through interactions with DLG [77] and hScrib [78], the respective human homologues of the

Drosophila proteins Discs Large and Scribble. DLG interacts with PDZ domain-binding residues at the C-terminus of APC, and the two proteins form a complex with hScrib at lateral regions of cell-cell contact in canine kidney epithelial cells [78].

Finally, APC localizes to leading edges of migrating cells in a microtubule-dependent manner [79]. This function of APC maps not to its N-terminal Armadillo domain, but to

C-terminal regions of APC that directly contact the microtubule network [80].

Specifically, amino acids 2200-2400 are enriched for basic residues that facilitate microtubule binding and that promote polymerization [80] and bundling [81, 82] of microtubules in vitro. The APC C-terminal region also binds the microtubule-interacting protein EB1, which specifically interacts with and localizes APC to microtubule plus ends [83]. These microtubule-related functions of APC are proposed to drive the formation of membrane protrusions and influence the balance between adhesion and motility [84]. APC functions in mouse fibroblasts as an RNA-binding protein that localizes key RNA molecules to cell protrusions with consequences for cell migration

[85]. In dying cells, APC may be required to target microtubules to the appropriate side of the cell, promoting proper extrusion from the apical rather than the basal side of the epithelium [86]. Collectively, these studies of APC at the cell periphery argue for roles

12 in linking the actin cytoskeleton with the microtubule network and in the establishment of cell polarity and suppression of invasive behaviors [87].

APC in mitotic spindle dynamics and genomic stability.

Colorectal cancers segregate into two mutually exclusive categories based on exhibiting genomic instability at either the microsatellite level (due defects in mismatch repair) or the chromosomal level [88, 89]. APC mutations are particularly well-represented within the subset of colorectal cancers that feature widespread chromosomal damage [90], and also play an important role in this phenotype [91]. The ability of APC to interact with the microtubule network facilitates its localization to the kinetochore [91, 92], a protein structure that mediates the attachment of microtubules to sister chromatids during mitosis. Murine cells with Apc mutations exhibit chromosomal abnormalities as well as mitotic spindles characterized by numerous microtubules lacking proper connections to the kinetochore [91]. APC silencing in human colorectal cancer cells similarly decreases inter-kinetochore tension during metaphase, and results in defective progression through mitosis [93]. The APC-interacting protein EB1 collaborates with APC to regulate the mitotic spindle, and the loss of this APC-dependent mechanism facilitates errors in chromosome alignment without halting cell division [94].

APC is modified during mitosis by phosphorylation [95], including by the spindle checkpoint kinases Bub1 and BubR1, suggesting that its contributions to mitotic spindle integrity and chromosome segregation are accompanied by a role in the spindle checkpoint [92]. Interestingly, overexpression of a truncated APC protein exerts a

13 dominant negative effect that compromises the spindle checkpoint [96], in which defects are often linked to chromosomal instability. These functional data collectively suggest that truncating APC mutations drive genomic instability at the chromosomal level through both loss-of-function and gain-of-function mechanisms that occur independently of its role in canonical WNT signaling.

APC controls DNA replication and cell cycle progression.

In addition to its role in the spindle checkpoint and mitotic progression, APC controls the cell cycle by regulating the G1/S transition. Various APC-deficient colorectal cancer cell lines stably transfected to express wild-type APC exogenously exhibit dramatic changes in doubling time and inhibition of G1/S phase progression [97]. Overexpression of RB pathway components such as Cyclin D1/CDK4, Cyclin E/CDK2, E1A, and E2F overrides cell cycle inhibition by APC, suggesting that APC re-expression restores the G1/S checkpoint [98, 99]. Negative regulation of canonical WNT signaling by APC contributes to this function by downregulating targets including Cyclin D1 itself [36, 37].

The physical interaction of APC with other proteins such as DLG may also be involved, as the C-terminal DLG-binding residues within APC are required for complete inhibition of S-phase entry in a mouse fibroblast cell model, while DLG overexpression in itself is sufficient to arrest cells at the G1/S transition [100]. The interaction of the APC C- terminus with A/T-rich DNA also has been shown to block entry into or progression through S-phase by blocking DNA replication [101, 102]. These findings are consistent with evidence that defective G1/S-phase progression due to APC overexpression is only

14 partially alleviated by co-transfection of a constitutively active mutant β-catenin [99].

APC thus contributes to the regulation of cell cycle progression through a combination of

WNT-dependent and WNT-independent mechanisms.

Pro-apoptotic functions of APC.

APC exhibits a gradient of increasing expression in the luminal direction along the colorectal crypt-villus axis [46], coinciding with mature, non-proliferative areas of the crypt where apoptotic cell death and cell shedding occur. Functional evidence linking

APC to apoptosis began to accumulate with the observation that exogenously restoring

APC expression in an APC-deficient colon cancer cell line triggered a 10-fold increase in the proportion of apoptotic cells [41]. In addition, overexpression of WNT-inhibitory

PDZ domain-containing peptides from the Dishevelled protein induce apoptosis in an

APC-dependent manner [103]. The link between APC and apoptosis is mediated at least in part by its role in canonical WNT signaling, whose targets include the BIRC5 gene that encodes the anti-apoptotic Survivin protein [104].

The ability of APC to sensitize cells to apoptosis is also partially WNT-independent

[105]. Caspase family members proteolytically cleave APC in apoptotic cells [106], producing an N-terminal fragment that localizes to mitochondria and interacts with hTID-

1 to promote caspase activity and cell sensitivity to apoptosis [107]. While truncated

APC proteins in colorectal cancer also exhibit mitochondrial localization, they appear to exert anti-apoptotic effects instead, as their knockdown promotes apoptosis and mitochondrial membrane permeability [108]. Truncated APC proteins interact with and

15 promote mitochondrial localization of the anti-apoptotic BCL2 protein [108]. The role of

APC loss in shifting the balance between apoptosis and survival therefore is mediated by multiple mechanisms, both direct and indirect.

The role of APC in differentiation.

Just as the pattern of APC expression along the colorectal crypt-villus axis indicates its role in apoptosis, the coincidence of APC expression with areas of mature, functional epithelium provided the first clue that APC affects differentiation as well. This luminal upper half of the colorectal crypt is populated with a mixture of terminally-differentiated columnar epithelial cells, goblet cells that produce mucus and neuroendocrine cells, all of which are derived from multipotent stem cells at the base of the crypt. Apc loss in the mouse small intestine disrupts commitment to all three cell fates while driving commitment to the basally-located Paneth cell lineage through a WNT-dependent mechanism [109]. Apc likely drives differentiation through suppression of canonical

WNT signaling as well, as mice deficient in the canonical WNT signaling transcription factor Tcf7L2 exhibit dramatic and lethal defects in the proliferation of intestinal stem cells [110]. Consistent with these findings, canonical WNT signaling activates the expression of Lgr5, a critical marker of intestinal stem cells [111].

Interestingly, loss of Apc in the zebrafish gut produces a differentiation defect that occurs prior to nuclear β-catenin accumulation or changes in proliferation [112]. This may result from the loss of an Apc function in promoting proteasome-dependent degradation of the transcriptional co-repressor Ctbp1 [113]. Upon Apc loss, Ctbp1 suppresses the

16 transcription of genes involved in retinoic acid biosynthesis [113], a process required for differentiation in the zebrafish gut [112, 114]. Differentiation in this model is accomplished through the ability of retinoic acid to antagonize the expression of demethylase genes, resulting in the methylation of gene promoters critical to maintaining progenitor-like phenotypes [115]. APC similarly regulates CtBP1 expression in human colon cancer cell lines [113], indicating that the zebrafish mechanism by which it drives differentiation through retinoic acid biosynthesis may be present in the human colorectal epithelium as well. Since differentiation and cell migration along the crypt-villus axis are coupled processes [116], APC may help link cell fate and cell migration in the colorectal epithelium.

Summary

In conclusion, the APC tumor suppressor protein performs multiple functions that contribute to its role in preventing colorectal tumorigenesis. It is difficult to assign relative importance to individual functions, other than to point out that APC contributions to relevant phenotypes such as cytoskeletal regulation, cell cycle progression, chromosomal stability, sensitivity to apoptosis and differentiation are mediated by multiple mechanisms, both WNT-mediated and WNT-independent. Dominant negative effects of truncated APC proteins further counteract wild-type APC functions by promoting migration, chromosomal instability and evasion of apoptosis.

Chapter 2: Thesis Rationale and Research Objectives

The present study is part of a larger effort to further our understanding of the molecular events underlying colorectal tumorigenesis in order to address obstacles to the effective treatment of advanced disease. Of particular concern is a lack of molecular markers to predict long-term vs. short-term survival among patients whose disease has spread to nearby lymph nodes (Stage III) or to other organs (Stage IV). The vast majority of colorectal cancer cures are accomplished surgically; even liver metastases are resected in the absence of evidence for wider dissemination. Disease recurrence and mortality of patients originally diagnosed as “Stage II” strongly correlates with the presence of lymph node micrometastases detected later on [123]. This observation reflects multiple challenges in the treatment of colorectal cancer, the first of which is the impossibility of detecting the full extent of tumor spread with absolute certainty. The other challenge is that approved chemical and antibody-based agents for colorectal cancer are limited in number (12 in total as of 2016) and represent only five different mechanisms of action.

These agents can be effective as adjuvant treatments, but provide a limited set of options for long-term growth suppression in metastatic patients who can also develop disease resistance.

The question of whether primary tumors contain molecular evidence to predict metastatic

(or micrometastatic) disease has been a matter of some debate. Some insight has been gained through the recent characterization of the metastasis-associated in colon cancer 1

(MACC1) gene. This gene is aberrantly expressed specifically in malignant cells relative to normal and benign tissues, and its encoded protein drives migratory, invasive and metastatic behaviors as a transcriptional activator of the hepatocyte growth factor (HGF) /

Met signaling pathway [3]. High MACC1 expression in colorectal cancer is a marker that predicts multiple metastases and poor survival [124]. High MACC1 staining at the center of a tumor correlates highly with the presence of individual cells and clusters of cells

(tumor buds) in the tumor microenvironment at the invasive front, while MACC1 staining at the invasive front correlates with the presence of distant metastases [125].

This example argues that the metastatic potential of a colorectal tumor can be defined as early as the onset of invasion and that the markers of metastasis can be detected within the primary tumor.

Researchers have characterized key genetic events that occur during colorectal carcinogenesis, defining multiple pathways by which tumor development can progress.

Certain genetic changes are associated with poor prognosis, including activating mutations in the KRAS2 gene, which serve as prognostic markers of multiple metastases

[124]. APC mutations occur very early on in the process of tumorigenesis [2], yet exhibit an unequal pattern of distribution among recognized colorectal cancer subtypes, each of which is associated with a different prognosis [88]. APC loss produces profound changes in gene expression characterized through transcriptional profiling of tumors [126-128] as

19 well as studies of the related canonical WNT signaling pathway through both transcription-oriented [39, 45] and mechanistic (chromatin immunoprecipitation-based)

[129-131] experiments. These studies provide highly valuable lists of canonical WNT signaling targets that have informed the search for additional markers of metastasis and potential therapeutic targets.

Some limitations in the design of these large-scale previous studies have allowed certain targets of APC transcriptional control to remain unidentified or uncharacterized. Most studies precede the field’s gradual acceptance that certain target genes are repressed by canonical WNT signaling [71, 132, 133], and instead report primarily on WNT-activated targets with increased expression following APC loss. Another limitation is the repeated experimental use of canonical WNT signaling activation by expression of either β-catenin or TCF/LEF family members as a surrogate for APC loss. The assumption that all of the transcriptional changes that occur upon APC loss are mediated by canonical WNT signaling overlooks the idea that APC performs anti-tumorigenic functions through multiple WNT-dependent and WNT-independent mechanisms. These additional WNT- independent functions may explain the high prevalence of APC mutations and their more sustained representation across all stages of colorectal cancer relative to those mutations constitutively stabilizing β-catenin [17].

Recent evidence that chromatin-associated APC protein interacts with and removes β- catenin from WNT-activated genomic sites [58, 60] prompted this study to test the possibility that chromatin-associated APC may collaborate with other transcription factor complexes to control transcription. I hypothesized that APC recruitment to other target

20 genes could be mediated by protein-protein interactions with a transcription factor or co- factor other than β-catenin. An additional idea is that the S(T)PXX motifs within APC

(which are known to interact with A/T-rich DNA and influence DNA replication [101,

134]) might localize APC to specific binding sites within the genome independently of other DNA-binding proteins driving APC to TCF/LEF binding sites.

This project began with the research objective of identifying APC-associated genomic sites comprehensively in the chromatin fraction and comparing these observations with gene expression profiling data from mouse colon adenomas with either Apc deficiency

(from ApcMin/+ mice) or constitutive β-catenin activation (from mice treated with azoxymethane / dextran sulfate sodium). While both types of mouse adenomas would exhibit transcriptional effects of activated canonical WNT signaling, they would differ in the expression of WNT-independent Apc targets. I carried out similar gene expression profiling of human colon cancer cells with endogenous stabilized mutant β-catenin in the presence or absence of siRNA-based knockdown of APC. The results of these studies would include novel or previously-overlooked targets of APC transcriptional control that could be regulated by chromatin-associated APC protein independently of canonical

WNT signaling. The success of the project would enable us to segregate individual changes in gene expression following APC loss into groups, based on the different mechanisms of transcriptional control by APC. Our long-term goal was to assess the ability of specific subsets of APC targets to predict prognosis, particularly for those targets regulated independently of canonical WNT signaling.

Chapter 3: Chromatin-associated APC tumor suppressor protein

antagonizes canonical WNT repression of the MALL gene.

Abstract

Mutation of the APC tumor suppressor gene occurs in a high percentage of colorectal tumors and is considered the central event driving tumor initiation in the large intestine. The APC protein performs multiple functions including negative regulation of the canonical WNT signaling pathway by both cytoplasmic and nuclear mechanisms.

Intriguing reports that APC interacts with chromatin to repress key WNT-activated targets prompted this study’s use of chromatin immunoprecipitation and next-generation sequencing to identify more than 6,000 genomic peaks associated with the APC protein in a colon cancer cell line wild-type for APC. Target selection was performed by comparison to whole transcriptome sequencing data from APC-deficient and APC-wild- type colon cancer cell lines and mouse colon tumors. 175 transcripts whose expression changes upon APC loss were linked to genomic regions physically associated with the

APC protein. Motif analysis of APC-associated genomic peaks identified DNA binding sites for the TCF7L2 and AP-1 transcription factors that were overrepresented.

Luciferase reporter assays validated some targets and demonstrate that APC antagonizes canonical WNT signaling not only at WNT-activated genes such as PHLDB2 (Pleckstrin

Homology Like Domain Family B Member 2), but also at WNT-repressed genes such as

MALL (Mal, T-Cell Differentiation Protein-Like). The first intron of the MALL locus mediates transcriptional activation by the AP-1 transcription factor, repression by the canonical WNT signaling pathway, and reversal of that repression by APC through a β- catenin-dependent mechanism.

Introduction

Biallelic mutations of the APC gene initiate the development of a high percentage of colorectal cancers [5-7]. APC encodes a multi-purpose protein whose functions include negative regulation of the canonical WNT signaling pathway [47, 48]. The APC protein inactivates canonical WNT signaling by limiting the availability of β-catenin [24], a licensing factor whose presence in the nucleus modifies how transcription factors of the

TCF/LEF family regulate gene transcription [34]. APC interacts with β-catenin in a cytoplasmic complex that facilitates β-catenin degradation [20-23].

Biallelic APC mutations in cells of the colorectal epithelium result in β-catenin accumulation and constitutive activation of canonical WNT signaling [31], which in turn promotes proliferation, survival and maintenance of an undifferentiated state [36, 40-42].

Constitutive activation of canonical WNT signaling is a common feature of early colorectal tumors, occurring not only as a result of loss-of-function mutations to APC but also through gain-of-function mutations to β-catenin that confer resistance to degradation

[48]. Acquisition of biallelic mutations of APC or a single gain-of-function mutation to

β-catenin that results in its constitutive stabilization appear to be mutually exclusive events [19], either of which triggers the development of colorectal adenomas aberrantly expressing the genes encoding c-Myc, Cyclin D1 and other canonical WNT targets [35-

37].

Just as cytoplasmic APC interacts with β-catenin to promote its degradation, nuclear APC interacts with β-catenin to antagonize canonical WNT signaling further. Nuclear APC 23 facilitates both the export of β-catenin to the cytoplasm [51-53] and the removal of β- catenin from specific genomic loci [58]. The interaction of APC with chromatin- associated β-catenin negatively regulates the expression of the MYC, AXIN2, DKK1 and

SP5 genes [58, 60], four well-known targets of canonical WNT signaling.

The observation that colorectal cancer development can be initiated either by mutations inactivating APC or mutations activating β-catenin has led to the prevailing view that negative regulation of canonical WNT signaling is the predominant function by which

APC prevents tumorigenesis. On the other hand, the percentage of biallelic APC mutations remains relatively constant across the spectrum from early to late-stage colorectal tumors, while activating mutations within β-catenin are better represented in early adenomas than in colorectal malignancies [17]. It is possible that although either type of mutation initiates tumor development, APC mutations may support tumor progression along a more efficient pathway due to the loss of its additional functions.

