USP22 controls multiple signaling pathways that are essential for vasculature formation in the mouse placenta

Development of the mouse placenta requires specification of the trophoblast lineage at early stages of embryogenesis. The trophoblast cells that overlie the inner cell mass form the extra-embryonic ectoderm, and those away from the inner cell mass form the ectoplacental cone, immediately after implantation (Simmons et al., 2007). The extra-embryonic ectoderm gives rise to the chorionic epithelium, which will make contact with the extra-embryonic mesoderm (allantois) and eventually give rise to the labyrinth layer. Chorion-derived cells form trophoblasts and allantois-derived cells form endothelial cells. These two cell types undergo extensive intertwining and branching to create the labyrinth that facilitates nutrient exchange between mother and fetus. Defects in this process cause embryonic lethality between embryonic day (E) 9.5 and E13.5 of development (Rossant and Cross, 2001; Watson and Cross, 2005). Not surprisingly, these events require exquisite spatial and temporal control of gene expression programs, mediated by complex interactions between genetic and epigenetic factors, as mutations in these factors have detrimental effects among multiple placental layers (Hemberger, 2010). However, the molecular mechanisms that are responsible for many developmental events in the growing placenta are not well defined, as the placenta is still commonly overlooked in the phenotypic characterization of mice that bear embryonic lethal mutations, leading to an under-representation of the genes that are involved in regulating this pivotal organ (Perez-Garcia et al., 2018). Nonetheless, mouse genetic approaches have provided some important mechanistic insights to placental development, as mutations in several signaling molecules, transcription factors and epigenetic regulators cause lethality due to placental defects (Rossant and Cross, 2001; Watson and Cross, 2005).

The highly conserved SAGA complex plays important roles at multiple steps of gene transcription. SAGA houses two enzymatic activities, the Gcn5 histone acetyltransferase and the USP22 deubiquitinase (Koutelou et al., 2010), which alter histone acetylation and ubiquitination (also known as ubiquitylation) levels to regulate the expression of multiple genes (Bonnet et al., 2014; Hirsch et al., 2015). Genome-wide analyses in yeast revealed multiple gene-specific regulators that associate with SAGA, and a high dynamic range of transcripts that are regulated by SAGA, which provides evidence that SAGA is involved in cellular response events that require highly efficient transcript formation (Venters et al., 2011; Vinayachandran et al., 2018). In mice, global loss of Gcn5 causes loss of mesodermal tissues in early embryogenesis (Xu et al., 2000). Lineage-specific Gcn5 deletions have revealed its roles in neural stem cell differentiation in vivo and in vitro (Martinez-Cerdeno et al., 2012). Gcn5 also regulates retinoic acid signaling in the developing mouse diencephalon (Wilde et al., 2017) and affects multiple components of FGF signaling during embryoid body differentiation (Wang et al., 2018).

In contrast to Gcn5, the functions of USP22 during development are poorly defined. Usp22 was first described as a member of an 11-gene ‘death from cancer’ signature that was defined by gene expression microarrays (Glinsky et al., 2005). Overexpression of USP22 has since been observed by several groups in multiple cancer types (reviewed by Wang and Dent, 2014), but no clear picture has yet emerged for how this deubiquitinase might contribute to oncogenesis. The biochemical activity of USP22 against ubiquitinated histones H2B or H2A is well characterized, and USP22 also has non-histone substrates, including TRF1 (also known as TERF1) (Atanassov et al., 2009), FBP1 (Atanassov and Dent, 2011) and SIRT1 (Lin et al., 2012). When and where USP22 activity is required to regulate these or other substrates in vivo is still unclear. Usp22 deletion causes embryonic lethality in mice (Lin et al., 2012), whereas mice that harbor a hypomorphic allele of Usp22 are viable but exhibit a reduced body size (Kosinsky et al., 2015). However, the molecular basis of these phenotypes is not well defined.

To gain more insights into USP22 functions, we performed a detailed analysis of the cause of death of Usp22 null embryos. Our findings reveal that USP22 plays important roles in placenta development that are tied to multiple signaling pathways driven by TGFβ and several receptor tyrosine kinases, including VEGFR (KDR), HGFR (MET) and PDGFR (PDGFRB).

We took advantage of a β-galactosidase marker driven by the Usp22 promoter in a gene-trap allele (www.genetrap.org) (Fig. 1A and Fig. S1) to define Usp22 expression patterns within the embryo. At E6.5, Usp22 is expressed in the anterior epiblast (presumptive neuroectoderm), as well as in the posterior epiblast. A second site of expression initiates in the ectoplacental cone and is maintained in extra-embryonic ectoderm and mesoderm at E7.25. At E9, Usp22 mRNA is present throughout the brain, neural tube, heart, allantois and chorionic plate. At later stages, Usp22 expression is gradually restricted to the developing forebrain and midbrain, to cephalic and dorsal root ganglia, as well as to internal organs including the heart, testis, intestine and ribs (Fig. S2).

The Usp22 expression pattern indicates potential functions in extra-embryonic tissues early in development. Close examination of Usp22-deficient embryos revealed no abnormalities in postimplantation events, including gastrulation and neurulation. Mutant embryos survived until 11.5-14.5 days postcoitum, but died shortly thereafter, with extensive hemorrhages and signs of potential hypoxia-induced stress that are consistent with severe defects in extra-embryonic tissues (Gnarra et al., 1997) (Fig. 1B,C). Hypoxic stress response was confirmed upon RNA-seq of total RNA collected from embryos at E10.5, as several genes (Adm, Ndrg1, Car9, Egln3, P4ha1, Ier3, Pfkp) in this pathway were upregulated upon USP22 loss (Majmundar et al., 2010) (Fig. S3). These same homozygous mutant phenotypes were obtained using two independent Usp22 gene-targeting technologies (Figs S1,S4,S5) and were consistent across different genetic backgrounds (Fig. 1C and Fig. S6).

