Forkhead Box F1 (FOXF1) is an essential effector of the
PAX3/FOXO1 oncogene in human alveolar rhabdomyosarcoma
A dissertation submitted to the Division of Research and Advanced Studies of the University of Cincinnati in partial fulfillment of the requirements for the degree of
Doctor of Philosophy
in the Graduate Program of Molecular and Developmental Biology in the College of Medicine on April 9, 2019
by David E. Milewski B.A. Miami University (2010)
Dissertation Committee Dr. Tanya Kalin, MD PhD (Chair) Dr. Vladimir Kalinichenko, MD PhD Dr. Joseph Pressey, MD Dr. Doug Millay, PhD Dr. Lionel Chow, MD PhD
P a g e | 2
Abstract
Rhabdomyosarcoma (RMS) is an aggressive soft tissue neoplasm that displays characteristic of incomplete skeletal muscle differentiation. RMS can be subcategorized based on the presence of specific genetic alterations which can be associated with adverse patient outcome. The recurrent t(2;13) translocation defines the alveolar subtype of rhabdomyosarcoma (aRMS) and gives rise to a PAX3/FOXO1 fusion protein. RMS patients with this translocation have the poorest outcomes, despite the absence of cooperating mutations. Multiple studies have defined a gene signature associated with fusion-positive aRMS. While it is presumed that these genes may contribute to the malignant phenotype of aRMS, little mechanistic evidence exists for their role in aRMS. In this work, we study the biological and molecular contributions of Forkhead Box F1 (FOXF1), one of the most consistently and uniquely upregulated genes in fusion-positive aRMS. Using a combination of cell lines and newly established patient derived xenograft (PDX) models, we uncover a central role for FOXF1 in the pathogenesis of aRMS. We found that FOXF1 is uniquely and specifically induced in aRMS due to direct transcriptional regulation by PAX3/FOXO1 at distal enhancer elements. Knockdown or knockout of FOXF1 in aRMS revealed an important role for promoting the proliferation and survival of aRMS cells. Moreover, loss of FOXF1 was associated with widespread induction of spontaneous myogenic differentiation despite the presence of the PAX3/FOXO1 oncogene. In a primary human myoblast model, expression of
FOXF1 alone was sufficient to block terminal myogenic differentiation and bypass myoblast senescence. Mechanistically, we found that FOXF1 establishes a unique epigenetic landscape by co-regulating enhancers with PAX3/FOXO1, MYOD1 and MYOG. In conclusion, we have identified the transcriptional activation of FOXF1 by PAX3/FOXO1 is a critical event for tumorigenesis and endows unique molecular mechanisms to fusion positive RMS. P a g e | 3
P a g e | 4
Acknowledgements I have had fantastic support from those inside and outside the laboratory. I would first like to thank my mentor, Dr. Tanya Kalin, for her strong support over the years. She has given me the freedom to pursue my research interests, even though some of them were outside the scope of the lab. She has always given great feedback on data and helped keep me focused on the goals for my projects. My committee has also helped steer me in the right direction and were always willing to meet to discuss ideas about the project.
I would also like to thank the lab members of the Kalin and Kalinichenko labs. They have made a great lab environment where we can learn from each other and have a fun while we do our work.
I would also like to thank Dr. Joseph Pressey, Dr. Sara Szabo, Dr. Brian Turpin for their trust and support with establishing the sarcoma PDX program at CCHMC. It has been such a unique, impactful experience for my career development and has really inspired me.
Most importantly, I would like to thank my parents, Dr. Allan and Mary Kay Milewski, and my wonderful wife Maribeth Milewski for their support of my research. They have always been understanding of the long hours I spent in lab, gave me good advice, and made sure I relaxed and had fun when I was out of lab. They have helped make this such a fun experience.
P a g e | 5
Table of Contents
Title Page ...... 1
Abstract ...... 2
Acknowledgements ...... 4
Table of Contents ...... 5
List of Figures ...... 7
List of Tables ...... 10
Chapter 1: Introduction ...... 12
I. Overview of Rhabdomyosarcoma...... 12
a) Histological subtypes and clinical correlates ...... 12
b) Genomic Landscape of Rhabdomyosarcoma ...... 15
II. Identification of a fusion-positive aRMS gene signature ...... 18
III. Forkhead Box F1 (FOXF1) ...... 20
a) Features of the FOXF1 gene locus...... 20
b) Structural anatomy of the FOXF1 protein...... 22
d) Phenotypes of FOXF1+/- humans and FOXF1-/- mice...... 23
e) Expression of FOXF1 in human cancer...... 24
Chapter 2: Tumor modeling using patient-derived xenografts (PDX): CCHMC Sarcoma PDX Initiative ...... 26
I. Abstract ...... 26
II. Introduction ...... 26
III. Materials and Methods ...... 31
IV. Results ...... 37
V. Discussion ...... 51
Chapter 3: Transcriptional regulation of the FOXF1 gene locus in aRMS...... 53 P a g e | 6
I. Abstract ...... 53
II. Introduction ...... 53
III. Materials and Methods ...... 54
IV. Results ...... 57
V. Discussion ...... 69
Chapter 4: Identification of FOXF1 as positive regulator of cell proliferation and a potent repressor of terminal myogenic differentiation ...... 71
I. Abstract ...... 71
II. Introduction ...... 71
III. Materials and Methods ...... 73
IV. Results ...... 74
V. Discussion ...... 92
Chapter 5: Genome-wide binding of FOXF1 in aRMS uncovers cooperative enhancer activation with PAX3/FOXO1 and myogenic regulatory factors ...... 94
I. Abstract ...... 94
II. Introduction ...... 94
III. Materials and Methods ...... 96
IV. Results ...... 98
V. Discussion ...... 106
Chapter 6: Conclusions and Discussion ...... 110
I. Working model of FOXF1 function in aRMS...... 110
II. Understanding the potent anti-myogenic properties of FOXF1...... 111
III. The relationships between PAX3/FOXO1, PAX3, and FOXF1...... 113
References ...... 119
P a g e | 7
List of Figures
Chapter 1:
Figure 1. Improvement in RMS patient survival in retrospective study...... 13
Figure 2. Histological subtypes of rhabdomyosarcomas in pediatric and young adults...... 14
Figure 3. FOXF1 is only expressed in aRMS tumors...... 20
Figure 4. FOXF1 exon 2 3’UTR contains hot spots of conservation...... 20
Figure 5. FOXF1 is located in gene cluster and has nearby long noncoding RNAs (LINCRNA)...... 21
Figure 6. Location of the DNA binding domain in the FOXF1 protein...... 23
Chapter 2:
Figure 1. Spectrum of sarcoma patient diagnoses...... 38
Figure 2. Engrafted and non-progressing tissue xenografts over long term...... 41
Figure 3. Pan-human antibody effectively distinguishes human tumor cells from host (mouse) tumor microenvironment...... 42
Figure 4. Preservation of hallmark histological features between paired patient and PDX rhabdomyosarcomas...... 43
Figure 5. Retention of driver mutation in fusion positive aRMS...... 44
Figure 6. Differential expression of stem cell marker (PAX7) and differentiating myoblast (Myogenin) in human rhabdomyosarcoma...... 48
Figure 7. Cancer testis antigen (CTA) and immune checkpoint profiling of PDX sarcomas. .... 50
P a g e | 8
Chapter 3:
Figure 1. FOXF1 is only upregulated in the presence of the fusion gene...... 57
Figure 2. FOXF1 requires PAX3/FOXO1 for expression in aRMS...... 59
Figure 3. PAX3/FOXO1 binds to active enhancers near the FOXF1 gene locus...... 60
Figure 4. The enhancer RNA LINC01082 correlates with FOXF1 and PAX3/FOXO1...... 62
Figure 5. The enhancer RNA LINC01082 correlates with FOXF1 in aRMS and across human tissues...... 63
Figure 6. Strong enhancer element downstream of FOXF1 is newly acquired in evolutionary history of vertebrates...... 66
Figure 7. FOXF1 is a direct transcriptional target of the PAX3/FOXO1 oncogene by activation of FOXF1 enhancer elements...... 69
Chapter 4:
Figure 1. FOXF1 knockdown in Rh30 aRMS cells reduces proliferation...... 75
Figure 2. Design and validation of guide RNAs targeting the human FOXF1 gene locus...... 76
Figure 3. Schematic diagram of FOXF1 deletion strategy in human aRMS cell line Rh18...... 77
Figure 4. Identification of four FOXF1 null clones...... 78
Figure 5. Significant reduction in Rh18 tumor size after FOXF1 deletion...... 80
Figure 6. Reduced tumor cell proliferation in FOXF1-KO tumors...... 81
Figure 7. RNAseq analysis identifies induction of mature myogenic genes following FOXF1 loss...... 84
Figure 8. Deletion of FOXF1 promotes spontaneous myogenic differentiation in vivo...... 85
Figure 9. FoxF1 is a potent repressor of skeletal muscle differentiation in primary human skeletal muscle myoblasts...... 89 P a g e | 9
Figure 10. FoxF1 expression is sufficient to sustain long-term proliferative capacity of primary human skeletal muscle myoblasts...... 92
Chapter 5:
Figure 1. FoxF1 binds at distal enhancers which are active in aRMS but not eRMS...... 98
Figure 2. Majority of FOXF1 binding sites are also bound by PAX3/FOXO1, MYOD1 and/or MYOG...... 101
Figure 3. FOXF1 bound enhancers are associated with aRMS signature genes...... 102
Figure 4. FoxF1 binding with PAX3/FOXO1 and myogenic regulatory factors is essential for aRMS-specific enhancers activation...... 106
Chapter 6:
Figure 1. Schematic diagram showing the establishment of the epigenomic landscape which gives rise to the unique molecular and biological properties exhibited by alveolar rhabdomyosarcoma...... 110
Figure 2. FOXF1 is not expressed during myogenic specification and migration in the embryonic limbs and trunk...... 114
Figure 3. FOXF1 and PAX3 are co-expressed in the mouse head at E13.5 but not in the skeletal muscle lineage...... 116
Figure 4. FOXF1 and PAX3 are co-expressed in the mouse tongue mesenchyme but not myogenic progenitors at E14.5...... 118
P a g e | 10
List of Tables
Chapter 1
Table 1. FOXF1 expression is correlated with fusion positive RMS...... 19
Chapter 2
Table 1. Engraftment rates for bone and soft tissue tumors in previous studies...... 30
Table 2. Engraftment rate of implanted patient tissue...... 39
Table 3. Characteristics of established rhabdomyosarcoma PDX models...... 40
Chapter 3
Table 1. Oligonucleotide pairs for cloning of shRNA and guide RNAs...... 56
Table 2. Antibodies used in this study...... 56
P a g e | 11
P a g e | 12
Chapter 1: Introduction
I. Overview of Rhabdomyosarcoma
Rhabdomyosarcoma (RMS) is a high-grade soft tissue neoplasm which displays features of incomplete skeletal muscle differentiation. RMS is the most commonly diagnosed soft tissue tumor in children with an overall annual incidence of ~350 newly diagnosed patients in the US.
While rhabdomyosarcoma remains a difficult clinical challenge, progress has been made over the past several decades. A large scale retrospective analysis of 987 children with newly diagnosed rhabdomyosarcoma found an increased 5-year overall survival from 52.7% to 61.8% between 1975 and 2005 (1). Despite these advances, retrospective sub-classification of tumors based on histology, genetics, and other risk factors revealed cohorts of patients who have had little or no improvements in outcome over this same period. A series of “high-risk” features have been defined and are currently used for risk stratification of RMS patients. Interestingly, many of these high-risk features are repeatedly observed in a single entity: alveolar rhabdomyosarcoma, a subtype with a recurrent PAX3/FOXO1 or PAX7/FOXO1 rearrangement and a propensity for metastatic spread.
This chapter will provide an in-depth summary of the biology and genetics of rhabdomyosarcoma with a focus on alveolar rhabdomyosarcoma. a) Histological subtypes and clinical correlates
Rhabdomyosarcoma typically displays one of four histologies: embryonal, spindle/sclerosing, alveolar, and pleomorphic. Each of these histologies is associated with unique P a g e | 13 genetics, anatomical location, and clinical behavior.
For these reasons, the histological classification of
RMS tumors has served as an effective, but incomplete, indicator of disease prognosis.
Furthermore, patient with certain tumor histologies have benefited from improved risk stratification and Figure 1. Improvement in RMS overall quality of disease management resulting in patient survival in retrospective study. improved survival (Fig. 1). Patients with other Retrospective survival analysis of 987 tumor histologies and genetics, however, have children with newly diagnosed observed no significant improvements in outcome rhabdomyosarcoma from 1975 to 2005. for several decades. (adapted from Ognjanovic et al., 2009). i.) Embryonal: Embryonal histology (eRMS) is observed in roughly 60% of pediatric and
adolescent RMS patients. This subtype can display a wide range of cell morphologies.
The tumor cells typically have hyperchromatic nuclei with abundant eosinophilic
cytoplasm and occasionally cross striations (Fig. 2, left). These cells a generally organized
in a loose myxoid pattern although solid patterns can also be observed. eRMS tumors are
most commonly found in the head and neck region, genitourinary tract. The latter is
associated with the botryoid variant of eRMS in which tumor cells grow in a grapelike
appearance beneath the epithelial mucosa. The median age at diagnosis for eRMS is 6.5
years old. The prognosis for these patients is relatively good with a 5-year overall survival
of 70%(2). P a g e | 14
Embryonal Spindle/Sclerosing Alveolar
Figure 2. Histological subtypes of rhabdomyosarcomas in pediatric and young adults. H&E staining of patient embryonal (left), spindle/sclerosing PDX (middle), and patient alveolar (right) rhabdomyosarcomas. Arrow in ssRMS tumor denotes sclerosing patter
ii.) Spindle/sclerosing. Spindle/sclerosing RMS (ssRMS) is newly established subtype
which was previously grouped with eRMS tumors. This distinct tumor histology is
characterized by large spindled tumor cells. These tumors have abundant extracellular
matrix deposition and can create sclerosing patterns of hyaline matrix (Fig. 2, middle).
These tumors can arise in various locations with the trunk and head and neck primary
locations as the most common(3, 4) . Due to the relative rarity of this tumor, little is known
about its prognosis. The largest study to date found 87.5% overall survival among 16
pediatric patients and generally ssRMS have favorable prognoses in children(3). In
adolescence and young adults, however, very poor outcomes have been observed and c
attributed to distinct genetic alterations (discussed next section) (3).
iii.) Alveolar. Alveolar RMS (aRMS) represents roughly 30% of all rhabdomyosarcomas.
aRMS tumor cells display a small round blue cell morphology. The tumor cells are
typically organized as discohesive tumor nests surrounded by fibrovasular septae, giving
the appearance of a pseudo-alveolar structure (Fig. 2, right). In particularly rapid growing P a g e | 15
aRMS tumors, the fibrous septae are absent resulting in a solid pattern of aRMS. These
tumors tend to originate in extremities and trunk but primary tumors in the prostate, head
and neck, paratesticular, and chest wall are also observed. The prognosis for alveolar
rhabdomyosarcoma is very poor with only a 5-year overall survival of 50%(2, 5). A
significant portion (39%) of aRMS patients present with distant metastases at diagnosis(6).
Metastatic sites include lungs, lymph nodes, soft tissue, and bone marrow (7). These
patients have a dismal prognosis with only 15% surviving five years(8). b) Genomic Landscape of Rhabdomyosarcoma
The underlying genetic basis for rhabdomyosarcoma is still under active investigation.
Chromosome karyotyping, whole exome and whole genome sequencing has led the identification of several recurrent genetic alterations in rhabdomyosarcoma. While some alterations are observed across rhabdomyosarcoma histological subtypes, most of the “driver” mutations are associated with a single distinct histology.
i.) Embryonal: eRMS tumors are genomically complex and the driving mutation is not
clearly identified in most tumors. These tumors have frequency copy number changes and
aneuploidy with an average number of 17.8 non-synonymous somatic mutations per
tumor(9). Among the known oncogenic mutations, activation of Ras (N-Ras, K-Ras, H-
Ras), FGFR4, or PIK3CA are the most commonly mutated genes (9-11). Activating
mutations in CTNNB1 or BCOR have also been reported although their incidence is much
lower. Tumor suppressors TP53, PTEN, or NF1 are also inactivated at diagnosis in a small
percentage of patients. Common copy number alterations include LOH of 11p15.5
encompassing the IGF2 gene locus is found in nearly half of all eRMS tumors (9, 12). Whole P a g e | 16 chromosome gains and losses can be observed, with gain of chromosome 8 in 46% of eRMS patients(13). ii.) Spindle/sclerosing: Sequencing of ssRMS patient tumors had led to the identification of two distinct groups. The first is a variant typically observed in infants (0-2yo) and is characterized by one of many fusion genes such as SRF-NCOA2 (14), VGLL2-CITED2 (3), and VGLL2-NCOA2(3). These tumors have generally favorable outcomes although data is limited. ssRMS in adolescents and young adults, however, is quite different. These tumors typically contain a MYODL122R activating mutation and/or PIK3CA mutations and have very poor clinical outcomes with less than 30% (3, 15, 16). The current understanding of the
MYODL122R mutation is that this amino acid substitution confers MYC-like transcriptional activity as these two are related bHLH paralogs(17). iii.) Alveolar: aRMS tumors have strikingly simple genomes with an average number of
6.4 non-synonymous somatic mutations per tumor with less than half of these detectable in transcribed genes(9). Whole genome sequencing has also identified tumors in which the t(2;13) rearrangement (PAX3/FOXO1) was the only genomic abnormality. Despite a low frequency of non-synonomyous somatic mutations, aRMS have copy number changes which likely contribute to disease. The chromosomal region 12q13-q14 is frequency amplified in aRMS (10%) which includes the CDK4 and the 2p24 amplicon that includes
MYCN are frequently observed and correlates with overexpression of these oncogenes(9,
18).
The hallmark genomic alteration observed in aRMS is the t(2;13)(q35;q14) or t(1;13)(p36;q14) rearrangement which results in an in-frame fusion of PAX3 or PAX7 to
FOXO1, respectively. These translocations are observed in 60% (Pax3/FoxO1) and 20% P a g e | 17
(Pax7/FoxO1)(19) of all aRMS tumors and function as the central oncogenic feature in aRMS. The breakpoints for these rearrangements are remarkably consistent, with intron 7 of PAX3 or PAX7 to intron 1 of FOXO1. This generates an in-frame fusion gene that contains the N-terminus and DNA binding domains of PAX3/7 proteins and the C-terminal transactivation domain of FoxO1, resulting in a potent chimeric transcription factor(20-23).
Early analysis of these genes demonstrates that these chimeric transcription factors bind
DNA with similar affinities to wildtype PAX3/7 but are capable of activating transcription a thousand times higher than their wildtype counterparts(21, 22). ChIPseq analysis of
PAX3/FOXO1 in aRMS tumor cells demonstrates that well characterized wildtype PAX3 binding sites (MYF5 enhancer) are also bound and activated by PAX3/FOXO1, further supporting the notion that the fusion genes have similar transcriptional targets as the wildtype proteins(24).
Little is known about the similarities and differences between the PAX3/FOXO1 and
PAX7/FOXO1 fusion genes. PAX3 and PAX7 are highly conserved proteins and the fusion genes have high structural similarities. Both PAX3 and PAX7 are involved in skeletal muscle development, although PAX3 is more involved in myogenic specification and migration while PAX7 is involved in maintaining stemness of skeletal muscle satellite cells(25). In aRMS tumors, the PAX7/FOXO1 fusion gene is almost always genomically amplified while only 9% of PAX3/FOXO1 fusion genes are amplified(26). Some evidence also suggests that the PAX7/FOXO1 fusion has a slightly better prognosis than
PAX3/FOXO1, although this difference is not well established(27-29).
Rare alternative fusion genes have also been identified in aRMS such as the
PAX3/NCOA1(30), PAX3/NCOA2(30), PAX3/INO80D(9), and PAX3/FOXO4(31). These P a g e | 18
alternative fusions make up less than 5% of aRMS tumors and nothing is known about their
biological mechanisms. Approximately 15-20% of all aRMS tumors lack any fusion gene.
Gene expression unsupervised hierarchical clustering of RMS tumors has demonstrated
that these fusion negative aRMS tumors are indistinguishable from eRMS no clustering
with fusion positive aRMS(13). Multiple large-scale retrospective studies accounting for
fusion status have observed that the presence of a PAX fusion gene, not aRMS histology,
is the single best predictor of patient outcome (19, 28, 32, 33).
II. Identification of a fusion-positive aRMS gene signature
Rhabdomyosarcomas are defined by the striated skeletal muscle features observed in tumor cells. Although the patterns of expression can differ, all rhabdomyosarcomas express myogenic markers such as myogenin, MyoD1, desmin, and muscle-specific actin. Despite this similarity, tumors which contain the PAX3/FOXO1 oncogene are notably more aggressive and are associated with poor outcomes. Beyond the genetic basis for this difference, nothing was known about what makes fusion positive tumors unique.
Multiple gene expression studies have profiled RMS patient tumors in an attempt to identify biomarkers of prognosis and discover disease contributing genes. Unsupervised hierarchical clustering of these data led to the identification of a recurrent pattern of gene expression which was unique to fusion positive tumors. Some differences exist between studies, likely due to differences in analyses or reflecting variation among patient cohorts. A small list of upregulated genes span all datasets, suggesting that these genes are the most consistently and uniquely upregulated genes in fusion positive RMS. Forkhead Box F1 (FOXF1) is one of the most consistently and uniquely upregulated genes in aRMS patients (Table 1). The three largest RMS patient gene expression studies, totaling 195 RMS patients, all identified FOXF1 as one of the P a g e | 19
most consistently upregulated mRNAs in aRMS. Analysis of FOXF1 protein expression in aRMS
was confirmed by immunostaining and western blot analysis of aRMS primary tumors and tumor
xenograft models (Fig. 3). Altogether, the robust statistical significance of this upregulation across
patients suggests i.) most aRMS tumors, if not all, express FOXF1 and ii.) FOXF1 is completely
absent in eRMS tumors.