Functional investigations of APC at the protein level have uncovered diverse interacting partners, subcellular localizations and functions, many of which are not linked to β- catenin but nevertheless are anti-tumorigenic. These APC functions contribute to the establishment of proper cell polarity [85, 87], cytoskeletal organization [70, 80, 135], adhesion and migration [69, 76, 86], control of DNA replication and cell cycle progression [101, 102], mitotic chromatin compaction and genomic stability [91, 136-

138], sensitization to apoptosis [105-108], and differentiation [113, 139]. The importance of these individual APC functions relative to one another remains a subject of considerable interest.

The present study was motivated by recent reports of a chromatin-associated APC function and was designed to identify a more complete list of targets whose transcription is regulated by chromatin-associated APC. The study’s anticipated results included the 24 discovery of novel targets of canonical WNT signaling, as well as target genes whose expression is controlled by chromatin-associated APC independently of canonical WNT signaling. The actual outcomes did not include a definitive answer as to the existence of

WNT-independent targets of chromatin-associated APC. On the other hand, our data identify a novel shared target of canonical WNT signaling and chromatin-associated

APC, the MALL gene, that is unusual in that its transcription is directly repressed by canonical WNT signaling and activated in the presence of APC. Within one of the genomic regions mediating this effect, WNT- and APC-sensitivity are dependent on a binding site for the AP-1 transcription factor. This finding was highly unexpected, but consistent with the observation that TCF7L2 and AP-1 binding sites are highly enriched and co-occurring in many APC-associated genomic peaks identified in our screen. The implications of these findings include the importance of sorting canonical WNT targets into categories based on functional criteria such as the magnitude and direction of their transcriptional response to changes in canonical WNT signaling, as well as mechanistic criteria such as the presence or absence of nearby binding sites for TCF7L2 and/or AP-1.

WNT-repressed targets in particular may be interesting markers of prognosis or treatment outcome that have been overlooked in some previous studies. These results further suggest that AP-1 modulation may represent a novel strategy for targeting the expression of certain canonical WNT target genes, and that expression levels of WNT-repressed targets such as MALL could be prognostic indicators.

Results

α-APC ChIP-seq identified more than 6,000 APC-associated genomic peaks.

Chromatin immunoprecipitation of APC was performed using the colon cancer cell line

HCT-116, which expresses wild-type APC but also a degradation-resistant point mutant of β-catenin that constitutively activates canonical WNT signaling. Next-generation sequencing identified more than 6,000 APC-associated genomic peaks enriched by chromatin immunoprecipitation of APC, observed by comparison of two ChIP replicates to their respective input material. Peak calling thresholds were adjusted to optimize signal:background ratio (Fig. 5A), and peaks were selected based on overlap in ChIP-seq replicates 1 and 2 (Fig. 5B), which show similar degrees of enrichment.

Previous reports informed the interpretation of these data by identifying four genes

(MYC, AXIN2, SP5 and DKK1) regulated by chromatin-associated APC [58, 60]. These genomic loci served as positive controls for the anti-APC ChIP-seq (Fig. 6). Peaks were observed near three out of four positive control transcription start sites (MYC, AXIN2 and

DKK1). The absence of peaks associated with the SP5 gene suggests that the current

ChIP-seq datasets provide a partial list of APC-associated peaks, and that the combination of enrichment and peak counts may not have been sufficient to reach saturating depth. MYC, AXIN2 and DKK1 peaks were used to evaluate the success of

ChIP-seq by comparison of their locations to those of equivalent peaks within a published anti-β-catenin ChIP-seq dataset generated from the same cell line [129] (comparison not shown). One of the anti-APC peaks observed near MYC shared 80% overlap with a peak from the anti-β-catenin study, while a second MYC-associated peak was adjacent but non- overlapping. Of the five anti-APC peaks observed near AXIN2, three overlapped completely with equivalent peaks from the anti-β-catenin dataset, while two overlapped

26 partially (30% and 91%, respectively). The anti-β-catenin study did not identify a peak near DKK1 that would enable comparison [129]. The similarities in these data were consistent with the published conclusion that APC association with chromatin is mediated by its direct interaction with β-catenin [58].

Figure 5: Anti-APC ChIP-seq exhibited high signal:background ratio and correlation between replicates.

Peak selection for α-APC ChIP-seq was performed to maintain both a high ratio of signal to noise and significant correlation between similarly-enriched replicates 1 and 2. (A) A heat map depicting 4-kb genomic regions centered on each overlapping peak (identified in both replicates 1 and 2) shows high signal (red) in each ChIP-seq sample and low background (blue) in each matching input sample. (B) A scatter plot comparing the peak scores (in log10 scale) from ChIP-seq replicate 1 (x-axis) and ChIP-seq replicate 2 (y- axis) indicated that overlapping peaks observed in both replicates generally had similar peak scores in both samples (Pearson correlation coefficient ρ = 0.8246).

Figure 6: Anti-APC ChIP-seq exhibited genomic peaks near positive control loci.

Sequencing data for 4 positive control loci visualized using the Integrative Genomic Viewer confirmed that 3 out of 4 genes known to be regulated by APC at the chromatin level were located near genomic peaks (indicated by red boxes) enriched by anti-APC chromatin immunoprecipitation in multiple replicates. A third replicate was not included in Figures 5A and 5B but showed greater enrichment / higher peaks than the other replicates, although replicates 1 and 2 exhibited stronger correlation.

Binding sites for AP-1 and several promoter-associated transcription factors were enriched in APC peaks.

Motif analysis of anti-APC ChIP peak sequences identified recurring transcription factor binding motifs. CisFinder [140], MEME-ChIP [141] and RSAT [142] algorithms were used to perform preliminary analyses using their default settings. Because of the known association of APC with β-catenin and its transcription factor binding partner TCF7L2, it was anticipated that many peaks would contain the consensus TCF7L2 binding site. The three algorithms gave inconsistent results, with only MEME-ChIP detecting significant enrichment of binding sites for TCF7L2 (with a p-value of 3.1 x 10-23). On the other hand, all three algorithms detected enrichment of motifs consistent with binding sites for the AP-1 transcription factor, with MEME-ChIP providing a p-value of 8.3 x 10-123 for the TGAVTCAY motif. The highly regarded MatInspector algorithm [143] (from the

Genomatix Software Suite) detected enrichment of binding sites for a number of transcription factors, including NRF1, SP1, ZBTB14 (ZFP161), EGR-1, USF1/2, E2F3 and E2F1 (Fig. 7, yellow). Common denominators for all these motifs is a known pattern of association with promoters and high G/C content. Enrichment of SP1, EGR-1 and

USF1/2 binding sites had similarly been detected by MEME-ChIP and RSAT. Following this initial analysis of all α-APC peak sequences, MatInspector was subsequently used to analyze a subset of peaks that each contained one or more predicted TCF7L2 binding sites. This subset of TCF7L2+ peaks showed enrichment of binding sites for a variety of transcription factors related to the AP-1 complex (Fig. 7, blue). All AP-1-related binding sites were variations of the TPA response element (TGASTCA). The TCF7L2, AP-1,

NRF1, SP1, ZBTB14, EGR-1, USF1/2, E2F3 and E2F1 binding sites were identified as candidates for further investigation of their potential to mediate APC recruitment and sensitivity. 29

Figure 7: Motif analysis of ChIP-seq peaks detects significant enrichment of binding sites for TCF7L2 and other diverse transcription factors.

Motif analysis of all peaks using the MatInspector algorithm (Genomatix Software Suite) detected significant enrichment of predicted binding sites for a number of transcription factors that associate with promoters (highlighted in yellow). Predicted TCF7L2 binding sites (highlighted in green) served as a positive control (indicated with arrows) whose enrichment by α-APC ChIP was expected. Motif analysis of a subset of peaks containing predicted TCF7L2 binding sites detected enrichment of binding motifs for non-promoter associated transcription factors (highlighted in blue), many of which serve as components of the dimeric AP-1 complex. Predicted binding sites from both groups were selected for further study based on these observations.

Only the TCF7L2 transcription factor binding site was sufficient to mediate APC- sensitive transcription.

Three consecutive repeats of each transcription factor binding site of interest were cloned into the pGL3-promoter luciferase reporter vector (upstream of the firefly luciferase gene). Available positive (TOPFLASH) and negative (FOPFLASH) control firefly luciferase reporter vectors each contained six consecutive repeats of either wild-type or mutant TCF7L2 binding sites [47]. These firefly luciferase plasmids were co-transfected with a renilla luciferase control plasmid into HCT-116 cells that had been previously transfected with either scrambled or anti-APC siRNA. APC knockdown results in at least a two-fold enhancement of the ability of the TCF7L2 positive control to drive luciferase expression, while other transcription factor binding motifs show little or no sensitivity to the presence or absence of APC (Fig. 8). These results do not exclude the possibility that one of these transcription factor binding sites mediates transcriptional sensitivity to APC in the genomic context of a complete regulatory element.

The lowest p-value subsets of anti-APC ChIP peaks are highly enriched for TCF7L2 and

AP-1 binding sites.

Following this initial screen, α-APC ChIP peak sequences were examined for the highest- confidence permutations of each transcription factor binding motif (Fig. 9). The highest- confidence permutations of AP-1 (TGASTCA [144]) and TCF7L2 (WWCAAAG [129]) binding sites occur in 27.0% and 27.6% of all peaks, respectively. Those for SP1

(KGGGCGGRRY [145], 16.1%), USF1/2 (CACGTG [146], 15.0%), NRF1

(GCGCRYGCGC [147], 4.0%), and EGR-1 (GCGKGGGCG [148], 3.6%) occur less frequently. Occurrences of each binding site were then counted within various subsets of peaks that were grouped according to p-value. Motif analysis of these subsets was 31 designed to test whether the most significant peaks were especially enriched for a particular cis-motif. TCF7L2 and AP-1 show the most consistent and dramatic trends, occurring in 70% and 50% of the 100 highest-confidence peaks, respectively. Although all six motifs of interest show a trend of increasing enrichment as lower-confidence peaks are filtered out, these results strongly suggest that TCF7L2 and/or AP-1 binding sites play a role in APC recruitment.

Figure 8: Most transcription factor binding sites of interest are relatively APC- insensitive outside of their genomic context.

Luciferase reporter assays tested the hypothesis that motifs of interest recruit APC to chromatinized DNA, resulting in a change in transcriptional output. Candidate transcription factor binding sites were selected from motif analysis summarized in Figure 7. Three consecutive repeats of each predicted transcription factor binding site were cloned into the multiple cloning site of the pGL3-promoter vector, using the restriction endonucleases Kpn I and Bgl II. The resulting firefly luciferase constructs were co- transfected with a renilla luciferase plasmid into HCT-116 cells previously transfected either with scrambled siRNA (siSCR, in purple) or anti-APC siRNA (siAPC, in red). The negative control (FOPFLASH, green arrow) and positive control (TOPFLASH, red arrow) indicated that the anti-APC siRNA successfully upregulated canonical WNT signaling, while the constructs exhibited minimal changes in transcription following APC silencing.

Figure 9: TCF7L2 and AP-1 binding sites preferentially occur in the highest- confidence subset of peaks.

Motif analysis was repeated using only the highest-confidence versions of 6 transcription factor binding sites of interest (panels A-F). The percentage of anti-APC ChIP-seq peaks containing either zero (blue) or at least one (red) transcription factor binding site was examined in seven different subgroups of peaks. The highest confidence peaks are enriched for the most important transcription factor binding sites. This trend was observed among all six transcription factor binding sites, but was strongest for the TCF7L2 (A) and AP-1 (B) binding sites, and to a lesser extent for USF1/2 (C). The subset of peaks located near known APC-sensitive genes (right column) was defined by the data in Figure 4 and also showed a high co-occurrence of TCF7L2 and AP-1 binding sites. 34

Motif analysis data for anti-APC ChIP are similar to published data for anti-β-catenin and anti-TCF7L2 ChIP.

Our motif analysis data were consistent with previous reports that APC associates with chromatin through its interaction with β-catenin [58], which in turn interacts with the transcription factor TCF7L2. Enrichment of the AP-1 motif was also consistent with the literature, as canonical WNT signaling shares transcriptional targets with the AP-1 signaling pathway [149, 150]. Previous ChIP-seq experiments targeting either β-catenin

[129] or TCF7L2 [130] have similarly found enrichment of AP-1 binding motifs.

Published studies of TCF7L2 [130] and AP-1 [151] binding sites separately report relatively poor association with promoters and greater association with more distant enhancer regions further from transcription start sites. These observations are consistent with the broad genomic distribution patterns observed for APC in this study (Fig. 10).

The AP-1 component c-Jun physically interacts with TCF7L2 in HCT-116 cells, in which they function together to regulate the JUN promoter in a β-catenin-dependent manner

[149]. Finally, AP-1 components coimmunprecipitate with β-catenin and co-regulate expression of the TCF7L2 target genes MYC and CCND1 [150].

Figure 10: Anti-APC peaks exhibit broad genomic distribution rather than close association with promoter elements.

α-APC peaks exhibited a broad genomic distribution, with greater than 50% of peaks located more than 5,000 bp from a transcription start site.

TCF7L2 and AP-1 binding sites frequently occur within the same peaks or near the same genes.

To test the hypothesis that TCF7L2 and AP-1 binding sites might collaborate to regulate transcription, their co-occurrence was examined within a high-confidence subset of peaks

(Fig. 11). The subset of 918 α-APC peaks was selected by their location near 406 genes encoding transcripts that were demonstrated in this study to be sensitive to the presence/absence of APC in HCT-116 cells. Peaks associated with genes transcriptionally activated following APC loss (Fig. 11A, 11C) showed the highest

36 occurrence and co-occurrence with TCF7L2 and AP-1 binding sites. Peaks associated with genes transcriptionally repressed following APC loss contained fewer binding sites for both transcription factors. However, those peaks which contained TCF7L2 binding sites often contained AP-1 binding sites as well (Fig. 11B, 11D).

Figure 11: TCF7L2 and AP-1 binding sites co-occur in anti-APC ChIP-seq peaks located near genes whose transcription changes upon APC loss.

TCF7L2 and AP-1 occurrence in the subset of peaks located within 5000 bp of genes activated or repressed by the loss of APC. Many of these peaks contain TCF7L2 and/or AP-1 sites, especially those peaks associated with genes activated following APC loss (A compared to B). When multiple peaks for a single gene are grouped together (shown in C and D), the percentage of overlap between TCF7L2 and AP-1 binding sites is much greater. These data suggest that TCF7L2 and AP-1 binding sites are likely co-regulating the expression of these shared genes.

RNA-seq data identify 175 APC-responsive transcripts encoded by genes located near α-

APC peaks.

The genome-wide gene expression data referred to in the last column of Figure 9 were generated by RNA-seq from HCT-116 cells, either in the presence or absence of siRNA reducing APC expression (Fig. 12). These RNA-seq data were then used to filter ChIP- seq peaks in order to select individual targets for further study. 1,376 transcripts exhibiting significant sensitivity to the presence of APC were compared with a list of

APC ChIP-seq peaks located within 5000 bp upstream or downstream of transcription start sites. This limited subset of APC ChIP-seq peaks was chosen to accurately assign each predicted transcriptional regulatory element to the appropriate target gene. The overlap of the ChIP-seq and RNA-seq datasets yielded a list of 175 APC-responsive mRNAs encoded by genomic loci closely associated with APC ChIP-seq peaks. The list of target genes includes well-characterized targets of canonical WNT signaling such as

AXIN2, as well as novel targets (Fig. 15) with limited connections to canonical WNT signaling or inclusion on long lists of candidate targets generated by genome-wide studies.

Mouse tumor RNA-seq identifies some APC-sensitive transcripts as targets of canonical

WNT signaling.

These genes of interest were further screened using RNA-seq data from two mouse models of colon tumorigenesis (Fig. 13): one expressing wild-type Apc and a degradation-resistant mutant β-catenin (as a result of treatment with azoxymethane and dextran sulfate sodium (AOM/DSS) [152], confirmed in Fig. 14) and the other expressing a mutant form of Apc (ApcMin/+). As a result, both tumors exhibit activated canonical

WNT signaling but differ in their Apc status. Values depicted are fold changes in 38 expression, normalized to adjacent non-tumor tissue from the appropriate mouse. Of the

175 transcripts of interest, 43 showed at least 1.5-fold change in expression in both

ApcMin/+ and AOM/DSS colon tumors relative to adjacent tissue. These represented candidate target genes of canonical WNT signaling, sensitive not only to the presence or absence of Apc but also to the presence of degradation-resistant β-catenin. Overall, the

RNA-seq data from Apc-wild-type and Apc-deficient mouse colon tumors show a high degree of similarity between the tumor types and in many cases match their human orthologs as well (examples seen in Fig. 12). 16 other transcripts of interest change in expression only in AOM/DSS tumors, while 25 transcripts change in expression specifically in ApcMin/+ tumors. These latter groups were considered to include potential

β-catenin-independent APC targets. Target selection (Fig. 15) ultimately included loci of interest showing aberrant transcription in adenomas from both ApcMin/+ and AOM/DSS- treated mice, as well as targets aberrantly transcribed only in ApcMin/+-derived adenomas.

These mouse data will be discussed in greater detail in Chapter 4.

Figure 12: RNA-seq from HCT-116 cells identified transcripts that change in expression upon silencing of APC.