Given that Usp22 loss-of-function induces hypoxic stress in embryos and Usp22 is expressed in extra-embryonic tissues, we hypothesized that disruption of Usp22 may have deleterious consequences in the development of the placenta. The mouse placental labyrinth is the site of oxygen and nutrient exchange between the mother and the fetus, and from E11 onward it is the primary site of nutrient transport. Defects occurring in the labyrinth layer of Usp22 mutants could limit the transport of nutrients to the embryo, ultimately resulting in lethality. Therefore, we next analyzed Hematoxylin and Eosin (H&E)-stained sections that were obtained from wild-type and Usp22-deficient placentas. At embryonic day E14.5, when most Usp22 mutant embryos have died, the three layers of the placenta, including the labyrinth, the spongiotrophoblasts and the trophoblast giant cells in the junctional zone, and the maternal decidua were easily discernible (Fig. 1D) in both wild type and mutants. The trophoblast giant cell [marked with anti-proliferin (Prl2c2)] and spongiotrophoblast layers (marked with anti-TPBPA) appeared to be unaffected by Usp22 loss (Fig. 1D). However, the labyrinth layer of the Usp22 null placentas was abnormal. Examination of earlier time points (E12.5) confirmed that the cellular density of the labyrinth epithelium was decreased and the network of fetal vessels and maternal sinuses were dilated and poorly developed in the Usp22 mutants. H&E staining followed by computational morphometry also indicated that the number of fetal vessels in the labyrinth of Usp22 null embryos was significantly reduced compared with wild-type littermates (Fig. S7A,B, upper panels).

To determine whether the defects in Usp22 mutant placentas were restricted to a specific cell type, we analyzed expression of cell and lineage specific markers. PECAM1 staining, which is specific for endothelial cells, again indicated a striking defect in the formation of the vasculature of the labyrinth in the mutant (Fig. 1Ε, upper panel). The endothelial cells did not form normal tubular vessel structures, but instead clustered in unconnected, unstructured vessel-like formations. MCT4 (Slc16a3), a marker for syncytiotrophoblast (SynT)-II cells that exhibits a sharp, cell-surface expression in wild-type placentas (Moreau et al., 2014; Nagai et al., 2010), showed broad expression throughout the aberrant vessels in the Usp22 mutants (Fig. 1Ε, middle panel), which indicates misorganization of SynT-II cells. Smooth-muscle actin (ACTA2), which is normally expressed in mural cells that surround the fetal endothelial cells, was also misorganized in the mutants (Fig. 1E, lower panel). Altogether these findings indicate that USP22 is required for the functional interaction of the different cell types (endothelial cells, mural cells, syncytiotrophoblasts) that form the labyrinth fetal vessels.

To more precisely determine the primary cause of embryonic lethality in the Usp22 mutants, we defined the timing of death. Dead embryos were first observed at E10.5, and no live embryos were recovered past day E13.5-E14.5 (Fig. 1B). This time window, along with the phenotypes of the mutant embryos described above (Fig. 1C and Fig. S6), might be related to defects in cardiovascular development within the embryo or to defects in the maternal-embryonic circulatory system within the placenta. Therefore we examined expression of specific markers and morphological changes that are associated with such defects, focusing first on hallmark events within the placenta.

Placental development is marked by three stepwise events, beginning with the chorioallantoic fusion at E8.5. This process involves extension of the allantoic mesoderm from the embryo towards the chorionic plate to enable fusion. The interaction of adhesion molecules such as VCAM1 (Gurtner et al., 1995; Kwee et al., 1995) and its receptor α4 integrin (Yang et al., 1995) are essential for completion of fusion. Next, fetal blood vessels that are derived from the allantoic mesoderm and trophoblasts in the chorion form branches at E9, a process that depends on transcription factor Gcm1 (Anson-Cartwright et al., 2000). Finally, elaborate branching forms a functional labyrinth, in which fetal vessels are intertwined with trophoblasts in order to absorb nutrients and gases from the sinuses of maternal blood. Signaling cascades triggered by FGF family members, EGF and HGF are crucial for the regulation of these cellular interactions (Cross et al., 2003; Rossant and Cross, 2001).

To determine whether the labyrinth defects we observe in Usp22 mutant placentas result from deregulation of the above events, we collected embryos at E8.5-E9 and carefully observed the structure of the allantois fusing into the chorionic plate. We did not detect any obvious morphological differences in this process between wild-type and Usp22 mutant embryos (Fig. 2A). Evaluation of VCAM1 expression patterns in wild-type and Usp22 mutant placentas revealed normal fusion and protrusion of allantoic mesoderm in the chorionic plate at multiple branching points in both mutant and wild-type placentas (Fig. 2A). Further analysis of placenta architecture at E9 by staining of trophoblast giant cells (marked with anti-proliferin) and spongiotrophoblast layers (marked with anti-TPBPA) again indicated both layers were well formed in wild-type and Usp22 mutant placentas. Finally, no marked changes in either proliferation (as indicated by KI67 expression) or cell death rates (as indicated by TUNEL assay) were observed in E9 placental sections (Fig. 2B).