Table 1. FOXF1 expression is correlated with fusion positive RMS. Gene expression profiling
of 195 RMS patient tumors across three independent studies identify FOXF1 as one of the most
significantly upregulated transcripts.
Expression significantly Study Source correlated with fusion status Microarray on RMS Patients n=38 (23 Lae et al 2007(34) YES FP, 15 FN) Davicioni et al Microarray on RMS Patients n=128 (55 YES 2009(27) FP, 73 FN) Watchtel et al 2004 Microarray on RMS Patients n=29 (10 YES (30) FP, 19 FN)
(a) FOXF1 IHC (b) Human RMS xenograft tumor lysates
eRMS aRMS Patient eRMS Patient
FoxO1 Pax3/FoxO1
(C-term.) FoxO1
FoxF1
β-Actin Patient aRMS Patient P a g e | 20
Figure 3. FOXF1 is only expressed in aRMS tumors. (a) Immunohistochemical staining of human RMS patient tumors. Magnification 100x. (b) Western blot analysis of RMS xenograft models established from primary patient tumors and cell lines. The C-terminal FOXO1 antibody recognizes both the wildtype FOXO1 protein (75kda) and the PAX3/FOXO1 fusion gene (105kda).
III. Forkhead Box F1 (FOXF1)
a) Features of the FOXF1 gene locus.
STOP
Figure 4. FOXF1 exon 2 3’UTR contains hot spots of conservation. Nucleotide conservation
analysis across 100 vertebrate species identifies highly conserved non-coding regions in the
3’UTR. The human FOXF1 gene locus is located on chromosome 16:86544133-86548070 and is
composed of two exons and one intron. The majority of the FOXF1 coding region, including
the DNA binding domain, is located in exon 1. Exon 2 is mostly composed of a 3’ UTR
(~1,275bp) which represents more than half of the FOXF1 transcript. This region contains
hotspots of high conservation presumably containing target sites for microRNA or other post-
transcriptional regulatory mechanisms (Fig. 4). The FOXF1 gene promoter has a unique
structure approximately 2kb upstream of the FOXF1 transcriptional start site lies a long
noncoding RNA FENDRR. This noncoding transcript mirrors FOXF1 expression across tissues
and is likely due to bidirectional transcription from this promoter region.
P a g e | 21
Conservation
Figure 5. FOXF1 is located in gene cluster and has nearby long noncoding RNAs
(LINCRNA). (top) View of all transcripts produced around the FOXF1 locus. (bottom) FOXF1
shares a 2.0kb bi-directional promoter with FENDRR, a long-noncoding RNA.
The region 400kb upstream of FOXF1 and FENDRR lies a series of conserved enhancer
elements and long noncoding RNAs considered to function as enhancer RNAs (Fig. 5)(35, 36).
Deletion of these putative enhancer regions in humans results in decreased FOXF1 expression
and phenocopies loss-of-function mutations of the FOXF1 coding region (discussed Ch.1
sect.IIId). Two eRNAs have been identified to date as being correlated with enhancer activity
and FOXF1 expression. The first eRNA is termed LINC01081 and lies -225kb upstream of the
FOXF1 gene. Deletions of this region, including the LINC01081 gene and the enhancer located
at the LINC01081 TSS result in dramatically reduced FOXF1 expression and phenocopies
genomic disruption of the FOXF1 gene locus. Knockdown of LINC01081 results in reduced
transcriptional of the FOXF1 gene, suggesting that LINC01081 functions to facilitate enhancer
looping similar to other eRNAs. The second enhancer lies slightly more distal at -315kb
upstream of FOXF1. This enhancer and corresponding eRNA LINC01082 are also associated
with FOXF1 expression. Neither of these eRNA genes are found in the mouse, suggesting the
presence of human-specific gene regulation.
P a g e | 22
The hedgehog signaling pathway is the only known inducer of FOXF1 gene expression. Several
FOXF1 enhancers are bound and positively regulated by hedgehog pathway effectors GLI1 and
GLI2 to induce FOXF1 expression in the embryonic mesenchyme (37, 38). The location and
regulation of the other enhancer elements in this region have yet to be characterized. Lastly,
the human FOXF1 gene is paternally imprinted in some tissues resulting in exclusive expression
from the maternal allele(38). The FOXF1 locus is not imprinted in mice and likely explains why
only the FOXF1-/- mice can recapitulate the lethal phenotype observed in FOXF1+/- humans
(discussed Ch.1 sect.IIId). b) Structural anatomy of the FOXF1 protein.
The human FOXF1 protein is 379 amino acids long (NP_001442). FOXF1 has an N-terminal
winged helix DNA binding domain which is structurally similar to other Forkhead family
members. Amino acids 48-133 correspond to the Forkhead DNA binding domain with amino
acids Phe 84, Phe 85, Asn 94, Arg 97, His 98 and Lys 118 predicted to be involved in the actual
interaction with DNA (Fig. 6). The FOXF1 protein contains a c-terminal transactivation
domain(39) although the exact regulatory regions of this domain are not well defined. The
FOXF1 protein is highly conserved between mouse and human with 93% sequence similarity.
P a g e | 23
Figure 6. Location of the DNA binding domain in the FOXF1 protein. (top) Protein
conservation analysis for the human FOXF1 protein identified a highly conserved Forkhead
superfamily domain. (bottom) Prediction of six amino acids within the FOXF1 Forkhead
domain which are involved in protein-DNA recognition.
c) Expression in embryonic and adult tissues.
FOXF1 is exclusively expressed in mesenchymal cells. FOXF1 is expressed in most
stromal cell populations of the respiratory tract (trachea, lung), gastrointestinal tract (esophagus,
stomach, liver, gallbladder, pancreas, duodenum, small intestine, colon, bladder, anus), heart,
brain, and reproductive tract (uterus, vagina). FOXF1 is also found in the endothelial cells of
the yolk sac, lungs and retina as well as more restricted expression in smooth muscle cells of
the gastrointestinal and respiratory tracts. Lastly, FOXF1 is also expressed in chondrocytes of
the nasal septum, Meckel’s cartilage in the lower mandible, and vertebral bodies(40-42). d) Phenotypes of FOXF1+/- humans and FOXF1-/- mice.
Loss of one FOXF1 allele in humans or both alleles in mice is early embryonic lethal due to
extra-embryonic defects(41-44). In mice, FOXF1 haploinsufficiency can have a spectrum of
severities, but the development of many organs is severely impaired. Depending on the genetic
background, up to 90% of heterozygous animal die perinatally. In addition to the extra-
embryonic defects, loss of FOXF1 is associated with lung malformations including hypoplasia,
overall immaturity, vascular defects, and improper branching of the primary lung lobe resulting
in fusion(41, 42, 44).
In humans, the loss of maternally expressed FOXF1 is 100% lethal. Infants are generally
born alive but decease within several weeks due to a progressive neonatal respiratory failure.
This is congenital defect is classified as Alveolar Capillary Dysplasia with Misalignment of P a g e | 24
Pulmonary Veins (ACD/MPV) and is characterized by a loss of capillaries, thickening of
pulmonary arteries, and with misaligned pulmonary veins(45, 46). In addition to this, there are
severe cardiac and gastrointestinal malformations and usually an absence of gallbladder and/or
spleen. e) Expression of FOXF1 in human cancer.
The role of FOXF1 in human cancers is not well characterized. In carcinomas of the lung,
colon, prostate, pancreas, and breast, FOXF1 mRNA expression is decreased. Since FOXF1 is
only expressed in the mesenchyme of these tissues, much of this can be explained by the reduced
representation of these cells in bulk tumor samples giving the impression of downregulated
expression in tumor cells. Data from our lab demonstrates, however, that tumor associated
endothelial and stromal cells downregulate FOXF1 in the context of cancer.
FOXF1 is expressed in several sarcomas, however. Alveolar rhabdomyosarcoma,
gastrointestinal stromal tumor (GIST), Ewing-like sarcoma with BCOR/CCNB3
rearrangement, and clear cell sarcoma of the kidney all have consistently high expression of
FOXF1 and sporadic FOXF1 expression observed in synovial sarcoma. Out of these, only GIST
has been investigated for the role of FOXF1. In this study, FOXF1 was identified as an
upstream transcriptional activator of KIT and ETV1, both of which are the central drivers of
GIST tumors(47). FOXF1 functioned as a transcriptional activator by binding with ETV1 at loci
associated with Cajal cell (cell of origin) gene function and GIST identity. The authors claimed
that this was evidence of pioneer activity but failed to provide any evidence that FOXF1 can
directly open chromatin de novo. In our early investigation into the impact of FOXF1 in
rhabdomyosarcoma, we identified that FOXF1 promoted aRMS proliferation by overriding
(48) G1/S checkpoint proteins CDKN1A and CDKN1B . The clear association of FOXF1 and the P a g e | 25
PAX3/FOXO1 fusion protein, however, suggests a more intimate relationship between these two transcription factors and is the focus of the current study. This dissertation provides a detailed analysis into the upstream regulation and downstream consequences of FOXF1 expression in the context of the PAX3/FOXO1 fusion protein. P a g e | 26
Chapter 2: Tumor modeling using patient-derived xenografts (PDX):
CCHMC Sarcoma PDX Initiative
I. Abstract
Patient derived xenograft (PDX) models are a rapidly emerging platform for studying tumor biology and response to therapy. Traditional cell-line based xenograft approaches have largely failed to accurately predict clinical responses due to genetic and phenotypic drift during the culture process. PDX models resolve this shortcoming by allowing for a minimally selective in vivo system for expanding human tumor tissue. This approach also provides the broad genomic and biological heterogeneity seen within tumor types which is largely absent in transgenic tumor models. For these reasons, we established a PDX consortium with the goal of establishing and characterizing soft tissue and bone solid tumors. Since many of these sarcomas lack truly representative models for research and drug development, little progress has been made in developing targeted therapies. This chapter details the technical advancements, tumor data, and ongoing studies of the CCHMC Sarcoma PDX Initiative.
II. Introduction
Culture and xenografting of cancer cell lines has been the backbone of drug development. This approach has historically relied on the ability to establish long-term cancer cell cultures in a 2D in vitro system. Despite the malignant nature of tumor cells, particularly the uncontrolled proliferation and evasion of apoptosis, very few tumors are capable of giving rise to stable cell lines. Multiple large-scale efforts to establish human cell lines from aggressive solid tumors have demonstrated the low efficiency of this process. For example, found only 15 of 570 lung cancers P a g e | 27
(2.6%)(49), 2 of 44 breast cancers (4.1%)(50), and 6 of 81 pancreatic cancers (7.4%)(51) are capable of generating stable cell lines. The incongruity between their aggressive nature in patients and their inability to persist in culture demonstrates the significant shortcoming of cell-line based approaches. In addition to the strong negative selection of cells as the cell line is being established, genetic and phenotypic drift becomes a major concern. Accumulating evidence now shows that the poor representative nature of cancer cell line model systems are one of the leading impediments to successful drug development.
The use of immunocompromised mouse strains is required to prevent immuno-rejection of human tissue by the mouse host immune system. Tissue graft rejection is largely carried out by the T-cells as a component of the adaptive immune response and mice containing defects in T-cell development and maturation display a reduced capacity to reject human tissue upon xenograft.
While the first immunocompromised mouse strains such as the athymic nude mice or NOD-scid mice were derived from mice with a spontaneous mutation, emerging gene editing technologies have allowed for the rapid generation of strains with multiple, targeted mutations to the mouse genome. Common immunocompromised mouse strains, including those used in this study (*) are summarized below:
Athymic/nude. The athymic nude mouse strain was one of the first
immunocompromised mouse strains discovered. This mouse contained a
spontaneous mutation in the FOXN1 transcriptional factor. This mutation was
caused by a single base deletion in exon 3 which results in a frameshift mutation
and loss of function to the FOXN1 gene. The loss of the FOXN1 protein results in
faulty hair growth, lack of whiskers, and most notably the absence of a functional
thymus. The absence of the thymus results an absence of mature T-cells, although P a g e | 28 immature T-cells are still present. The lack of functional T-cells also results in reduced B-cell function. All other immune cell types are present and mostly functional which resulting in moderate immune-competence of nude mice.
NOD-Prkdcscid. These mice carry a spontaneous mutation in the protein kinase,
DNA activated, catalytic polypeptide gene PRKDC which results in the Severe
Combined ImmunoDeficiency (SCID) syndrome in mice on a NOD background.
The PRKDC gene encodes a kinase which is required for non-homologous end joining during DNA repair of double strand breaks. Since V(D)J recombination also requires this enzyme for repairing the double strand breaks, the immunoglobulin gene loci fail to undergo rearrangement. As a result, no mature B cells or T cells develop in Nod-scid mice. The remaining immune cell types are present but display reduced activity. The downside of using this background is that these mice are extremely sensitive to therapies which induce double stranded breaks
(radiation, PARP inhibitors, ect).
* NOD-Prkdcscid-Il2rgtm1Wjl (NSG). NSG were used throughout the current study for establishing PDX xenografts and were the preferred strain of mice. These mice are similar to the NOD-Prkdcscid mice but carry an additional loss of function mutation in the IL2Rgamma gene. The absence of this receptor component disrupts multiple signaling pathways and abolishes NK function. Since these mice lack functional T-cells, B-cells, and NK-cells, they have no cytolytic ability against foreign cells allowing for un-impeded human engraftment.
* NOD-Prkdcscid-Il2rgtm1Wjl-TGcmv-IL3,CSF2,KIT (NSGS). NSGS mice are ideal for
“humanization” of the mouse immune system. This modified NSG mouse contains P a g e | 29
a single transgenic cassette which encodes three human cytokines (IL3, GM-CSF,
and cKIT). These three factors do not cross react from mouse to human and are
required for human immune development, especially for myeloid cell types. The
addition of the human genes to these mice allow for transgenic expression of the
human isoforms and allow for engraftment, expansion, and multi-lineage
hematopoietic reconstitution. Note: Mature T-cells achieve full maturity around 3
months after reconstitution with human CD34+ progenitors, upon which mild
GVHD symptoms will develop in the mice.
* NOD-RAG1tm1Mom-Il2rgtm1Wjl-TGcmv-IL3,CSF2,KIT (NRGS). These mice are one
of the most versatile strains available. NSGS and NRGS mice have the same lack
of B, T, and NK cells and equally impaired macrophage and dendritic cell function.
They also both can fully support human immune cell development by the inclusion
of the human IL3/GM-CSF/KIT transgene. The NRGS mice are different by RAG1
mutation instead of the Prkdcscid mutation. The recombination activating gene 1
(RAG1) is also involved in double strand break repair similar to PRKDC. Unlike
PRKDC, however, RAG1 is required for V(D)J recombination but is dispensable
for non-homologous end joining. As a result, NRG or NRGS mice are significantly
more resistant to genotoxic drugs or radiation.
Direct xenografting of primary human tumor tissue is an attractive alternative to traditional cell-line based xenografts. With this approach, tumor tissue is directly implanted into immunocompromised mice at heterotopic or orthotopic sites. The tumor tissue engrafts to the host
(mouse) vascular and allowed to grow over the course of several months. These patient-derived xenograft (PDX) models avoid the strong artificial selection that takes place during in vitro culture P a g e | 30 and allows for the maximal retention of tumor cell heterogeneity and biology. These tumors can be “passaged” from mouse to mouse for multiple generations without significant changes in tumor cell genomic, transcriptomic, or biological heterogeneity. PDX tumor tissue can also be cryopreserved to maintain long-term stocks of human tumors. The engraftment rates for this approach can be variable, ranging for 20% to 70% (Table 1). Generally, recurrent/relapsed tumors have higher engraftment rates presumably due to increased genomic complexity and aggressiveness.
Sarcomas are a highly heterogeneous group of tumors which can occur across age groups.
While the incidence of sarcomas is greatly lower than that for carcinomas and hematopoietic malignancies, they still significantly contribute to overall patient mortality. Among the pediatric population, the overall 5-year survival rate for bone tumors and soft tissue tumors are 66% and
69% respectively, compared to 78.2% observed for all pediatric cancer patients(52). This disparity in survivorship between sarcoma and non-sarcoma patients can be attribute to multiple factors, including low tumor accessibility for surgical excision, availability of effective system therapies, and our overall understanding of the genetic and molecular drivers of these tumors.
Table 1. Engraftment rates for bone and soft tissue tumors in previous studies. Study Tumor Type Engraftment Rate (%) (4) Bauer et al Osteosarcoma 14/25 (56%) (5) Meyer et al Osteosarcoma 8/33 (24%) (6) Fujisaki et al Osteosarcoma 21/34 (62%) (7) Bruheim et al Osteosarcoma 11/55 (20%) (8) Stewart et al Bone and soft tissue sarcomas 67/148 (45%) (9) Hoffmann et al Bone and soft tissue sarcomas 31/82 (37.8%) (10) Meohas et al Bone and soft tissue sarcomas 11/16 (69%) Total: 163/393 (41.5%)
P a g e | 31
Accumulating evidence also shows that age is significantly correlated with patient outcomes.
Adolescent and young adult (AYA) sarcoma patients tend to have poorer outcomes compared to the same diagnoses in their pediatric counterparts. A massive retrospective study of over 30,000 sarcoma patients was carried out in Europe between 2000 and 2007. Comparison between pediatric (0-14) and adolescent and young adult (15-39) patients found significantly lower survivorship across all sarcoma types for AYA patients(53). Some of this can be attributed to differences in disease management, as some AYA patients receive treatment at adult centers who lack experience in treating these diseases. Biological differences likely also contribute to this disparity. Genomic profiling of sarcomas has demonstrated recurrent mutations which are more common among pediatric or AYA patients.
We and others have identified a clear need for improving our understanding and treatment of bone and soft tissue tumors, particularly within the AYA population. The overall goal of the current study is to harness the power of the PDX platform to establish human sarcoma models for research and drug development. Ancillary objectives include establishing PDX models from untreated and AYA patients. These sarcoma models will provide the foundation for basic research and for testing emerging therapies such as epigenetic or immune-based therapies.
III. Materials and Methods
Workflow for establishing sarcoma PDX models. Patients are identified based on confirmed or suspected diagnosis of sarcoma originating from bone or soft tissue. All tissue collection from patients are exclusively from procedures on the patient as part of their medical care. Only excess tissue not needed by pathology, biobanking, clinical trial protocols, or other purposes can be used for the purpose of establishing PDX models. Tissue which meets these requirements are collected in RPMI media supplemented with 2x antibiotics/antimiotics (GIBCO) and maintained on ice and P a g e | 32 immediately transferred to research. Once received, the tissue is processed and designated from high to low priority: transplant into mice > cryopreservation > RNA > histology > culture > protein. The methodology for each of these is detailed below.
Xenografting of primary tissue into mice. All transplantation studies were done in NSG
(Jackson #005557), NRG (Jackson #007799), or NSGS (Jackson #013062) mice. For the purpose of generating PDX models, all three strains are to be considered equivalent. The majority of models have been established using NSG mice. Athymic nude, Scid, or Rag mice were not used.
NSG-W (NSG+KITW41/W41) mice were used for several F2 PDX mice and no difference in engraftment rate was observed. Mice of ages between 4-8 weeks were preferred and male and female mice were used interchangeably. For subcutaneous grafts, a fragment of tissue with a maximum diameter of 2mm in any dimension were used. Mice were anesthetized using isofluorane and a 1cm incision was made along the midline of the mouse lower back and the subcutaneous interstitial tissue was gently separated from the dorsal muscles. The tissue fragment was placed on the far side, superior and lateral to the hip joint along the draining lymphatic blood vessel. Mice were sutured and in some instances, a subcutaneous injection of 0.1cc of 50%
Matrigel and PBS was injected at the implantation site to promote vascularization. For intramuscular injection, tissue was carefully sliced lengthwise until the diameter of the tissue was approximately 0.5 x 3mm. These dimensions allow for the tissue to be gently loaded up a 17ga needle and “reverse-biopsied” into the quadriceps of the mice. This approach is preferred for core needle biopsies.
Processing tissue for cryopreservation. Tissue was sliced or minced to a maximum of 1.5mm diameter in one dimension. On occasions in which significant amounts of tumor were received, the tumor tissue would be partially dissociated (see attached protocol) before cryopreservation. P a g e | 33
Tissue was placed in a vial with 10% FBS + 90% media, placed in a Mr. Frosty container, and stored at -80C for 24 hours. Frozen tissue was transferred to liquid nitrogen for long-term storage.
Processing tissue for RNA. Tissue is placed in a minimum of 10 volumes of room temperature
RNALATER storage solution. This solution rapidly penetrates tissue, denatures proteins
(RNAses) and preserves RNA integrity. Whenever possible, the tissue was a maximum of 1.5mm in one dimension to ensure good penetration of the tissue. The tissue was incubated overnight at
4C with gentle agitation. The following day, the RNALATER solution was carefully removed and the bare tissue was stored at -80C.
Processing tissue for histology. Tissue was fixed in 4% paraformaldehyde in PBS O/N at 4C with gentle agitation. After 24 hours, the tissue was washed twice with PBS, dehydrated and embedded in paraffin following typical tissue embedding procedures.
Processing tissue for cell culture. Tumor tissue was either minced or digested before plating.
Tumor cells were cultured on gelatin coated plates in RPMI media supplemented with 10% FBS,
1x amino acids, 10mM HEPES, 1x antibiotics/antimiotics, 10nM hydrocortisone, and 10nM β- estradiol. Fragmented tissue was allowed to attach to the dish over a maximum of 10 days before it was removed.
Processing tissue for protein. Tissue was snap frozen and stored at -80C. Protein lysates were prepared using RIPA buffer supplemented with 300mM NaCl final concentration, protease inhibitors, and phosphatase inhibitors.