More than 1,300 APC-sensitive transcripts were identified using whole transcriptome sequencing (RNA-seq) data from APC-wild-type (siSCR, in purple) and APC-deficient (siAPC, in red) HCT-116 cells. Depicted here are RNA-seq data for 22 of those transcripts that match anti-APC genomic peaks observed in the ChIP-seq dataset, as well as for 2 positive control transcripts (MYC and AXIN2) reported to be regulated by chromatin-associated APC. Expression values for each condition are normalized to corresponding values from a third condition in which HCT-116 cells were treated with transfection reagent only. Error bars are not included since only a single sequencing replicate was performed for each condition. Two positive control transcripts are indicated by red arrows, only one of which (AXIN2) exhibits the expected activation following APC silencing.

Figure 13: RNA-seq from two types of mouse colon adenoma identified transcripts that change in expression upon Apc loss of function or β-catenin gain of function.

RNA-seq data show a high degree of similarity between colon adenomas with wild-type Apc and activating mutations in the gene encoding β-catenin (from AOM/DSS-treated mice, in orange) and colon adenomas with biallelic Apc mutations (from ApcMin/+ mice, in red). Mouse transcripts are listed along the x-axis in the same order as their human counterparts in Fig. 12, with two gaps where mouse orthologs have not been identified. Many mouse transcripts were aberrantly regulated in the same direction as their human orthologs from Fig. 12. Expression values for each adenoma type have been normalized to corresponding values from adjacent colon tissue. Those marked with asterisks differ significantly in fold-change between the Apc-wild-type and Apc-deficient tumor types according to two-tailed Student’s T-test (p values of less than 0.05). Error bars represent standard deviation. Two positive control transcripts are indicated by red arrows, each exhibiting significant activation in adenomas from both AOM/DSS-treated and ApcMin/+ mice, as expected for targets activated by canonical WNT signaling.

Figure 14: Colon adenomas from AOM/DSS-treated mice encode a degradation- resistant mutant form of β-catenin.

Sanger sequencing confirms that colon adenomas from AOM/DSS-treated mice contain activating mutations in the gene (Ctnnb1) encoding β-catenin. The presence of this mutation strongly suggests that the adenomas are wild-type for Apc, as reported in a published study of the AOM/DSS model [152].

Figure 15: Target selection included peaks with and without predicted TCF7L2 binding sites.

24 α-APC genomic peaks were selected for further study based on their association with 22 genes of interest and 2 positive control genes (AXIN2 and MYC). A list of high- confidence predicted transcription factor binding sites within each peak is shown, reflecting a mixed group of TCF7L2-containing peaks and non-TCF7L2 peaks. Since many genes of interest are associated with multiple peaks, one peak was chosen based on lowest p-value and shortest distance to transcription start site. Red arrows indicate the two loci known to be transcriptionally regulated by chromatin-associated APC. Both contain predicted TCF7L2 binding sites.

Silencing β-catenin is not sufficient to distinguish β-catenin-dependent and -independent subsets of anti-APC peaks by ChIP-qPCR.

ChIP-qPCR experiments tested the hypothesis that APC interaction with β-catenin mediates its association with some but not all genomic peaks. qPCR primer sets were designed to amplify individual genomic peaks that included putative targets of canonical

WNT signaling as well as potential β-catenin-independent APC targets. ChIP was performed from HCT-116 cells transfected with either scrambled siRNA or siRNA silencing β-catenin, using antibodies either to β-catenin itself (Fig. 16) or to APC (Fig.

17). These data confirm that both anti-APC ChIP and anti-β-catenin ChIP enrich most of the genomic target sites of interest. Most targets exhibit reduced enrichment by anti-β- catenin ChIP following transfection with siRNA against β-catenin. However, the effect is less dramatic than anticipated. The effect of transfection with siRNA against β-catenin on the enrichment of targets by anti-APC ChIP is even more subtle. While the data are suggestive that at least some APC targets are β-catenin-dependent, the statistics are not sufficient for a definitive answer.

Figure 16: β-catenin silencing significantly abolishes β-catenin ChIP signal at most APC targets of interest.

15 anti-APC ChIP-seq peaks of interest were examined by ChIP-qPCR using an anti-β- catenin antibody. ChIP was performed from HCT-116 cells transfected with either scrambled siRNA (siSCR, in purple) or siRNA targeting β-catenin (siCTNNB1, in blue). α-satellite repeats (green arrow) were used as a negative control target, while AXIN2 intron 1 and MYC promoter (red arrows) were positive controls. Silencing β-catenin clearly reduced qPCR signal for most genomic targets from the anti-β-catenin ChIP. Error bars are based on standard deviation.

Figure 17: β-catenin silencing does not disrupt APC localization to targets sufficiently to distinguish β-catenin-dependent vs. β-catenin-independent anti-APC peaks.

15 anti-APC ChIP-seq peaks of interest were examined by ChIP-qPCR using an anti- APC antibody. ChIP was performed from HCT-116 cells transfected with either scrambled siRNA (siSCR, in purple) or siRNA targeting β-catenin (siCTNNB1, in blue). α-satellite repeats were used as a negative control target (green arrow), while AXIN2 intron 1 and MYC promoter were positive controls (red arrows). Silencing β-catenin had a more modest and variable effect on qPCR signal from the anti-APC ChIP. Signals from the AXIN2 intron 1 and GPRC5A enhancer were significantly reduced according to single-tailed Student’s T-test (with respective p-values of 0.017 and 0.044). Error bars are based on standard deviation.

Luciferase reporter assays screened genomic regions of interest for transcriptional response upon silencing of APC or β-catenin.

24 genomic peaks confirmed to be enriched by anti-APC immunoprecipitation were selected for further study (Fig. 15). This group included two positive control peaks known to be APC-associated and APC-responsive (AXIN2 intron 1 and MYC promoter).

Also included were novel candidate canonical WNT target genes and potential

TCF7L2/β-catenin-independent APC targets lacking predicted TCF7L2 binding sites

(Fig. 15). Each peak of interest (approximately 500–1,000 bp) was PCR-amplified and cloned into the pGL3-promoter firefly luciferase vector. These constructs were transfected into HCT-116 cells to measure their ability to drive luciferase expression either in the presence or absence of siRNA to APC. The AXIN2 intron 1 and MYC promoter served as positive controls. The pFOPFLASH and pTOPFLASH plasmids served as negative and positive controls, respectively, for responsiveness to canonical

WNT activation [47]. Normalization was performed to signal from a co-transfected

Renilla luciferase control, as well as to the signal from the empty parent vector. While many of the candidate luciferase constructs either failed to drive luciferase expression efficiently or showed only limited responsiveness to siRNA reducing the expression of

APC, several showed a strong response (Fig. 18) complemented by an opposite response to siRNA reducing the expression of β-catenin (Fig. 19). Complementary results from the two experiments were used to confirm regulation by canonical WNT signaling. The

WNT-activated PHLDB2 intron 1 and WNT-repressed MALL intron 1 were selected for further study.

Canonical WNT signaling activates a transcription-promoting genomic region from

PHLDB2 intron 1.

The 541-bp PHLDB2 intron 1 region was sufficient to activate luciferase transcription

(Figs. 18, 19). APC silencing by siRNA transfection increased luciferase activity from the PHLDB2 construct, indicating that this genomic region contributes to the activation of PHLDB2 transcription observed upon APC loss (expression data in Fig. 12). It is important to note that this effect is mediated not only by the loss of APC function as a direct regulator of β-catenin in the chromatin-like context of the luciferase construct, but also by the upregulation of β-catenin protein levels that typically follows the loss of cytoplasmic APC as a negative regulator of β-catenin stability. Conversely, silencing of

CTNNB1, which encodes β-catenin, decreased luciferase activity from the PHLDB2 construct, consistent with the conclusion that this region of interest acts a regulatory element activating PHLDB2 by the canonical WNT signaling pathway (Fig. 19).

APC loss in vitro and in vivo activates PHLDB2 transcription and represses MALL transcription.

Within the ChIP-seq data, the PHLDB2 and MALL gene loci are each associated with multiple genomic APC peaks. RNA-seq data indicate that the PHLDB2 gene exhibits the expression pattern of a stereotypical target of canonical WNT signaling, increasing in expression upon APC loss (Figs. 12, 13). Expression profiling of the mouse colon tumors confirmed that PHLDB2 increases in expression relative to adjacent non-tumor tissue, regardless of whether canonical WNT signaling is constitutively activated as a result of Apc loss of function or a gain of function point mutation of β-catenin (Fig. 13).

The MALL gene exhibits an opposite pattern with decreasing expression following APC loss in vitro and in vivo (Figs. 12, 13). These data match recent reports that canonical 48

WNT signaling activates the transcription of certain genes and simultaneously represses the transcription of others. Collectively, the results of our study suggest that chromatin- associated APC may function to reverse the effects of canonical WNT signaling on both activation and repression of targets.

Canonical WNT signaling represses a transcription-promoting genomic region from

MALL intron 1.

The 982-bp MALL intron 1 region is sufficient to activate luciferase transcription.

However, APC silencing by siRNA transfection decreased luciferase activity from the

MALL construct (Fig. 18). Consistent with this observation, silencing the expression of

β-catenin increased luciferase activity from the construct (Fig. 19). Both pieces of evidence indicate that this genomic region contributes to the activation of MALL following APC loss detected in the expression data (Figs. 12, 13). However, it is not possible to distinguish the extent to which this activation in the reporter assay was due to

APC downregulation of β-catenin levels through the cytoplasmic mechanism or APC removal of β-catenin from the chromatinized plasmid through the nuclear mechanism. It is likely that both cytoplasmic and chromatin-associated APC contribute to the MALL and

PHLDB2-related changes observed in the expression data and the reporter assays.

Figure 18: Luciferase reporter assays identified PHLDB2 intron 1 and MALL intron 1 as transcriptional elements that change in activity following APC silencing.

22 peaks of interest and 2 positive control peaks (AXIN2 intron 1 and MYC promoter, indicated by red arrows) were PCR-amplified and cloned into the pGL3-promoter firefly luciferase vector. These constructs were co-transfected with a renilla luciferase plasmid into HCT-116 cells previously transfected with either scrambled siRNA (siSCR, in purple) or siRNA targeting APC (siAPC, in red). Aside from the pTOPFLASH and MYC promoter positive controls (indicated by red arrows), three constructs were repressed following APC loss, while five constructs were activated following APC loss. pFOPFLASH and empty vector served as negative controls (indicated by green arrows). Errors bars are based on standard deviation, and p-values lower than 0.05 for 2-tailed Student’s T-test are indicated by asterisks.

Figure 19: Luciferase reporter assays identified PHLDB2 intron 1 and MALL intron 1 as transcriptional elements that change in activity following silencing of β- catenin expression.

In the reverse experiment, HCT-116 cells were transfected with either scrambled siRNA (siSCR, in purple) or siRNA targeting β-catenin (siCTNNB1, in blue) identifying 8 β- catenin-repressed constructs and 1 β-catenin-activated construct (aside from the AXIN2 intron 1 and pTOPFLASH controls, indicated by red arrows). Complementary results from Fig. 18 and this experiment were used to confirm regulation by canonical WNT signaling. The WNT-activated PHLDB2 intron 1 and WNT-repressed MALL intron 1 were selected for further study. pFOPFLASH and empty vector served as negative controls (indicated by green arrows) while the MYC promoter (indicated by a red arrow) was expected to serve as a positive control but showed little effect. Errors bars are based on standard deviation, and p-values lower than 0.05 for 2-tailed Student’s T-test are indicated by asterisks. 51

The activity and WNT-sensitivity of PHLDB2 intron 1 are abolished by mutation of two overlapping TCF7L2 binding sites, while the activity of MALL intron 1 is abolished by mutation of an AP-1 binding site.

APC-associated peaks within both the PHLDB2 and MALL genes contain putative

TCF7L2 binding motifs (Fig. 15). This suggests that APC-sensitivity is mediated at least in part through the TCF7L2 transcription factor and presumably their mutual interacting partner, β-catenin. Disruption by site-directed mutagenesis of both predicted TCF7L2 binding motifs within the intronic PHLDB2 region decreases the baseline luciferase activity of the construct, as well as its APC sensitivity (Fig. 20).

Mutation of the predicted TCF7L2 binding motif within the intronic MALL region does not significantly compromise either the high transcriptional activity or APC- responsiveness of the construct (Fig. 21). Subsequent mutations were introduced to disrupt four additional low-confidence TCF7L2 binding motifs within the intronic MALL region, each with limited impact on luciferase activity. Thus, the region is activated by

APC and repressed by β-catenin, but is insensitive so far to the mutation of its putative

TCF7L2 binding sites individually. Mutation of the AP-1 binding motif, in contrast, abrogates luciferase activity to near-baseline levels for the parent vector, abolishing APC sensitivity in the process.

Figure 20: PHLDB2 intron 1 activation by canonical WNT signaling is dependent upon two TCF7L2 binding sites.

Point mutations were generated within the PHLDB2 intron 1 luciferase construct to test whether transcriptional changes following silencing of APC or β-catenin were dependent on candidate transcription factor binding sites. Two overlapping high-confidence TCF7L2 binding sites (in yellow) were disrupted by mutation, as well as a low- confidence “near-SP1” binding site (in green). Transcriptional activity was measured in the presence of siRNA reducing expression of β-catenin (siβ-cat, in blue), scrambled siRNA (siSCR, in purple), or siRNA reducing expression of APC (siAPC, in red). Mutation of the two overlapping TCF7L2 binding sites abolished transcriptional activity as well as changes in transcription following silencing of APC or β-catenin expression.

Figure 21: Transcriptional activity of MALL intron 1 is dependent upon an intact AP-1 binding site.

Point mutations were generated within the MALL intron 1 luciferase construct to test whether transcriptional changes following silencing of APC or β-catenin were dependent on candidate transcription factor binding sites. High-confidence TCF7L2 (yellow) and AP-1 (red) binding sites were disrupted by mutation, as well as low-confidence “near- TCF7L2” (pink) binding sites. Transcriptional activity was measured in the presence of siRNA knocking down β-catenin (siβ-cat, in blue), scrambled siRNA (siSCR, in purple), or siRNA knocking down APC (siAPC, in red). Mutation of individual TCF7L2 and near-TCF7L2 binding sites did not significantly alter transcription, while mutation of the AP-1 binding site mutation abolishes transcriptional activity as well as changes following silencing of APC or β-catenin expression.

Discussion

Our results indicate that the chromatin-associated fraction of the APC tumor suppressor protein associates with genomic regions that are more numerous than anticipated and only partially identified. Analyses of these genomic regions detect recurring patterns that bear similarities to ChIP-seq studies of other known components of the canonical WNT signaling pathway. For example, more than half of these peaks are more than 5 kilobases away from transcription start sites, and binding sites for the TCF7L2 and AP-1 transcription factors are both clearly enriched among APC-associated regions. Both observations resemble published ChIP-seq results for β-catenin [129], a known co- regulator of TCF7L2 transcription factor complexes. These similarities are consistent with the known physical interaction between APC and β-catenin, the reported colocalization of both proteins to target sites in the chromatin fraction, and the interpretation that APC can be recruited to chromatin by its binding to β-catenin [58].

However, these findings do not conclusively answer other interesting questions such as whether β-catenin localizes to AP-1 sites independently of TCF7L2, or whether APC can be recruited to chromatin either by its known ability to interact with DNA [134] or by interactions with proteins other than β-catenin. The frequent co-occurrence of TCF7L2 and AP-1 binding sites within the same peaks or within neighboring peaks associated with the same gene make it especially difficult to distinguish their separate effects on

APC recruitment. This question is further complicated by the challenging nature of identifying functional transcription factor binding sites. Whether the genomic targets of

APC, β-catenin, and TCF7L2 completely overlap remains a question of ongoing interest and investigation.

A valuable outcome of this study is a list of high-confidence direct targets of APC, identified using a combination of mechanistic (ChIP-seq) and functional (RNA-seq) data. 55

While at least one other study has used a similar combined approach to generate an important list of WNT-activated target genes [130], the present study provides convincing evidence that canonical WNT signaling represses target genes in human colorectal cancers as well. This pattern contrasts with most well-characterized canonical

WNT targets but corresponds to a small group of targets found in other model systems such as Drosophila [153, 154], chick [155], mouse [71, 132] and even human melanocytes [133]. The results of this study argue that APC antagonizes canonical WNT signaling at repressed target genes and enhances their expression in the colorectal epithelium. Regardless of whether canonical WNT signaling activates or represses transcription of a particular gene, APC most likely exerts an antagonistic effect by mediating the removal of β-catenin (Fig. 22).

What determines whether a target gene is activated or repressed by canonical WNT signaling? The higher prevalence of TCF7L2 and AP-1 binding sites within the WNT- activated subset of genes (Fig. 11) suggests that there is a coordinate function of both factors on expression. The crippling effect of an AP-1 site mutation in the 1st intron of the MALL gene (Fig. 21) reveals that AP-1 is largely responsible for the high transcriptional activity of this regulatory element and that a more subtle effect of TCF7L2 might be exerted by modulating AP-1 binding or subsequent AP-1-dependent steps. If canonical WNT signaling exerts its transcriptional effects largely by modifying AP-1 activity on shared target genes, activated targets may be those where TCF7L2 promotes

AP-1 binding or coordinates long-distance interactions between AP-1 proteins from multiple binding sites at the same promoter (Fig. 22). This would be consistent with the observation that many target genes (including PHLDB2) have multiple associated peaks, some with either TCF7L2 or AP-1 binding sites, and some with both. These regulatory elements likely interact with one another and converge on the promoter to co-regulate 56 transcription. According to this model (Fig. 22), WNT-repressed targets might be those where TCF7L2 competes with AP-1 binding to adjacent sites or interferes with long- distance interactions between AP-1 proteins and their intended target promoters.