We next examined the cardiovascular system of the embryo, primarily focusing on development of the heart. Poor cardiac function or circulatory obstruction usually leads to edema, providing a good indicator of underlying heart failure (Conway et al., 2003). No edema was observed in the mutant embryos at E11.5-E12.5, which indicates normal heart function (Fig. 1C and Fig. S6). Moreover, no morphological differences were observed between wild-type and mutant embryos at E9-E10, and heartbeat and blood flow were evident in both genotypes, with no evidence of pericardial edema (Fig. 2C). As heart looping is normally completed, with formation of four heart chambers, by E10, we examined myocardium (by phalloidin staining) and endocardium (by staining for PECAM1) formation in wild-type and Usp22 mutant hearts at E11. We observed normal formation of all four heart chambers (right and left ventricle, right and left atrium) as well as the interventricular septum (Fig. 2D) in all embryos. Between E9 and E10 the myocardial tube is lined by the inner endothelium and outer myocardium, from which the trabeculation is formed. We examined ventricular myocardial trabeculation by multiphoton imaging of wild-type and Usp22 mutant hearts in whole embryos stained with phalloidin to visualize the myocardium and anti-PECAM1 to visualize the endocardium (Fig. S8) at E10. No abnormalities were observed in the organization of ventricular wall cardiomyocytes (myocardium) or endocardial cells in the Usp22 mutant hearts.

The strong expression of Usp22 in the developing allantois and the chorion in wild-type embryos, in combination with the absence of early functional heart defects (normal heart beat, blood flow, lack of edemas) in the mutants, indicates an essential role of Usp22 function in establishing embryonic-maternal circulation.

To further define the origin of the defects in the Usp22 mutants, we examined expression of trophoblast lineage differentiation markers at earlier time points. At E11, no reduction in expression of genes that are specific for trophoblast giant cells (Prl2c2, Prl3d1) or spongiotrophoblasts (Tpbpa, Prdm1) was detected. However, reduced expression of genes that are specific for labyrinth trophoblast cells, specifically syncytiotrophoblasts (Mct4, Cebpa), as well as for endothelial cells (Pecam1) and sinusoidal trophoblast giant cells (Ctsq), was observed (Fig. 3A), which indicates defective differentiation and/or specification of several cell lineages in Usp22 null placentas.

Usp22 depletion in cultured cells has been associated with G1/M cell cycle arrest (Gennaro et al., 2018; Zhang et al., 2008b), proliferation defects (Atanassov and Dent, 2011), p53-dependent apoptosis (Lin et al., 2012) and blockage of stem cell differentiation (Sussman et al., 2013). Although no defects in proliferation or increased cell death were observed at E9.5 (Fig. 2B), we further examined cell cycle progression in the mutants by examining expression of several marker genes (Ccna2, Ccne1, Ccne2, Cdkn1c) at E11. Minimal changes in expression of these genes were detected in the mutants at this stage (Fig. 3B). Surprisingly, apoptotic markers such as Cdkn1a and Puma (Bbc3) were downregulated, rather than upregulated, in the mutant placentas, and expression of anti-apoptotic markers such as Bcl2 and Bcl6 were increased (Fig. 3B). These findings indicate that the placental abnormalities observed at E11 are not due to either decreased proliferation or increased cell death. At later times, after E12.5, a significant reduction in dividing labyrinth trophoblasts and an increase in apoptosis specifically in fetal vessels was observed (Fig. S7). These effects are likely secondary to the abnormal organization of fetal vessels that we observed at earlier stages (E11).

Interestingly, expression of the genes that encode several factors that induce vascular formation in the labyrinth, including Vegfa, Vegfr3 (Flt4), which is also expressed in lymphatic endothelial cells, Hif2a (Epas1), Hif3a and Adm were downregulated in the Usp22 mutant placentas at E11 (Fig. 3C). In contrast, histological staining for representative markers of hypoxic stress (HIF2A, VEGFA) showed no obvious difference between wild-type and mutant placentas (Fig. 3D), which suggests that the hypoxic stress response was not yet induced at this early time point, consistent with a lack of induction of Hif1a or Hif1b (Arnt) RNA expression (Fig. 3C). Expression of Cited2, a major transcriptional regulator of labyrinth vasculature that also responds to induced hypoxia, was similarly unaltered (Moreau et al., 2014; Withington et al., 2006). Altogether, these results indicate that labyrinth development initiates within the placentas of Usp22 mutants, but development is halted or delayed at a subsequent step, resulting in undervascularization.

To define the molecular mechanisms that underlie the defects in labyrinth vasculogenesis in Usp22 mutants, we isolated RNA from extra-embryonic structures of wild-type and mutant placentas at E9.5 to capture early events and performed RNA sequencing (RNA-seq). We identified 1158 genes as upregulated and 1035 genes as downregulated in the Usp22-deficient placentas [P<0.01, false discovery rate (FDR)<0.05, fold change≥2].

Gene ontology (GO) terms for physiological functions and diseases enriched for genes that were strongly downregulated in the Usp22 mutant placentas included processes that are related to vessel formation (vascular lesion, angiogenesis, vasculogenesis) and cancer (advanced malignant tumor, metastasis, angiogenesis of tumor) (Fig. 4A). In contrast, processes that are related to concentration and transport of lipids, as well as to the quantity of cholesterol in the blood, were consistently upregulated (Fig. 4A). The GO terms for molecular and cellular functions enriched for the downregulated genes highlighted processes including cellular movement, death, invasion, adhesion and formation of cell protrusions, which are all highly consistent with vessel formation failure (Fig. 4B). The GO terms for quantity of myeloid cells and for gene expression were upregulated. Ingenuity Pathway Analysis (IPA) software (Qiagen; www.ingenuity.com/products/ipa) was used for the in-depth processing of the RNA-seq data, and the above-mentioned GO categories were verified by two more GO analysis software packages (DAVID v6.7, PANTHER v11.0) (Fig. 4C,D).