Tumor Dissociation Protocol P a g e | 34
Reagents:
− RPMI
− 100x Antibiotics + antimiotic: ThermoFisher Cat. #15240-062
− Liberase TM: resuspended to 5mg/mL with water, aliquot and freeze -20C
Roche Cat. #05401127001
− DNASE I: Sigma Cat.# DN25
- ACK Lysing Buffer: Gibco cat. #A1049201
Dissociation media:
RPMI
Anti-anti (2x Final)
Liberase TM (20µl/mL Final)
DNASE I (0.5mg/mL Final)
Note: Make dissociation media fresh, pre-cool to 4C
Tumor Dissociation:
1. Harvest tumor, immediately take 1-2 small (2mm) fragments for RNA put into
RNALATER on ice
2. (optional) Cross-sectional slice through whole tumor for histology (4% PFA O/N)
3. Keep tumor in dissociation media in 50mL conical tube on ice until ready to proceed.
a. General media volume is ~10mL dissociation media per 1cm diameter tumor
4. Carefully remove tumor from dissociation media, place in plastic dish, and mince tumor
finely (<1mm fragments) with blade P a g e | 35
a. Thinnest blades possible, i.e. scalpel blades
b. Avoid “chopping,” but try to slice the tissue
c. For fibrotic/necrotic or calcified tumors, cut into larger pieces (1-3mm) and pre-
digest for 20 minutes in a tissue culture plate to soften the tissue before further
cutting into 1mm fragments.
5. Carefully scoop minced tissue back into dissociation media in 50mL conical tube.
6. Incubate in a 37C water bath for 25-60 minutes, swirling the tissue every 5 minutes
a. Time is dependent on tumor density, matrix composition, ect.
b. For “permeabilizing” tissue for freezing, use 30 minutes
7. Optional: intact tissue can be further minced if any fragments are too large (>3mm). If
fragments are minced further, incubate tissue for an additional 5 minutes to allow complete
DNASE treatment of cells killed by additional cutting.
8. Add equal volume of complete media (RPMI+10% FBS+anti or equivalent), and pellet at
200g x 5 min at 4C
9. Optional: If tumor was highly vascular and bloody, resuspend pellet in 5mL of ACK-lysis
buffer, incubate 5 min at RT, and re-pellet. We have not noted any adverse affects of
leaving RBCs.
10. Gently resuspend tissue in complete media using a cut pipette tip.
a. For cryopreservation: 20 pellet volumes of media is a good starting point.
Resuspend the tissue well and add an equal volume to 2x Freezing media (80%
FBS, 20% DMSO). Mix well by inverting and keep on ice until all tissue is
aliquoted. Place tubes in a Mr. Frosty container (or equivalent) and put in -80C
for 24-72 hours. Transfer to liquid nitrogen. P a g e | 36
b. For culture: gently resuspend tissue in 10mL of complete media. Mix by inverting
for 10 seconds, place upright for 3 seconds, and collect all supernatant and transfer
to a tissue culture dish (usually one to three 10cm dishes). This unsettled fraction
represents single cells and small tissue fragments (<1000 cells) which all can
readily adhere in vitro. Coating of dishes is generally recommended (gelatin,
collagen, fibronectin). Media components will need to be optimized.
Viability of single cells can be anywhere from 60% to >90%, mostly depending on the tumor.
Cell clusters/tissue fragments will adhere to the plate and cells will migrate out. It is common to keep these tissue chunks in the plate for >10 days as more and more cells migrate out. Every day, for the first 3 days, wash 3x with PBS and replace media. Any dislodged tissue fragments are collected, set aside, and added back to plate after washing.
P a g e | 37
IV. Results
Patient summary. A total of 63 tissues from 59 patients were acquired 63 Tissues (59 patients) from July 2016 to March 2019. Of these 59 patients, 51 had a confirmed diagnosis of sarcoma. The other diagnoses included squamous cell carcinoma 54 Sarcomas (n=3), hepatoblastoma (n=1), lipoblastoma (n=1), atypical smooth muscle (50 patients) tumor (n=1), neurofibroma (n=1), nasopharyngeal carcinoma (n=1), and B-cell lymphoblastic lymphoma (n=1). (Note: only sarcoma cases will be described 43 Implanted herein). The tissue was taken from patients by a variety of methods, both core needle biopsies (n=24), excisional resections (n=16), incisional resection (n=10), or fluid collection (n=5). The spectrum of sarcoma patient diagnoses in this study reflects their frequency in the pediatric population, with the majority of patients with Ewing sarcoma, osteosarcoma or rhabdomyosarcoma (Fig. 1).
Implantation and growth characteristics. Tumor implantation was performed at both heterotopic and orthotopic sites. In general, two mice were implanted with each tumor tissue: one mouse with two subcutaneous xenografts on each flank and a second mouse with a single intramuscular injection of tumor. No significant differences in engraftment rates were observed between subcutaneous and intramuscular transplants, although the intramuscular grafts tended to grow at a faster rate. It is unclear if this is attributed to differences in implantation technique or to the contribution of the microenvironment (vasculature).
The growth rates and overall success rates varied significantly across tumor types. The fastest observed growth rates following implantation occurred from a locally relapsed telangiectatic
P a g e | 38
Sarcoma Patient Diagnoses (n=50) DSRCT EWS: Ewing Sarcoma SS ADM CS OS: Osteosarcoma MRT RMS: Rhabdomyosarcoma ASPS 1 1 1 EWS Undiff/Unclass.: undifferentiated, 1 1 unclassified sarcomas (includes MPNST 2 epithelioid sarcomas) n=13 GCT 2 SCCOHT: Small cell carcinoma of the 2 ovary, hypercalcemic type SCCOHT GCT: Germ cell/yolk sac tumor 2 MPNST: Malignant peripheral nerve sheath tumor n=5 Undiff./ ASPS: Alveolar soft part carcinoma n=11 MRT: Malignant Rhabdoid tumor Unclass. n=8 SS: Synovial Sarcoma DSRCT: Desmoplastic small round cell OS tumor RMS CS: Chondrosarcoma ADM: Adimantinoma Figure 1. Spectrum of sarcoma patient diagnoses. Pie chart showing the frequency and diagnosis of patients with tissue for PDX modeling. osteosarcoma (35 days to completion) and a bone metastasis from an untreated fusion positive alveolar rhabdomyosarcoma (25 days to completion). The longest time to PDX establishment was from a locally recurrent spindle/sclerosing rhabdomyosarcoma (303 days; 9 months and 28 days) and a highly calcified osteosarcoma lung metastasis (249 days; 8 months and 4 days). These two had stagnant growth for 4 months following implantation, after which they slowly began to progress. Given the possibility that some PDX models may take this amount of time to demonstrate progressive growth, we have defined a 9 month timeline from implantation to allow for tumor progression.
Overall engraftment efficiency. Out of 54 sarcomas in our study, 43 tumor tissues were implanted into mice (Table 2). The other 11 tissues were cryopreserved immediately or contained poor viability unsuitable for transplantation. As of March 2019, 35 of 43 implanted tissues have P a g e | 39 conclusive (finished) results for the first generation in mice. From the 35 completed tissue implants, 26 of implanted tissues (74.3%) successfully generated a progressively growing PDX model within 9 months. This success rate is significantly higher than what has been achieved by other sarcoma PDX consortiums (Table 1). These 26 established PDX models represented a wide range of diagnoses. All rhabdomyosarcoma tumor tissues (7 of 7; 100%) generated PDX models
(Table 3). Roughly 2/3 of osteosarcomas and Ewing sarcomas successfully generated PDX models. Out of the 26 PDX models we established, 13 (50%) were taken from diagnostic biopsies of tumors with no prior exposure to cytotoxic therapies. The establishment of 13 untreated PDX models is one of the main accomplishments for this work and will allow for clear insight into underlying genomic and molecular features of these tumors.
Table 2. Engraftment rate of implanted patient tissue. *Established PDX defined as the progressive growth of tissue within nine months of transplantation. Tissues without progressive growth at nine months, even if still it has remained viable, were considered failed grafts.
Tissues implanted Established Established & Diagnosis Success rate (including PDX* Treatment Naïve ongoing) EWS 11 6 of 9 68% 4 OS 11 5 of 6 83% 1 RMS 8 7 of 7 100% 3 Undiff./Unclass. 3 0 of 3 0% - GCT 2 2 of 2 100% 2 SCCOHT 2 1 of 2 50% - MPNST 2 2 of 2 50% 1 SS 1 1 of 1 100% - DSCRT 1 1 of 1 100% 1 MRT 1 1 of 1 100% 1 ASPS 1 0 of 1 0% - Total 43 26 of 35 74.3% 13
P a g e | 40
Table 3. Characteristics of established rhabdomyosarcoma PDX models. Clinical and histopathological classification of rhabdomyosarcoma patients with established PDX models.
PDX Histology Anatomical Location Treated Stage Genomics
SARC Spindle/sclerosing Bicep Y I Unknown 160718 SARC Lymph node adjacent Embryonal Y III NRAS 171030 to stomach SARC Embryonal Nasal cavity Y II NRAS 180612 SARC Alveolar Cheek Y II Pax3/FoxO1 170206 SARC Alveolar Bone Marrow N IV Pax3/FoxO1 170503 SARC Alveolar Chest Wall N III Pax3/FoxO1 180409 SARC Alveolar Parameningeal N IV Pax3/FoxO1 180613
Once established, all tumors were serial passaged to at least a second generation (F2) in mice to ensure the stability of the PDX model and expand the amount of tumor tissue for future study.
All mice with established F1 PDX models that were re-transplanted had successful engraftment and established an F2 PDX model (57 of 57; 100%). We also tested the success rate of our cryopreservation protocol. Seven different cryopreserved PDX tumor models were thawed and reinjected after at least one month of storage in liquid nitrogen. All seven (100%) cryopreserved
PDX tissues successfully generated a PDX model when implanted into mice. The time to engraftment was delayed by roughly 30% of what occurred when fresh tissue was used. This is presumably due to a reduction in tissue viability during the freezing process and the handling of P a g e | 41
the tissue a. Whenever a tumor did not progress after the nine month time point, we looked for
evidence of residual viable human tissue at the injection site. In roughly 1/3 of all failed grafts,
we were able to find clear evidence of engrafted human tumor cells which failed to proliferate to
generate progressive tumor graft (Fig. 2.).
(a) (b)
Figure 2. Engrafted and non-progressing tissue xenografts over long term. (a) H&E staining
of biopsy from a mouse 6 months after implantation of a sclerosing rhabdomyosarcoma. The
confirmation of tumor cells was verified using myogenin IHC (inset) but low proliferative activity.
(b) H&E staining of a nasopharyngeal carcinoma at 8 months after implantation. A surprising
number of tumor cells were present in this 1mm mass but failed to progressively grow.
In vitro culture of PDX or patient-derived tumor cells varied dramatically between tumor
types. Embryonal rhabdomyosarcomas, synovial sarcoma, MPNST, rhabdoid, and germ cell
tumors were the most efficient in vitro growing tumors. Although not a sarcoma, cells derived
from a hepatoblastoma metastatic site also rapidly grows in vitro. In an effort to standardize in
vitro growth conditions and assess tumor growth patterns, a co-immunofluorescent approach was
used for PDX-derived tumor cultures. Fibroblast contamination is an issue for many cultures P a g e | 42 and becomes a significant problem for slow-growing cultures. While it is unclear if the presence of fibroblasts impact the growth rate of the tumor cells, fibroblasts can become the dominant cell type in long term cultures (3+ weeks). To determine the tumor purity of cultures, cells were co- stained with alpha smooth muscle actin (mouse fibroblast) and human nuclear antigen (tumor cells) (Fig. 3).
(a) Alveolar (b) Rhabdomyosarcoma PDX Alveolar Rhabdomyosarcoma PDX (Intramuscular) (2 week culture) Pan-human αSMA
Low mag. Low
DAPI Merged
human nuclei IHC IHC nuclei human
-
Pan
High mag. High
Figure 3. Pan-human antibody effectively distinguishes human tumor cells from host
(mouse) tumor microenvironment. (a) IHC using a pan-human nucleoli antibody on an intramuscular alveolar rhabdomyosarcoma PDX. Note the absence of staining in adjacent host skeletal muscle (top) and intra-tumoral host stromal/endothelial cells (bottom). (b) Co- immunofluorescence of a PDX primary culture shows significant contamination with tumor- associated stromal cells (αSMA). P a g e | 43
FN-RMS FP-RMS
SARC 171030 SARC 180612 SARC 170206 SARC 180409
Patient Tumor Patient
F1 PDX Tumor PDX F1
Figure 4. Preservation of hallmark histological features between paired patient and PDX rhabdomyosarcomas. H&E stain of paired patient and first generation (F1) PDX tumors.
Characterization of PDX models. While the majority of PDX tumor tissue was cryopreserved, tissue samples from each PDX tumor processed for histological and RNA evaluation. Of the 26 established PDX models, 15 (58%) have patient tumor tissue kept for RNA and 11 (42%) for histology. This allows for detailed histological and molecular characterization of our models. H&E were performed on all F0 and F1 tumors and, when applicable, were cross- referenced. We found that tumor histologies were highly correlated with the original sample, even within tumors with the same diagnosis but unique histological features. For example, we profiled four rhabdomyosarcoma patient + PDX tumor pairs (Fig. 3). Both fusion negative RMS retained P a g e | 44
SARC 170206 Pax3 FoxO1
SARC 170503 Pax3 FoxO1
Figure 5. Retention of driver mutation in fusion positive aRMS. Sanger sequencing
results of the PDX models from two known fusion positive alveolar
rhabdomyosarcomas. Both show retention presence of the PAX3/FOXO1 splice
junction produced by the t(2;13)(q35;q14) translocation. their classic embryonal tumor histologies of small cells in a loose myxoid pattern. Moreover, two fusion positive aRMS tumors demonstrated a uniform small round blue cell morphology. One of these patient tumors (SARC 170206) had a more cellular and dense morphology which was reflected in the PDX model. A second patient whose tumor had a significant stromal component in the tumor also showed a similar stromal component in the resulting PDX model. We also validated the presence of the PAX3/FOXO1 fusion transcript in two aRMS PDX models (Fig. 4).
These results demonstrate a retention of the patient tumor histopathological features in the PDX model.
The stem cell marker PAX7 and differentiating myoblast maker MYOG are differentially expressed between aRMS and eRMS. Histological analyses of RMS patient tumors has demonstrated that while all rhabdomyosarcomas contain MYOG+ cells, fusion positive aRMS has significantly higher percentage and intensity of positive cells. Despite being widely used as a surrogate marker for fusion status, no one has interpreted the significance of this high MYOG P a g e | 45 expression in aRMS. Since MYOG is normally expressed as myoblasts exit the cell cycle to differentiate, this would suggest that almost all aRMS cells have accumulated at this transition point but obviously fail to differentiate which leaves them in a stalled, proliferative state.
Conversely, eRMS tumors have a rare to moderate frequency of MYOG expression which suggests that eRMS cells are not stalled at this same myogenic state. We stained our 7 RMS PDX models and 4 human cell line RMS xenografts (n=11 total) for MYOG and the myogenic stem cell marker
PAX7 to profile the spectrum of differentiation between aRMS and eRMS. While all eRMS tumors robustly expressed PAX7, aRMS xenografts had little or no PAX7 expressing cells (Fig. 5 a-c). This demonstrates that while fusion positive tumors are more mature in the spectrum of differentiation, they are clinically more aggressive. It also implies that eRMS and aRMS employ different mechanisms to block myogenic differentiation.
P a g e | 46
P a g e | 47
P a g e | 48
Figure 6. Differential expression of stem cell marker (PAX7) and differentiating myoblast
(Myogenin) in human rhabdomyosarcoma. (a) Schematic diagram of important transcription factors that govern myoblast differentiation status(54). (b) Fusion negative RMS xenograft expression of PAX7, MYOG, and FOXF1. (c) Fusion positive RMS xenograft expression of PAX7,
MYOG, and FOXF1.
PDX profiling for preclinical immunotherapy development. In addition to serving as a representative model for basic biology research, PDX models accurately reflect sensitivities to various therapies. Even though our PDX mouse models lack functional immune systems, they can be “humanized” by the transfer of human immune progenitors, T-cells, or other immune cell populations. We hope to apply our knowledge of immune-tumor targeting to a wide spectrum of pediatric solid tumors. Not only would this allow a new frontier for therapy, but it would avoid many of the long term morbidities associated with chemotherapy and radiotherapy. To identify tumors which may be selectively targeted by the immune system, we characterized expression of several cancer testis antigens (CTAs). This class of tumor antigens are proteins exclusively expressed in the human reproductive tract which are aberrantly expressed in tumor cells due to epigenetic disease mechanisms. Cells expressing these antigens can then be eradicated from the body using CTA-directed immunotherapy, minimizing on-target toxicities to vital organs in the body.
We profiled 15 different PDX tumors for seven different known cancer testis antigens. All 15 of these PDX models have patient matched peripheral blood samples stored and available for research. Since the tumor and immune cells are from the same patient, we hope to identify endogenous T cells which are reactive to CT antigen found on the same patient’s tumor. Following in vitro T cell expansion, we can perform autologous T cell infusions to test the ability of these P a g e | 49
cells to exert anti-tumor affects. This screening approach will identify the CT antigen which will
be used for T cell expansion for each patient. We found a variable expression of CTAs across
PDX tumor types (Fig. 6). PRAME had the broadest level of expression across PDX models with
consistent expression in eRMS (3 of 3) and osteosarcoma (2 of 2). The synovial sarcoma PDX
served as a positive control for many of these markers as this disease exhibits consistent, high
expression of CTAs. One unexpected result was that SARC 160718, a recurrent sclerosing PDX,
showed high expression of nearly all CTAs. This has not been described as a highly immunogenic
tumor so these results identify a possibility for immune targeting of this subtype of tumor.
MRT MPNST RMS EWS OS SS
NY-ESO1
PRAME
SSX2
MAGEA1
MAGEA3
MAGEA4
MAGEA9
GAPDH
P a g e | 50
160718 171030 180612 170206 180409 180613 170410 170808 171003 180404 170127 170911 170320 161031 180801 NY-ESO1 +++ +++ ++ PRAME +++ +++ +++ + + +++ +++ +++ + +++ +++ SSX2 +++ + +++ +++ + MAGEA1 +++ + MAGEA3 +++ +++ +++ +++ +++ + MAGEA4 +++ ++ ++ ++ ++ + +++ MAGEA9 +++ +++
MRT MPNST RMS EWS OS SS
CD80
CD86
CD274/PDL1
CD273/PDL2
GAPDH
Figure 7. Cancer testis antigen (CTA) and immune checkpoint profiling of PDX sarcomas.
RT-PCR analysis of expression of seven different cancer testis antigens and tumor immune
checkpoint inhibitors. Top: representative agarose gel images. Bottom: relative expression
intensity. P a g e | 51
V. Discussion
Sarcomas are poorly understood due to their relatively low incidence and complex genomic and biological characteristics. While progress has been made in treating most sarcoma patients with the use of cytotoxic therapies, little progress has been made in our understanding of the underlying disease mechanisms. This lack of knowledge has impeded the development of representative animal models and the application of targeted therapies. In this work, we have sought to establish PDX models from patients with bone and soft tissue solid tumors. Twenty-six unique PDX models from nine different types of pediatric soft tissue and bone tumors were established and characterized as part of this work. Many of these PDX models also have patient- matched samples for validation and tumor profiling studies. Our histological and molecular characterization of these models revealed a preservation of patient tumor core features.
Three main cohorts of PDX models were established during this study (Ewing sarcoma, osteosarcoma, and rhabdomyosarcoma). Relevant to this dissertation, seven different rhabdomyosarcoma PDX models were established (3 fusion negative + 4 fusion positive). We have used these RMS PDX models in conjunction to the cell line xenograft models to generate a panel of 11 human RMS xenograft models. We profiled them based on Myogenin, PAX7, and
FOXF1. Myogenin is expressed in all RMS tumors, but ranges from rare to moderate positivity in fusion negative RMS to strong diffuse in most cells in fusion positive RMS. Since MYOG is normally expressed during terminal myogenic differentiation, we wondered if this expression is anti-correlated with the classic myogenic stem cell marker PAX7. These transcription factors were mutually exclusive in individual tumor cells (not shown) and broadly anti-correlated. The robust expression of PAX7 and low expression of MYOG in fusion negative tumors resembles a more stem-like state of the tumor cells. Fusion positive aRMS, however, have only rare PAX7+ cells P a g e | 52 with nearly every tumor cell expressing MYOG. This suggests that fusion positive tumor cells are uniformly stalled at the Myogenin transition state. Furthermore, this broadly suggests that
PAX3/FOXO1 positive tumors employ different mechanisms to block myogenic differentiation.
In addition to establishing multiple PDX models for individual tumor types, we also have diverse sarcoma PDX models. Given the genetic and biological diversity of these sarcomas, identifying mutual biomarkers, disease mechanisms, and therapeutic targets would allow for more broad applications. Immunotherapy was one of the focal points for the study and can be applied to tumors independent of their anatomical location, genetic background, or biology. Since these sarcomas can be so genetically and biologically diverse, we looked for common patterns of immunotherapy susceptibility. By screening cancer testis antigen expression in our PDX models we were able to identify tumors with high expression of one or more CTA. The use of autologous
T cell transplant targeting CTAs has been successfully used to selectively kill tumor cells in melanoma and synovial sarcoma with multiple clinical trials underway. Ongoing experiments with these newly established models will hopefully provide proof of concept for screening individual sarcoma patient tumors to identifying CTA expression for immunotherapy. P a g e | 53
Chapter 3: Transcriptional regulation of the FOXF1 gene locus in aRMS.