Figure 22: WNT-mediated transcriptional activation (A) and repression (B) involving TCF7L2, AP-1, β-catenin, and APC.

(A) Our data suggest models in which activation of canonical WNT signaling upregulates transcription of the PHLDB2 gene through a previously-characterized mechanism in which β-catenin (blue) binds to the TCF7L2 transcription factor (yellow), promoting DNA bending that brings transcriptional machinery (possibly including AP-1, in red) into closer association with the proximal promoter. APC (green) disrupts this association by mediating the removal of β-catenin from the complex. (B) Activation of canonical WNT signaling downregulates AP-1-dependent transcription of the MALL gene through a less-common mechanism in which β-catenin binds to the TCF7L2 transcription factor to disrupt the interaction of transcriptional machinery with the proximal promoter. APC may relieve this disruption by removing β-catenin from the complex. 58

The MALL gene was confirmed (Figs. 20, 21) as a direct target repressed by canonical

WNT signaling. These data imply that other targets with similar expression patterns are similarly regulated candidates. Their value as potential markers for prognosis or treatment may be of interest. MALL itself has recently been reported as a marker whose low levels of expression in colorectal cancer predict recurrence, metastasis, and poor outcome [156]. Future experiments will test the hypothesis that aberrant repression of

MALL occurs as early as the adenoma stage of tumorigenesis, persists into later-stage disease, and is associated with APC loss.

Finally, this study strongly suggests that targets of the canonical WNT signaling pathway should be stratified into several categories, based on criteria such as the direction or degree of their response to activation of the pathway. The high rates of co-occurrence of

TCF7L2 and AP-1 binding sites observed in ChIP-seq data further suggest that the presence or absence of AP-1 binding sites may be an additional criterion by which to sort target genes. Interestingly, the AP-1 component c-Jun is required for the full phenotype of Apc loss, as introduction of Jun mutations into the ApcMin/+ mouse model lead to lower polyp number, reduced polyp size, and longer life span [149]. Jun mutations do not exert the same effects on other mouse models of colon tumorigenesis, however [157]. The present study suggests that AP-1 binding sites make large contributions to the baseline activity of some of these transcriptional elements, such as the one found in the first intron of the MALL gene. In these cases, transcription of genes predominantly controlled by

AP-1 may simply be fine-tuned by changes in canonical WNT signaling. These observations suggest that modulation of the AP-1 pathway may be a potential approach to therapeutic intervention to reduce the contribution of shared TCF7L2 / AP-1 target genes to colorectal tumorigenesis.

Materials and Methods

Chromatin immunoprecipitation and next-generation sequencing

For each ChIP reaction, 6 x 106 HCT-116 cells were seeded in a 150-mm dish. After 24 hours, 60-70% confluent cells were crosslinked with 1% formaldehyde for 10 minutes before being quenched by the addition of 0.125M glycine and incubation for 2 minutes.

Cells were washed once and collected by scraping into Dulbecco’s PBS, and the pellet was snap frozen and stored at -80oC. The thawed pellet was later resuspended in 10 mM

HEPES pH 7.5, 10 mM EDTA, 0.5 mM EGTA, 0.75% Triton X-100, 1 mM PMSF with mammalian protease inhibitor cocktail (Sigma-Aldrich) and incubated 10 minutes at 4oC with rotation. The resulting pellet was resuspended in 10 mM HEPES pH 7.5, 200 mM

NaCl, 1 mM EDTA, 0.5 mM EGTA, 1 mM PMSF with mammalian protease inhibitor cocktail and incubated 10 minutes with rotation. The resulting pellet was resuspended in

1 mL of Lysis Buffer (25 mM Tris-HCl pH 7.5, 150 mM NaCl, 5 mM EDTA, 0.1% SDS,

1% Triton X-100, 0.5% Sodium Deoxycholate, 1 mM PMSF with mammalian protease inhibitor cocktail). Probe-based sonication was performed at 4oC by 30 pulses of 10- seconds each at 35% amplitude, spread out over a period of 40 minutes total. In order to further reduce the average fragment size to ~1kb, the resulting supernatant was first divided into 300 μL aliquots. Each aliquot was mixed with 100 μL of 2 M sucrose, 320 mM KCl, 80 mM HEPES pH 7.5, 14 mM ATP, 24 mM CaCl2 and 4 µL of 50 U/µL micrococcal nuclease (Affymetrix, Inc.) and incubated 15 minutes at room temperature, then quenched by mixing with 45 µL of 200 mM EDTA, 20 mM EGTA on ice. Samples were diluted by addition of an equal volume of 16.7 mM Tris-HCl pH 8, 167 mM NaCl, 60

1.2 mM EDTA, 0.01% SDS, 1.1% Triton X-100, 1 mM PMSF, with mammalian protease inhibitor cocktail (Sigma-Aldrich). Input material was pre-cleared by 2-hour incubation with 5 μg rabbit IgG antibody and 40 μL of pre-equilibrated Protein G Dynabead slurry

(Thermo Fisher Scientific) at 4oC with rotation. 100 μL of the resulting supernatant was saved as “pre-cleared input”, and the remainder was used for α-APC ChIP in combination with 10 μg of α-APC antibody (catalog # A300-981A, Bethyl Laboratories, Inc.). This reaction was incubated overnight at 4oC with rotation, then mixed with 40 μL of pre- equilibrated Protein G Dynabead slurry and incubated for 3 hours at 4oC with rotation.

Beads were collected by magnet and washed twice with RIPA wash buffer (50 mM Tris-

HCl pH 8.0, 150 mM NaCl, 1 mM EDTA, 0.1% SDS, 1% NP-40, 0.5% sodium deoxycholate, 1 mM PMSF, and mammalian protease inhibitor cocktail), twice with

High-Salt RIPA wash buffer (50 mM Tris-HCl pH 8.0, 500 mM NaCl, 1 mM EDTA,

0.1% SDS, 1% NP-40, 0.5% sodium deoxycholate, 1 mM PMSF, and mammalian protease inhibitor cocktail), twice with LiCl Wash Buffer (50 mM Tris-HCl pH 8.0, 250 mM LiCl, 1 mM EDTA, 1% NP-40, 0.5% sodium deoxycholate, 1 mM PMSF, and mammalian protease inhibitor cocktail), and twice with TE buffer (10 mM Tris-HCl pH

8.0, 1 mM EDTA, 1 mM PMSF, and mammalian protease inhibitor cocktail). All washes were performed with cold buffers and accompanied by 5 minute rotation at 4oC. Elution was performed by the addition of 210 μL room temperature Elution Buffer (50 mM Tris-

HCl pH 8, 10 mM EDTA, 1% SDS) to the washed beads, and 45 minute incubation at

65oC with occasional gentle resuspension of beads. The supernatant was isolated away from the beads, mixed with 16 μL of 5M NaCl, and incubated ~12 hours at 65oC to

61 reverse crosslinks. 210 μL TE was then added, while the pre-cleared input sample was thawed and combined with 300 μL TE. 8 μL of 10 mg/mL RNase A was mixed with each sample, followed by 2 hours incubation at 37oC. 4 μL of 25 mg/mL Proteinase K was then mixed with each sample, followed by 2 hours incubation at 55oC. Samples were twice extracted with phenol:chloroform:isoamyl alcohol (25:24:1), once extracted with chloroform, and then mixed with 1/10th volume sodium acetate pH 5.2 and 60 μg glycogen for ethanol precipitation. Yields from multiple parallel α-APC ChIP reactions were pooled to obtain the 10 ng required for ChIP-seq library preparation. Also,

QIAGEN’s PCR purification kit was used to clean up reactions prior to library preparation. Three separate biological replicates were performed. ChIP-seq library preparation was performed using the NEBNext ChIP-Seq Library Prep Master Mix Set for Illumina, using adaptors AD005 and AD019 (New England Biolabs) for input and

ChIP libraries, respectively. Following the adaptor ligation step, size selection was performed by E-gel (Life Technologies, Inc.) to obtain fragments ranging in size from

300-400 bp. Next-generation sequencing (50 bp, single-end) was performed by the OSU

Genomics Shared Resource using a HiSeq 2500 instrument (Illumina, Inc.).

ChIP-seq data analysis

The OSU Bioinformatics Shared Resource performed sequencing data analysis for 3

ChIP-seq samples and their respective inputs. It was confirmed that 97-98% of reads passed quality control filters, and the Burrows-Wheeler Aligner was used to generate bam alignment files using UCSC hg19. The Model-based Analysis of ChIP-Seq

(MACS2) tool [158, 159] was used to perform peak-calling and generate peak scores, p- values and false discovery rates (%) for each assigned peak. The peak score threshold was adjusted to lower the noise observed in input files. The Refgene database was used to calculate distances to transcription start sites and assign peaks to genes whose transcription start sites were located within 5000 bp upstream or downstream. Peaks within genomic regions were visualized by uploading .wig files into the Integrative

Genomics Viewer software [160, 161]. R software was used to identify overlapping peaks present in multiple replicates, to generate scatter plots comparing the peak scores of these shared peaks in two replicates, and to calculate correlation coefficients to express the degree of similarity between replicates.

Motif analysis of ChIP-seq data

CisFinder [140], MEME-ChIP [141], Regulatory Sequence Analysis Tools (RSAT)

[142], and MatInspector [143] algorithms were each used with standard/default settings to perform initial analyses of all ChIP-seq peak sequences (minus strand) from FASTA files. MatInspector then performed analysis of a FASTA file containing only the

“TCF7L2-positive” subset of peaks confirmed to include at least one WWCAAAG motif

(or its reverse complement). FASTA files were then searched manually for the highest- confidence versions of each transcription factor binding site of interest, counting occurrences of each binding site across all ChIP-seq peaks and across smaller subsets of peaks generated by sequentially raising the p-value threshold for peak calling.

Whole transcriptome profiling of HCT-116 cells

HCT-116 cells were seeded in 6-well plates 16-20 hours in advance, at a concentration of

1.0 x 105 cells per well. Cells were transfected on consecutive days with pooled siRNA

(Dharmacon) targeting APC (L-003869-00-0005), scrambled sequence (non-targeting pool, D-001810-10-05), or no siRNA (mock transfection). Each transfection followed

Dharmacon’s instructions for the transfection reagent (Dharmafect 2, T-2002-01) and used 4 μL of Dharmafect 2 and 5 μL of 10 μM siRNA stock solution. Media was changed 6 hours after each transfection. Cells were harvested 48 hours after the second transfection by washing with Dulbecco’s PBS, addition of 500 μL Trizol (Life

Technologies catalog # 15596-026), and scraping. 50 μg glycogen (RNase-free) was mixed into each sample. A standard Trizol isolation protocol was then performed [162], with final resuspension in nuclease-free water. Single-read library preparation was performed using the TruSeq RNA Library Preparation Kit (Illumina catalog # RS-122-

2001). Libraries were quantified by Agilent Bioanalyzer and next-generation sequencing was performed by the OSU Genomics Shared Resource using a HiSeq 2500 instrument

(Illumina, Inc.). Initial data processing was performed by the OSU Bioinformatics

Shared Resource. Data analysis used the CuffLinks software program to assign reads to transcripts, perform quantification and calculate statistical significance [163]. Transcript levels were expressed in FPKM units and p-values and q-values were calculated based on fold changes. Differences between the anti-APC siRNA condition and either the scrambled siRNA or mock-transfected conditions were considered statistically significant for q values less than 0.05.

Whole transcriptome profiling of ApcMin/+ and AOM/DSS mouse colon tumors

Whole mouse colons were harvested from sacrificed ApcMin/+ and azoxymethane- and dextran sulfate sodium-treated (AOM/DSS) mice (all from the C57BL/6 background).

Tissues were immediately stored in RNAlater solution (Thermo Fisher Scientific, catalog

# AM7020) to protect RNA quality. Tumors were excised from the colons of 3 ApcMin/+ mice, and an adjacent sample of “normal” colon tissue from was obtained from 1 of them. 3 colonic tumors were similarly obtained from mice treated with AOM/DSS, and an adjacent sample of AOM/DSS-treated colon tissue was obtained from 1 of them.

Tissues were washed twice in Dulbecco’s PBS to remove residual RNAlater solution and then submerged in 500 μL each of Trizol (Life Technologies catalog # 15596-026).

Tissues were homogenized into Trizol using an OMNI TH tissue homogenizer at room temperature over a period of 1-2 minutes each, until the majority of tissue fragments were no longer visible. 50 μg glycogen (RNase-free) was mixed into each sample. A standard

Trizol isolation protocol was then performed [162], with the following modification:

Because the mixture appeared cloudy following the addition of isopropanol (due to the precipitation of residual salts from the RNAlater solution), a 50% mixture of nuclease free water and isopropanol was added until the solution became clear. RNA was finally resuspended in nuclease-free water. Preparation of 8 single-read libraries (3 for ApcMin/+ tumors, 3 for AOM/DSS tumors, and 1 each for ApcMin/+ and AOM/DSS adjacent tissues) was performed as described previously, using the TruSeq RNA Library Preparation Kit

(Illumina catalog # RS-122-2001). Next-generation sequencing was performed by the

OSU Genomics Shared Resource using a HiSeq 2500 instrument (Illumina, Inc.). Data

65 processing and analysis were performed by the OSU Bioinformatics Shared Resource, with quantification of transcripts expressed in FPKM units. Data from each tumor were normalized to its corresponding adjacent tissue. Differentially expressed transcripts between Apc-deficient (ApcMin/+) and Apc-wild-type (AOM/DSS) colon tumors were identified by using a two-tailed Student’s T-test to compare the three tumors of each type.

AOM/DSS tumors were confirmed to express degradation-resistant β-catenin as previously reported [152], strongly implying that they were likely Apc-wild-type. This was done by first reverse-transcribing an RNA sample from each AOM/DSS tumor and from the adjacent colon sample using SuperScript II reverse transcriptase (Thermo Fisher

Scientific, catalog # 18064022). The Ctnnb1 gene was then PCR amplified using the primers Ctnnb1S (GCTGACCTGATGGAGTTGGA) and Ctnnb1AS

(GCTACTTGCTCTTGCGTGAA), and Sanger-sequenced from each side using the same primers to examine codons 32 through 41. The point mutations observed were all consistent with the previous report [152].

ChIP-qPCR with silencing of β-catenin expression

Chromatin immunoprecipitation with α-APC was performed as previously described, while ChIP for β-catenin was similar except for the substitution of 10 μL of rabbit α-β- catenin antibody (Cell Signaling Technology catalog # 8480S). Cleanup of ChIP reactions was performed using the QIAGEN PCR purification kit, and 0.5-2 μL of the 50

μL elution volume was sufficient for qPCR amplification using most primer sets of interest. Approved primer sets had each been previously optimized to confirm that CT

66 values of 30 cycles or fewer could be obtained within a linear range and with a single observed melting temperature. This volume of template was adjusted to a total volume of

6 μL using nuclease-free water, combined with 4 μL total of diluted qPCR primers

(please see Fig. 26 for sequences and stock concentrations), and mixed with 10 μL of 2X

Power SYBR Green PCR master mix (Thermo Fisher Scientific catalog # 4367659). For each ChIP reaction, corresponding input material was purified in parallel and analyzed by qPCR using 5-20 ng of purified input DNA as template. qPCR reactions were performed by BIORAD iCycler Real-Time PCR Detection System. qPCR reactions were performed as zthree technical replicates, while measurements from three independent biological replicates were statistically analyzed by one-tailed Student’s T-test.

Generation of luciferase constructs by direct ligation, PCR-based cloning and PCR mutagenesis

Luciferase Reporter constructs for transcription factor binding sites of interest were generated by annealing the complementary primer sets (Fig. 23), ligating them directly

(using sticky ends on the antisense primer) into the Kpn I- Bgl II-cut pGL3-promoter vector, and confirming the desired insertion by Sanger sequencing. Each primer set contains three consecutive repeats of the binding site of interest, separated by 7-14 bp spacing linkers similar to those found within the positive control pTOPFLASH construct

[47].

Genomic regions of interest were PCR amplified using Pfu Ultra II Fusion HS polymerase (Agilent Technologies), column-purified using the QIAGEN PCR

67 purification kit and digested simultaneously with the appropriate pair of restriction enzymes. Following a second round of column purification, products were ligated into the compatibly cut pGL3-promoter vector. Restriction enzymes pairs varied depending on the target, with Kpn I and Bgl II used for the majority of clones. Primer sets are listed in Fig. 24 with asterisks indicating those clones generated using different restriction enzyme pairs (one asterisk for Sac I / Bgl II, two for Kpn I / Xho I, or three for Sac I /

Xho I). Genomic regions of interest were cloned into the multiple cloning site upstream of the firefly luciferase gene regardless of whether they were originally located in promoters, enhancers, or first introns.

PCR mutagenesis was performed using Pfu Ultra II Fusion HS polymerase, 20 cycles of amplification and the mutagenic PCR primer sets listed in Fig. 25. Briefly, the

WWCAAAG motifs predicted to be bound by TCF7L2 were mutated to GCCAAAG.

Near-TCF7L2 binding sites conforming to the WWVWAAR motif were mutated to

GCVWAAR. TGASTCA motifs predicted to be bound by AP-1 were mutated to

TCASTGA. KGGGCGGRRY motifs predicted to be bound by SP1 were mutated to

KGTTCGGRRY.