Upstream regulator analysis defined by IPA and Gene Set Enrichment Analysis (GSEA) indicated inhibition of a number of signaling pathways in the mutants (Fig. 4E,F), with the TGFβ1 pathway being the most affected (z-score: −2.900, P=4.9E-37) (Figs 4E and 5A). Tgfb1 is required for yolk sac hematopoiesis and endothelial differentiation in mice (Dickson et al., 1995). Similar to Usp22 mutants, defects in vasculogenesis in Tgfb1 mutants affect endothelial cell differentiation, rather than specification (Dickson et al., 1995). TGFβ1 receptors are also required for proper vasculogenesis (Larsson et al., 2001; Oshima et al., 1996). Expression of Tgfb1 itself is not affected in Usp22 mutants, but Tgfbr1 was significantly downregulated, as were Tgfb1 downstream transducers including Smad2 and Smad3 (Fig. 5B). Expression of Smad1 and Smad5, the two receptor-regulated Smads that do not respond to Tgfbr1, was not affected (data not shown).

IPA analysis of upstream regulators also indicated inhibition of several receptor tyrosine kinase (RTK) pathways in the Usp22 mutants, including those that are responsive to Erbb2 (z-score: −3.369, P=1.9E-21), EGF (z-score: −4.616, P=5.2E-16), HGF (z-score: −2.051, P=1.23E-10) and PDGF (z-score: −4.560, P=3.9E-09) (Figs 4E and 5C; Tables S1-S5). Several of these receptor tyrosine kinases, including HGFR, EGFR and PDGFR (Dackor et al., 2009; Ohlsson et al., 1999; Ueno et al., 2013), are essential for proper development of the trophoblast compartment of the labyrinth (Lindahl et al., 1997; Uehara et al., 1995). These signaling pathways mediate cellular interactions that underlie branching morphogenesis. Importantly, the block in labyrinth development that is caused by mutations in these proteins occurs after the initiation of chorioallantoic morphogenesis, consistent with the Usp22 mutant labyrinth phenotype. Moreover, downregulation of multiple cytoplasmic transducers of these RTKs, including Gab1, Gab2, Grb2, Akt1 and Pdpk1, was observed in the Usp22 mutants, which provides an explanation for the inhibition of multiple RTK pathways (Fig. 5D).

Consistent with a role for USP22 in regulating labyrinth development, Usp22 (as indicated by β-galactosidase staining) is expressed in E9.5 placentas in the labyrinth trophoblasts of the chorion and in the mesenchymal cells of the allantois (Fig. 6A). In contrast, Usp22 expression is low in the trophoblast giant cell layer and in the spongiotrophoblast layers, which were not affected in the mutants at this stage, as confirmed by histological analysis and by analysis of expression of genes that are specific for these layers (data not shown).

To further connect the defects we observed in fetal vessel formation with defective TGFβ1 and RTK signaling in Usp22 mutant placentas, particularly the disorganization of endothelial cells and labyrinth trophoblast loss, we examined expression of specific markers for vascular formation that are responsive to these pathways at E9.5. Examination of sections of Usp22 mutant placentas at E9.5 indicates that embryonic vessels are unable to sprout properly into the labyrinth layer (Fig. 6B). Endothelial cells normally migrate towards angiogenic signals and then multiply and form a migration column. Migration columns lead to endothelial cell differentiation, with changes in cell shape and adhesion to form a lumen. PECAM1 and vimentin staining suggest that the endothelial cells in Usp22 mutant placentas that originate from the allantois do not migrate into the chorionic epithelium as efficiently as the wild-type cells, which results in the failure of lumen formation (Fig. 6B, upper and middle panel). The chorionic epithelium undergoes branching morphogenesis by changing the shape of cells in the basal layer that express the transcription factor Gcm1. These cells exit the cell cycle and begin to involute to create the primary villi into which fetal vessels will grow. Staining for β-catenin (Ctnnb1), which defines the chorionic trophoblasts, indicates decreased formation of villi in the Usp22 mutant labyrinths (Fig. 6B, lower panel).

Mutations in either Gcm1 or Cebpa cause delayed labyrinth development (Simmons et al., 2008). Gcm1, Cd9 and Cebpa are all downstream targets of HGF signaling in the labyrinth trophoblasts, and these genes were all downregulated in the Usp22 mutant (Fig. 6C, upper panel). Labyrinth trophoblast cells originate from the chorion, and their mitotic activity is stimulated by HGF after contact with the allantoic mesenchyme at ∼E10 (Ueno et al., 2013). Decreased mitotic activity of the trophoblasts in the labyrinth is also consistent with attenuation of HGF signaling that is observed in Usp22 mutant placentas, as verified by decreased expression of genes that are specific for both SynT-I [Syna, Mct1 (Slc16a1)] and SynT-II (Synb) (Fig. 6C, middle panel). Mct4 is the only syncytiotrophoblast marker to be upregulated in the mutant, consistent with Mct4 induction by hypoxic stress (Kay et al., 2007). The deregulation of HGF signaling was further verified by measuring expression of additional SynT markers (Dlx3, Ovol2, Nr6a1), as well as cell polarity markers such as Crb3 (Fig. 6C, lower panel).

Increased expression of endothelial markers (Pecam1) and decreased expression of markers of mural cells (Acta2) was also observed in the Usp22 mutant placentas, consistent with an imbalance of PDGF and TGFβ signaling (Fig. 6D, upper panel). PDGF receptors play a broad role in vascular development of yolk sac and placenta, as well as other tissues. Embryonic lethality in PDGF mutants is due to yolk sac and placental defects (Hellstrom et al., 1999), specifically in mural cells (French et al., 2008). Endothelial cells and mural cells of the placenta both express Tgfbr1 (Larsson et al., 2001), and embryonic lethality at ∼E9.5 of Tgfbr1 null embryos is attributed to a vascular defect both in the yolk sac and in the placenta. Nonetheless, both Usp22 and Tgfbr1-deficient placentas completed initial steps of branching morphogenesis, but vessels failed to sprout, which indicates defective angiogenesis. We also observe a migratory defect of the endothelial cells of the allantois, probably due to changes in expression of Mmp9, Plau, Lama3 and Cfl1 (Fig. 6D, lower panel). Altogether, these observations indicate that the Usp22 mutant phenotype likely reflects an accumulation of defects in multiple signaling pathways, not only that of TGFβ. The migration defects we observe in Usp22 mutant endothelial cells inside the labyrinth layer are consistent with defective TGFβ signaling, as indicated by our RNA-seq data, whereas the defects in mural cells and chorionic trophoblast cells likely result from deregulation of HGF and PDGF signaling.