I. Abstract
This study and others have identified FOXF1 as one of the most uniquely and consistently upregulated mRNAs in patient aRMS tumors. The expression of FOXF1 in this context is particularly interesting. Despite the fact that RMS is considered a myogenic disease, FOXF1 has no known associations with the skeletal muscle lineage. In this chapter, we detail the expression of in RMS. We found that FOXF1 expression correlates not with alveolar histology, but specifically the presence of the PAX3 – FOXO1 rearrangement. The PAX3/FOXO1 fusion gene is both necessary and sufficient for FOXF1 expression. Mechanistically, PAX3/FOXO1 positively regulates the FOXF1 gene locus by activating two distal enhancer elements.
II. Introduction
FOXF1 is a highly conserved Forkhead transcription factor which is expressed in various mesenchymal cell populations in the respiratory and genitourinary tract. This includes a range of mesenchymal cell types including smooth muscle cells, fibroblasts, endothelial cells, adipose, and cartilage (refer to Chapter 1). FOXF1 expression is absent in the skeletal muscle lineage in mouse and human, which highlights the peculiarity of FOXF1 expression in this strongly myogenic tumor.
The aberrant expression of FOXF1 is also different than other mechanisms of cancer
“overexpression” in which tumor cells retain features of their cell of origin, thus, are passively overrepresented in bulk tumor tissue samples. Skeletal muscle satellite cells have been considered P a g e | 54 a presumptive cell of origin for many rhabdomyosarcomas, although there currently is no consensus on the cellular origins of rhabdomyosarcoma. Since FOXF1 is not expressed in satellite cells, the alternative mechanism is that FOXF1 expression is induced at some point during malignant transformation.
If FOXF1 is expressed de novo in aRMS tumors, then it must be regulated by a transcriptional network which is consistently activated in aRMS. Since most (80%) of aRMS tumors contain the
PAX3/FOXO1 or PAX7/FOXO1 fusion gene, it is possible that these transcription factors are directly or indirectly involved in FOXF1 activation. These PAX3/FOXO1 fusion genes functions as potent transcriptional activators due to the addition of the FOXO1 transactivation domain in the fusion protein and could directly activate the FOXF1 locus. If this is true, we hypothesized that the remaining 20% of aRMS tumors lacking a FOXO1 rearrangement would not express FOXF1 despite the presence of an alveolar tumor histology. The uncharacterized relationship between
FOXF1 expression and fusion status will indicate whether FOXF1 is induced specifically downstream of a FOXO1 rearrangement or conversely, is downstream of a pathway which is associated with the general small round blue cell and alveolar histology.
III. Materials and Methods
Cell line and patient-derived RMS xenografts. All conducted mouse experiments were reviewed and approved by the Cincinnati Children’s Hospital Medical Center Institutional Animal
Care and Use Committee. Cell lines RD, Rh18, and Rh30 were a kind gift from Dr. Timothy Cripe
(Nationwide Children’s, OH) and Rh4 was a kind gift from Dr. Javed Khan (NCI, MD). RD and
Rh4 were cultured in DMEM (GIBCO) supplemented with 10% FBS and antibiotics. ARMS cell lines Rh18 and Rh30 were cultured in RPMI-1640 (GIBCO) supplemented with 10% FBS and antibiotics. Tumor xenografts were performed by injecting 500k tumor cells resuspended in 50% P a g e | 55 matrigel (Corning) into the quadriceps of 4-8 week old NODscid IL2Rγnull mice. Mice were sacrificed after 8 weeks. Patient-derived RMS xenografts were established by subcutaneous and/or intramuscular transplantation of freshly biopsied tumor tissue. Mice were sacrificed when tumors reached 15mm in their greatest diameter. shRNA and CRISPR-mediated repression. An shRNA was designed and cloned based off a previously validated siRNA approach. Oligonucleoties were annealed and cloned into the pLKO plasmid (Addgene #8453) in the EcoRI and AgeI sites. For the CRISPR repression studies, guide
RNAs were designed using Guide RNAs were cloned into the pLV hU6-sgRNA hUbC-dCas9-
KRAB-T2a-Puro plasmid in the BsmBI site which will allow for expression of the gRNA, nuclease dead Cas9 fused to the KRAB repression domain, and a puromycin resistance casette (Addgene
#71236). The sequences for the cloning primers are shown in Table 1. Transductions were performed in triplicate in 10cm plates. 24 hours after transduction, cells were selected with
1.5ug/ml puromycin for 48 hours. Cells were given an additional 24 hours without puromycin to avoid any artifacts associated with puromycin treatment. Cells were harvested and protein and
RNA were extracted for western blotting or qRT-PCR, respectively. A list of antibodies is provided in Table 2.
P a g e | 56
Table 1. Oligonucleotide pairs for cloning of shRNA and guide RNAs.
Oligonucleotide Application Oligonucleotide sequence 5’ to 3’ name shRNA Scramble cloning CCGGCCTAAGGTTAAGTCGCCCTCGCTCGAGCGAGGGCGACTTAACCTTAGGTTTTTTG Forward primers Scramble AATTCAAAAACCTAAGGTTAAGTCGCCCTCGCTCGAGCGAGGGCGACTTAACCTTAGG Reverse shPax3/FoxO1 CCGGCCTCTCACCTCAGAATTCAATCTCGAGATTGAATTCTGAGGTGAGAGGTTTTTG Forward shPax3FoxO1 AATTCAAAAACCTCTCACCTCAGAATTCAATCTCGAGATTGAATTCTGAGGTGAGAGG Reverse -315kb gRNA CRISPRi CACCGCGAAAAGAGCTTAGCGGTTC Forward -315kb gRNA AAACGAACCGCTAAGCTCTTTTCGC Reverse +8kb gRNA CACCGTCAGTAACCAGGTGTAGGAA Forward +8kb gRNA AAACTTCCTACACCTGGTTACTGAC Reverse FoxF1 CDS CACCGCCATGGACCCCGCGTCGTC Forward FoxF1 CDS AAACGACGACGCGGGGTCCATGGC Reverse
Table 2. Antibodies used in this study. Antibody Target Manufacturer Catalog Number FoxF1 R&D Systems AF4798 FoxO1 (C-terminus) Cell Signaling Technology 2880 Pax3 (N-terminus) R&D Systems MAB2457 β-Tubulin ThermoFisher Scientific PA5-16863
ChIPseq and RNAseq analyses. Previously published PAX3/FOXO1, P300 and H3K27Ac
ChIPseq datasets were downloaded from GSE19063 and GSE83726 and analyzed using
Biowardrobe. Sequence conservation analysis was done on USCS genome browser. A previously published RNAseq dataset (GSE66533) from 58 RMS patients was analyzed using GEO2R to identify FOXF1 and LINC01082 expression in RMS. A second analyzed, published RNAseq dataset discriminating expression based on both histology and fusion status was used to report relative expression values of FOXF1 (from ArrayExpress accession no. E-MEXP-121). P a g e | 57
IV. Results
FOXF1 lies genetically downstream of the Human RMS microarray 1400 PAX3/FOXO1 fusion gene. FOXF1 mRNA p < 0.001 1200 expression in rhabdomyosarcoma is only observed in 1000 800
tumors with alveolar histology. Gene expression 600 mRNAexpression 400
profiling of RMS patient tumors revealed that 200
(relative (relative signal intensity FOXF1 0 occasional aRMS tumors were transcriptomically eRMS aRMSn aRMSp (n=15) (n=5) (n=10) similar to embryonal rhabdomyosarcomas. Figure 1. FOXF1 is only upregulated Interestingly, these outliers lacked a PAX/FOXO1 in the presence of the fusion gene. rearrangement and had favorable patient outcomes Gene expression microarray of patient similar to eRMS tumors. This suggests that the fusion RMS tumors grouped based on status of RMS, not histology, is correlated with the histology and fusion status (aRMSn = unique tumor biology. To determine if FOXF1 also alveolar histology fusion negative; correlated with fusion status, we analyzed FOXF1 aRMSp = alveolar histology fusion expression in a previously published dataset which positive). Data from ArrayExpress segregated patients on both histology and fusion accession no. E-MEXP-121. status(30). FOXF1 expression was exclusively upregulated in fusion positive aRMS, with little or no expression in fusion negative aRMS or eRMS (Fig. 1). This demonstrates that FOXF1 expression correlates with the presence of the
PAX3/FOXO1 fusion and more aggressive clinical behavior.
We analyzed expression of FOXF1, PAX3 (N-terminus), and FOXO1 (C-terminus) in an aRMS patient tumor. The majority of FOXF1 expressing tumor cells also expressed the peptide corresponding to the PAX3/FOXO1 fusion protein presenting the possibility that PAX3/FOXO1 P a g e | 58
P a g e | 59
Figure 2. FOXF1 requires PAX3/FOXO1 for expression in aRMS. (a) Co-immunofluorescence of Pax3 (N-terminus), FoxO1 (C-terminus), and FoxF1 in a patient aRMS containing the t(2;13)(q35;q14) translocation. (b) Expression of FOXF1 in PAX3/FOXO1 loss- and gain-of- function in human RMS cell lines. (c) (top) shRNA targeting the PAX3 and FOXO1 junction.
Scramble and shPax3/FoxO1 shRNAs were cloned into lentiviral plasmids and transduced on human aRMS cell lines Rh4 and Rh18. qRT-PCR was performed following puromycin selection and values were normalized to β-actin (n=3). Data reported as mean ± SEM, ** p<0.01.
(c)Representative Western blots of cell lysates after Pax3/FoxO1 knockdown.
transcriptionally activates FOXF1 (Fig. 2a). Analysis of PAX3/FOXO1 loss- and gain-of-function
datasets demonstrated that PAX3/FOXO1 is required for FOXF1 expression (Fig. 2b). Using a
shRNA spanning the PAX3/FOXO1 junction, we generated a lentiviral knockdown of
PAX3/FOXO1 in aRMS cell lines Rh4 and Rh18. FOXF1 mRNA and protein significantly
decreased demonstrating that PAX3/FOXO1 directly or indirectly activates FOXF1 expression
(Fig. 2c,d). Of note, the wildtype FOXO1 total protein levels are increased following
PAX3/FOXO1 knockdown, suggesting the existence of a negative regulatory mechanism between
PAX3/FOXO1 and wildtype FOXO1 (Fig. 2d).
PAX3/FOXO1 binds to and activates two distinct FOXF1 enhancers. PAX3/FOXO1 is a
potent transcriptional activator that binds at distal enhancer elements instead of gene promoters(24,
55). To determine if the PAX3/FOXO1 bound to enhancers that regulate the FOXF1 locus, we first
mapped the location of aRMS-specific active enhancers around the FOXF1 locus by ChIPseq for
H3K27Ac in Rh18 and with previously published datasets(55). Of the four protein coding genes at
the FOXF1 locus (FOXF1, MTHFSD, FOXC2, FOXL1), only FOXF1 is transcribed at high levels P a g e | 60
chr16:86,118,400- 86,680,700
10 Pax3/FoxO1 (Rh4) 0 10 P300 (Rh4) 0 10 H3K27Ac (Rh4) 0 10 H3K27Ac (RD) 0
LINC01082 FOXF1
Pax3/FoxO1 p300 H3K27Ac (aRMS)
H3K27Ac (eRMS) Figure 3. PAX3/FOXO1 binds to active enhancers near the FOXF1 gene locus. Browser snapshot
of PAX3/FOXO1 and P300 binding in Rh4 cells are associated with H3K27Ac in Rh4 (aRMS) but not
RD (eRMS) human cell lines. The upstream binding site is 315kb upstream of the FOXF1 TSS and is
associated with the promoter of a lincRNA LINC01082. The second binding site falls 8kb downstream
of the FOXF1 3’UTR. ChIPseq datasets from GSE19063 and GSE83726.
(note: low expression of MTHFSD is observed in aRMS but higher in eRMS). Since no other
genes are located within this 1Mb region, it is likely that any proximal active enhancer elements
correspond to the FOXF1 gene. Across our datasets, we identified three distinct regions with high
enrichment of H3K27Ac in aRMS but little to no H3K27Ac in eRMS (Fig. 3).
Characterization of the -315kb upstream enhancer. The first active enhancer is 315kb
upstream of the FOXF1 gene and is associated with a long noncoding RNA transcript named
LINC01082 (Fig 3). The H3K27Ac marks extend through the gene body, suggestive of
transcription of LINC01082 at this active enhancer. We screened 12 human RMS xenograft
models for expression of LINC01082. We found that while none of the fusion negative RMS P a g e | 61 tumors had expression of LINC01082, all fusion positive RMS tumors showed robust expression
(Fig. 4a). Linear regression analysis showed a correlation (R2=0.898) between LINC01082 and
FOXF1 expression demonstrating a proportional the level of enhancer activation and LIINC01082 expression to FOXF1 expression. Since this enhancer and associated eRNA transcript correlate with FOXF1 expression, we hypothesized that LINC01082 must also be exclusively expressed in fusion positive RMS similar to FOXF1. We analyzed a previously published microarray dataset of 58 RMS patients which classified tumors based on fusion positivity and not histology(56).
LINC01082, like FOXF1, was significantly overexpressed in fusion positive but not fusion negative RMS (Fig. 4b). We also found that LINC01082 mirrored FOXF1 expression across a panel of 53 human tissue types (Fig. 5). Altogether, the upstream 315kb enhancer and associated eRNA transcript LINC01082 are distinctly associated with activation of the FOXF1 gene. P a g e | 62
Figure 4. The enhancer RNA LINC01082 correlates with FOXF1 and PAX3/FOXO1. (a) qRT-
PCR of FOXF1 and LINC01082 mRNA levels in human RMS orthotopic xenograft models. (Top) log scale of fold change relative to a spindle-sclerosing PDX which has detectible (negligible)
FOXF1 mRNA but not protein. Data reported as mean ± STDEV. (b) Microarray data of fusion negative (FN-RMS) and fusion positive (FP-RMS) RMS patient tumors show that both the FoxF1 mRNA and the LINC01082 eRNA are specifically upregulated in FP-RMS tumors. Data from GEO series accession no. GSE66533. (c) qRT-PCR on Rh4 and Rh18 cells with knockdown for
Pax3/FoxO1. Values were normalized to β-actin (n=3). Data reported as mean ± SEM. P a g e | 63
(a) Human Tissues – FOXF1 mRNA
(b) Human Tissues – LINC01082 mRNA
Figure 5. The enhancer RNA LINC01082 correlates with FOXF1 in aRMS and across human tissues. Expression of FOXF1and LINC01082 mRNA across human tissues. Data and graphic from https://gtexportal.org dbGaP Accession phs000424.v7.p2.
P a g e | 64
Characterization of the +8kb downstream enhancer. The remaining two enhancer elements fall within the first intron of a lincRNA FENDRR immediately upstream of the FOXF1 locus and a second enhancer is roughly 8kb downstream of the FOXF1 3’UTR. The latter enhancer is a significantly stronger enhancer as determined by H3K27Ac enrichment. The nucleosome depleted region, and presumptive binding site for transcription factors, is a genomic element acquired recently in evolutionary history (Fig. 6a,b). When analyzing this sequence, we realized that this enhancer element consists of six sequential 43bp repeats with slight variation in the sequence (Fig.
6c). Notably, the fourth repeat of this sequence which is located at the center of the nucleosome depleted region has a single base-pair variant which results in a perfect E-Box motif. This motif is found at PAX3/FOXO1, MYOD1, and MYOG binding sites and presents the possibility that one of these factors activates this enhancer. P a g e | 65
(a)
Rh4
SCMC aRMS H2K27Ac aRMS
Chimp Gorilla Orangutan Rhesus Baboon Marmoset Mouse Dog Elaphant Platypus Lizard Conserv. Zebrafish
3’ site Active enhancer (b)
New enhancer sequence (c) 3’ site enhancer sequence: GTGGTGGTCTCTGCACTGGCCCTTTCTTCGTAATTCTGTTCCACCGCTGTTCTCGGCCTTTCTGATTTCTCGGTGATTCTGTTCCGCCGC TGTTCTCGGCCTTTCTGATTTCTCGGTGATTCTGTTCCACCGCTGTTCTCGGCCTTTCTGATTTCTCGGTGATTCTGTTCCACAGCTGTT CTCGGCCTTTCTGATTTCTCAGTGATTCTGTTCCGCTGCTGTTCTCAGCCTTTCTGATTTCTCGGTGATGCTGTTCCACCGCTGTTCTCG GCCTTTCTGATTTGCTGTCTTGTGTCTC
3’ site enhancer sequence are repetitive fragments:: 5’- CTTCGTAATTCTGTTCCACCGCTGTTCTCGGCCTTTCTGATTT CTCGGTGATTCTGTTCCGCCGCTGTTCTCGGCCTTTCTGATTT CTCGGTGATTCTGTTCCACCGCTGTTCTCGGCCTTTCTGATTT CTCGGTGATTCTGTTCCACAGCTGTTCTCGGCCTTTCTGATTT CTCAGTGATTCTGTTCCGCTGCTGTTCTCAGCCTTTCTGATTT CTCGGTGATGCTGTTCCACCGCTGTTCTCGGCCTTTCTGATTT -3’ E-box
P a g e | 66
Figure 6. Strong enhancer element downstream of FOXF1 is newly acquired in evolutionary history of vertebrates. (a) Top: H3K27Ac location around the FOXF1 locus in two human fusion positive aRMS cell lines. Bottom: Sequence conservation across species. Conserved sequence shown in green, non-conserved is blank. (b) Evolutionary tree with boxed regions highlighting the phylogenetic location of vertebrates in panel a. Shaded blue box includes vertebrates containing new enhancer sequence. (c) Sequence of the 3’ enhancer region. Bottom: The same enhancer sequence but presented as an alignment to showing that the whole enhancer is composed of short
43bp repeats.
PAX3/FOXO1 binds to and activates these enhancers to induce FOXF1 expression. To determine if any of these binding sites were bound to PAX3/FOXO1, we analyzed existing
PAX3/FOXO1 ChIPseq datasets performed in the aRMS Rh4 cell line(24). We found that all three enhancers were bound by PAX3/FOXO1 and that these binding events were reproduced in two biological replicates. These exact binding sites were associated with P300 binding and strong
H3K27Ac demonstrating that they are functioning as active enhancers. Notably, the
PAX3/FOXO1 binding at the 3’ site is among the most enriched binding sites of PAX3/FOXO1 genome-wide (replicate 1 ranked 34th, replicate 2 ranked 17th).
As previously shown, knockdown of PAX3/FOXO1 results in decreased FOXF1 mRNA (Fig.
2c). qRT-PCR analysis showed that PAX3/FOXO1 is also required for inducing expression of the
LINC01082 eRNA and presumably activation of this -351kb enhancer element (Fig. 7b). Lastly, to test if these active enhancers regulate FOXF1 expression, we used a CRISPRi system to locally repress each enhancer element. This system uses a nuclease dead Cas9 protein fused to the
Krüppel associated box (KRAB) domain which induces local repression through histone P a g e | 67 methylation(57, 58). We designed gRNA targeting either the -315kb upstream or 3’ downstream enhancer elements and exon 1 of FOXF1 as a positive control (Fig. 7c). Targeted repression of either enhancer element resulted in a significant decrease in FOXF1 expression. Altogether, we have identified that FOXF1 is only expressed in fusion-positive RMS due to the direct transcriptional regulation by the PAX3/FOXO1 protein at FOXF1 enhancer elements.
P a g e | 68
P a g e | 69
Figure 7. FOXF1 is a direct transcriptional target of the PAX3/FOXO1 oncogene by
activation of FOXF1 enhancer elements. (a) ChIPseq of PAX3/FOXO1 binding (two
technical replicates) and active enhancer makers in the aRMS cell line Rh4. (b) qRT-PCR
expression of LINC01082 after shRNA knockdown of PAX3/FOXO1. (c) qRT-PCR of FoxF1
mRNA levels after repression of enhancers using CRISPRi. Rh4 and Rh18 cells were
transduced with a lentivirus containing a dCas9:KRAB and a gRNA targeting the 315kb
upstream binding site, the 8kb downstream binding site, or the FoxF1 coding sequence (CDS)
as a positive control. Values were normalized to β-actin (n=3). Data reported as mean ± SEM.
* p<0.05; ** p<0.01.
V. Discussion
FOXF1 is one of the most uniquely and consistently upregulated genes in aRMS. We have identified that FOXF1 expression does not necessarily correlate with alveolar histology, but rather the presence of the PAX3/FOXO1 fusion protein. Data gathered from gain and loss of function experiments in multiple tumor models demonstrated that PAX3/FOXO1 is both necessary and sufficient for FOXF1 expression. Using PAX3/FOXO1 ChIPseq data, we identified several
FOXF1 enhancer elements which are bound and activated by PAX3/FOXO1. One of these binding sites activates expression of an enhancer RNA LINC01082, which mirrors FOXF1 expression in fusion positive aRMS and in normal human tissues. A second site is an evolutionarily new sequence downstream of the FOXF1 gene locus and represents one of the most enriched
PAX3/FOXO1 binding sites in the Rh4 cell line. Together, these two enhancer elements function additively to promote FOXF1 gene expression. P a g e | 70
We have demonstrated on a mechanistic level that FOXF1 is directly regulated by the
PAX3/FOXO1 protein. While this is not too surprising given the strong correlation of fusion status and FOXF1 expression, it does provoke some interesting questions. PAX3/FOXO1 is thought to retain the complete function of the paired and homeobox domains of the normal PAX3 protein.
EMSA and ChIPseq results have demonstrated that the PAX3/FOXO1 binding specificities are similar to the wildtype PAX3 protein, although the PAX3/FOXO1 protein has relatively lower affinity (21, 59). If motif preference is equivalent, than it is possible that wild-type PAX3 may regulate the FOXF1 locus in some context (discussed in Chapter 6). There is no evidence in the literature that PAX3 regulates the FOXF1 locus and there also are no reports that PAX3 mutant mice have altered FOXF1 expression. Despite these speculations, it is clear that PAX3/FOXO1 directly binds to and activates FOXF1 enhancers in aRMS. Since all fusion positive tumors studied to date also express FOXF1, it is likely that this transcriptional regulation is strongly selected for during the course of malignant transformation by PAX3/FOXO1.