Luciferase Reporter Assays

Sequence-perfect luciferase constructs were prepared at levels of purity compatible with cell culture by QIAprep spin midiprep kit (QIAGEN catalog # 27104). HCT-116 cells were seeded in 6-well plates 16-20 hours in advance, at a concentration of 1.5 x 105 cells per well. Cells were transfected on consecutive days with pooled siRNA (Dharmacon)

68 targeting APC (L-003869-00-0005), targeting CTNNB1 (encoding β-catenin, L-003482-

00-0010), or with scrambled sequence (non-targeting, D-001810-10-05). Each transfection followed Dharmacon’s instructions for the transfection reagent (Dharmafect

2, T-2002-01) and used 4 μL of Dharmafect 2 and 5 μL of 10 μM siRNA stock solution.

Media was changed 6 hours after each transfection. 12 hours after the second siRNA transfection, cells were trypsinized and re-seeded into 96-well plates (black with clear bottom) at a density of 25,000 cells per well. Cells were co-transfected 12-hours post- plating with a firefly luciferase construct (9 ng per well) and a renilla luciferase construct

(1 ng per well) using 0.5 μL of lipofectamine 2000 per well. Media was changed 6 hours post-transfection. Cell lysis to initiate the luciferase assay took place 24 hours after this final transfection, and coincided with a previously-optimized time point 48 hours after the second siRNA transfection, when the respective expression levels of APC or β- catenin proteins are at their lowest. Luciferase assays were performed by following the instructions of the Promega Dual-Luciferase Reporter Assay System (catalog # E1910), with measurements made on a Veritas luminometer. Raw data were processed by determining the ratio of firefly luciferase signal to renilla luciferase signal for each well, and normalizing to the average firefly:renilla ratio across all wells transfected with empty vector (pGL3-promoter). Each firefly luciferase construct of interest was measured in three independent experiments and two-tailed Student’s T-test was used to calculate p- values to test the hypothesis that each construct was APC-sensitive or β-catenin-sensitive, relative to its scrambled siRNA control.

Figure 23: Primer Sets for Luciferase Cloning (Data in Fig. 8)

Figure 24: Primer Sets for Luciferase Cloning (Data in Figs. 18, 19)

no asterisks = cloned using Kpn I / Bgl II * = cloned using Sac I / Bgl II ** = cloned using Kpn I / Xho I *** = cloned using Sac I / Xho I 71

Figure 25: Mutagenic Primers for Luciferase Constructs (Data in Figs. 20, 21)

Figure 26: ChIP qPCR Primer Sets (Data in Fig. 16, 17)

Chapter 4: Mouse Adenoma Studies

Introduction

The advent of personalized medicine has motivated the increasing stratification of colorectal tumors into subtypes distinguished by molecular characteristics and associated with different prognoses. Activating mutations in the KRAS2 gene are clearly associated with poor prognosis [124], while the relationship between the biallelic APC mutations and outcome is less clear. At least one study has failed to detect significant differences in prognosis between APC-deficient and APC-wild-type colorectal tumors [16]. Some debate has arisen over whether biallelic APC mutations are equivalent to gain-of-function mutations in the CTNNB1 gene encoding β-catenin, as the two types of mutations occur mutually exclusively and both activate canonical WNT signaling. APC loss additionally contributes to colorectal tumorigenesis through a variety of WNT-independent mechanisms. Mutational data show that as smaller adenomas progress to larger adenomas and adenocarcinomas, CTNNB1 mutations become less prevalent while APC mutations remain well-represented [17]. This difference may reflect the contribution of

WNT-independent APC functions to the prevention of tumor progression. Nuclear APC

74 and chromatin-associated APC protein clearly interact with β-catenin to inhibit canonical

WNT signaling, but may mediate WNT-independent transcriptional changes as well.

Previous transcriptional profiling studies of human and murine colorectal tumors have assumed the equivalence of APC and CTNNB1 mutations in their effects on gene transcription, and subtle differences between APC and CTNNB1 mutant colorectal tumors have not yet been examined specifically. We and others have previously reported a transcriptional profiling of adenomas from four murine models of intestinal cancer, which focused on the shared characteristics of activated canonical WNT signaling in colon adenomas from ApcMin/+ and azoxymethane- and dextran sulfate sodium- (AOM/DSS) treated mice versus colon adenomas from models with aberrant Smad / TGFβ1 signaling

[128, 173].

This study examines the hypothesis that chromatin-associated APC mediates some transcriptional changes independently of canonical WNT signaling and that those differences in gene expression can be identified by a comparison of transcriptional profiles of mouse colon adenomas bearing either biallelic Apc mutations (from ApcMin/+ mice) or an activating Ctnnb1 mutation (from mice treated with AOM/DSS). Whole transcriptome profiling (RNA-seq) of three colon adenomas from each mouse model was performed and normalized to adjacent non-tumor colon tissues. A list of up- and down- regulated transcripts was generated for each type of adenoma. Transcriptional changes in

Apc-deficient adenomas exhibit significant overlap with their counterparts in AOM/DSS adenomas, but also include unique aberrations particularly related to the regulation of apoptosis and the cytoskeleton.

Results

Gene expression profiling data were generated by next-generation sequencing of cDNA libraries from mouse colon adenomas with either biallelic mutations of Apc (from

ApcMin/+ mice) or Ctnnb1 mutations that render β-catenin resistant to degradation (from

Apc+/+ mice treated with AOM/DSS). Mutations in Ctnnb1 were confirmed by PCR and

Sanger sequencing (Fig. 12) consistent with previous studies of the AOM/DSS model of colon tumorigenesis [152]. Both types of mouse adenomas were characterized by transcriptional evidence of activated canonical WNT signaling. while they would differ in the expression of WNT-independent targets of Apc protein in the chromatin fraction.

RNA-seq quantified transcript levels in colon adenomas from ApcMin/+ and AOM/DSS- treated mice and fold-changes were calculated for each type relative to adjacent colon tissue from the same mouse model. Transcripts that changed by at least 2-fold were compiled into a list. Venn diagrams (Fig. 27) allowed comparison of the lists from the

AOM/DSS-derived and ApcMin/+-derived adenomas. Transcripts increased in expression in adenomas are represented in Fig. 27A while those decreased in expression in adenomas are represented in Fig. 27B.

Figure 27: Colon adenomas from AOM/DSS mice (C57BL/6) and ApcMin/+ mice (C57BL/6) exhibit significant overlap in their transcriptional profiles.

RNA-seq analysis was used to quantify transcript levels in colon adenomas from ApcMin/+ and AOM/DSS-treated mice. Fold-changes were calculated for each adenoma type relative to adjacent colon tissue from the same mouse model. Venn diagrams were generated by comparing two lists of transcripts that changed significantly by at least 2- fold in each adenoma type. Shared changes were identified between transcripts that increased (A) in each adenoma type and that decreased (B) in each adenoma type. Significant overlap (in purple) is observed between the transcriptional profiles of colon adenomas from AOM/DSS mice (in red) and from ApcMin/+ mice (in blue).

Transcriptional changes unique to adenomas from ApcMin/+ mice and to adenomas from

AOM/DSS-treated mice were identified. Non-overlapping transcripts unique to adenomas derived from ApcMin/+ mice (represented by the areas of Fig. 27 in blue) were subjected to more rigorous removal of candidates that changed similarly in adenomas from AOM/DSS mice, albeit with less dramatic fold-changes. Transcripts that changed by at least 2-fold in adenomas from ApcMin/+ mice and by 1.5- to 2-fold in the same direction in AOM/DSS-derived adenomas were ruled out. The removal of these partially overlapping candidates yielded 763 transcripts increased and 502 transcripts decreased only in ApcMin/+-derived adenomas. These 1265 transcripts were subjected to gene ontology using Ingenuity Pathway Analysis software (Fig. 28) from QIAGEN. For comparison, 961 transcripts that changed at least 2-fold in adenomas from AOM/DSS mice but less than 1.5-fold in adenomas from ApcMin/+ mice were identified and subjected to pathway analysis (Fig. 29). As an additional control, 1019 transcripts that increased by at least 2-fold in both adenoma types and 728 transcripts that decreased by at least 2-fold in both adenoma types were subjected to pathway analysis (Fig. 30).

The results in Figure 30 highlight the similarity in transcriptional changes in the two types of adenomas, in that functions associated with canonical WNT signaling

(highlighted in yellow) stand out from the pathway analysis targets shared by both types.

Both colon adenomas are characterized by activation of canonical WNT signaling and transcriptional changes are associated with epithelial cell proliferation, differentiation and apoptosis.

Our data (Figure 29) also highlight a key characteristic of the adenomas that form in the

AOM/DSS-treated model as these tumors are driven by inflammation induced by dextran sulfate sodium. The ApcMin/+ model of tumorigenesis is not inflammation-driven.

Inflammation-related pathways (highlighted in gray) dominate the gene ontology of the

AOM/DSS-specific transcript list (Fig. 29) and are reflected to some extent in the pathway analysis of the Apc-deficient adenomas (Fig. 28). Differences in immune infiltration and inflammation may be associated with some of the transcriptional changes that distinguish the two adenoma types (Fig. 29).

Adenomas from ApcMin/+ mice (Fig. 28) exhibit unique transcriptional changes consistent with non-transcriptional tumor suppressor functions of the Apc protein. Apc promotes apoptosis both indirectly (by negatively regulating the expression of pro-survival canonical WNT signaling targets such as survivin [104]) and directly (by localizing to the mitochondria and promoting caspase activity [105-107]). Additionally, Apc drives the formation of membrane protrusions by its interaction with cytoskeletal components including microtubules [84] and by its ability to bind key RNA molecules and promote their localization to cell protrusions where they are subsequently translated [85]. This latter mechanism influences cell migration [85]. The aberrant expression of transcripts involved in regulation of apoptosis and the formation of cell protrusions in adenomas from ApcMin/+ mice suggests that these two pathways may become dysregulated in Apc- deficient adenomas by mechanisms both dependent on and independent of canonical

WNT signaling.

Figure 28: Colon adenomas from ApcMin/+ mice exhibit unique transcriptional changes related to apoptosis and cell protrusion pathways.

763 transcripts increased and 502 transcripts decreased only in ApcMin/+-derived adenomas. These were combined into a single list and subjected to gene ontology analysis using Ingenuity Pathway Analysis software. Pathways with the greatest statistical significance generally segregated into those linked with Apc tumor suppressor function in previous reports (highlighted in yellow) and those linked with immune cell infiltration and inflammation (highlighted in gray). Those linked to Apc function are involved in regulating apoptosis and the formation of cell protrusions.

Figure 29: Colon adenomas from AOM/DSS-treated mice exhibit unique transcriptional changes related to immune infiltration and inflammatory pathways.

961 transcripts that changed at least 2-fold in adenomas from AOM/DSS mice but less than 1.5-fold in adenomas from ApcMin/+ mice were identified and subjected to pathway analysis. Transcriptional changes unique to adenomas from AOM/DSS mice were associated with inflammatory pathways (indicated by gray highlighting) and were not associated with known functions of Apc (indicated by yellow highlighting). The latter group were largely absent from transcriptional changes unique to adenomas from AOM/DSS mice, at least in comparison to more statistically significant trends related to inflammation.

Figure 30: Colon adenomas from ApcMin/+ and AOM/DSS-treated mice both exhibit transcriptional changes linked to pathways associated with Apc tumor suppressor functions.

1019 transcripts increased by at least 2-fold in both adenoma types and 728 transcripts decreased by at least 2-fold in both adenoma types were subjected to pathway analysis. Transcriptional changes shared by both adenomas were significantly associated with pathways linked to known Apc tumor ssuppressor functions (highlighted in yellow) and to a lesser extent with pathways involved in immune cell infiltration and inflammation (highlighted in gray).

Two panels of representative transcripts summarize differences in gene expression between adenomas from ApcMin/+ and AOM/DSS mice in the pathways regulating apoptosis (Fig. 31) and the formation of cell protrusions (Fig. 32). The canonical WNT signaling target gene Myc was included in both panels as a control that confirms that the canonical WNT pathway is similarly activated in both types of adenoma. These panels do not depict other known targets of canonical WNT signaling, however four known

WNT targets are present on the lists implicated by the pathway analysis: Fn1, Birc5, Fos and Yap1. Transcripts linked to apoptosis (Fig. 31) are mostly anti-apoptotic regulators such as Ereg [174] and Fasn [175] increased in colon adenomas from ApcMin/+ mice relative to adjacent colon tissue, but also include some decreased pro-apoptotic regulators such as Slc20A1 [176]. Most of the transcripts linked to the formation of cell protrusions in Apc-deficient colon adenomas are increased relative to the adjacent colon tissue, indicating that chromatin-associated Apc may function as a transcriptional repressor whose loss promotes their transcription.

Figure 31: Colon adenomas from ApcMin/+ mice increase expression of anti- apoptotic transcripts and decrease pro-apoptotic transcripts.

Twelve representative transcripts were selected from among those aberrantly expressed in adenomas from ApcMin/+ mice based on their involvement in the regulation of apoptosis and the detection of APC-associated genomic peaks near their transcription start sites (by ChIP-seq in human colon cancer cells reported in Chapter 3). Expression is shown in each type of mouse adenoma (in orange and red), normalized to their respective adjacent colon tissue controls (in blue and purple). Error bars depict standard deviation and are shown only for the three adenomas, as adjacent colon tissue controls were analyzed as single samples without replicates. The Student’s T-test was performed to assess statistically significant differences in fold-change between adenomas from ApcMin/+ mice and adenomas from AOM/DSS mice. All comparisons are statistically significant except for Myc and Ereg. The Myc transcript (indicated by a green arrow) is included as a negative control similarly changed in both adenoma types.

Figure 32: Colon adenomas from ApcMin/+ mice exhibit increases and decreases of transcripts related to the formation of cell protrusions.

Fourteen representative transcripts were selected from among those aberrantly expressed in adenomas from ApcMin/+ mice based on their involvement in regulating the formation of cell protrusions and the detection of APC-associated genomic peaks near their transcription start sites (by ChIP-seq in human colon cancer cells reported in Chapter 3). Expression is shown in each type of mouse adenoma (in orange and red), normalized to their respective adjacent colon tissue controls (in blue and purple). Error bars depict standard deviation and are shown only for the three adenomas, as adjacent colon tissue controls were analyzed as single samples without replicates. The Student’s T-test was performed to assess statistically significant differences in fold-change between adenomas from ApcMin/+ mice and adenomas from AOM/DSS mice. All comparisons are statistically significant except for Myc, Srgap2, Mid1 and Reln. The Myc transcript (indicated by a green arrow) is included as a negative control similarly changed in both adenoma types.

Discussion

This gene expression analysis of colon adenomas from two mouse models of WNT- activated intestinal tumorigensis, one Apc-deficient and one β-catenin-mutated, has identified 1,747 transcripts (1,019 increased and 728 decreased) that are similarly dysregulated in both types of adenoma. These include numerous known targets of canonical WNT signaling as well as novel direct and indirect targets of the pathway. The high degree of overlap confirms previous reports that both adenoma models exhibit dramatic transcriptional changes due to activation of the canonical WNT signaling pathway [128, 173]. The utility of these data in the process of identifying and characterizing novel WNT targets was briefly discussed in Chapter 3 and demonstrated in

Figure 13.

The known canonical WNT signaling target gene Myc, included in Figures 31 and 32 as a control, is representative of the largest trend of targets in the canonical WNT pathway similarly increased in expression in both types of adenoma. Similarities between both adenoma types in the aberrant expression of other known targets of canonical WNT signaling are summarized in Figures 33 and 34. Their expression is comparable between adenomas from ApcMin/+ and AOM/DSS-treated mice, indicating that the presence or absence of chromatin-associated Apc does not lead to meaningful differences in expression levels of canonical WNT signaling target genes between adenomas with Apc deficiency (those from ApcMin/+ mice) and those with degradation-resistant β-catenin

(those from AOM/DSS-treated mice).

This gene expression analysis also identified 763 transcripts that increased and 502 transcripts that decreased in expression only in ApcMin/+-derived adenomas. Some of these targets are associated with pathways of immune infiltration and inflammation, reflecting the inflammation-driven mechanism in adenomas from the AOM/DSS mice.

Other transcripts aberrantly expressed in ApcMin/+-derived adenomas are associated with regulation of apoptosis and the formation of cell protrusions, two pathways with well- characterized links to Apc function and relevance to tumorigenesis. We hypothesize that some of these transcripts include some WNT-independent targets regulated by chromatin- associated Apc protein.

Apc may control the abundance of some of these RNAs independently of transcription, especially in the context of cell protrusions where Apc localizes and functions as an

RNA-binding protein [85]. In a mouse fibroblast cell line, Apc promotes the localization of key RNA molecules to cell protrusions where they are translated as a part of a mechanism that controls migration [85]. Of five transcripts that physically associate with

Apc [85], four do not show a difference in expression in our adenoma data relative to normal tissues (Rab13, Pkp4, Ankrd25 and Inpp1, data not shown), while the Kank2 transcript is repressed only in AOM/DSS-derived adenomas. These data collectively do not support the conclusion that the RNA-binding function of APC is mediating changes in transcript abundance in these adenomas. The presence of APC-associated genomic sites near some of the relevant transcription start sites (detected by ChIP-seq studies presented in Chapter 3) suggests that Apc contributes to the formation of cell protrusions

87 and to cell migration through a chromatin-associated mechanism in addition to its published transcription-independent role.