The vascular defects we observe in the Usp22 mutants establish an intriguing role for USP22 in the functional organization of endothelial cells, pericytes and labyrinth trophoblasts. Endothelial cells and pericytes are both crucial for the formation and function of the developing vasculature, but as our mutant allele affects USP22 expression in both tissues, we could not discriminate whether USP22 functions are most important in one or both of these cell types. To address this issue, we took an in vitro approach, differentiating wild-type and Usp22 null mouse embryonic stem cells (ESCs) towards a mesendodermal fate, gradually inducing differentiation towards vascular progenitors, and then finally to endothelial cells and pericytes (Fig. 7A and Fig. S9A). Generation of endothelial cells and pericytes was confirmed both by morphological observation (Fig. 7B,D and Fig. S9B) and by immunostaining with endothelial- [VEGFR2 (KDR), PECAM1] and pericyte-specific markers [NG2 (CSPG4), ACTA2, PDGFRβ] (Fig. 7C,E and Fig. S11A,B).

Both wild-type and Usp22^−/− early passage (passage 2-3) and low confluency endothelial cells and pericytes proliferated well. Closer observation of Usp22^−/− endothelial cells showed that VEGFR2 and PECAM1 markers were expressed, but were mislocalized to the cytoplasm rather than the plasma membrane (Fig. 7C and Fig. S10A). At passage 5, the proliferative capacity of the mutant endothelial cells was reduced, and increased numbers of cells in G1 were observed (Fig. S10B,C). In contrast, Usp22^−/− pericytes expressed NG2 similarly to wild-type cells but were more proliferative, and the cell population clearly contained a higher proportion of progenitor cells that did not express NG2. These findings suggest that loss of Usp22 limits pericyte differentiation, as fewer cells make it to mature pericytes, but those that do, express NG2 normally. Quantification of highly proliferative cells showed a significantly larger number of H3S10ph-positive cells in the Usp22^−/− pericytes compared with the wild-type cells, but analysis of DNA content indicated no difference in overall cell cycle progression (Fig. S10E,F). These results indicate that USP22 loss affects endothelial cell propagation but not specification. In contrast, USP22 deletion strongly impacts pericyte differentiation. Boyden chamber assays adapted for endothelial cells revealed that USP22 mutant induced endothelial cells have a dramatic decrease in migratory capacity relative to wild-type cells (Fig. S11C,D), which is consistent with our observation of a migratory defect of endothelial cells in USP22 mutant labyrinths (Fig. 6B).

Pericytes and endothelial cells work together to form vessels, so we next performed a tube formation assay to assess functionality of these cell types. We co-cultured different combinations of wild-type or Usp22^−/− endothelial cells and pericytes at early passages (passage 2-3). As expected, the combination of wild-type endothelial cells with wild-type pericytes led to formation of endothelial sprouts, which were visualized by immunostaining with VEGFR2-specific antibodies, which clearly outlined the formation of tubes (Fig. 7F). However, no combination that included either Usp22^−/− endothelial cells or pericytes produced these structures, which indicates loss of USP22 in either or both cell types severely limits functional interactions between endothelial cells and pericytes, thereby crippling vessel formation.

The phenotypes observed above may reflect deregulation of signal integration by endothelial cells and pericytes in the absence of Usp22, as was observed in the mutant placentas. Specifically, deregulation of VEGF signaling in induced endothelial cells would not inhibit their specification, but rather would cause a propagation defect (Schuh et al., 1999), as we observed in the Usp22 mutant endothelial cells. Deregulation of PDGF and TGFβ signaling in vascular progenitors is known to hinder their capacity to differentiate into pericytes (Hirschi et al., 1998), again consistent with the phenotype of Usp22 null progenitors.

Assessment of both RTK and TGFβ signaling intermediates in Usp22 null endothelial cells revealed diminished levels of phosphorylated ERK1/2 and AKT in the Usp22^−/−-induced endothelial cells, as well as phosphorylated cofilin (Fig. 8A), which is consistent with our observed effects on endothelial cell proliferation, survival and shape. Levels of signaling regulators and intermediates that were identified to be transcriptionally downregulated in Usp22^−/− placentas were also strongly downregulated in the Usp22^−/− endothelial cells, including Gab1, Gab2, PKCα (PRKCA), PLCγ1 and Smad2/3 (Fig. 8B). Interestingly, the levels of VEGFR2 and HGFR receptors were also affected in the Usp22^−/−-induced endothelial cells (Fig. 8C). Our results identify USP22 as an indispensable regulator of signal integration in developing endothelial cells and pericytes.

Our findings reveal functions for USP22 in the development of the mouse placenta through effects on crucial signaling pathways. The requirement of USP22 for proper execution of multiple different signaling pathways likely explains the severity of the placental defects we observe in the Usp22 mutants and may provide new insights to the highly aggressive nature of tumors that overexpress USP22.