P a g e | 71
Chapter 4: Identification of FOXF1 as positive regulator of cell proliferation and a potent repressor of terminal myogenic differentiation
I. Abstract
FOXF1 expression is completely dependent on PAX3/FOXO1. Such high PAX3/FOXO1 enrichment at the FOXF1 3’ enhancer locus, conversely, suggests that PAX3/FOXO1 function may be dependent on FOXF1 function. In this chapter, we detail the phenotype of aRMS tumor cells in the absence of FOXF1. We identify that loss of FOXF1 is associated with reduced proliferation and increased apoptosis. More strikingly, we find that aRMS cells lacking FOXF1 spontaneously differentiate in vitro and in vivo. In a primary human myoblast model, we find that expression of FOXF1 is sufficient to prevent terminal differentiation and senescence of myoblasts.
Overall, FOXF1 activation by the PAX3/FOXO1 protein is essential for promoting aRMS proliferation and blocking terminal myogenic differentiation.
II. Introduction
Retrospective analysis of rhabdomyosarcoma patients has led to the important realization that patients containing the FOXO1 rearrangement have significantly lower survival than those without a FOXO1 rearrangement. The fusion status of RMS patients now is one of the single best predictors of patient outcome. Despite both fusion negative and fusion positive tumors displaying similar markers of skeletal muscle differentiation, it is presumed that unique mechanisms of tumorigenesis must be present in fusion positive tumors. P a g e | 72
Gene expression profiling rhabdomyosarcoma patients led to the discovery of a gene signature which correlates with fusion status. While there is some variation between studies in the composition of the signature, a consistent core set of genes is reproducibly upregulated in fusion positive tumors. These genes are associated with a wide range of cellular processes from signal transduction to transcriptional regulation. Most of these genes, oddly, do not have clearly defined roles in cancer and even less is known about their functional importance in aRMS tumor cells.
Since this group of genes is so consistently expressed in aRMS, it is unlikely due to chance.
The expression of these genes may represent a certain cell “state” which is characteristic of aRMS cells in the presence of the fusion. This possibility is supported by high enrichment of neural genes such as OLIG2, ABAT, NRCAM, CNR1, ASS1, and TFAP2B as part of the gene signature.
Alternatively, these genes may be induced downstream of the PAX3/FOXO1 fusion to provide functions essential for tumorigenesis. This is an interesting possibility as the fusion positive RMS tumors have characteristically low number of nonsynonymous somatic mutations which suggests
PAX3/FOXO1 relies on “normal” mechanisms of proliferation and survival instead of a genetic basis.
FOXF1 is one of the most significantly upregulated gene in fusion positive aRMS. We have shown that FOXF1 is a direct transcriptional target of PAX3/FOXO1 (see Chapter 3) which explains how FOXF1 expression is only observed in presence of the fusion protein. We tested if the activation of FOXF1 in aRMS is important for any aspect of aRMS tumor cell biology. We used a combination of gain-and-loss of function approaches to define the biological contribution of FOXF1. FOXF1 loss resulted in significantly reduced proliferation of aRMS tumor cells in vitro and in vivo. Gene expression profiling of cells immediately after FOXF1 loss showed activation of the skeletal muscle differentiation program. In vivo, FOXF1 deletion results in P a g e | 73 widespread tumor cell differentiation. To test if FOXF1 could block differentiation in the absence of the PAX3/FOXO1 oncogene, we used a gain-of-function approach in a primary human skeletal muscle myoblast model. We found FOXF1 expression alone was sufficient to block myoblast differentiation. Furthermore, FOXF1-expressing myoblasts did not senesce after extended culture or after growth factor depletion. These data demonstrate that FOXF1 significantly contributes to the biology of aRMS tumors by promoting cell proliferation and blocking the myogenic differentiation process.
III. Materials and Methods
CRISPR deletion of FOXF1 in Rh18. For CRISPR/Cas9-mediated knockout studies, gRNA were cloned into the AgeI and EcoRI sites of the pSpCas9(BB)-2A-GFP (PX458) plasmid
(Addgene #48138). Rh18 WT cells were transiently transfected using Viafect Transfection
Reagent (Promega). One day post-transfection, GFP+ transfected cells were isolated by Flow
Assisted Cell Sorting (FACS) and plated as single-cells into 96 well plates. Single cell clones were expanded and screened for the presence of loss-of-function mutations. After clonal expansion, genomic DNA was isolated from each clone and a 339bp fragment encompassing the gRNA target site was PCR amplified and cloned into the pMINIT plasmid (NEB). Since
CRISPR/Cas9-mediated mutagenesis introduces random mutations at each allele, a minimum of six colonies were picked and Sanger sequenced to identify and verify the mutation at each allele.
Tumor Xenograft studies. Human tumor cell lines (Rh30, Rh18) were injected into immunocompromisd mice to generate human tumor xenografts. 500k cells were resuspended in
50% Matrigel (50ul final) and injected into the quadriceps of the highly immunodeficient mouse strain NOD.Cg-PrkdcscidIl2rgtm1Wjl/SzJ (NSG). After 8 weeks, mice were sacrificed and tumor P a g e | 74 volume was calculated using (lenth x width2)/2 formula. For BrdU studies, mice were injected with 1mg BrdU four hours before sacrificing.
RNAseq. Rh18 cells were transduced with control or FOXF1 exon1 targeted dCAS9-KRAB
(Chapter 3). Cells were briefly puromycin selected, allowed to recover for 24 hours, and send for sequencing. EBseq analysis was used to determine differentially expressed genes. Gene ontology was performed using ToppFun.
Human skeletal muscle myoblast growth and differentiation assays. Human Skeletal Muscle
Myoblasts (HSMM) were purchased from LONZA (Cat. #CC-2580) and cultured according to the manufacturer’s instructions. To establish stable FOXF1 overexpressing and control lines, HSMMs were transduced with pSD44-tagRFP or pSD44-FOXF1HA:tagRFP lentiviral supernatants.
Transduced cells were selected with puromycin for 48 hours and used for studies. All experiments were conducted on cells cultured for under five passages. For longitudinal cell outgrowth assays,
4x104 myoblasts were plated into 35mm dishes. Every 2-4 days the cells were trypsinized, counted and replated to maintain an optimal cell density. Myoblast differentiation was performed by culturing myoblasts in DMEM (GIBCO) supplemented with 2% horse serum for 5 days.
Movies of myoblast differentiation and proliferation were visualized using Incucyte Zoom (Essen
BioScience).
IV. Results
Generation of a FOXF1 knockout in Rh18 cells using CRISPR/Cas9. Our initial characterization of the role of FOXF1 in aRMS demonstrated that FOXF1 depletion by shRNA reduced tumor cell proliferation by 30%-50% (Fig. 1). Despite this effect on the tumor cells, we suspected that the Rh30 cell line may not be the best representative model. Known regulators of P a g e | 75
(a) (b) (c)
Rh30 2.5 Scramble ) 1.2 3 Xenograft 2.25 shFoxF1 6000 1 cells) 2 6 5000
mRNA mRNA 1.75 0.8 1.5 4000 * 0.6 1.25 ** 3000 1 FOXF1 0.4 2000
Expression 0.75 0.2 0.5 1000
0 (10 Number Cell 0.25 Tumor Volume (mm Volume Tumor Relative Relative 0 0 0 1 2 3 4 Day Figure 1. FOXF1 knockdown in Rh30 aRMS cells reduces proliferation. Stable lentiviral knockdown in Rh30 cells. (a) qRT-PCR knockdown efficiency of FOXF1 mRNA normalized to beta- actin. (b) In vitro cell proliferation assay of Rh30 cells after FOXF1 knockdown with three biological replicates per time point. (c) Tumor volumes 60 days after intramuscular injection of Rh30 cells (n=8 per group).
aRMS biology such as CXCR4 are not expressed in Rh30. Expression of myogenic markers myogenin and desmin are also not highly expressed, in stark contrast to aRMS patient samples.
Genomic sequencing of this cell line also showed an unusually high mutation rate with 269 nonsynonymous somatic mutations(60), in stark contrast to what is typically observed in aRMS with an average nonsynonymous somatic mutation rate of 6.4 per tumor (9). In these studies, we also did not see significant downregulation of FOXF1 in tumor lysates, indicating that over time our shRNA knockdown efficiency had decreased.
To circumvent these concerns, we adapted our model system. The Rh18 cell line is a fusion- positive cell line which highly expresses FOXF1 (nearly three-fold higher than in Rh30). To test P a g e | 76
FOXF1 Gene Locus
Figure 2. Design and validation of guide RNAs targeting the human FOXF1 gene locus. Four guide RNAs were clone: two targeting the FOXF1 promoter and two targeting the coding region of exon 1. gRNA pairs were transfected into HEK293T cells. Genomic DNA was extracted 48 hours later and PCR screened for the relative frequency of deletion between the gRNA cut sites
(~180bp).
the importance of FOXF1 in these cells, we used a CRISPR/Cas9 gene editing approach to
knockout the FOXF1 gene in Rh18. Guide RNAs targeting the promoter and first exon of the
FOXF1 gene were designed, cloned, and tested in HEK293T (Fig. 2). Guide RNA #4, which will
introduce a double-strand break at position c.89 of the FOXF1 coding region, was cloned into a
GFP containing plasmid with CAS9 and transiently transfected into Rh18 cells (Fig. 3). After 24
hrs, GFP+ cells were isolated by FACS and plated as single cells into 96 well plates. These cells
were clonally expanded. This process took 110 days until the last clone (clone 10) reached 5x106
cells. The quickest growing clone (clone 1) took 51 days to reach this point. In stark contrast, this P a g e | 77
Figure 3. Schematic diagram of FOXF1 deletion strategy in human aRMS cell line Rh18. Guide
RNA #4 was cloned into pSpCas9(BB)-2A-GFP and transiently transfected into Rh18 cells. Twenty- four hours post-transfection, transfected cells were purified using FACS isolation of GFP high cells.
These cells were then plated at a single-cell dilution and clonally expanded.
experiment was repeated with Rh18 parental cells in which 28 clones reached 5x106 cells within
30 days.
Western blot lysates were prepared from 10 isolated FOXF1-CRISPR clones. Of the 10
clones, 6 had noticeable reduced FOXF1 protein levels and 4 of these had undetectable FOXF1
protein suggesting the presence of frameshift or significant truncation of the protein (Fig. 4a). To
verify that these clones had loss of function mutations, we used Sanger sequencing. Since
CRISPR/Cas9 mutagenesis introduces random mutations at each allele, we needed to analyze each
allele separately. Since Sanger sequencing does not produce full length sequence reads, we PCR
amplified and cloned the FOXF1 alleles into plasmids. This allows for a single FOXF1 allele to
be analyzed separately. Since the cloning of the alleles is random, we anticipate that the same P a g e | 78
(a) * * * * Clone: WT 1 2 3 4 5 6 7 8 9 10
FoxF1
β-tubulin
(b)
Figure 4. Identification of four FOXF1 null clones. (a) Protein lysates from ten Rh18 FOXF1-
CRISPR clones were screened by Western blot for evaluating FOXF1 protein levels. Clones 2,
5, 9 and 10 had undetectable FOXF1 protein. (b) Mutation analysis of FOXF1 alleles in clones
2, 5, 9, and 10 by Sanger sequencing.
P a g e | 79 allele will be cloned multiple times and repeated mutations should be discarded. Using this approach, we found three unique mutations in each of the four FOXF1-CRISPR clones. We also excluded the likelihood of a PCR artifact by using a high fidelity polymerase and by validating the mutation across multiple plasmids. This indicates the presence of three FOXF1 alleles in the Rh18 cell line. Of all of these alleles, 11 of the 12 were confirmed frameshift mutations (Fig. 4b). A third unique mutation was not identified in FOXF1-CRISPR clone 10. This may be due to either a large deletion that exceeds the PCR primer binding site. It is also possible that an identical mutation was created through homologous recombination using the mutant allele as a repair template. In support of this, the c.33-94 deletion was observed in most of the sequenced PCR products.
Reduced growth of Rh18 aRMS tumors in the absence of FOXF1. All four FOXF1-KO
Rh18 clones had a reduction in cell proliferation in vitro. To test the in vivo effect of complete
FOXF1 loss, we injected parental and FOXF1-KO tumor cells into the quadriceps of immunocompromised mice. After eight weeks, FOXF1-KO tumors were significantly smaller than the Rh18 WT tumor controls (Fig. 5a,b). FOXF1 protein was also undetectable by Western blot or immunohistochemistry, validating the absence of FOXF1 in our cells (Fig. 5c,d). qRT-
PCR profiling of tumor cells for proliferation markers showed a significant reduction in proliferation markers such as FOXM1 and cyclins (Fig. 6a). Tumor suppressors CDKN1A and
CDKN1B were significantly upregulated in all FOXF1 KO clones, consistent with FOXF1’s role in repressing their expression (48). BrdU incorporation was also reduced in FOXF1 KO tumors
(Fig. 6b). These data demonstrate that fusion positive aRMS requires FOXF1 for tumor growth in vivo.
P a g e | 80
(a) (b) Rh18 Tumor Volume
Rh18 intramuscular xenograft 4000 )
3 3000
2000 * * 1000
(mm volume Tumor * Rh18 WT FOXF1 KO 2 0 WT KO 2 KO 5 KO 9 FOXF1 IHC (c) WT FOXF1 KO 2 (d)
Mouse #: 1 2 3 4 1 2 3
FoxF1
WT Rh18 Lamin A/C
FOXF1 2 KO FOXF1
Figure 5. Significant reduction in Rh18 tumor size after FOXF1 deletion. (a) Representative image of intramuscular tumor size in a competitive transplant model. (b) Tumor volume measurements of
Rh18 tumors. Rh18 WT n=8; KO 2, 5, 9 per group n=4. Data reported as mean ± SEM. (c) Western blot analysis of protein lysates from Rh18 WT or FOXF1 KO clone 2 tumors. (d)
Immunohistochemistry of FOXF1 in Rh18 WT or FOXF1 KO clone 2 tumors. Magnification 200x.
P a g e | 81
(a) 6 WT Rh18 Tumor Lysates KO 2 5 KO 5 KO 9 4
3
2 Relative mRNA expression expression mRNA Relative 1
0 FOXM1 CCNE1 CCNE2 CCND1 CDKN1A CDKN1B (b)
140 … WT FoxF1 KO 120
+ cells + 100
BrdU 80 **
60 BrdU 40
20 Mean number Mean 0 WT KO
Figure 6. Reduced tumor cell proliferation in FOXF1-KO tumors. (a) qRT-PCR analysis of proliferation markers in FOXF1-KO tumor lysates. N=3 tumors per group. Values are normalized to actin. Data reported as mean ± SEM. (b) Representative immunohistochemistry of BrdU incorporation in Rh18 xenografts. A minimum of 5 random 20x fields were quantified. WT: n=2;
FOXF1-KO: n=2 KO2 + n=2 KO9. Data reported as mean ± SEM.
P a g e | 82
Acute FOXF1 loss in vitro is associated with upregulation of mature skeletal muscle genes.
The loss of FOXF1 has a dramatic effect on aRMS tumor growth. To understand the underlying changes which occur when FOXF1 is absent from aRMS cells, we performed RNAseq. We used the CRISPR-KRAB repression approach to provide an acute and potent knockdown FOXF1 (Fig.
7a,b). A total of 150 and 177 genes were significantly downregulated or upregulated, respectively
(Fig 7c). Analysis of genes downregulated after FOXF1 knockdown included genes previously implicated in tumor growth, invasion and metastasis such as matrix metalloproteinase 3 (MMP3), fibroblast growth factor 7 (FGF7), transcription factor AP-2β (TFAP2β) and the RMS oncogene vestigial like family member 2 (VGLL2) (Fig. 7e) (3, 61-63). While these downregulated genes were not distinctly associated with a biological pathway, the genes upregulated after FOXF1 knockdown were highly enriched in mature skeletal muscle fiber genes (Fig. 7d,f). Multiple troponin and myosin genes were among the top upregulated genes after FOXF1 loss, indicating the maturation and, potentially, terminal differentiation of aRMS tumor cells.
P a g e | 83
(a) FOXF1 mRNA 1.2 (b) 1 0.8 0.6 FoxF1
mRNA 0.4
Fold change in change Fold 0.2 β-Tubulin 0 ControlFOXF1 KD Rh18 FOXF1 KD RNAseq (c) (d) GO - Cellular Component FOXF1 MMP3 MYL1 myofilament 3.82 17 sarcomere 3.82 MYLPF 13.4 TNNI1 contractile fiber 3.96
) striated muscle thin filament 4.09 FGF7 TNNC1 9.8 actin cytoskeleton 4.13 pvalue AGRN VGLL2 MYBPH contractile fiber part 4.27 6.2 troponin complex
Log10( 4.30 - TNNC2 2.6 TFAP2B sarcolemma 4.71 extracellular matrix 5.72 1 -10 -7.5 -5 -2.5 0 2.5 5 7.5 10 0 1 2 3 4 5 6 Upregulated (-log10pvalue) (e) Log2(FC) Downregulated genes Control FoxF1 KD EBseq Results Gene ID Gene Description (TPM) (TPM) Fold change P-Value ART3 ADP-ribosyltransferase 3 26.26 3.47 -6.46 P<1.10E-17 FOXF1 Forkhead box F1 76.94 12.92 -5.47 P<1.10E-17 MMP3 Matrix metalloproteinase-3 258.75 50.2 -4.74 P<1.10E-17 MAGED4B MAGE family member D4B 78.14 15.39 -4.69 P<1.10E-17 RMST Rhabdomyosarcoma 2 associated transcript 22.98 5.36 -4.07 P=8.26E-13 FGF7 Fibroblast growth factor 7 25.06 8.11 -2.84 P=6.50E-11 RXRG Retinoid X receptor gamma 65.25 20.99 -2.79 P=5.51E-11 DOCK10 Dedicator of cytokenesis 10 22.42 7.36 -2.54 P=1.84E-08 VGLL2 Vestigial like family member 2 51.09 20.14 -2.38 P=4.63E-07 TFAP2B Transcription factor AP-2beta 18.64 8.92 --1.92 P=1.9E-03 (f) Upregulated genes Control FoxF1 KD EBseq Results Gene ID Gene Description (TPM) (TPM) Fold change P-Value MYL1 Myosin light chain 1 35.15 192.48 +5.87 P<1.10E-17 DMBT1 Deleted in malignant brain tumors 1 4.11 19.16 +4.97 P<1.10E-17 TNNC1 Troponin C1, slow skeletal and cardiac type 9.1 33.77 +4.01 P=2.17E-10 MYBPH Myosin binding protein H 10.1 22.53 +4.01 P=6.96E-06 MYLPF Myosin light chain, phosphorylatable, fast 30.54 104.62 +3.73 P=4.88E-15 skeletal muscle TNNI1 Troponin I1, slow skeletal type 82.24 253.53 +3.01 P=1.37E-13 TNNC2 Troponin C2, fast skeletal type 18.91 41.04 +2.36 P=3.25E-04 AGRN Agrin 11 23.54 +2.32 P=3.46E-07
P a g e | 84
Figure 7. RNAseq analysis identifies induction of mature myogenic genes following FOXF1 loss.
(a) qRT-PCR quantification of FoxF1 mRNA knockdown efficiency using CRISPRi against the FoxF1 coding sequence. Normalized to β-actin. Data reported as mean ± STDEV. (b) Western blot of knockdown efficiency of FoxF1 CRISPRi cells. β-tubulin was used as a loading control. (c) Volcano plot of differentially expressed genes 72 hr. after FOXF1 knockdown in Rh18 cells. (d) Gene ontology analysis of genes significantly upregulated after FOXF1 knockdown. (FC>1.5; p<0.05). (e,f) Select disease-relevant genes downregulated (e and upregulated (f) genes after FoxF1 knockdown.
P a g e | 85
(a) (b) MyH IHC Rh18 Tumor Lysates WT FOXF1 KO 2 FOXF1 KO 5 FOXF1 KO 9 Mouse #: 1 2 3 1 2 3 1 2 3 1 2 3
MyH Rh18 WT Rh18 FoxF1
PAX3/ FOXO1
β-tubulin FOXF1 KO 9 KO FOXF1 (c) (d)
Human Nuclei MyH 6 Rh18 Tumor RNA WT * KO 2 * 5 KO 5 KO 9 4 * * 3 * Human MyH Merged 2 * * ** *** * * * 1 ***
0 Fold change in mRNA expression mRNAin change Fold
Figure 8. Deletion of FOXF1 promotes spontaneous myogenic differentiation in vivo. (a)
Western blot analysis of Rh18 tumor lysates (3 mice per group). β-tubulin is used as a loading control.
(b) Representative staining of myosin heavy chain in Rh18 tumors. (c) Representative images of co- immunofluorescent staining of intramuscular Rh18 xenografts with the mature skeletal muscle marker MyH+ and a pan-human nucleoli antibody. (d) qRT-PCR on RNA isolated from Rh18 tumors.
(n=4 mice per group). Data reported as mean ± SEM. * p<0.05.
P a g e | 86
Alveolar rhabdomyosarcoma tumors lacking FOXF1 spontaneously differentiate in vivo.