Transcripts aberrantly expressed only in adenomas from ApcMin/+ mice include a number of novel targets previously linked to apoptosis (Fig. 31). APC-associated genomic peaks near their respective transcription start sites (from ChIP-seq studies presented in Chapter

3) suggests that chromatin-associated Apc may directly control their transcription, with most examples suggesting Apc-mediated repression. Future experiments will determine the extent to which the effect of Apc on the abundance of these transcripts is due to

WNT-independent as opposed to WNT-dependent mechanisms. The list of transcripts aberrantly expressed only in ApcMin/+ adenomas includes Birc5, known to be repressed by

Apc through negative regulation of canonical WNT signaling [104]. This indicates that some transcriptional targets of canonical WNT signaling are among the transcripts aberrantly expressed only in adenomas from ApcMin/+ mice. Subsequent studies in colon cancer cell lines will be required to better characterize the mechanism(s) by which Apc regulates these novel transcripts involved in apoptosis as well as those functioning in the formation of cell protrusions.

Chapter 5: Thesis Summary and Future Directions

The findings in this thesis support several interesting conclusions and suggest a number of future investigations that will be discussed in this chapter. Further experiments fall into three general categories, the first of which will determine the degree of correlation between clinical outcomes and expression levels of novel canonical WNT target genes from this study. Another set of experiments will seek to further characterize the mechanistic differences that distinguish WNT-activated from WNT-repressed targets.

Finally, we plan to revisit one of our original hypotheses by characterizing the APC- associated genomic elements that control the transcription of several candidate WNT- independent target genes.

Correlating low expression of WNT-repressed targets to colorectal cancer prognosis.

The characterization of a transcriptional element from the MALL locus demonstrates for the first time in a human colorectal cancer model that canonical WNT signaling directly represses transcription of certain targets. This mechanism of canonical WNT repression has been reported in several other model systems, and these new findings are relevant to other transcripts of interest previously reported to decrease upon activation of canonical

WNT signaling in colorectal tissues (such as E-cadherin [71]). For the vast majority of

89 these transcripts, direct repression by the pathway has been proposed but not demonstrated. Obvious links between such candidate transcripts and clinically relevant phenotypes like migration raise the point that future studies should test the correlation between low expression of these targets and colorectal cancer staging and prognosis.

Low MALL expression, for example, is correlated with recurrence, metastasis and poor survival in one cohort [156]. This example provides a proof of principle that canonical

WNT repression of targets following early APC loss can persist into later stages of disease. Recommended follow-up studies could use public datasets from the Cancer

Genome Atlas (TCGA) to observe the relationship between low MALL expression and stage, as well as its correlation with clinical variables including relapse and survival. The staining pattern of MALL protein in colorectal epithelial tissue sectioned along the colorectal crypt axis would also be of interest. MALL should exhibit an expression pattern similar to that of APC, which is absent from the proliferating / progenitor region near the base of the crypt, and forms a gradient of increasing expression in the lumenal direction that peaks among the differentiated epithelial cells [46]. This preliminary experiment would establish a basis for subsequent studies of MALL distribution within individual tumors (comparing its levels in the center versus at the leading edge) and among tumors of different stages.

Other candidate transcripts / loci from our datasets that respond similarly to changes in canonical WNT signaling (Fig. 33) will be subjected to in silico validation as well. We hypothesize that these transcripts are poorly expressed in tumors with APC mutations, and that their low levels persist into the later stages of disease. The correlation between

90 expression levels and disease stage would also enable us to identify candidates that might help predict degree of metastatic dissemination with greater accuracy than current methods of establishing staging. A long-term goal of these studies would be to develop a panel of similarly-regulated targets with low expression levels in colorectal cancer that collectively predict prognosis.

Figure 33: Candidate genes repressed by canonical WNT signaling

All transcripts identified in human cell line RNA-seq and mouse colon tumor RNA-seq decreasing in abundance following APC loss / activation of canonical WNT signaling (including MALL). These were selected from a larger group of candidates using the close association of one or more α-APC ChIP-seq peaks with their transcription start sites. Data from the HCT-116 human colon cancer cell line (from single replicates) are expressed as fold-changes relative to a mock-transfected negative control, while colon adenoma data (from three of each adenoma type) are expressed as fold-changes relative to adjacent tissue, plus or minus one standard deviation from the mean. Adenomas from ApcMin/+ mice generally exhibit an acquired loss-of-function mutation in the second allele of Apc, in addition to the inherited Min allele [15]. AOM/DSS colon adenomas are generated by treatment with Azoxymethane and Dextran Sulfate Sodium and exhibit activating mutations in the gene encoding β-catenin (Fig. 12).

Dissecting the mechanism of TCF7L2 repression of MALL-like targets activated by

AP-1.

Luciferase data from this project provide evidence that transcriptional activation of the

MALL locus is dependent upon a putative AP-1 binding motif (TGASTCA), while 92 activation of canonical WNT signaling counteracts this effect and represses MALL transcription. Our mutational analysis has not yet succeeded in identifying which

TCF7L2 binding site or combination of binding sites within the MALL locus mediates this latter effect. However, one explanation for our data would be that TCF7L2 binding to its target site(s) is facilitated by the earlier binding of the AP-1 transcription factor to an adjacent site. This would be consistent not only with the high degree to which transcriptional activity is dependent on the AP-1 site, but also with the high frequency of co-occurrence between TCF7L2 and AP-1 binding sites in APC-associated genomic regions. The fact that we have not yet identified the DNA basis of canonical WNT regulation of the MALL locus does not exclude the possibility that the collaboration works in reverse, with TCF7L2 binding to its target site(s) first and facilitating subsequent AP-1 binding to an adjacent site. This mechanism might be more consistent with reports that TCF family members are architectural transcription factors that function by bending DNA and facilitating binding opportunities for other factors [164]. In either case, the presence of β-catenin would disrupt AP-1 binding or activity. This hypothesis could be tested by chromatin immunoprecipitation of the AP-1 component c-Jun in the presence or absence of siRNA targeting β-catenin, followed by qPCR to look for enrichment of the shared MALL intron 1 target site. To test the hypothesis that either

TCF7L2 or AP-1 facilitates the binding of the other to an adjacent site, future experiments could include siRNA-based knockdown of either c-Jun or TCF7L2, followed by ChIP of the other protein and qPCR for the MALL intron 1 locus. These knockdown

93 experiments could also be followed by RNA isolation and qPCR to test the correlation between transcription factor binding and MALL expression.

TCF7L2 and AP-1 collaboration in the regulation of WNT-activated gene targets.

Putative binding sites for both TCF7L2 and AP-1 occur far more frequently within α-

APC peaks associated with WNT-activated genes than within peaks associated with

WNT-repressed genes. Our experiments with luciferase studies of the MALL intron 1 transcriptional element suggest a model of an antagonistic interaction between TCF7L2 and AP-1 pathways among WNT-repressed genes. However, the nature of TCF7L2 and

AP-1 collaboration in the much larger group of WNT-activated genes has not yet been investigated. The hypothesis that they act synergistically could be examined by cloning a new group of transcriptional elements (candidates shown in Fig. 34) into firefly luciferase expression vectors and examining their transcriptional activity in the presence / absence of siRNA targeting APC or β-catenin. Constructs are predicted to be more active upon

APC silencing and less active upon β-catenin silencing. Point mutations disrupting the

TCF7L2 binding site(s) would abolish the construct’s sensitivity to APC or β-catenin silencing; the effects of disrupting the AP-1 binding sites are more difficult to predict.

Examining AP-1 binding site mutants under conditions where APC or β-catenin is silenced would test whether TCF7L2 binding sites influence transcription independently of AP-1. TCF7L2 and AP-1 binding sites are predicted to exert additive effects on transcriptional activation, so that mutations abolishing either will diminish the baseline activity of the constructs. Chromatin immunoprecipitation experiments analyzed by

94 qPCR would also be useful to test whether APC association with transcriptional elements of interest is dependent upon the presence of TCF7L2 or AP-1 proteins. The order of

TCF7L2 and AP-1 recruitment could be examined by similar ChIP experiments in combination with siRNA silencing either TCF7L2, β-catenin or c-Jun. In comparison with a more complete analysis of the MALL locus, these experiments would help shed light on the mechanistic differences between canonical WNT activation versus repression of different target genes.

Figure 34: Candidate genes activated by canonical WNT signaling

All transcripts identified in human cell line RNA-seq and mouse colon tumor RNA-seq increasing in abundance following APC loss / activation of canonical WNT signaling (including AXIN2). These were selected from a larger group of candidates using the close association of one or more α-APC ChIP-seq peaks with their transcription start sites. Data from the HCT-116 human colon cancer cell line (from single replicates) are expressed as fold-changes relative to a mock-transfected negative control, while colon adenoma data (from three of each adenoma type) are expressed as fold-changes relative to adjacent tissue, plus or minus one standard deviation from the mean. Adenomas from ApcMin/+ mice generally exhibit an acquired loss-of-function mutation in the second allele of Apc, in addition to the inherited Min allele [15]. AOM/DSS colon adenomas are generated by treatment with Azoxymethane and Dextran Sulfate Sodium and exhibit activating mutations in the gene encoding β-catenin (Fig. 12). 96

Negative and positive feedback loops in the canonical WNT signaling pathway.

Surprising differences were exhibited by the three lists of genes identified (1) in the human cell line ChIP-seq, (2) in the human cell line RNA-seq and (3) in the mouse colon tumor RNA-seq, despite their common purpose to discover novel targets of APC / canonical WNT signaling. The human cell line RNA-seq and mouse colon tumor RNA- seq experiments in particular shared a relatively small degree of overlap between their two lists of APC-responsive transcripts. For example, while the ChIP-validated mouse colon tumor expression data consisted mostly of transcripts activated by Apc loss, the

ChIP-validated expression data from the human cell line consisted mostly of transcripts that were repressed nu APC loss. On the other hand, those transcripts identified in both experiments were generally changed in the same direction in response to APC loss, so that their overlap was not accidental. It can be concluded from these observations that each experimental model of APC function or APC loss has unique advantages and limitations, exposed by comparison to one another as well as to other published datasets.

One source of variability may be that human cancer cell lines like HCT-116 carry a huge number of mutations that come not only from the effects of in vitro culture, but also from their origins in an advanced tumor. Another source of variability may be that in the human cell model APC was being silenced in a cellular context that already featured constitutive activation of canonical WNT signaling due to a degradation-resistant mutant form of β-catenin [165]. According to the just-right hypothesis of canonical WNT signaling [166], one of the expected consequences of hyperactivating the pathway in an already-optimized context might be to induce negative feedback. This might explain why

97 we only observed an increase in the abundance of one of our positive control transcripts, the negative feedback mediator AXIN2 [167], while levels of the MYC transcript remained constant.

The idea of feedback also arises when considering the collaboration between the TCF7L2 and AP-1 transcription factors at shared target genes. The known targets of canonical

WNT signaling include the JUN and FOSL1 genes [168], each of which encodes a component of the dimeric AP-1 transcription factor. By activating transcription of these two genes, canonical WNT signaling could form a feedback loop that regulates shared targets of TCF7L2 and AP-1 by both direct and indirect mechanisms. Shared targets like

MALL that are repressed by β-catenin/TCF7L2 and activated by AP-1 would be directly repressed (by β-catenin/TCF7L2) but indirectly activated by canonical WNT signaling

(through upregulation of JUN and FOSL1). On the other hand, shared targets that are activated by both β-catenin/TCF7L2 and AP-1 would be positively controlled by the same feedback loop. Canonical WNT signaling would activate them not only directly through TCF7L2 but also indirectly through upregulation of JUN and FOSL1.

Collectively, the evidence from this study in support of positive and negative feedback loops suggests that the canonical WNT signaling pathway is heavily self-regulated and self-tuned, and that the idea of a binary WNT pathway with an active and an inactive state is an oversimplification. While colorectal cancer cells achieve the “just-right” level of canonical WNT signaling through careful selection of biallelic APC mutations [166], healthy colorectal progenitor cells may achieve this by fine-tuning based on negative and positive feedback mechanisms.

The genome-wide data collected here are consistent with a feedback loop by which canonical WNT signaling modulates the activity of its collaborator AP-1. α-APC peaks were observed in the promoters of the JUNB and FOS genes, and in the first introns of

FOSL1 and FOSL2. On the other hand, an enhancement of AP-1 activity was not observed following APC knockdown in HCT-116 cells (Fig. 8). In addition, none of the genes listed above exhibited a statistically significant change in expression following

APC knockdown (RNA-seq data not shown). To test whether modulation of canonical

WNT signaling can modify AP-1 levels / activity, the luciferase reporter vector containing multiple AP-1 binding sites could be co-transfected at different time points after knockdown of either the APC or β-catenin proteins. The expression of the JUN,

JUNB, FOS, FOSL1 and FOSL2 genes could also be measured by qPCR at these same time points.

Revisiting original hypotheses that have proven difficult to test.

The two key RNA-seq experiments performed at the beginning of this project were designed to identify the transcriptional changes that distinguish APC-deficient colorectal cancers from their counterparts with wild-type APC but activated canonical WNT signaling due to a degradation-resistant mutant form of β-catenin. In the process, we tested the hypothesis that APC loss (particularly from the chromatin fraction) results in both WNT-mediated and WNT-independent transcriptional changes. The mouse RNA- seq experiment did in fact identify a particularly promising list of transcriptional changes that distinguish APC-deficient adenomas from their APC-wild-type counterparts (Figs.

24, 27) in addition to longer lists of transcriptional changes shared by both sets of adenomas (Figs. 33, 34). The candidate transcripts aberrantly expressed in ApcMin/+ adenomas only were further narrowed down by rigorous validation criteria such as requirements for similar changes in abundance upon APC loss in human cells as well as the association of one or more α-APC ChIP-seq peaks with their respective transcription start sites. These candidates of interests have not yet been studied further because they failed subsequent screening steps due to technical problems encountered in either ChIP- qPCR validation or luciferase reporter studies. Although the list of transcripts in Fig. 35 includes false positives (known canonical WNT targets) such as Fn1 [169] and Yap1

[170], other promising transcripts such as Atf3 (encoding an AP-1 component) and Ereg

(encoding an EGFR ligand) remain highly interesting candidates for further study at the mechanistic level. Further experiments in human cell lines would test the hypothesis that

APC-associated regulatory elements located near these genes mediate transcriptional changes in an APC-dependent but β-catenin-independent manner. Suggested experiments include luciferase studies to characterize the transcriptional activities of an

APC-associated enhancer region upstream of ATF3 and an APC-associated region within the first intron of EREG. If the sequences are sufficient to activate luciferase transcription, then the effect of siRNA-based knockdown of APC or β-catenin would be measured. Mutational analysis of predicted transcription factor binding sites within them would follow.

100

Figure 35: Candidate transcripts aberrantly expressed only in ApcMin/+-derived adenomas

Expression data for transcripts that changed significantly in abundance in ApcMin/+ colon adenomas but not in AOM/DSS-treated colon adenomas relative to their respective adjacent tissues. All were selected from a larger group of candidates based on a significant, similar change observed in HCT-116 cells following APC knockdown, and the close association of one or more α-APC ChIP-seq peaks with their transcription start sites. Data from the HCT-116 human colon cancer cell line (from single replicates) are expressed as fold-changes relative to a mock-transfected negative control, while colon adenoma data (from three of each adenoma type) are expressed as fold-changes relative to adjacent tissue, plus or minus one standard deviation from the mean. Adenomas from ApcMin/+ mice generally exhibit an acquired loss-of-function mutation in the second allele of Apc, in addition to the inherited Min allele [15]. AOM/DSS colon adenomas are generated by treatment with azoxymethane and dextran sulfate sodium and exhibit activating mutations in the gene encoding β-catenin (Fig. 12).

101

Another key question is the value of APC mutational status as a marker for colorectal cancer prognosis. Previous studies have failed to detect a strong correlation between

APC mutations and outcome in colorectal cancer as a whole [16]. On the other hand, the high proportion of biallelic APC mutations in all stages of disease has been cited as evidence that they drive colorectal cancer progression down a particularly efficient pathway relative to mutually exclusive point mutations that activate β-catenin, which are more frequently detected in small adenomas and less often in larger, malignant tumors

[17]. APC mutations are also unequally distributed across the recognized subsets of colorectal cancer, and the microsatellite-unstable group characterized by the best prognosis exhibits the lowest frequencies of APC mutations and the highest occurrence of mutations activating β-catenin [90]. One explanation for the previous failure to correlate

APC status with prognosis might be the frequent occurrence of APC mutations in all subsets of colorectal cancer, so that the great variability between the outcomes associated with individual subsets generates a high level of noise when they are grouped together.

One strategic approach to re-test our original question of whether APC loss drives tumorigenesis more efficiently than β-catenin activation would be to control for subset.

The group of colorectal cancers characterized by microsatellite instability might provide a level playing field on which to compare APC-deficient cancers with those expressing mutant β-catenin. This comparison should be performed in datasets from the Cancer

Genome Atlas that have available clinical outcomes [5].

In conclusion, this project leaves some of its original questions unanswered, but has yielded evidence to support an ongoing search for APC-controlled molecular markers and

102 therapeutic targets. APC loss leads to the aberrant expression of target genes whose transcription is directly activated by canonical WNT signaling as well as atypical transcriptional targets that are directly repressed by canonical WNT signaling.