The effects of USP22 loss on VEGFA signaling is especially intriguing as this pathway transmits signals that are important for proliferation, migration and differentiation of vascular progenitors. VEGFA predominantly binds two receptors, VEGFR1 and VEGFR2. VEGFR2-deficient mice lack endothelial and hematopoietic cells (Shalaby et al., 1995) and die between E8.5 and E9.5 (Shalaby et al., 1997). Nevertheless, VEGFR2-deficient cells differentiate into hematopoietic and endothelial cells in culture, suggesting that VEGFR2 is required mostly for the subsequent migration and propagation of the specified cells (Schuh et al., 1999).

We used differentiated ESCs to probe the role of USP22 in the signaling cascades that are activated during lineage specification of VEGFR2-positive mesodermal progenitor cells because migration is not necessary in this context (Hirashima et al., 1999; Orlova et al., 2014; Yamashita et al., 2000). We clearly observed that Usp22 loss does not prevent vascular progenitor differentiation towards endothelial cells, as endothelial-specific markers such as PECAM1 and VE-cadherin (Cdh5) are expressed in the mutant induced endothelial cells. However, these factors exhibit abnormal localization away from the cytoplasmic membrane, and VEGFR2 itself is mislocalized and reduced in levels. The stability and internalization of VEGFR2 could be affected directly by USP22-dependent deubiquitination or indirectly by mislocalization of VE-cadherin (Lampugnani et al., 2006). Several USP deubiquitinating enzymes have been directly implicated in transmembrane receptor deubiquitination (Berthouze et al., 2009; Zhang et al., 2012). Our future work will address whether USP22 also plays such a role.

Our observation of strong decreases in phosphorylation of AKT and ERK1/2 in Usp22 null induced endothelial cells further indicates defective execution of VEGF-driven pathways. VEGFR2 induces proliferation of endothelial cells through the activation of ERK1/2 (Shu et al., 2002) and regulates cell survival through inhibition of pro-apoptotic pathways via the activation of PI3K/AKT phosphorylation (Gerber et al., 1998). A more detailed molecular analysis of the induced endothelial cells gave us further insight in the mechanistic role USP22 plays, as we found that several crucial signaling cascade effectors such as Gab1, Gab2, PKCα, PLCγ1 and PDPK1 were greatly downregulated in the absence of USP22 (model in Fig. 8D). Most of these molecules were also identified in the genome-wide transcriptional profiling of the placenta that we performed, which indicates that these factors are likely directly regulated by USP22, either transcriptionally or post-transcriptionally. USP22 has both histone and non-histone substrates, and it also functions in non-transcription related processes (Atanassov and Dent, 2011; Atanassov et al., 2009). The phenotypes we observe here, then, may reflect a combination of gene expression changes and changes in levels or activities of non-histone proteins that are regulated directly by USP22-mediated deubiquitination. Definition of such changes will require additional study.

Examination of monoubiquitination of histone H2B (H2Bub1) in Usp22 null endothelial cells revealed a decrease, rather than the expected increase, in the levels of ubiquitinated H2B relative to wild-type cells (Fig. S12). This counterintuitive result has been observed before (Atanassov et al., 2016), and it reflects redundancies and cross-regulation of two other USPs that are highly related to USP22: USP27x and USP51. USP27x and USP51 do not integrate into the SAGA complex; the phenotype of our Usp22 null mice indicate that these deubiquitinating enzymes (DUBs) cannot compensate for the loss of USP22 in the placenta. However, it is possible that USP22 loss allows greater association of Eny2 and Atxn7L3 with the other two DUBs, leading to a gain-of -function for USP27x or USP51. Additional studies will be required to define the relevant substrates of USP22 and the relative roles of these three USPs in RTK and TGFβ signaling.

A few previous studies have identified putative links between USP22 and signaling pathways in cancer cells. For example, USP22 was reported to play a role in proliferation of ovarian cancer cells that was correlated with regulation of TGFβ1 function (Ji et al., 2015). In nasopharyngeal carcinoma cells, USP22 functions in cell cycle progression were linked to modulation of the AKT/GSK3b/cyclin pathway (Zhuang et al., 2015). Interestingly, USP22 overexpression in lung adenocarcinoma invasion was correlated with upregulation of TGFβ1 and induction of epithelial-to-mesenchymal transitions (Hu et al., 2015). However, in another setting, diabetic nephropathy, USP22 overexpression in primary glomerular mesangial cells was linked to inhibition of TGFβ1 and fibronectin (Huang et al., 2015).

Unfortunately, studies to date have yielded a confusing amalgam of possible roles for USP22 in oncogenesis. Several studies have focused on potential roles for USP22 in regulating cell cycle progression through effects on oncogenes such as BMI1 (Liu et al., 2012), c-MYC (MYC) (Zhang et al., 2008a), SIRT1 (Lin et al., 2012), cyclin B1 (Lin et al., 2015), cyclin D1 (Gennaro et al., 2018) and KDM1A (Zhou et al., 2016), or on tumor suppressors such as p53 (Trp53) (Ding et al., 2015; Lin et al., 2012; Lv et al., 2011). The previously defined 11-gene ‘death from cancer’ signature that includes USP22 also includes BMI1 and downstream targets of this polycomb group protein. BMI1 overexpression was proposed to be the driving force that underlies these cancers, as BMI1 is required for self-renewal and it confers a cancer stem cell-like phenotype when overexpressed in both hematopoietic and epithelial tumors (Lessard and Sauvageau, 2003; Molofsky et al., 2003; Park et al., 2003). Our data do not exclude the possibility that USP22 might also contribute to a cancer stem cell-like phenotype when overexpressed in cancers together with BMI1. However, our results also indicate that USP22 overexpression likely has effects that are independent of BMI1.

USP22 overexpression in many different types of aggressive cancers makes it an attractive target for therapy development and, as such, it is essential to understand the full range of USP22 functions in both normal cells and in tumors. Genetic studies, as presented here, can provide unbiased definitions of the roles of USP22 in both developmental and disease contexts. Future work in mice will determine whether USP22 overexpression on its own is sufficient to drive or facilitate tumor formation or progression. Future analyses of human placentas are also needed to determine whether USP22 loss contributes to diseases such as preeclampsia or intrauterine growth restriction.