Transcriptomic profiling of FOXF1 depletion in aRMS showed a widespread activation of mature skeletal muscle genes. We investigated if this upregulation signifies a terminal myogenic differentiation process of aRMS cells in the absence of FOXF1. We assessed the level of myogenic differentiation taking place in our FOXF1-KO tumor xenografts. Western blot and immunohistochemistry showed widespread induction of skeletal muscle myosin heavy chain, a known marker of terminal myogenic differentiation (Fig. 8a,b). To rule out that these MyH+ cells were host derived, we performed co-immunofluorescence using a human specific nuclear antibody
(Fig. 8c). Gene expression profiling with human specific primers showed a broad upregulation of mature skeletal muscle genes (Fig. 8d). Thus, loss of FOXF1 induces myogenic differentiation of aRMS tumor cells
To remove the contribution of oncogenic mutations to the function of FOXF1 in myogenic differentiation, we used an established human skeletal muscle myoblast (HSMM) primary culture system. Since FOXF1 is absent in human myoblasts, we cloned the human FOXF1 cDNA into a lentiviral vector with a weak pGK promoter to obtain physiological levels of expression. Early passage primary HSMM were transduced and puromycin selected for one passage to obtain >90% pure culture. We verified FOXF1 expression by western blot and qRT-PCR (Fig. 9a,b).
After generation of the stable cell cultures, no significant phenotypic differences were observed. Upon switching to differentiation media, however, FOXF1-OE myoblasts migrated into small clusters of cells with reduced cytoplasm. This was in contrast with control myoblasts which elongate and fuse to form myotubes (Fig. 9c). Immunostaining for myosin heavy chain and desmin showed a significant reduction in the number of differentiated cells in FOXF1-OE HSMM (Fig.
9d). qRT-PCR profiling of control HSMM showed robust expression of skeletal muscle P a g e | 87 differentiation markers Troponin C1 (TNNC1), Myosin heavy chain 14 (MYH14), Troponin T1
(TNNT1), and Muscle Creatine Kinase (CKM) after in vitro differentiation (Fig. 9e). The induction of these genes was significant reduced in FOXF1 expressing cells (Fig. 9e). These results demonstrate that in the absence of any oncogenic mutations, the presence of FOXF1 is sufficient to block human myoblast differentiation. P a g e | 88
(a) (b) Growth Diff. Empty - Growth E OE E OE 10000 FOXF1 FoxF1 OE - Growth Empty - Differentiation FoxF1 OE - Differentiation FoxF1 5000
change in Fold Lamin A/C expression mRNA 0 4 Days Diff. 9 Days Diff.
(c) 0 Days Diff
Empty
FoxF1 OE FoxF1
Differentiated human skeletal muscle myoblasts (d)
MyH Desmin FoxF1 Merged
Control
MyH Desmin FoxF1 Merged
FoxF1 OE FoxF1
(e) 1500 MYOG 7500 CKM 2000 TNNC1 Empty - Growth FoxF1 OE - Growth 1500 Empty - Differentiation 1000 5000 FoxF1 OE - Differentiation 1000 500 2500
500
Fold change in Fold mRNA expression mRNA 0 0 0
15 MYH14 1500 TNNT1
10 1000
5 500
0 0 P a g e | 89
Figure 9. FoxF1 is a potent repressor of skeletal muscle differentiation in primary human skeletal muscle myoblasts. (a,b) Lentiviral overexpression of human FoxF1 in human skeletal muscle myoblasts (HSMM) in growth media or after 5 days in differentiation media. (a) Western blot on HSMM transduced with empty (“E”) pSD44 lentiviral plasmid or pSD44 containing the human FoxF1 cDNA “OE”. Lamin A/C was used as a loading control. (b) qRT-PCR for FoxF1 overexpression in HSMM. Values were normalized to β-actin. (c) Atypical morphological changes of FoxF1 expressing primary human skeletal muscle myoblasts during in vitro differentiation. (d)
Co-immunofluorescent staining of intermediate (desmin) and late (MyH) skeletal muscle differentiation markers with FoxF1 on HSMM after 9 days in differentiation media. (e) qRT-PCR of skeletal muscle differentiation markers on HSMM after 5 days in differentiation media. Values were normalized to β-actin (n=3). Data reported as mean ± SEM.
FOXF1 maintains the proliferative capacity of primary human myoblasts in vitro. The proliferative capacity of FOXF1-OE HSMM was also assessed using a long-term outgrowth assay in growth media. While control HSMM began to senesce after several passages, FOXF1-OE
HSMM maintained exponential growth for >4 weeks at a rate equivalent to first-passage HSMM
(Fig. 10a). We considered this positive influence on the cell cycle as a potential explanation for
FOXF1’s anti-differentiation effects, since cell-cycle withdrawal is required for myogenic differentiation(28). Immunofluorescent staining with Ki67 after 5 days in differentiation media showed that both control and FOXF1-OE HSMM had fully withdrawn from the cell cycle, demonstrating a proliferation-independent mechanism (Fig. 10b). Interestingly, FOXF1-OE
HSMM can rapidly return to the cell cycle after re-introduction of growth media while control
HSMM have differentiated or senesced (Fig. 10b). Altogether, these data demonstrate that FOXF1 P a g e | 90 expression is sufficient to block myogenic differentiation and that the mechanism is independent of FOXF1’s role in promoting cell proliferation.
P a g e | 91
(a) Primary myoblast outgrowth assay Empty 100000000 FoxF1 OE
1000000
10000 (thousdands)
100 Total Cell NumberCell Total
1 0 5 10 15 20 25 30 Time (days) (b) Human Skeletal Muscle Myoblasts MyH Ki-67 DAPI 100% *
80% Control
nuclei(%) 60% -
Growth media 40%
MyH Ki-67 DAPI MyH
+ 20%
OE - Ki67 0%
Empty FoxF1 OE FoxF1
MyH Ki-67 DAPI
After 6 days in Control differentiation Not Detected
MyH Ki-67 DAPI
media
OE
- FoxF1
MyH Ki-67 DAPI 100% ***
80% Control
nuclei(%) 60% After 2 days -
40% returned to growth MyH Ki-67 DAPI MyH + 20%
media OE - Ki67 0%
EMPTY FOXF1 OE FoxF1
P a g e | 92
Figure 10. FoxF1 expression is sufficient to sustain long-term proliferative capacity of primary human skeletal muscle myoblasts. Analysis of proliferative capacities of HSMM expressing FoxF1. (a) Cell transformation assay of HSMM after FoxF1 expression. Cells were maintained in growth media at optimal cell densities by re-plating every 48-72 hours. (b) Ki67 proliferation indices of HSMM in growth or differentiation media. Only non-differentiated (MyH-) cells were included for quantification using a minimum of 10 random 20x fields (n=3). * p<0.05; *** p<0.001.
V. Discussion
FOXF1 is an essential downstream effector of the PAX3/FOXO1 protein in aRMS.
Knockdown or knockout of FOXF1 in human aRMS cell lines severely impairs their growth in
vitro and in vivo. This is effect is independent of a larger phenotype in which loss of FOXF1
sensitizes aRMS cells to spontaneous, terminal myogenic differentiation. While not all tumor cells
differentiation upon FOXF1 loss, it appears that the threshold is significantly reduced in the
absence of FOXF1. This effect may could be amplified even further by inclusion of cytotoxic
therapies. We also found that in unmanipulated aRMS PDX models, FOXF1 expression is absent
in the occasional differentiated tumor cells. This suggests that the absence of FOXF1 is required
for aRMS myogenic differentiation.
Using primary human skeletal muscle myoblasts, we found that simply expressing FOXF1 was
sufficient to completely block myogenic differentiation and bypass senescence. This primary
culture system demonstrates that even in the absence of any contributing oncogenic alterations,
FOXF1 expression alone can alter the phenotype of human myoblasts. This model supports the
role for FOXF1 in providing a potent block on terminal myogenic differentiation. Since FOXF1 P a g e | 93 is absent from the skeletal muscle lineage in mouse and human, this result is somewhat surprising.
FOXF1 transcriptional activities are limited by its ability to bind to DNA and regulate the local epigenetic landscape. If FOXF1 does not perform this in other context, how can FOXF1 exert this potent anti-myogenic blockade on cells? Perhaps FOXF1 binds to Forkhead motifs for anti- myogenic genes which are normally regulated by other Forkhead transcription factors.
Alternatively, perhaps FOXF1 induces a competing epigenetic and transcriptional landscape which is incompatible with complete myogenic differentiation. These possibilities will be investigated in detail in the following chapter (Chapter 5). P a g e | 94
Chapter 5: Genome-wide binding of FOXF1 in aRMS uncovers cooperative enhancer activation with PAX3/FOXO1 and myogenic regulatory factors
I. Abstract
Since PAX3/FOXO1 exhibits its oncogenic effects through epigenetic regulation, it is likely that additional transcriptional regulators are required. FOXF1 is one of the few transcriptional regulators which is upregulated only in the presence of the PAX3/FOXO1 fusion. The phenotype of FOXF1 gain or loss of function resembles what has been shown for the PAX3/FOXO1 oncogene, suggesting that they these two factors may work in parallel. We performed ChIPseq to understand the genome wide binding distribution of FOXF1 in aRMS. FOXF1 bound to the canonical RTAAAYA motif at distal enhancers. Unexpectedly, the majority of FOXF1 binding sites were co-occupied by PAX3/FOXO1, MYOD1 and MYOG. These sites display strong active enhancer marks in aRMS but not in eRMS. Genes nearby these binding sites are enriched in neural development with significant overlap with aRMS signature genes. Lastly, we show that FOXF1 binding is required for full activation of these co-regulated enhancers. These findings demonstrate that FOXF1 is an integral component of PAX3/FOXO1 transcriptional regulation and provides mechanistic insight as to why FOXF1 loss severely impacts aRMS biology.
II. Introduction
With the lack of significant additional genetic alterations, aRMS cells rely almost entirely on the function of PAX3/FOXO1 to acquire the malignant characteristics of cancer. While nearly all cancers display epigenetic dysregulation to some extent (global hypomethylation, P a g e | 95 hypermethylation of tumor suppressors, ect), cancers which are caused by mutations in transcription factors and epigenetic regulators are distinctively reliant on altered epigenetic states.
The low mutation burden in aRMS tumors suggests that PAX3/FOXO1 must induce a broad but robust epigenetic programming of the cells which is sufficient to promote proliferation while blocking differentiation and apoptosis.
Previous studies have shown that aRMS cells cannot survive without the transcriptional activity of PAX3/FOXO1, demonstrating that the core disease mechanisms lie in PAX3/FOXO1’s downstream targets(64). This essential transcriptional function of PAX3/FOXO1 can also be a unique therapeutic opportunity. While transcription factors are currently undruggable for multiple reasons, there are small molecules which can broadly inhibit chromatin readers and writers and are preferentially toxic to tumor cells reliant on abnormal epigenetic state. PAX3/FOXO1 directly activates enhancers to abnormally high amounts, making them uniquely sensitive to pharmacological inhibition. Small molecules targeting bromodomains (ex. JQ-1, iBET-151 and
OTX-015) can uniquely interfere with over-active enhancer/promoter coupling and gene expression by the PAX3/FOXO1 fusion. Treatment of rhabdomyosarcomas with JQ-1, a BRD4 inhibitor, is effective in inhibiting PAX3/FOXO1 function and ultimately the growth and survival of human aRMS(55, 65).
The outlying, fundamental question of aRMS is how exactly does PAX3/FOXO1 establish the enhancer landscape. Previous work has shown that myogenic regulatory factors MYOD1 and
MYOG also bind at enhancer loci with PAX3/FOXO1(55). The collective binding of these transcription factors at enhancers is associated with strong transcriptional activation. While this collaborative binding between these three myogenic factors (PAX3, MYOD1, MYOG) appears to be a broad component of PAX3/FOXO1 function, a few important questions remain: all P a g e | 96 rhabdomyosarcomas express MYOD1 and MYOG, so do these TFs function differently in the presence of PAX3/FOXO1 and its target genes? Do MYOD1 and MYOG have a universal function in RMS and the PAX3/FOXO1 fusion has an independent, parallel transcriptional mechanism?
The aRMS gene signature is also associated with a unique epigenetic signature based on
H3K27Ac ChIPseq studies. Beyond the PAX3/FOXO1 protein, only a few other transcriptional factors are exclusively expressed in aRMS. FOXF1 is uniquely expressed in aRMS. Although it has no known relevance to skeletal myogenesis, it is possible that FOXF1’s epigenetic effects contribute to the unique gene expression of aRMS cells. The general transcriptional activities of
FOXF1 are not well understood. Two FOXF1 ChIPseq studies have been performed to date. The first showed that FOXF1 can function as a transcriptional activator or repressor in a mouse transformed fetal lung mesenchyme cell line(66). The second showed FOXF1 bound to and activated enhancers in human gastrointestinal stromal tumor (GIST) cell lines(47).
III. Materials and Methods
FOXF1 ChIP-Seq and analysis. FOXF1 ChIP-Seq was performed on a human aRMS cell line Rh18. Briefly, cells were fixed in 0.8% formaldehyde, quenched with glycine, and cells were washed with PBS. Fixed cells were collected, lysed and sonicated using a S220 Focused-
Ultrasonicator (Covaris). Immunoprecipitation of FoxF1 was performed with a previously validated anti-FoxF1 antibody (R&D Systems)(66). Input and FOXF1 -ChIP DNA fragments were purified using a PCR cleanup kit (Epoch). Each sample was sequenced using a HiSeq 2500 sequencing system (Illumina) with 20 million single end reads. FOXF1 ChIP-seq and previously published RMS ChIP-seq datasets (from GSE19063, GSE83726) were analyzed using
BioWardrobe (67, 68). Reads were mapped against human assembly GRCh37 (hg19) and peak P a g e | 97 calling was performed by MACS against input DNA. De novo motif discovery analysis on FOXF1 islands was performed using MEME-ChIP (69). Quantitative comparison between transcription factor datasets was performed using MAnorm (70). Overlapping peaks were quantified and ranked using normalized read densities represented as a-values (“A-value” = 0.5 × log2 (Read density in sample 1 × Read density in sample 2). Identification of islands containing all TFs was performed using the HOMER mergePeaks function.
Dual Luciferase Assays. The ERRFI1 enhancer element was PCR amplified using Q5 polymerase (NEB) and cloned upstream of the minimal promoter of the luciferase reporter plasmid pGL4.23luc2/minP (Promega). Introduction of point mutations to the presumptive consensus binding sites was performed using the Q5 Site Directed Mutagenesis (NEB) according to the manufactures instructions. RD or Rh4 cells were transfected with a Renilla control plasmid and the respective pGL4.23 enhancer construct using Viafect transfection reagent (Promega). 48 hours after transfection, reporter activity was quantified using the Dual-Luciferase Reporter Assay
System (Promega). Enhancer activity is reported as Firefly luciferase activity normalized to
Renilla luciferase. The following primers were used for cloning and mutagenesis of the binding site:
SLC45A1 enhancer Forward ATATGGTACCCAGAAAGCTAGCAGCTCCGCAT
SLC45A1 enhancer Reverse ATATCTCGAGGGCTGAGCACCTCTCTGAGACT
FoxF1 mutagenesis Forward ATTCAGTTCTCTGCAGAACACTGTGAAAGGC
FoxF1 mutagenesis Reverse TTTAGAAAATGAATTCCTTTATCC
E-Box mutagenesis Forward TATGTTTCTTGGATCCCCGATTTACCAGCTC
E-Box mutagenesis Reverse CAGTTCTGTGTTTCAAAATG P a g e | 98
Pax3 mutagenesis Forward CTGAGGGACAAAGCTTCTGACCGATTCTCTCTGG
Pax3 mutagenesis Reverse CCAAAATGAACTCAGCCTATTG
IV. Results
FOXF1 binds at enhancers active in aRMS but not in eRMS. To identify the underlying
mechanism, we analyzed the genome-wide binding of FOXF1 by performing FOXF1 ChIP-seq in
Rh18 cells. A total of 13,157 FOXF1 peaks were identified. FOXF1 bound almost exclusively
within gene introns (43.85%) and in intergenic regions (40.62%) defined as >20kb from the nearest
gene (Fig. 1a). A FOXF1 ChIP-seq biological replicate using Rh4 cells identified the same binding
sites, although this dataset had low overall enrichment. FOXF1 binding in Rh18 and Rh4 cells
was associated with active enhancer marks P300 and H3K27Ac in both Rh18 and Rh4 cells (Fig.
1b). Interestingly, little to no enrichment of H3K27Ac was observed at these same loci the eRMS
cell line RD (Fig. 1b). These results establish that FOXF1 genome-wide binding is associated with
enhancers active in aRMS but not eRMS. (a) (b) FoxF1 peak distribution Input FOXF1 P300 H3K27Ac (13,157 peaks) 7.46% 4.48% 3.60%
Upstream 40.6… Promoter Exon 43.85% Intron Intergenic Rh18 Rh18 Rh4 Rh4 Rh18 Rh4 RD 10kb (eRMS) Figure 1. FoxF1 binds at distal enhancers which are active in aRMS but not eRMS. (a) Genomic distribution of FoxF1 binding in human aRMS cell line Rh18. Promoter = ± 1kb from TSS, Upstream =
1kb to 20kb from TSS, Intergenic >20kb from TSS. (b) Heatmap of FOXF1, P300 and H3K27Ac ChIPseq data in aRMS cell lines Rh4, Rh18 or eRMS cell line RD.
P a g e | 99
Discovery of PAX3/FOXO1, MYOD1, and/or MYOG at most FOXF1 binding sites genome-wide. De novo motif discovery on FOXF1 peaks identified the canonical Forkhead
RTAAAYA motif as the most significantly enriched motif, verifying the quality of our ChIP (Fig.
2a)(66). The second most enriched motif was the CAGCTGYB E-Box motif which is known to be enriched at PAX3/FOXO1, MYOD1 and MYOG binding sites in aRMS (Fig. 2a). Using available
ChIPseq datasets, we looked at the presence of these transcription factors at FOXF1 binding sites.
Out of all FOXF1 bound regions, 36% of these sites also had PAX3/FOXO1 bound (Fig. 2b).
Similarly, 26% and 34% of all FOXF1 binding sites were bound by MYOD1 or MYOG, respectively.
The myogenic transcription factors PAX3/FOXO1, MYOD1 and MYOG can have shared binding sites in myoblasts and RMS. To determine if FOXF1 binding sites contained multiple myogenic TFs, we performed a clustered analysis with PAX3/FOXO1, MYOD1 and MYOG. We identified 1,615 FOXF1 binding sites in which PAX3/FOXO1, MYOD1, and MYOG were all bound in overlapping fashion (Fig. 2b,). In total, 7,226 FOXF1 binding sites (55% of all FOXF1 binding) are co-occupied by at least one of these transcription factors. These data demonstrate that
FOXF1 exhibits widespread binding of enhancers in tandem with the core transcriptional machinery of aRMS.
This clustering analysis displays the frequency in which a binding region is shared but it does not provide information about the extent of FOXF1 binding at each peak. Generally, high ChIP- seq peak enrichment is indicative of high affinity binding and/or a higher percentage of cells with the TF bound. To understand if FOXF1 enrichment level (not peak frequency) was higher in any cluster, we analyzed the number of tags found at each FOXF1 binding site in each cluster. Pairwise P a g e | 100 comparison revealed several clusters which had significantly higher FOXF1 peaks than any other peaks (Fig. 2c). Clusters 2, 6, and 8 all had significantly higher FOXF1 peaks compared to any of
(a) (b) # sites ChIPseq Cluster (% FOXF1 sites) Top enriched motifs in FOXF1 peaks Forkhead 5931 C1: FOXF1 only (45.1%) p-value: 4.1e-2544 # peaks: 9372/13157 (71%) E-Box FOXF1 & 2087 C2: PAX3/FOXO1 (15.9%) p-value = 1.7e-606 # peaks: 3059/13157 (23.2%) FOXF1 & 468 C3: MYOD1 (3.6%)
(c) FOXF1 peak quality between clusters FOXF1 & 853 C4: (p-values) MYOG (6.5%)
Significance Least FOXF1, 230 C5: PAX3/FOXO1
& MYOD1 (1.7%)
FOXF1, 853 C6: PAX3/FOXO1 (6.5%) & MYOG
FOXF1, 1110 C7: MYOD1 &
MYOG (8.4%) Most
(d) FOXF1 peak intensity 1615 C8: ALL 4 TFs C1 C2 (12.3%) C6 C8 5kb
-2500 0 2500 P a g e | 101
Figure 2. Majority of FOXF1 binding sites are also bound by PAX3/FOXO1, MYOD1 and/or
MYOG. (a) De novo motif analysis of the top two enriched motifs within FoxF1 ChIPseq peaks.
Analysis was performed using MEME-ChIP. (b) Analysis of FOXF1 ChIPseq peaks by cluster analysis to identify shared ChIPseq binding sites. (left columns) Each of 13,157 FOXF1 binding sites were assigned to a single cluster based on the presence of additional transcription factors.(middle columns) Heatmaps representing the read densities of all sites in each cluster.
Heatmaps are centered around the FOXF1 peak and ordered high (top) to low (bottom) of FOXF1 peak reads. (right column) Total number and proportion of FOXF1 sites in each cluster.
the other clusters. Besides FOXF1 binding, the common factor between these clusters is
PAX3/FOXO1. Overall, FOXF1 binds to enhancers most frequently with PAX3/FOXO1 and that these co-bound sites also display the highest level of FOXF1 enrichment.
FOXF1 binds to aRMS-specific enhancer elements to drive gene signature. To determine the relevance of FOXF1-bound enhancers, we performed gene-set enrichment on genes proximal to FOXF1 binding sites. FOXF1 associated genes were strongly associated with the fusion positive aRMS gene signature such as FGFR2, ELMO1, CNR1, TFAP2B, and SPATS2L (Fig. 7g). We also performed this analysis on FOXF1 sites co-bound with PAX3/FOXO1 or with PAX3/FOXO1,
MYOD1 and MYOG. With both of these additional iterations, we observed significant enrichment of aRMS signature genes. For visual representation of these enhancers, we looked at the ELMO1 gene locus. ELMO1 is an upregulated gene in the fusion-positive gene signature and functions to promote rac-dependent cell migration (71-74). The ELMO1 gene locus contains multiple FOXF1 peaks throughout the gene locus, some of which are shared by PAX3/FOXO1 and the MRFs (Fig. P a g e | 102
7h). Expression of ELMO1 is decreased after FOXF1 knockdown, indicating that FOXF1
positively regulates these enhancer sites.