Mechanisms of WNT activation and WNT repression share components such as β- catenin, TCF7L2 and an emerging collaborator AP-1. The striking degree of overlap between the predicted binding sites of TCF7L2 and AP-1 raises the possibility that AP-1 inhibitors (currently still in development [171]) may demonstrate the ability to reverse an important subset of transcriptional abnormalities resulting from APC loss. This alternative is particularly interesting given that many of the more promising inhibitors of canonical WNT signaling target early steps of the pathway and therefore are less effective in the absence of functional APC [172].

103

References

1. Howlader N, N.A., Krapcho M, Miller D, Bishop K, Altekruse SF, Kosary CL, Yu M, Ruhl J, Tatalovich Z, Mariotto A, Lewis DR, Chen HS, Feuer EJ, Cronin KA (eds). SEER Cancer Statistics Review, 1975-2013, National Cancer Institute. Bethesda, MD. 2016.

2. Jen, J., et al., Molecular determinants of dysplasia in colorectal lesions. Cancer Res, 1994. 54(21): p. 5523-6.

3. Stein, U., et al., MACC1, a newly identified key regulator of HGF-MET signaling, predicts colon cancer metastasis. Nat Med, 2009. 15(1): p. 59-67.

4. Linardou, H., et al., Assessment of somatic k-RAS mutations as a mechanism associated with resistance to EGFR-targeted agents: A systematic review and meta-analysis of studies in advanced non-small-cell lung cancer and metastatic colorectal cancer. Lancet Oncol, 2008. 9(10): p. 962-72.

5. Comprehensive molecular characterization of human colon and rectal cancer. Nature, 2012. 487(7407): p. 330-7.

6. Powell, S.M., et al., APC mutations occur early during colorectal tumorigenesis. Nature, 1992. 359(6392): p. 235-7.

7. Miyoshi, Y., et al., Somatic mutations of the APC gene in colorectal tumors: Mutation cluster region in the APC gene. Hum Mol Genet, 1992. 1(4): p. 229-33.

8. Groden, J., et al., Identification and characterization of the familial adenomatous polyposis coli gene. Cell, 1991. 66(3): p. 589-600.

9. Nishisho, I., et al., Mutations of chromosome 5q21 genes in FAP and colorectal cancer patients. Science, 1991. 253(5020): p. 665-9.

10. Joslyn, G., et al., Identification of deletion mutations and three new genes at the familial polyposis locus. Cell, 1991. 66(3): p. 601-13. 104

11. Kinzler, K.W., et al., Identification of FAP locus genes from chromosome 5q21. Science, 1991. 253(5020): p. 661-5.

12. Giardiello, F., Gastrointestinal polyposis syndromes and hereditary nonpolyposis colorectal cancer, in Gastrointestinal Cancers: Biology, Diagnosis, and Therapy, R. AK, Editor 1995, Lippincott-Raven: Philadelphia. p. 367-377.

13. Ichii, S., et al., Detailed analysis of genetic alterations in colorectal tumors from patients with and without familial adenomatous polyposis (FAP). Oncogene, 1993. 8(9): p. 2399-405.

14. Levy, D.B., et al., Inactivation of both APC alleles in human and mouse tumors. Cancer Res, 1994. 54(22): p. 5953-8.

15. Luongo, C., et al., Loss of Apc+ in intestinal adenomas from Min mice. Cancer Res, 1994. 54(22): p. 5947-52.

16. Conlin, A., et al., The prognostic significance of K-ras, p53, and APC mutations in colorectal carcinoma. Gut, 2005. 54(9): p. 1283-6.

17. Samowitz, W.S., et al., Beta-catenin mutations are more frequent in small colorectal adenomas than in larger adenomas and invasive carcinomas. Cancer Res, 1999. 59(7): p. 1442-4.

18. Mirabelli-Primdahl, L., et al., Beta-catenin mutations are specific for colorectal carcinomas with microsatellite instability but occur in endometrial carcinomas irrespective of mutator pathway. Cancer Res, 1999. 59(14): p. 3346-51.

19. Sparks, A.B., et al., Mutational analysis of the APC/beta-catenin/Tcf pathway in colorectal cancer. Cancer Res, 1998. 58(6): p. 1130-4.

20. Orford, K., et al., Serine phosphorylation-regulated ubiquitination and degradation of beta-catenin. J Biol Chem, 1997. 272(40): p. 24735-8.

21. Behrens, J., et al., Functional interaction of an axin homolog, conductin, with beta-catenin, APC, and GSK3beta. Science, 1998. 280(5363): p. 596-9.

22. Sakanaka, C., J.B. Weiss, and L.T. Williams, Bridging of beta-catenin and glycogen synthase kinase-3beta by axin and inhibition of beta-catenin-mediated transcription. Proc Natl Acad Sci U S A, 1998. 95(6): p. 3020-3.

105

23. Aberle, H., et al., Beta-catenin is a target for the ubiquitin-proteasome pathway. EMBO J, 1997. 16(13): p. 3797-804.

24. Munemitsu, S., et al., Regulation of intracellular beta-catenin levels by the adenomatous polyposis coli (APC) tumor-suppressor protein. Proc Natl Acad Sci U S A, 1995. 92(7): p. 3046-50.

25. Yost, C., et al., The axis-inducing activity, stability, and subcellular distribution of beta-catenin is regulated in Xenopus embryos by glycogen synthase kinase 3. Genes Dev, 1996. 10(12): p. 1443-54.

26. Ikeda, S., et al., Axin, a negative regulator of the Wnt signaling pathway, forms a complex with GSK-3beta and beta-catenin and promotes GSK-3beta-dependent phosphorylation of beta-catenin. EMBO J, 1998. 17(5): p. 1371-84.

27. Yamamoto, H., et al., Axil, a member of the Axin family, interacts with both glycogen synthase kinase 3beta and beta-catenin and inhibits axis formation of Xenopus embryos. Mol Cell Biol, 1998. 18(5): p. 2867-75.

28. Seeling, J.M., et al., Regulation of beta-catenin signaling by the B56 subunit of protein phosphatase 2A. Science, 1999. 283(5410): p. 2089-91.

29. Gao, Z.H., et al., Casein kinase I phosphorylates and destabilizes the beta-catenin degradation complex. Proc Natl Acad Sci U S A, 2002. 99(3): p. 1182-7.

30. Su, L.K., B. Vogelstein, and K.W. Kinzler, Association of the APC tumor suppressor protein with catenins. Science, 1993. 262(5140): p. 1734-7.

31. Rubinfeld, B., et al., Loss of beta-catenin regulation by the APC tumor suppressor protein correlates with loss of structure due to common somatic mutations of the gene. Cancer Res, 1997. 57(20): p. 4624-30.

32. Rubinfeld, B., et al., Binding of GSK3beta to the APC-beta-catenin complex and regulation of complex assembly. Science, 1996. 272(5264): p. 1023-6.

33. Liu, C., et al., Control of beta-catenin phosphorylation/degradation by a dual- kinase mechanism. Cell, 2002. 108(6): p. 837-47.

34. Behrens, J., et al., Functional interaction of beta-catenin with the transcription factor LEF-1. Nature, 1996. 382(6592): p. 638-42.

106

35. He, T.C., et al., Identification of c-MYC as a target of the APC pathway. Science, 1998. 281(5382): p. 1509-12.

36. Shtutman, M., et al., The cyclin D1 gene is a target of the beta-catenin/LEF-1 pathway. Proc Natl Acad Sci U S A, 1999. 96(10): p. 5522-7.

37. Tetsu, O. and F. McCormick, Beta-catenin regulates expression of cyclin D1 in colon carcinoma cells. Nature, 1999. 398(6726): p. 422-6.

38. Yamamoto, Y., et al., Overexpression of orphan G-protein-coupled receptor, Gpr49, in human hepatocellular carcinomas with beta-catenin mutations. Hepatology, 2003. 37(3): p. 528-33.

39. Van der Flier, L.G., et al., The intestinal Wnt/TCF Signature. Gastroenterology, 2007. 132(2): p. 628-32.

40. Fevr, T., et al., Wnt/beta-catenin is essential for intestinal homeostasis and maintenance of intestinal stem cells. Mol Cell Biol, 2007. 27(21): p. 7551-9.

41. Morin, P.J., B. Vogelstein, and K.W. Kinzler, Apoptosis and APC in colorectal tumorigenesis. Proc Natl Acad Sci U S A, 1996. 93(15): p. 7950-4.

42. Chandra, S.H., et al., A common role for various human truncated adenomatous polyposis coli isoforms in the control of beta-catenin activity and cell proliferation. PLoS One, 2012. 7(4): p. e34479.

43. Dow, L.E., et al., Apc restoration promotes cellular differentiation and reestablishes crypt homeostasis in colorectal cancer. Cell, 2015. 161(7): p. 1539- 52.

44. Strater, J., et al., In situ detection of enterocytic apoptosis in normal colonic mucosa and in familial adenomatous polyposis. Gut, 1995. 37(6): p. 819-25.

45. van de Wetering, M., et al., The beta-catenin/TCF-4 complex imposes a crypt progenitor phenotype on colorectal cancer cells. Cell, 2002. 111(2): p. 241-50.

46. Senda, T., et al., Adenomatous polyposis coli (APC) plays multiple roles in the intestinal and colorectal epithelia. Med Mol Morphol, 2007. 40(2): p. 68-81.

107

47. Korinek, V., et al., Constitutive transcriptional activation by a beta-catenin-Tcf complex in APC-/- colon carcinoma. Science, 1997. 275(5307): p. 1784-7.

48. Morin, P.J., et al., Activation of beta-catenin-Tcf signaling in colon cancer by mutations in beta-catenin or APC. Science, 1997. 275(5307): p. 1787-90.

49. Zhang, F., R.L. White, and K.L. Neufeld, Phosphorylation near nuclear localization signal regulates nuclear import of adenomatous polyposis coli protein. Proc Natl Acad Sci U S A, 2000. 97(23): p. 12577-82.

50. Galea, M.A., A. Eleftheriou, and B.R. Henderson, ARM domain-dependent nuclear import of adenomatous polyposis coli protein is stimulated by the B56 alpha subunit of protein phosphatase 2A. J Biol Chem, 2001. 276(49): p. 45833-9.

51. Henderson, B.R., Nuclear-cytoplasmic shuttling of APC regulates beta-catenin subcellular localization and turnover. Nat Cell Biol, 2000. 2(9): p. 653-60.

52. Neufeld, K.L., et al., APC-mediated downregulation of beta-catenin activity involves nuclear sequestration and nuclear export. EMBO Rep, 2000. 1(6): p. 519- 23.

53. Rosin-Arbesfeld, R., F. Townsley, and M. Bienz, The APC tumour suppressor has a nuclear export function. Nature, 2000. 406(6799): p. 1009-12.

54. Rosin-Arbesfeld, R., et al., Nuclear export of the APC tumour suppressor controls beta-catenin function in transcription. EMBO J, 2003. 22(5): p. 1101-13.

55. Zhang, F., R.L. White, and K.L. Neufeld, Cell density and phosphorylation control the subcellular localization of adenomatous polyposis coli protein. Mol Cell Biol, 2001. 21(23): p. 8143-56.

56. Fagman, H., et al., Nuclear accumulation of full-length and truncated adenomatous polyposis coli protein in tumor cells depends on proliferation. Oncogene, 2003. 22(38): p. 6013-22.

57. Davies, J.R., et al., Potential link between the NIMA mitotic kinase and nuclear membrane fission during mitotic exit in Aspergillus nidulans. Eukaryot Cell, 2004. 3(6): p. 1433-44.

58. Sierra, J., et al., The APC tumor suppressor counteracts beta-catenin activation and H3K4 methylation at Wnt target genes. Genes Dev, 2006. 20(5): p. 586-600.

108

59. Hamada, F. and M. Bienz, The APC tumor suppressor binds to C-terminal binding protein to divert nuclear beta-catenin from TCF. Dev Cell, 2004. 7(5): p. 677-85.

60. Choi, S.H., et al., alpha-Catenin interacts with APC to regulate beta-catenin proteolysis and transcriptional repression of Wnt target genes. Genes Dev, 2013. 27(22): p. 2473-88.

61. Anderson, C.B., K.L. Neufeld, and R.L. White, Subcellular distribution of Wnt pathway proteins in normal and neoplastic colon. Proc Natl Acad Sci U S A, 2002. 99(13): p. 8683-8.

62. Kouzmenko, A.P., et al., Truncation mutations abolish chromatin-associated activities of adenomatous polyposis coli. Oncogene, 2008. 27(36): p. 4888-99.

63. Zeineldin, M., et al., A knock-in mouse model reveals roles for nuclear Apc in cell proliferation, Wnt signal inhibition and tumor suppression. Oncogene, 2012. 31(19): p. 2423-37.

64. Rosin-Arbesfeld, R., G. Ihrke, and M. Bienz, Actin-dependent membrane association of the APC tumour suppressor in polarized mammalian epithelial cells. EMBO J, 2001. 20(21): p. 5929-39.

65. Kawasaki, Y., et al., Asef, a link between the tumor suppressor APC and G-protein signaling. Science, 2000. 289(5482): p. 1194-7.

66. Watanabe, T., et al., Interaction with IQGAP1 links APC to Rac1, Cdc42, and actin filaments during cell polarization and migration. Dev Cell, 2004. 7(6): p. 871-83.

67. Sudhaharan, T., et al., Rho GTPase Cdc42 is a direct interacting partner of Adenomatous Polyposis Coli protein and can alter its cellular localization. PLoS One, 2011. 6(2): p. e16603.

68. Aberle, H., H. Schwartz, and R. Kemler, Cadherin-catenin complex: Protein interactions and their implications for cadherin function. J Cell Biochem, 1996. 61(4): p. 514-23.

69. Faux, M.C., et al., Restoration of full-length adenomatous polyposis coli (APC) protein in a colon cancer cell line enhances cell adhesion. J Cell Sci, 2004. 117(Pt 3): p. 427-39.

109

70. Hulsken, J., W. Birchmeier, and J. Behrens, E-cadherin and APC compete for the interaction with beta-catenin and the cytoskeleton. J Cell Biol, 1994. 127(6 Pt 2): p. 2061-9.

71. Jamora, C., et al., Links between signal transduction, transcription and adhesion in epithelial bud development. Nature, 2003. 422(6929): p. 317-22.

72. Crawford, H.C., et al., The metalloproteinase matrilysin is a target of beta-catenin transactivation in intestinal tumors. Oncogene, 1999. 18(18): p. 2883-91.

73. Brabletz, T., et al., Beta-catenin regulates the expression of the matrix metalloproteinase-7 in human colorectal cancer. Am J Pathol, 1999. 155(4): p. 1033-8.

74. Mahmoud, N.N., et al., Apc gene mutation is associated with a dominant- negative effect upon intestinal cell migration. Cancer Res, 1997. 57(22): p. 5045- 50.

75. Wong, M.H., et al., Forced expression of the tumor suppressor adenomatosis polyposis coli protein induces disordered cell migration in the intestinal epithelium. Proc Natl Acad Sci U S A, 1996. 93(18): p. 9588-93.

76. Nelson, S.A., et al., Tumorigenic fragments of APC cause dominant defects in directional cell migration in multiple model systems. Dis Model Mech, 2012. 5(6): p. 940-7.

77. Matsumine, A., et al., Binding of APC to the human homolog of the Drosophila discs large tumor suppressor protein. Science, 1996. 272(5264): p. 1020-3.

78. Takizawa, S., et al., Human Scribble, a novel tumor suppressor identified as a target of high-risk HPV E6 for ubiquitin-mediated degradation, interacts with adenomatous polyposis coli. Genes Cells, 2006. 11(4): p. 453-64.

79. Nathke, I.S., et al., The adenomatous polyposis coli tumor suppressor protein localizes to plasma membrane sites involved in active cell migration. J Cell Biol, 1996. 134(1): p. 165-79.

80. Munemitsu, S., et al., The APC gene product associates with microtubules in vivo and promotes their assembly in vitro. Cancer Res, 1994. 54(14): p. 3676-81.

110

81. Zumbrunn, J., et al., Binding of the adenomatous polyposis coli protein to microtubules increases microtubule stability and is regulated by GSK3 beta phosphorylation. Curr Biol, 2001. 11(1): p. 44-9.

82. Mogensen, M.M., et al., The adenomatous polyposis coli protein unambiguously localizes to microtubule plus ends and is involved in establishing parallel arrays of microtubule bundles in highly polarized epithelial cells. J Cell Biol, 2002. 157(6): p. 1041-8.

83. Su, L.K., et al., APC binds to the novel protein EB1. Cancer Res, 1995. 55(14): p. 2972-7.

84. Iizuka-Kogo, A., A. Shimomura, and T. Senda, Colocalization of APC and DLG at the tips of cellular protrusions in cultured epithelial cells and its dependency on cytoskeletons. Histochem Cell Biol, 2005. 123(1): p. 67-73.

85. Mili, S., K. Moissoglu, and I.G. Macara, Genome-wide screen reveals APC- associated RNAs enriched in cell protrusions. Nature, 2008. 453(7191): p. 115-9.

86. Marshall, T.W., et al., The tumor suppressor adenomatous polyposis coli controls the direction in which a cell extrudes from an epithelium. Mol Biol Cell, 2011. 22(21): p. 3962-70.

87. Bellis, J., et al., The tumor suppressor Apc controls planar cell polarities central to gut homeostasis. J Cell Biol, 2012. 198(3): p. 331-41.

88. Cheng, Y.W., et al., CpG island methylator phenotype associates with low-degree chromosomal abnormalities in colorectal cancer. Clin Cancer Res, 2008. 14(19): p. 6005-13.