Animals were kept in regulated facilities, monitored daily and all procedures that involved animal handling were performed in accordance with the approved Institutional Animal Care and Use Committee (IACUC) protocols at the University of Texas MD Anderson Cancer Center.

We used two gene-trap ESC lines (RRS377 and XP0208) made by the Gene-Trap Consortium (www.genetrap.org). In both lines, the gene trap cassette was integrated within the first intron of Usp22, as validated by Southern blot analyses (Fig. S1A,B,D,E). We further mapped the exact integration site in RRS377 line to 428 bp downstream of exon 1 (Fig. S1C), whereas the integration site in XP0208 was 4075 bp downstream of exon 1 (Fig. S1F). We injected both lines into mouse blastocysts to generate mice bearing the gene trap alleles. Aggregation chimeras were generated with 129Ola ESCs using wild-type C57BL/6 morulae as previously described (Nagy et al., 1993). Two founder male chimeras with 80% or greater chimerism were backcrossed to wild-type C57BL6 females to produce progeny that were heterozygous for the βgeo allele. Heterozygous mice were intercrossed to produce homozygotes. Among 687 live pups born to mating between heterozygous parents from line RRS377, zero homozygous mutants (Usp22^−/−) were found (Fig. S4A). Among 208 pups born to mating between heterozygous parents from line XP0208, only 15 homozygous mutants were observed, significantly less than expected from Mendelian ratios (Fig. S4D). As Usp22 transcript (Fig. S4B) and protein (Fig. S4C) were undetectable in tissues from these embryos, the Usp22 RRS377 allele is likely a null allele and will be referred to here as Usp22^−/− for simplicity. On the contrary, homozygous mutant pups bearing the XP0208 allele had reduced but detectable levels of Usp22 mRNA (Fig. S4E), which indicates that this allele is hypomorphic. These mice, referred to as Usp22^hyp/hyp, exhibited obvious growth retardation (Fig. S4F,G). To further confirm that loss of USP22 results in embryonic lethality, we designed and validated a pair of zinc-finger nucleases (ZFN) that targeted the start codon of Usp22. Injection of 741 zygotes (Fig. S5A,B) resulted in creation of five founder lines. Two lines were selected for additional study, and among 241 pups born to mating between heterozygous parents from these lines, no homozygous mutants were observed, which verified that the Usp22 gene is essential for mouse embryo survival (Fig. S6A,C).

Genomic DNA was prepared from embryonic yolk sac or mouse tail. Tissues were lysed through incubation with Proteinase K (x Manufacturer x) overnight at 55°C, followed by phenol extraction. DNA was precipitated by adding isopropanol and was washed in 70% ethanol. All embryos and mice were genotyped with specific primers that distinguished the mutant allele from the wild-type allele (Table S7). Mutation detection analysis in the ZFN alleles was performed using the Cel-I assay (Surveyor Mutation Detection Kit, Transgenomic). Briefly, 10 µl of PCR products, produced with primers that flanked the deletion generated by the ZFN pairs, were incubated using the following program: 95°C for 10min, a gradient from 95°C to 85°C with a decrease of 2°C every sec and a second gradient from 85°C to 25°C with a decrease of 0.1°C per sec. We then added 1 μl each of nuclease S and enhancer (Transgenomic) to digest the above reaction at 42°C for 30 min. The mixture was resolved on a 10% polyacrylamide TBE gel (Bio-Rad).

Genomic DNA (6 μg) was digested overnight in 30 units of restriction enzyme [SacI (New England Biolabs) for USP22 RRS377 allele and NcoI (New England Biolabs) for USP22 XP0208 allele] made up to 35 μl using 1× NEBuffer (1.1 for SacI and 3.1 for NcoI), and an additional 10 units of enzyme were added for 2-3 h on the second day. Samples were loaded on 0.8% agarose gels in Tris-acetate-EDTA (TAE) buffer and run at 26 V overnight. The gels were then denatured for 1 h in 1.5 M NaCl/0.5 M NaOH, neutralized for 40 min in 0.1 M Tris-HCl (pH 7.5)/0.5 M NaCl, washed with 20× SSC, and transferred for 24 h with 20× SSC on positively charged membranes (Roche). Probes were labeled with DIG-dUTP by PCR using the PCR DIG Probe Synthesis Kit (Roche). The membranes were UV-crosslinked, rinsed with water, air dried and prehybridized and hybridized with a 5′ probe or 3′ probe according to the manufacturer's instructions, using the DIG High Prime DNA Labeling and Detection Starter Kit II (Roche). The probes were heat denatured, added into the hybridization buffer and membranes were rotated overnight at temperatures calculated according to the manufacturer's instructions. The washing buffer (0.2× SSC, 0.5% SDS) was pre-warmed to 68°C, and the membranes were washed twice for 15 min at 68°C under shaking. Next, the probe-bound DNA fragments were detected using the DIG High Prime DNA Labeling and Detection Starter Kit II and X-ray films were exposed to the probed membrane to obtain the desired signal, following the manufacturer's instructions.

Conceptuses were dissected at various gestational ages: noon of the day that a vaginal plug was detected was defined as E0.5. For X-gal staining, specimens were fixed for 15 to 30 min in 1% formaldehyde, 0.2% glutaraldehyde, 0.02% NP-40, 5 mM EGTA, 2 mM MgCl₂, 0.1 M sodium phosphate (pH 7.3). Specimens were stained whole for 4-24 h at 37°C in 0.1% X-gal, 5 mM K₃Fe(CN)₆, 5 mM K₄Fe(CN)₆ in buffer [0.02% NP-40, 0.01% deoxycholate, 2 mM MgCl₂, 0.1 M sodium phosphate (pH 7.3)]. Following X-gal staining, conceptuses were post-fixed with 4% paraformaldehyde (PFA) for 12 h at 4°C.