FOXF1 binding at aRMS signature genes
chr7:36,820,000- ELM01 37,527,000 Rh18
FoxF1 (Rh18) 50 FoxF1 (Rh4) 40 Pax3/FoxO1 (Rh4)
MyoD1 (Rh4) 30
Myogenin (Rh4) 20 P300 (Rh4)
Expression mRNA (TPM) 10
H3K27Ac (Rh4) ELMO1 H3K27Ac (RD) 0 Control FoxF1 KD 200kb aRMS Enhancer
Figure 3. FOXF1 bound enhancers are associated with aRMS signature genes. (a) ToppFun
gene ontology analysis of genes proximal to the FOXF1 binding sites. Genes proximal to all
FOXF1 peaks or those shared with PAX3/FOXO1, MYOD1, and MYOG were analyzed to identify
datasets in which these genes are co-expressed. (b) (left) Browser view of one of the genes
ELMO1 which was identified as a FOXF1 regulated gene and is part of the aRMS signature found P a g e | 103 by ToppFun analysis. (right) RNAseq expression of ELMO1 after FOXF1 knockdown in Rh18 cells.
FOXF1 functions co-operatively with PAX3/FOXO1, MYOD1, and MYOG to maximize enhancer activation. Lastly, we sought to clarify the contribution of each individual transcriptional factor when they co-regulate the same enhancer element. We studied an enhancer element upstream of the ERRFI1 gene locus whose gene product directly activates the AKT pathway and is associated with tumor proliferation and chemoresistance (75). This ERRFI1 enhancer is characterized by overlapping binding of FOXF1, PAX3/FOXO1, MYOD1, and
MYOG (Fig. 4a). Both this enhancer element and its associated target gene ERRFI1 are specifically active in Rh4 (aRMS) but not RD (eRMS), suggesting that the unique expression of
PAX3/FOXO1 and/or FOXF1 may be required for activation. In support of the latter, knockdown of FOXF1 in Rh18 and Rh4 cells reduces expression of ERRFI1.
To test this, an 826bp DNA fragment encompassing this enhancer element was cloned behind a minimal promoter driving expression of a firefly luciferase reporter. When transfected into Rh4 cells, this enhancer increased reporter expression by 304x more than the minimal promoter alone
(Fig. 4c). This same plasmid had no enhancer activity when transfected into the eRMS cell line
RD. This indicates that the cloned fragment retained enhancer activity that is reflective of the endogenous locus. To dissect the contribution of each transcription factor to enhancer activity, we generated a series of single and compound point mutations in the predicted FOXF1,
MYOD1/MYOG, and PAX3/FOXO1 binding sites (Fig. 4d). Point mutations to the Forkhead or
E-Box motifs decreased enhancer activity by roughly 50%. Combined mutagenesis of the
Forkhead and the E-Box motifs further reduces the enhancer activity by 97% of the activity of the P a g e | 104
WT enhancer, demonstrating that FOXF1, MYOD1 and MYOG function additively at these enhancers. Mutagenesis of the paired-domain consensus site alone dramatically reduced enhancer activity, suggesting that PAX3/FOXO1 sits atop the transcriptional hierarchy at these sites and is essential for enhancer activation. While FOXF1, MYOD1, and MYOG are not sufficient to activate this enhancer in the absence of PAX3/FOXO1 binding, they strongly amplify the transcriptional output. In conclusion, we found that the FOXF1 locus is directly activated by the
PAX3/FOXO1 protein and FOXF1, MYOD1, MYOG and the PAX3/FOXO1 proteins function collaboratively to establish a unique epigenetic landscape which promotes proliferation and blocks differentiation in alveolar rhabdomyosarcoma.
P a g e | 105
(a)
chr1: 8,050,000 - ERRFI1 8,260,000
FoxF1 (Rh18)
FoxF1 (Rh4)
Pax3/FoxO1 (Rh4)
MyoD1 (Rh4)
P300 (Rh4)
H3K4me3 (Rh4)
H3K27Ac (Rh4)
H3K27Ac (RD) 50kb aRMS Enhancer Forkhead (RTAAAYA) E-Box (CAGCTGYB) Pax3 (GTCAYGS) Chr1: 8,197,835 CAGTTCTATAAACAACA……CTTCAGCTGCCGAT ..……ACACCGTGACTGACCGA Chr1: 8,198,095
(b) (c) Rh18 Scramble 1.2 400 Rh18 shFoxF1 pGL4-Empty 304x 1 Rh4 Scramble 350 pGL4-ERRFI1 enh. Rh4 shFoxF1 300 0.8 250 0.6 200 expression 0.4 150 100 Fold Change in mRNA in Change Fold 0.2 50 1x 0.64x 1x 0 Foldchange reporter activity 0 FoxF1 ERRFI1 RD Rh4 (d) Minimal ERRFI1 Enhancer element promoter Relative luciferase activity (Luc/Ren) 0 1 2 3 4
Luc 1x Luc 1061x Luc 558x Luc 464x Luc 22x Luc 29x Luc 16x
Luc 2x
P a g e | 106
Figure 4. FoxF1 binding with PAX3/FOXO1 and myogenic regulatory factors is essential for aRMS-specific enhancers activation. (a) (top) Active aRMS-specific enhancer 187kb upstream of ERRFI1. (bottom) Identification of Forkhead, E-Box and Pax3/FoxO1 motifs corresponding to the ChIP-Seq peaks. (b) qRT-PCR of Rh18 and Rh4 cells with stable knockdown of FoxF1. Values were normalized to β-actin (n=3). Data reported as mean ± SEM. (c) An 826bp fragment corresponding to the ERRFI1 enhancer element was cloned into pGL4.23 plasmid containing a minimal promoter-driven luciferase reporter. RD (eRMS) and Rh4 (aRMS) were transfected with pGL4 and renilla control plasmids and a dual luciferase assay was performed after 48 hours. .
Normalized luciferase activity was reported as mean ± SEM (n=3). (d) Dual luciferase assay in
Rh4 cells with the pGL4-ERRFI1 enhancer plasmid following single or compound point mutations at the Forkhead (orange), E-Box (green) or Pax (blue) binding motifs. Normalized luciferase activity was reported as mean ± SEM (n=3).
V. Discussion
The mechanisms of how PAX3/FOXO1 rewires the aRMS epigenome is an ongoing area of investigation. Since so few mutations are observed in aRMS tumor cells, oncogenic pathways may be activated by transcriptional overexpression rather than activating mutations of pathway components. Conversely, genes which counteract malignant transformation must be inhibited by transcriptional, post-transcriptional, or post-translational mechanisms rather than by genetic loss.
All of these mechanisms are dependent on PAX3/FOXO1, either directly or indirectly. Loss of
PAX3/FOXO1 has a catastrophic effect on the aRMS epigenome and ensuing biological phenotype. Expression of PAX3/FOXO1 in fibroblasts, myoblasts, or eRMS cells is sufficient to induce some of the gene signature and epigenetic features of aRMS cells. These observations P a g e | 107 suggest that PAX3/FOXO1 and its downstream target genes are sufficient to establish and maintain a unique epigenome in aRMS.
FOXF1 is one of the core transcriptional targets of the PAX3/FOXO1 as concluded by i.) the consistency of FOXF1 expression in aRMS, ii.) the FOXF1 3’ enhancer is one of the most enriched
PAX3/FOXO1 binding sites in aRMS iii.) the requirement of PAX3/FOXO1 to maintain FOXF1 expression in aRMS, and iv.) the ability of PAX3/FOXO1 to induce FOXF1 expression in non- aRMS cells (see Chapter 3). We also have shown that FOXF1 is essential for promoting proliferation and providing a potent block on myogenic differentiation (see Chapter 4). In this chapter, we profiled the genome-wide binding of FOXF1 to understand the molecular mechanism(s) of how FOXF1 exerts its effects on aRMS cells. ChIPseq analysis of the endogenous FOXF1 protein in aRMS cells showed that FOXF1 bound DNA at the consensus
Forkhead motif RTAAAYA at enhancer elements associated with aRMS signature genes. A significant portion of FOXF1 binding sites (55%) had co-occupancy of PAX3/FOXO1, MYOD1, and/or MYOG. This is a very unexpected result considering that FOXF1 and these myogenic transcription factors aren’t co-expressed in any cell type that we are aware of other than aRMS tumor cells.
These myogenic factors, including PAX3/FOXO1, are centrally enriched at FOXF1 binding sites with an average distance of 17bp from the center of the peak and the Forkhead motif. The co-regulation of these enhancers in such a precise but widespread manner poses some interesting questions. One explanation is that while FOXF1 binds to these RTAAAYA enhancer motifs, they are actually normally bound by a different Forkhead transcription factor. Many Forkheads transcription factors such as FOXA1, FOXA2, FOXL1, FOXD1, and FOXF2 also bind to the
RTAAAYA motif. The converse possibility is also intriguing. The majority of these co-regulated P a g e | 108 enhancers are not associated with proximal myogenic genes. Instead, the nearby genes are associated with neural development and many of these are actually aRMS signature genes.
Perhaps the myogenic factors, rather, are binding aberrantly at E-boxes for neural genes due to the presence of FOXF1 binding. This could explain why MYOD1 and MYOG fail to induce myogenic differentiation if they are being recruited away from their normal myogenic binding sites.
The FOXF1 binding sites were clustered based on the transcription factor composition of the enhancer site. Approximately 45% of FOXF1 binding sites were bound by FOXF1 alone and had little evidence of P300 recruitment and enhancer activation. This observation may tell us something about how FOXF1 functions as a transcription factor. First, the binding without activation can be evidence of transcriptional repression. While it is possible that FOXF1 binding alone can serve as a transcriptional repressor, the binding of FOXF1 with PAX3/FOXO1,
MYOD1, and MYOG is associated with strong active enhancer marks. Second, this may indicate pioneer activity is a major mechanism of transcriptional regulation by FOXF1. Related Forkheads
FOXA1 and FOXA2 are well characterized pioneer transcription factors which can bind to repressed heterochromatin but require additional co-factors for establishing an active enhancer.
Our luciferase assay experiments occur in a nucleosome-free plasmid system. Since loss of
FOXF1 binding to nascent DNA reduced enhancer activity, FOXF1 is capable of activating enhancers.
The collaborative binding of FOXF1 and PAX3/FOXO1 is significantly more common than with FOXF1 and MYOD1 or MYOG. Approximately 36% of all FOXF1 binding sites are also bound by PAX3/FOXO1. In addition to being the most frequently shared binding events, the enrichment of each FOXF1 binding site is significantly higher at PAX3/FOXO1 binding sites then at any other clusters. This means that in aRMS cells, the majority of FOXF1 protein is binding at P a g e | 109 enhancers with PAX3/FOXO1. The binding of FOXF1 with PAX3/FOXO1 at so many enhancers helps explain why loss of FOXF1 recapitulates so many aspects of PAX3/FOXO1 loss including reduced cell proliferation with induction of apoptosis and differentiation.
In summary, we found that FOXF1 binds to enhancers with PAX3/FOXO1, MYOD1 and
MYOG in aRMS and that this binding activates thousands of enhancers which are mostly inactive in eRMS. This unique enhancer activation is associated with aRMS signature genes. P a g e | 110
Chapter 6: Conclusions and Discussion
I. Working model of FOXF1 function in aRMS.
PAX3/FOXO1
FOXF1 MYOD1
MYOG FOXF1
Clustered binding at Promote cell proliferation aRMS-specific enhancers Inhibit myogenic differentiation Establish aRMS-specific gene signature Figure 1. Schematic diagram showing the establishment of the epigenomic landscape which gives rise to the unique molecular and biological properties exhibited by alveolar rhabdomyosarcoma. The Pax3/FoxO1 fusion protein directly activates the FOXF1 locus by activating two distal enhancer elements. Upon expression, FoxF1 prevents myogenic differentiation and promotes proliferation by binding cooperatively with Pax3/FoxO1, MyoD1, and MyoG at distal enhancer elements and transcriptionally activating genes essential for aRMS tumorigenesis.
The goal of this work has been to provide a detailed molecular analysis of the upstream regulation and downstream consequences of FOXF1 expression in human alveolar rhabdomyosarcoma. This study was initiated after the identification of a unique gene signature which is associated with the presence of a PAX/FOXO rearrangement and poor patient prognosis. P a g e | 111
We found that PAX3/FOXO1 is necessary and sufficient to induce FOXF1 expression by activating two distal FOXF1 enhancer. This first enhancer is located -315kb upstream of FOXF1 and is associated with an enhancer lincRNA LINC01082 which mirrors FOXF1 expression in aRMS PDX models and normal human tissues. The second enhancer is located 8kb downstream of the FOXF1 3’UTR and is one of the most highly enriched PAX3/FOXO1 binding sites in aRMS.
Once activated, FOXF1 binds to enhancers with PAX3/FOXO1, MYOD1, and MYOG which gives rise to a unique transcriptional program which includes known aRMS oncogenes. Without
FOXF1, aRMS tumor cells are highly sensitive to spontaneous myogenic differentiation which occurs rapidly after acute FOXF1 knockdown and is sustained long term in FOXF1 knockout aRMS cells (Fig. 1).
II. Understanding the potent anti-myogenic properties of FOXF1.
One of the most important findings of this work is the identification of FOXF1 as a potent inhibitor of myogenic differentiation. Loss of FOXF1 sensitizes aRMS to terminal myogenic differentiation without having to also inhibit PAX3/FOXO1. Acute knockdown of FOXF1 led to the immediate upregulation of mature skeletal markers including multiple troponin, myosin, and actin isoforms.
In vivo, we were able to visualize widespread spontaneous myogenic differentiation in FOXF1-
KO cells. In the absence of genetic perturbation, spontaneously differentiated aRMS cells naturally lose FOXF1 expression. To remove the contribution of collaborating oncogenes in aRMS cells, we also tested the influence of FOXF1 in primary human skeletal muscle myoblasts.
We found that FOXF1 expression was sufficient to completely block myogenic differentiation and a parallel phenotype in sustaining the long-term proliferative capacity of the myoblasts.
The results in the myoblast model provide important insight into how FOXF1 might be exerting the anti-differentiation effects in aRMS. First, primary myoblasts are primed for P a g e | 112 myogenic differentiation and readily do so upon growth factor depletion. Second, these cells have limited proliferative capacity with significant senescing beginning after only 10-15 population doublings. Despite this restricted biology of a primary human myoblast culture, FOXF1 expression alone completely alters the phenotype of the cells. Notably, FOXF1’s impact on myogenesis does not require the presence of oncogenic mutations or the epigenetic dysregulation found in tumor cells.
Since FOXF1 alone can exert the anti-myogenic effects on cells, FOXF1 must be doing so independent of PAX3/FOXO1 function. This is assuming that the FOXF1 mechanisms are shared between myoblasts and aRMS tumor cells. ChIPseq analysis in aRMS showed that FOXF1 does not bind to and repress mature skeletal muscle genes even though they are highly upregulated after
FOXF1 loss.
An alternative mechanism would be through interfering with Myogenin function. In normal myoblasts, Myogenin expression is transient and functions to drive the transition from proliferating myoblast to a terminally differentiated myocyte. As discussed in Chapter 1, fusion positive aRMS cells display strong and diffuse nuclear staining for Myogenin which distinguishes it from fusion negative RMS. The strong and diffuse staining pattern of Myogenin is strikingly similar to what is observed for TP53 mutant tumors in loss of function, in-frame mutations have strong diffuse staining for TP53(76, 77). Furthermore, the majority of aRMS cells express MYOG and not the myogenic stem cell marker PAX7 which demonstrates that most aRMS cells are accumulating at this bottleneck during differentiation.
Previous ChIPseq studies demonstrated that Myogenin is capable of binding DNA at a typical
E-box motif. However, it is not known if these binding sites represent the typical binding distribution of Myogenin. Our FOXF1 ChIPSeq studies showed nearly 5,000 binding sites that P a g e | 113
FOXF1 and Myogenin bind together. With such a significant frequency of shared binding, it begs the question if Myogenin is recruited to these loci in the presence of FOXF1. In support of this, very few of the genes near these enhancers are known Myogenin transcriptional targets or related to myogenesis as determined by gene ontology.
In order for us to identify any pathological functions of Myogenin in aRMS, future studies are needed. The phenotype of MYOG knockout mice is lethal with a near complete loss of mature skeletal muscle(78-80). The severe phenotype of Myogenin knockout mice is quite different than
MYOD1 knockout mice which are phenotypically normal due to functional compensation by related bHLH MYF5(81). Although the early mouse knockout studies have demonstrated that
Myogenin is essential for terminal myogenic differentiation, no mechanisms have been demonstrated by genome-wide binding. This data will be important for understanding the roles of
Myogenin in normal and pathological myogenesis.
III. The relationships between PAX3/FOXO1, PAX3, and FOXF1.
PAX3 is essential for the specification and migration of myogenic precursors during embryonic development. PAX3 performs a similar role in neural crest cells during development.
In PAX3 knockout mice, both specification and migrations processes do not take place and leads to the absence of muscle and severe neural tube defects, respectively. The PAX3/FOXO1 fusion gene is thought to regulate the same enhancers as wildtype PAX3 with several well-characterized
PAX3 bound enhancers also bound by PAX3/FOXO1(24). Despite these apparent binding similarities, PAX3/FOXO1 endows cells with a malignant potential. One explanation for this is that PAX3/FOXO1 has stronger activating abilities due to the gain of the FOXO1 transactivation domain(20-23). A second is the t(2;13)(q35;q14) rearranged allele acquires new enhancer elements which drives transcription of PAX3/FOXO1 at a higher level than the normal PAX3 allele(82). P a g e | 114
We identified that the FOXF1 gene is a direct transcriptional target of PAX3/FOXO1. Since
no link has been established between FOXF1 and PAX3, we performed an embryro-wide
characterization to identify overlapping regions of expression in development and the adult. The
vast majority of known PAX3 expressing cells had undetectable FOXF1 expression, including the
(a) (b) Pax3 Pax7 FoxF1
Dermomyotome DM DM (DM) DM
LB LB LB LB
Limb bud (LB)
Figure 2. FOXF1 is not expressed during myogenic specification and migration in the
embryonic limbs and trunk. (a) Diagram showing specification and migration of PAX3
expressing stem cells from the dermomyotome (DM) and out to the limb bud (LB). Image adapted
from https://plasticsurgerykey.com/wp-content/uploads/2016/02/. (b) IHC for PAX3, PAX7, and
FOXF1 in the dermomyotome and limb bud during embryonic myogenesis. Sections from E9.5
wildtype mouse embryo.
PAX3 expressing myogenic precursors in early development (Fig. 2). In the mouse embryo head,
however, we were able to identify a population of cells in the tongue that expressed both PAX3
and FOXF1 (Fig. 3,4). These mesenchymal cells are derived from the cranial neural crest
mesenchyme and migrate ventrally to form the brachial arches(83). Some of these cells will commit P a g e | 115 to the myogenic lineage and give rise to the striated skeletal muscle of the tongue while many of these cells will mature into various fibroblast populations.
The contrast between PAX3+FOXF1+ cells and the myogenic lineage (PAX7+, MYH+) might be significant. While this is a single cell population in development, it demonstrates that PAX3 and FOXF1 co-expressing cells are distinctly absent in the myogenic lineage which PAX3 expressing cells alone are. It is possible that the expression of FOXF1 is sufficient to repress myogenic differentiation of these neural crest cells in a manner similar to aRMS. While we do not think that aRMS is precisely recapitulating this cell population, it is possible that aRMS broadly resembles multipotent neural crest cells. In support of this, the majority of genes uniquely upregulated in aRMS are involved in neurogenesis with significant overlap with cranial neural crest derived cell types. Future experiments could test if PAX3 regulates the FOXF1 enhancers in this context and if FOXF1 deletion within this cell population results in myogenic differentiation.
P a g e | 116
(a) FoxF1 Pax3
MyH Pax7
(b) FoxF1 Pax3
MyH Pax7
Figure 3. FOXF1 and PAX3 are co-expressed in the mouse head at E13.5 but not in the skeletal muscle lineage. (a). Coronal, serial sections were stained for FOXF1, PAX3, PAX7 and MyH in a wildtype mouse head at E13.5. PAX7 and MyH mark the myogenic progenitors and early myocytes, respectively. (b) High magnification image of the mouse tongue in these same sections show significant overlap in the FOXF1 and PAX3 expression domains. P a g e | 117
(a) FoxF1 Pax3 Pax7
Sk.M. Sk.M. Sk.M.
(b) Pax3 FoxF1 Merged
Pax7 FoxF1 Merged P a g e | 118
Figure 4. FOXF1 and PAX3 are co-expressed in the mouse tongue mesenchyme but not myogenic
progenitors at E14.5. (a) Immunohistochemistry on sagittal serial sections on a mouse head at E14.5 showing overlap in FOXF1 and PAX3 but not PAX7 expression domains. (b) (top row)
Coimmunofluorescence showing the co-expression of FOXF1 and PAX3 in the same cells. (bottom row)
The FOXF1 positive mesenchyme is interspersed between the skeletal muscle but shows no co- expression with the muscle satellite marker PAX7. P a g e | 119
References
1. Yang L, Takimoto T, Fujimoto J. Prognostic model for predicting overall survival in children and adolescents with rhabdomyosarcoma. BMC Cancer. 2014;14:654.
2. Ognjanovic S, Linabery AM, Charbonneau B, Ross JA. Trends in childhood rhabdomyosarcoma incidence and survival in the United States, 1975-2005. Cancer.
2009;115(18):4218-26.
3. Alaggio R, Zhang L, Sung YS, Huang SC, Chen CL, Bisogno G, et al. A Molecular
Study of Pediatric Spindle and Sclerosing Rhabdomyosarcoma: Identification of Novel and
Recurrent VGLL2-related Fusions in Infantile Cases. Am J Surg Pathol. 2016;40(2):224-35.