89. Goel, A., et al., The CpG island methylator phenotype and chromosomal instability are inversely correlated in sporadic colorectal cancer. Gastroenterology, 2007. 132(1): p. 127-38. 90. Samowitz, W.S., et al., APC mutations and other genetic and epigenetic changes in colon cancer. Mol Cancer Res, 2007. 5(2): p. 165-70.

91. Fodde, R., et al., Mutations in the APC tumour suppressor gene cause chromosomal instability. Nat Cell Biol, 2001. 3(4): p. 433-8.

92. Kaplan, K.B., et al., A role for the Adenomatous Polyposis Coli protein in chromosome segregation. Nat Cell Biol, 2001. 3(4): p. 429-32.

111

93. Dikovskaya, D., et al., Loss of APC induces polyploidy as a result of a combination of defects in mitosis and apoptosis. J Cell Biol, 2007. 176(2): p. 183-95.

94. Green, R.A., R. Wollman, and K.B. Kaplan, APC and EB1 function together in mitosis to regulate spindle dynamics and chromosome alignment. Mol Biol Cell, 2005. 16(10): p. 4609-22.

95. Trzepacz, C., et al., Phosphorylation of the tumor suppressor adenomatous polyposis coli (APC) by the cyclin-dependent kinase p34. J Biol Chem, 1997. 272(35): p. 21681-4.

96. Green, R.A. and K.B. Kaplan, Chromosome instability in colorectal tumor cells is associated with defects in microtubule plus-end attachments caused by a dominant mutation in APC. J Cell Biol, 2003. 163(5): p. 949-61.

97. Groden, J., et al., Response of colon cancer cell lines to the introduction of APC, a colon-specific tumor suppressor gene. Cancer Res, 1995. 55(7): p. 1531-9.

98. Baeg, G.H., et al., The tumour suppressor gene product APC blocks cell cycle progression from G0/G1 to S phase. EMBO J, 1995. 14(22): p. 5618-25.

99. Heinen, C.D., et al., The APC tumor suppressor controls entry into S-phase through its ability to regulate the cyclin D/RB pathway. Gastroenterology, 2002. 123(3): p. 751-63.

100. Ishidate, T., et al., The APC-hDLG complex negatively regulates cell cycle progression from the G0/G1 to S phase. Oncogene, 2000. 19(3): p. 365-72.

101. Qian, J., et al., The APC tumor suppressor inhibits DNA replication by directly binding to DNA via its carboxyl terminus. Gastroenterology, 2008. 135(1): p. 152- 62.

102. Brocardo, M.G., J.A. Borowiec, and B.R. Henderson, Adenomatous polyposis coli protein regulates the cellular response to DNA replication stress. Int J Biochem Cell Biol, 2011. 43(9): p. 1354-64.

103. Zhang, Y., et al., Inhibition of Wnt signaling by Dishevelled PDZ peptides. Nat Chem Biol, 2009. 5(4): p. 217-9.

112

104. Zhang, T., et al., Evidence that APC regulates survivin expression: A possible mechanism contributing to the stem cell origin of colon cancer. Cancer Res, 2001. 61(24): p. 8664-7.

105. Steigerwald, K., et al., The APC tumor suppressor promotes transcription- independent apoptosis in vitro. Mol Cancer Res, 2005. 3(2): p. 78-89.

106. Qian, J., et al., Caspase cleavage of the APC tumor suppressor and release of an amino-terminal domain is required for the transcription-independent function of APC in apoptosis. Oncogene, 2007. 26(33): p. 4872-6.

107. Qian, J., et al., The mitochondrial protein hTID-1 partners with the caspase- cleaved adenomatous polyposis cell tumor suppressor to facilitate apoptosis. Gastroenterology, 2010. 138(4): p. 1418-28.

108. Brocardo, M., et al., Mitochondrial targeting of adenomatous polyposis coli protein is stimulated by truncating cancer mutations: Regulation of Bcl-2 and implications for cell survival. J Biol Chem, 2008. 283(9): p. 5950-9.

109. Andreu, P., et al., Crypt-restricted proliferation and commitment to the Paneth cell lineage following Apc loss in the mouse intestine. Development, 2005. 132(6): p. 1443-51.

110. Korinek, V., et al., Depletion of epithelial stem-cell compartments in the small intestine of mice lacking Tcf-4. Nat Genet, 1998. 19(4): p. 379-83.

111. Barker, N., et al., Identification of stem cells in small intestine and colon by marker gene Lgr5. Nature, 2007. 449(7165): p. 1003-7.

112. Nadauld, L.D., et al., Adenomatous polyposis coli control of retinoic acid biosynthesis is critical for zebrafish intestinal development and differentiation. J Biol Chem, 2004. 279(49): p. 51581-9.

113. Nadauld, L.D., et al., Adenomatous polyposis coli control of C-terminal binding protein-1 stability regulates expression of intestinal retinol dehydrogenases. J Biol Chem, 2006. 281(49): p. 37828-35.

114. Nadauld, L.D., et al., The zebrafish retinol dehydrogenase, rdh1l, is essential for intestinal development and is regulated by the tumor suppressor adenomatous polyposis coli. J Biol Chem, 2005. 280(34): p. 30490-5.

113

115. Jette, C., et al., The tumor suppressor adenomatous polyposis coli and caudal related homeodomain protein regulate expression of retinol dehydrogenase L. J Biol Chem, 2004. 279(33): p. 34397-405.

116. Cheng, H. and C.P. Leblond, Origin, differentiation and renewal of the four main epithelial cell types in the mouse small intestine. V. Unitarian Theory of the origin of the four epithelial cell types. Am J Anat, 1974. 141(4): p. 537-61.

117. Issa, J.P., Colon cancer: It's CIN or CIMP. Clin Cancer Res, 2008. 14(19): p. 5939- 40.

118. Shen, L., et al., Integrated genetic and epigenetic analysis identifies three different subclasses of colon cancer. Proc Natl Acad Sci U S A, 2007. 104(47): p. 18654-9.

119. Akiyama, Y., et al., Transforming growth factor beta type II receptor gene mutations in adenomas from hereditary nonpolyposis colorectal cancer. Gastroenterology, 1997. 112(1): p. 33-9.

120. Calin, G.A., et al., Genetic progression in microsatellite instability high (MSI-H) colon cancers correlates with clinico-pathological parameters: A study of the TGRbetaRII, BAX, hMSH3, hMSH6, IGFIIR and BLM genes. Int J Cancer, 2000. 89(3): p. 230-5.

121. Peltomaki, P., Lynch syndrome genes. Fam Cancer, 2005. 4(3): p. 227-32.

122. Deng, G., et al., BRAF mutation is frequently present in sporadic colorectal cancer with methylated hMLH1, but not in hereditary nonpolyposis colorectal cancer. Clin Cancer Res, 2004. 10(1 Pt 1): p. 191-5.

123. Liefers, G.J., et al., Micrometastases and survival in stage II colorectal cancer. N Engl J Med, 1998. 339(4): p. 223-8.

124. Ilm, K., et al., High MACC1 expression in combination with mutated KRAS G13 indicates poor survival of colorectal cancer patients. Mol Cancer, 2015. 14: p. 38.

125. Koelzer, V.H., et al., Heterogeneity analysis of Metastasis Associated in Colon Cancer 1 (MACC1) for survival prognosis of colorectal cancer patients: a retrospective cohort study. BMC Cancer, 2015. 15: p. 160.

114

126. Gaspar, C., et al., Cross-species comparison of human and mouse intestinal polyps reveals conserved mechanisms in adenomatous polyposis coli (APC)-driven tumorigenesis. Am J Pathol, 2008. 172(5): p. 1363-80.

127. Bowden, N.A., A. Croft, and R.J. Scott, Gene expression profiling in familial adenomatous polyposis adenomas and desmoid disease. Hered Cancer Clin Pract, 2007. 5(2): p. 79-96.

128. Reichling, T., et al., Transcriptional profiles of intestinal tumors in Apc(Min) mice are unique from those of embryonic intestine and identify novel gene targets dysregulated in human colorectal tumors. Cancer Res, 2005. 65(1): p. 166-76.

129. Bottomly, D., et al., Identification of {beta}-catenin binding regions in colon cancer cells using ChIP-Seq. Nucleic Acids Res, 2010. 38(17): p. 5735-45.

130. Hatzis, P., et al., Genome-wide pattern of TCF7L2/TCF4 chromatin occupancy in colorectal cancer cells. Mol Cell Biol, 2008. 28(8): p. 2732-44.

131. Yochum, G.S., R. Cleland, and R.H. Goodman, A genome-wide screen for beta- catenin binding sites identifies a downstream enhancer element that controls c- Myc gene expression. Mol Cell Biol, 2008. 28(24): p. 7368-79.

132. Kahler, R.A. and J.J. Westendorf, Lymphoid enhancer factor-1 and beta-catenin inhibit Runx2-dependent transcriptional activation of the osteocalcin promoter. J Biol Chem, 2003. 278(14): p. 11937-44.

133. Delmas, V., et al., Beta-catenin induces immortalization of melanocytes by suppressing p16INK4a expression and cooperates with N-Ras in melanoma development. Genes Dev, 2007. 21(22): p. 2923-35.

134. Deka, J., et al., The APC protein binds to A/T rich DNA sequences. Oncogene, 1999. 18(41): p. 5654-61.

135. Smith, K.J., et al., Wild-type but not mutant APC associates with the microtubule cytoskeleton. Cancer Res, 1994. 54(14): p. 3672-5.

136. Tighe, A., V.L. Johnson, and S.S. Taylor, Truncating APC mutations have dominant effects on proliferation, spindle checkpoint control, survival and chromosome stability. J Cell Sci, 2004. 117(Pt 26): p. 6339-53.

115

137. Radulescu, S., et al., Defining the role of APC in the mitotic spindle checkpoint in vivo: APC-deficient cells are resistant to Taxol. Oncogene, 2010. 29(49): p. 6418- 27.

138. Dikovskaya, D., et al., The adenomatous polyposis coli protein contributes to normal compaction of mitotic chromatin. PLoS One, 2012. 7(6): p. e38102.

139. Rai, K., et al., DNA demethylase activity maintains intestinal cells in an undifferentiated state following loss of APC. Cell, 2010. 142(6): p. 930-42.

140. Sharov, A.A. and M.S. Ko, Exhaustive search for over-represented DNA sequence motifs with CisFinder. DNA Res, 2009. 16(5): p. 261-73.

141. Machanick, P. and T.L. Bailey, MEME-ChIP: Motif analysis of large DNA datasets. Bioinformatics, 2011. 27(12): p. 1696-7.

142. Medina-Rivera, A., et al., RSAT 2015: Regulatory Sequence Analysis Tools. Nucleic Acids Res, 2015. 43(W1): p. W50-6.

143. Cartharius, K., et al., MatInspector and beyond: Promoter analysis based on transcription factor binding sites. Bioinformatics, 2005. 21(13): p. 2933-42.

144. Eferl, R. and E.F. Wagner, AP-1: A double-edged sword in tumorigenesis. Nat Rev Cancer, 2003. 3(11): p. 859-68.

145. Nagaoka, M., Y. Shiraishi, and Y. Sugiura, Selected base sequence outside the target binding site of zinc finger protein Sp1. Nucleic Acids Res, 2001. 29(24): p. 4920-9.

146. Luo, X. and M. Sawadogo, Antiproliferative properties of the USF family of helix- loop-helix transcription factors. Proc Natl Acad Sci U S A, 1996. 93(3): p. 1308-13.

147. Virbasius, C.A., J.V. Virbasius, and R.C. Scarpulla, NRF-1, an activator involved in nuclear-mitochondrial interactions, utilizes a new DNA-binding domain conserved in a family of developmental regulators. Genes Dev, 1993. 7(12A): p. 2431-45.

148. Christy, B. and D. Nathans, DNA binding site of the growth factor-inducible protein Zif268. Proc Natl Acad Sci U S A, 1989. 86(22): p. 8737-41.

116

149. Nateri, A.S., B. Spencer-Dene, and A. Behrens, Interaction of phosphorylated c- Jun with TCF4 regulates intestinal cancer development. Nature, 2005. 437(7056): p. 281-5.

150. Toualbi, K., et al., Physical and functional cooperation between AP-1 and beta- catenin for the regulation of TCF-dependent genes. Oncogene, 2007. 26(24): p. 3492-502.

151. Biddie, S.C., et al., Transcription factor AP1 potentiates chromatin accessibility and glucocorticoid receptor binding. Mol Cell, 2011. 43(1): p. 145-55.

152. Kohno, H., et al., Beta-catenin mutations in a mouse model of inflammation- related colon carcinogenesis induced by 1,2-dimethylhydrazine and dextran sodium sulfate. Cancer Sci, 2005. 96(2): p. 69-76.

153. Yang, X., et al., Decapentaplegic is a direct target of dTcf repression in the Drosophila visceral mesoderm. Development, 2000. 127(17): p. 3695-702.

154. Piepenburg, O., G. Vorbruggen, and H. Jackle, Drosophila segment borders result from unilateral repression of hedgehog activity by wingless signaling. Mol Cell, 2000. 6(1): p. 203-9.

155. Lei, Q., et al., Wnt signaling inhibitors regulate the transcriptional response to morphogenetic Shh-Gli signaling in the neural tube. Dev Cell, 2006. 11(3): p. 325- 37.

156. Wang, X., et al., Decreased MALL expression negatively impacts colorectal cancer patient survival. Oncotarget, 2016. 7(16): p. 22911-27.

157. Hasselblatt, P., et al., The role of the transcription factor AP-1 in colitis-associated and beta-catenin-dependent intestinal tumorigenesis in mice. Oncogene, 2008. 27(47): p. 6102-9.

158. Zhang, Y., et al., Model-based analysis of ChIP-Seq (MACS). Genome Biol, 2008. 9(9): p. R137.

159. Feng, J., T. Liu, and Y. Zhang, Using MACS to identify peaks from ChIP-Seq data. Curr Protoc Bioinformatics, 2011. Chapter 2: p. Unit 2 14.

160. Robinson, J.T., et al., Integrative genomics viewer. Nat Biotechnol, 2011. 29(1): p. 24-6.

117

161. Thorvaldsdottir, H., J.T. Robinson, and J.P. Mesirov, Integrative Genomics Viewer (IGV): High-performance genomics data visualization and exploration. Brief Bioinform, 2013. 14(2): p. 178-92.

162. Chomczynski, P., A reagent for the single-step simultaneous isolation of RNA, DNA and proteins from cell and tissue samples. Biotechniques, 1993. 15(3): p. 532-4, 536-7.

163. Trapnell, C., et al., Transcript assembly and quantification by RNA-Seq reveals unannotated transcripts and isoform switching during cell differentiation. Nat Biotechnol, 2010. 28(5): p. 511-5.

164. Giese, K., et al., Assembly and function of a TCR alpha enhancer complex is dependent on LEF-1-induced DNA bending and multiple protein-protein interactions. Genes Dev, 1995. 9(8): p. 995-1008.

165. Ilyas, M., et al., Beta-catenin mutations in cell lines established from human colorectal cancers. Proc Natl Acad Sci U S A, 1997. 94(19): p. 10330-4.

166. Albuquerque, C., et al., The 'just-right' signaling model: APC somatic mutations are selected based on a specific level of activation of the beta-catenin signaling cascade. Hum Mol Genet, 2002. 11(13): p. 1549-60.

167. Yan, D., et al., Elevated expression of axin2 and hnkd mRNA provides evidence that Wnt/beta -catenin signaling is activated in human colon tumors. Proc Natl Acad Sci U S A, 2001. 98(26): p. 14973-8.

168. Mann, B., et al., Target genes of beta-catenin-T cell-factor/lymphoid-enhancer- factor signaling in human colorectal carcinomas. Proc Natl Acad Sci U S A, 1999. 96(4): p. 1603-8.

169. ten Berge, D., et al., Wnt signaling mediates self-organization and axis formation in embryoid bodies. Cell Stem Cell, 2008. 3(5): p. 508-18.

170. Hoverter, N.P., et al., A WNT/p21 circuit directed by the C-clamp, a sequence- specific DNA binding domain in TCFs. Mol Cell Biol, 2012. 32(18): p. 3648-62.

171. Ye, N., et al., Small molecule inhibitors targeting activator protein 1 (AP-1). J Med Chem, 2014. 57(16): p. 6930-48.

118

172. Garber, K., Drugging the Wnt pathway: Problems and progress. J Natl Cancer Inst, 2009. 101(8): p. 548-50.

173. Kaiser, S., et al., Transcriptional recapitulation and subversion of embryonic colon development by mouse colon tumor models and human colon cancer. Genome Biol, 2007. 8(7): p. R131.

174. Sunaga, N., et al., Oncogenic KRAS-induced epiregulin overexpression contributes to aggressive phenotype and is a promising therapeutic target in non-small-cell lung cancer. Oncogene, 2013. 32(34): p. 4034-42.

175. Lupu, R. and J.A. Menendez, Pharmacological inhibitors off fatty acid synthase (FASN)--catalyzed endogenous fatty acid biogenesis: A new family of anti-cancer agents? Curr Pharm Biotechnol, 2006. 7(6): p. 483-93.

176. Salaun, C., et al., Identification of a novel transport-independent function of PiT1/SLC20A1 in the regulation of TNF-induced apoptosis. J Biol Chem, 2010. 285(45): p. 34408-18.

119