4. Kumar A, Singh M, Sharma MC, Bakshi S, Sharma BS. Pediatric sclerosing rhabdomyosarcomas: a review. ISRN Oncol. 2014;2014:640195.
5. Meza JL, Anderson J, Pappo AS, Meyer WH, Children's Oncology G. Analysis of prognostic factors in patients with nonmetastatic rhabdomyosarcoma treated on intergroup rhabdomyosarcoma studies III and IV: the Children's Oncology Group. J Clin Oncol.
2006;24(24):3844-51.
6. Perez EA, Kassira N, Cheung MC, Koniaris LG, Neville HL, Sola JE.
Rhabdomyosarcoma in children: a SEER population based study. J Surg Res. 2011;170(2):e243-
51.
7. de la Monte SM, Hutchins GM, Moore GW. Metastatic behavior of rhabdomyosarcoma.
Pathol Res Pract. 1986;181(2):148-52.
8. Weigel BJ, Lyden E, Anderson JR, Meyer WH, Parham DM, Rodeberg DA, et al.
Intensive Multiagent Therapy, Including Dose-Compressed Cycles of Ifosfamide/Etoposide and
Vincristine/Doxorubicin/Cyclophosphamide, Irinotecan, and Radiation, in Patients With High- P a g e | 120
Risk Rhabdomyosarcoma: A Report From the Children's Oncology Group. J Clin Oncol.
2016;34(2):117-22.
9. Shern JF, Chen L, Chmielecki J, Wei JS, Patidar R, Rosenberg M, et al. Comprehensive genomic analysis of rhabdomyosarcoma reveals a landscape of alterations affecting a common genetic axis in fusion-positive and fusion-negative tumors. Cancer Discov. 2014;4(2):216-31.
10. Shukla N, Ameur N, Yilmaz I, Nafa K, Lau CY, Marchetti A, et al. Oncogene mutation profiling of pediatric solid tumors reveals significant subsets of embryonal rhabdomyosarcoma and neuroblastoma with mutated genes in growth signaling pathways. Clin Cancer Res.
2012;18(3):748-57.
11. Chen L, Shern JF, Wei JS, Yohe ME, Song YK, Hurd L, et al. Clonality and evolutionary history of rhabdomyosarcoma. PLoS Genet. 2015;11(3):e1005075.
12. Bridge JA, Liu J, Qualman SJ, Suijkerbuijk R, Wenger G, Zhang J, et al. Genomic gains and losses are similar in genetic and histologic subsets of rhabdomyosarcoma, whereas amplification predominates in embryonal with anaplasia and alveolar subtypes. Genes
Chromosomes Cancer. 2002;33(3):310-21.
13. Williamson D, Missiaglia E, de Reynies A, Pierron G, Thuille B, Palenzuela G, et al.
Fusion gene-negative alveolar rhabdomyosarcoma is clinically and molecularly indistinguishable from embryonal rhabdomyosarcoma. J Clin Oncol. 2010;28(13):2151-8.
14. Mosquera JM, Sboner A, Zhang L, Kitabayashi N, Chen CL, Sung YS, et al. Recurrent
NCOA2 gene rearrangements in congenital/infantile spindle cell rhabdomyosarcoma. Genes
Chromosomes Cancer. 2013;52(6):538-50. P a g e | 121
15. Rekhi B, Upadhyay P, Ramteke MP, Dutt A. MYOD1 (L122R) mutations are associated with spindle cell and sclerosing rhabdomyosarcomas with aggressive clinical outcomes. Mod
Pathol. 2016;29(12):1532-40.
16. Agaram NP, Chen CL, Zhang L, LaQuaglia MP, Wexler L, Antonescu CR. Recurrent
MYOD1 mutations in pediatric and adult sclerosing and spindle cell rhabdomyosarcomas: evidence for a common pathogenesis. Genes Chromosomes Cancer. 2014;53(9):779-87.
17. Szuhai K, de Jong D, Leung WY, Fletcher CD, Hogendoorn PC. Transactivating mutation of the MYOD1 gene is a frequent event in adult spindle cell rhabdomyosarcoma. J
Pathol. 2014;232(3):300-7.
18. Barr FG, Duan F, Smith LM, Gustafson D, Pitts M, Hammond S, et al. Genomic and clinical analyses of 2p24 and 12q13-q14 amplification in alveolar rhabdomyosarcoma: a report from the Children's Oncology Group. Genes Chromosomes Cancer. 2009;48(8):661-72.
19. Barr FG, Smith LM, Lynch JC, Strzelecki D, Parham DM, Qualman SJ, et al.
Examination of gene fusion status in archival samples of alveolar rhabdomyosarcoma entered on the Intergroup Rhabdomyosarcoma Study-III trial: a report from the Children's Oncology Group.
J Mol Diagn. 2006;8(2):202-8.
20. Sublett JE, Jeon IS, Shapiro DN. The alveolar rhabdomyosarcoma PAX3/FKHR fusion protein is a transcriptional activator. Oncogene. 1995;11(3):545-52.
21. Fredericks WJ, Galili N, Mukhopadhyay S, Rovera G, Bennicelli J, Barr FG, et al. The
PAX3-FKHR fusion protein created by the t(2;13) translocation in alveolar rhabdomyosarcomas is a more potent transcriptional activator than PAX3. Mol Cell Biol. 1995;15(3):1522-35. P a g e | 122
22. Bennicelli JL, Fredericks WJ, Wilson RB, Rauscher FJ, 3rd, Barr FG. Wild type PAX3 protein and the PAX3-FKHR fusion protein of alveolar rhabdomyosarcoma contain potent, structurally distinct transcriptional activation domains. Oncogene. 1995;11(1):119-30.
23. Bennicelli JL, Edwards RH, Barr FG. Mechanism for transcriptional gain of function resulting from chromosomal translocation in alveolar rhabdomyosarcoma. Proc Natl Acad Sci U
S A. 1996;93(11):5455-9.
24. Cao L, Yu Y, Bilke S, Walker RL, Mayeenuddin LH, Azorsa DO, et al. Genome-wide identification of PAX3-FKHR binding sites in rhabdomyosarcoma reveals candidate target genes important for development and cancer. Cancer Res. 2010;70(16):6497-508.
25. Relaix F, Montarras D, Zaffran S, Gayraud-Morel B, Rocancourt D, Tajbakhsh S, et al.
Pax3 and Pax7 have distinct and overlapping functions in adult muscle progenitor cells. J Cell
Biol. 2006;172(1):91-102.
26. Duan F, Smith LM, Gustafson DM, Zhang C, Dunlevy MJ, Gastier-Foster JM, et al.
Genomic and clinical analysis of fusion gene amplification in rhabdomyosarcoma: a report from the Children's Oncology Group. Genes Chromosomes Cancer. 2012;51(7):662-74.
27. Davicioni E, Anderson MJ, Finckenstein FG, Lynch JC, Qualman SJ, Shimada H, et al.
Molecular classification of rhabdomyosarcoma--genotypic and phenotypic determinants of diagnosis: a report from the Children's Oncology Group. Am J Pathol. 2009;174(2):550-64.
28. Sorensen PH, Lynch JC, Qualman SJ, Tirabosco R, Lim JF, Maurer HM, et al. PAX3-
FKHR and PAX7-FKHR gene fusions are prognostic indicators in alveolar rhabdomyosarcoma: a report from the children's oncology group. J Clin Oncol. 2002;20(11):2672-9. P a g e | 123
29. Kubo T, Shimose S, Fujimori J, Furuta T, Ochi M. Prognostic value of PAX3/7-FOXO1 fusion status in alveolar rhabdomyosarcoma: Systematic review and meta-analysis. Crit Rev
Oncol Hematol. 2015;96(1):46-53.
30. Wachtel M, Dettling M, Koscielniak E, Stegmaier S, Treuner J, Simon-Klingenstein K, et al. Gene expression signatures identify rhabdomyosarcoma subtypes and detect a novel t(2;2)(q35;p23) translocation fusing PAX3 to NCOA1. Cancer Res. 2004;64(16):5539-45.
31. Barr FG, Qualman SJ, Macris MH, Melnyk N, Lawlor ER, Strzelecki DM, et al. Genetic heterogeneity in the alveolar rhabdomyosarcoma subset without typical gene fusions. Cancer
Res. 2002;62(16):4704-10.
32. Arnold MA, Anderson JR, Gastier-Foster JM, Barr FG, Skapek SX, Hawkins DS, et al.
Histology, Fusion Status, and Outcome in Alveolar Rhabdomyosarcoma With Low-Risk Clinical
Features: A Report From the Children's Oncology Group. Pediatr Blood Cancer. 2016;63(4):634-
9.
33. Selfe J, Olmos D, Al-Saadi R, Thway K, Chisholm J, Kelsey A, et al. Impact of fusion gene status versus histology on risk-stratification for rhabdomyosarcoma: Retrospective analyses of patients on UK trials. Pediatr Blood Cancer. 2017;64(7).
34. Lae M, Ahn EH, Mercado GE, Chuai S, Edgar M, Pawel BR, et al. Global gene expression profiling of PAX-FKHR fusion-positive alveolar and PAX-FKHR fusion-negative embryonal rhabdomyosarcomas. J Pathol. 2007;212(2):143-51.
35. Szafranski P, Dharmadhikari AV, Wambach JA, Towe CT, White FV, Grady RM, et al.
Two deletions overlapping a distant FOXF1 enhancer unravel the role of lncRNA LINC01081 in etiology of alveolar capillary dysplasia with misalignment of pulmonary veins. Am J Med Genet
A. 2014;164A(8):2013-9. P a g e | 124
36. Dello Russo P, Franzoni A, Baldan F, Puppin C, De Maglio G, Pittini C, et al. A 16q deletion involving FOXF1 enhancer is associated to pulmonary capillary hemangiomatosis.
BMC Med Genet. 2015;16:94.
37. Hoffmann AD, Yang XH, Burnicka-Turek O, Bosman JD, Ren X, Steimle JD, et al. Foxf genes integrate tbx5 and hedgehog pathways in the second heart field for cardiac septation. PLoS
Genet. 2014;10(10):e1004604.
38. Szafranski P, Dharmadhikari AV, Brosens E, Gurha P, Kolodziejska KE, Zhishuo O, et al. Small noncoding differentially methylated copy-number variants, including lncRNA genes, cause a lethal lung developmental disorder. Genome Res. 2013;23(1):23-33.
39. Credendino SC, Lewin N, de Oliveira M, Basu S, D'Andrea B, Amendola E, et al.
Tissue- and Cell Type-Specific Expression of the Long Noncoding RNA Klhl14-AS in Mouse.
Int J Genomics. 2017;2017:9769171.
40. Kalinichenko VV, Gusarova GA, Shin B, Costa RH. The forkhead box F1 transcription factor is expressed in brain and head mesenchyme during mouse embryonic development. Gene
Expr Patterns. 2003;3(2):153-8.
41. Kalinichenko VV, Zhou Y, Bhattacharyya D, Kim W, Shin B, Bambal K, et al.
Haploinsufficiency of the mouse Forkhead Box f1 gene causes defects in gall bladder development. J Biol Chem. 2002;277(14):12369-74.
42. Mahlapuu M, Enerback S, Carlsson P. Haploinsufficiency of the forkhead gene Foxf1, a target for sonic hedgehog signaling, causes lung and foregut malformations. Development.
2001;128(12):2397-406. P a g e | 125
43. Mahlapuu M, Ormestad M, Enerback S, Carlsson P. The forkhead transcription factor
Foxf1 is required for differentiation of extra-embryonic and lateral plate mesoderm.
Development. 2001;128(2):155-66.
44. Kalinichenko VV, Lim L, Stolz DB, Shin B, Rausa FM, Clark J, et al. Defects in pulmonary vasculature and perinatal lung hemorrhage in mice heterozygous null for the
Forkhead Box f1 transcription factor. Dev Biol. 2001;235(2):489-506.
45. Miranda J, Rocha G, Soares H, Vilan A, Brandao O, Guimaraes H. Alveolar Capillary
Dysplasia with Misalignment of Pulmonary Veins (ACD/MPV): A Case Series. Case Rep Crit
Care. 2013;2013:327250.
46. Slot E, Edel G, Cutz E, van Heijst A, Post M, Schnater M, et al. Alveolar capillary dysplasia with misalignment of the pulmonary veins: clinical, histological, and genetic aspects.
Pulm Circ. 2018;8(3):2045894018795143.
47. Ran L, Chen Y, Sher J, Wong EWP, Murphy D, Zhang JQ, et al. FOXF1 Defines the
Core-Regulatory Circuitry in Gastrointestinal Stromal Tumor. Cancer Discov. 2018;8(2):234-51.
48. Milewski D, Pradhan A, Wang X, Cai Y, Le T, Turpin B, et al. FoxF1 and FoxF2 transcription factors synergistically promote rhabdomyosarcoma carcinogenesis by repressing transcription of p21(Cip1) CDK inhibitor. Oncogene. 2017;36(6):850-62.
49. Sugaya M, Takenoyama M, Osaki T, Yasuda M, Nagashima A, Sugio K, et al.
Establishment of 15 cancer cell lines from patients with lung cancer and the potential tools for immunotherapy. Chest. 2002;122(1):282-8.
50. Pandrangi SL, Raju Bagadi SA, Sinha NK, Kumar M, Dada R, Lakhanpal M, et al.
Establishment and characterization of two primary breast cancer cell lines from young Indian breast cancer patients: mutation analysis. Cancer Cell Int. 2014;14(1):14. P a g e | 126
51. Kim MJ, Kim MS, Kim SJ, An S, Park J, Park H, et al. Establishment and characterization of 6 novel patient-derived primary pancreatic ductal adenocarcinoma cell lines from Korean pancreatic cancer patients. Cancer Cell Int. 2017;17:47.
52. Ries LAG MD, Krapcho M, Mariotto A, Miller BA, Feuer EJ, Clegg L, Horner MJ, Howlader N,
Eisner MP, Reichman M, Edwards BK. SEER Cancer Statistics Review, 1975-2004. National Cancer
Institute. Bethesda, MD.: SEER data submission; 2007 [Available from: https://seer.cancer.gov/csr/1975_2004/.
53. Trama A, Botta L, Foschi R, Ferrari A, Stiller C, Desandes E, et al. Survival of European adolescents and young adults diagnosed with cancer in 2000-07: population-based data from
EUROCARE-5. Lancet Oncol. 2016;17(7):896-906.
54. Olguin HC, Pisconti A. Marking the tempo for myogenesis: Pax7 and the regulation of muscle stem cell fate decisions. J Cell Mol Med. 2012;16(5):1013-25.
55. Gryder BE, Yohe ME, Chou HC, Zhang X, Marques J, Wachtel M, et al. PAX3-FOXO1
Establishes Myogenic Super Enhancers and Confers BET Bromodomain Vulnerability. Cancer
Discov. 2017;7(8):884-99.
56. Sun W, Chatterjee B, Wang Y, Stevenson HS, Edelman DC, Meltzer PS, et al. Distinct methylation profiles characterize fusion-positive and fusion-negative rhabdomyosarcoma. Mod
Pathol. 2015;28(9):1214-24.
57. Gilbert LA, Larson MH, Morsut L, Liu Z, Brar GA, Torres SE, et al. CRISPR-mediated modular RNA-guided regulation of transcription in eukaryotes. Cell. 2013;154(2):442-51.
58. Margolin JF, Friedman JR, Meyer WK, Vissing H, Thiesen HJ, Rauscher FJ, 3rd.
Kruppel-associated boxes are potent transcriptional repression domains. Proc Natl Acad Sci U S
A. 1994;91(10):4509-13. P a g e | 127
59. Lam PY, Sublett JE, Hollenbach AD, Roussel MF. The oncogenic potential of the Pax3-
FKHR fusion protein requires the Pax3 homeodomain recognition helix but not the Pax3 paired- box DNA binding domain. Mol Cell Biol. 1999;19(1):594-601.
60. Bulusu KC, Tym JE, Coker EA, Schierz AC, Al-Lazikani B. canSAR: updated cancer research and drug discovery knowledgebase. Nucleic Acids Res. 2014;42(Database issue):D1040-7.
61. Mehner C, Miller E, Khauv D, Nassar A, Oberg AL, Bamlet WR, et al. Tumor cell- derived MMP3 orchestrates Rac1b and tissue alterations that promote pancreatic adenocarcinoma. Mol Cancer Res. 2014;12(10):1430-9.
62. Huang T, Wang L, Liu D, Li P, Xiong H, Zhuang L, et al. FGF7/FGFR2 signal promotes invasion and migration in human gastric cancer through upregulation of thrombospondin-1. Int J
Oncol. 2017;50(5):1501-12.
63. Ebauer M, Wachtel M, Niggli FK, Schafer BW. Comparative expression profiling identifies an in vivo target gene signature with TFAP2B as a mediator of the survival function of
PAX3/FKHR. Oncogene. 2007;26(51):7267-81.
64. Fredericks WJ, Ayyanathan K, Herlyn M, Friedman JR, Rauscher FJ, 3rd. An engineered
PAX3-KRAB transcriptional repressor inhibits the malignant phenotype of alveolar rhabdomyosarcoma cells harboring the endogenous PAX3-FKHR oncogene. Mol Cell Biol.
2000;20(14):5019-31.
65. Bid HK, Phelps DA, Xaio L, Guttridge DC, Lin J, London C, et al. The Bromodomain
BET Inhibitor JQ1 Suppresses Tumor Angiogenesis in Models of Childhood Sarcoma. Mol
Cancer Ther. 2016;15(5):1018-28. P a g e | 128
66. Bolte C, Flood HM, Ren X, Jagannathan S, Barski A, Kalin TV, et al. FOXF1 transcription factor promotes lung regeneration after partial pneumonectomy. Sci Rep.
2017;7(1):10690.
67. Vallabh S, Kartashov AV, Barski A. Analysis of ChIP-Seq and RNA-Seq Data with
BioWardrobe. Methods Mol Biol. 2018;1783:343-60.
68. Kartashov AV, Barski A. BioWardrobe: an integrated platform for analysis of epigenomics and transcriptomics data. Genome Biol. 2015;16:158.
69. Machanick P, Bailey TL. MEME-ChIP: motif analysis of large DNA datasets.
Bioinformatics. 2011;27(12):1696-7.
70. Shao Z, Zhang Y, Yuan GC, Orkin SH, Waxman DJ. MAnorm: a robust model for quantitative comparison of ChIP-Seq data sets. Genome Biol. 2012;13(3):R16.
71. Brugnera E, Haney L, Grimsley C, Lu M, Walk SF, Tosello-Trampont AC, et al.
Unconventional Rac-GEF activity is mediated through the Dock180-ELMO complex. Nat Cell
Biol. 2002;4(8):574-82.
72. Grimsley CM, Kinchen JM, Tosello-Trampont AC, Brugnera E, Haney LB, Lu M, et al.
Dock180 and ELMO1 proteins cooperate to promote evolutionarily conserved Rac-dependent cell migration. J Biol Chem. 2004;279(7):6087-97.
73. Davicioni E, Finckenstein FG, Shahbazian V, Buckley JD, Triche TJ, Anderson MJ.
Identification of a PAX-FKHR gene expression signature that defines molecular classes and determines the prognosis of alveolar rhabdomyosarcomas. Cancer Res. 2006;66(14):6936-46.
74. Rapa E, Hill SK, Morten KJ, Potter M, Mitchell C. The over-expression of cell migratory genes in alveolar rhabdomyosarcoma could contribute to metastatic spread. Clin Exp Metastasis.
2012;29(5):419-29. P a g e | 129
75. Cairns J, Fridley BL, Jenkins GD, Zhuang Y, Yu J, Wang L. Differential roles of
ERRFI1 in EGFR and AKT pathway regulation affect cancer proliferation. EMBO Rep.
2018;19(3).
76. Yemelyanova A, Vang R, Kshirsagar M, Lu D, Marks MA, Shih Ie M, et al.
Immunohistochemical staining patterns of p53 can serve as a surrogate marker for TP53 mutations in ovarian carcinoma: an immunohistochemical and nucleotide sequencing analysis.
Mod Pathol. 2011;24(9):1248-53.
77. Iggo R, Gatter K, Bartek J, Lane D, Harris AL. Increased expression of mutant forms of p53 oncogene in primary lung cancer. Lancet. 1990;335(8691):675-9.
78. Nabeshima Y, Hanaoka K, Hayasaka M, Esumi E, Li S, Nonaka I, et al. Myogenin gene disruption results in perinatal lethality because of severe muscle defect. Nature.
1993;364(6437):532-5.
79. Rawls A, Morris JH, Rudnicki M, Braun T, Arnold HH, Klein WH, et al. Myogenin's functions do not overlap with those of MyoD or Myf-5 during mouse embryogenesis. Dev Biol.
1995;172(1):37-50.
80. Hasty P, Bradley A, Morris JH, Edmondson DG, Venuti JM, Olson EN, et al. Muscle deficiency and neonatal death in mice with a targeted mutation in the myogenin gene. Nature.
1993;364(6437):501-6.
81. Rudnicki MA, Braun T, Hinuma S, Jaenisch R. Inactivation of MyoD in mice leads to up-regulation of the myogenic HLH gene Myf-5 and results in apparently normal muscle development. Cell. 1992;71(3):383-90. P a g e | 130
82. Vicente-Garcia C, Villarejo-Balcells B, Irastorza-Azcarate I, Naranjo S, Acemel RD,
Tena JJ, et al. Regulatory landscape fusion in rhabdomyosarcoma through interactions between the PAX3 promoter and FOXO1 regulatory elements. Genome Biol. 2017;18(1):106.
83. Xu J, Liu H, Lan Y, Aronow BJ, Kalinichenko VV, Jiang R. A Shh-Foxf-Fgf18-Shh
Molecular Circuit Regulating Palate Development. PLoS Genet. 2016;12(1):e1005769.