Conceptuses were fixed in 4% PFA for 30-120 min at 4°C, equilibrated in 15% sucrose in PBS for 4 h followed by 30% sucrose in PBS for 12 h at 4°C. Samples were post-fixed in 10% neutral buffered formalin for 24-48 h then transferred to 70% ethanol, dehydrated in an ethanol series to 100% ethanol, cleared in xylene substitute and infiltrated with paraffin. Samples were then embedded in paraffin, sectioned at 4 μm and put onto slides for staining with H&E or for immunoperoxidase staining with different antibodies (Table S6). For every marker analyzed at least three biological replicates were used, unless stated otherwise. Slides stained with H&E were further analyzed using surface morphometry, as described in Philipp et al. (2002).

Embryos and placentas of different developmental stages were carefully dissected and snap frozen. Total RNA from embryos or tissues was extracted using RNeasy Plus Mini Kit (Qiagen). qRT-PCRs were measured using the 7500 Fast Real-Time PCR System (Applied Biosystems) using the SYBR Green method. The sequences for the gene-specific primers used are listed in Table S8. The relative expression of RNAs was calculated using the comparative Ct method with Gapdh as an internal control. Data are presented as normalized individual data points. Significant P values were corrected using the Benjamini–Hochberg (BH) method.

Embryos were carefully dissected, snap frozen, pulverized with a mortar and pestle in liquid nitrogen and resuspended in RIPA buffer. The extracts were dissolved in Laemmli buffer and boiled at 95°C for 5 min. The lysates (10-20 μg of total protein) were analyzed using western blotting with the anti-USP22 antibody described in Atanassov et al. (2016). Protein extraction from cultured cells was carried out using RIPA buffer. Western blots were carried out using 20-30 µg of protein. The antibodies used are listed in Table S6.

For the experiment, total RNA from E9.5 wild-type and Usp22 mutant placentas was extracted using the RNeasy Plus Mini Kit. The purified RNA was treated with TURBO DNase (Thermo Fisher Scientific) and used for library preparation. Briefly, 1 µg of total RNA was used to prepare RNA-seq libraries with the Illumina TruSeq Stranded Total RNA Kit according to the manufacturer's protocol. The libraries were then sequenced using 2×75 base paired-end protocol on the Illumina HiSeq 2500; 35-77 million pairs of reads were generated per sample. Each pair of reads represents a cDNA fragment from the library. For mapping, the reads were mapped to the mouse genome (mm10) using TopHat (version 2.0.10) (Kim et al., 2013). By reads, the overall mapping rate is 85-94%; 73-92% fragments have both ends mapped to the mouse genome. For differential expression, the number of fragments in each known gene from the RefSeq database (Pruitt et al., 2014) (downloaded from UCSC Genome Browser on May 23, 2014) was enumerated using htseq-count from HTSeq package (version 0.6.0) (Anders et al., 2015). Genes with fewer than ten fragments in all the samples were removed before differential expression analysis. The differential expression between conditions was statistically assessed using the R/Bioconductor package DESeq (version 1.18.0) (Anders and Huber, 2010). Genes with FDR ≤0.05 and length >200 bp were described as differentially expressed. For the gene function and pathway analysis, data were analyzed using IPA software. Heat maps were generated using Cluster v3.0 and Java TreeView v1.1.6r4.

For RNA-seq functional enrichment analysis, the GSEA pre-rank test was used to test the enrichment of gene sets for differentially expressed genes in mutant versus wild-type placentas (Subramanian et al., 2005). DAVID v6.7 (Huang et al., 2007) and PANTHER v11.0 (Mi et al., 2013) were also used to perform functional enrichment analysis of differentially expressed genes for target pathway analysis. Significant P values were adjusted using the Bonferroni method.

ESCs were routinely grown on gelatin-coated plates in Dulbecco's Modified Eagle's medium (DMEM)/high glucose medium supplemented with 15% (v/v) ESC-screened fetal bovine serum, 2 mM L-glutamine, 0.1 mM non-essential amino acids, 1% (v/v) penicillin/streptomycin, 0.1 mM β-mercaptoethanol, 1000 U/ml of LIF/ESGRO, 1 μM PD0325901 and 3 μM 1-Azakenpaullone, and passaged with 0.25% trypsin every 2-3 days. Induced endothelial cells and pericytes were generated using the protocol of Orlova et al. (2014) with some modifications for mouse ESCs. Briefly, mouse ESCs were cultured on ColIV-coated plates for 2 days in DFNK medium supplemented with B27, N2 and ROCK inhibitor (Tocris Bioscience) (2.5 µg/ml). On day 2 the medium was replaced with MEDF (DMEM-based low serum medium) for 24 h, then supplemented with activin A (25 ng/ml) for two additional days to induce differentiation towards mesendoderm. On day 6, medium was replaced with MEDF supplemented with VEGF 164 (50 ng/ml) and SB431542 (10 µM) (Stemcell Technologies) to induce differentiation towards vascular progenitors. The medium was replaced every day after removing the dead cells carefully using a PBS wash. On day 10 medium was replaced using an endothelial differentiation medium that was supplemented with VEGF 164 (30 ng/ml) and basic FGF (20 ng/ml) (Sigma-Aldrich) for 3-4 days. Finally, on day 14 the medium was replaced with pericyte differentiation medium that was supplemented with PDGF-BB (4 ng/ml) and TGFβ3 (2 ng/ml) for 3-4 more days to induce pericyte fate. Induced endothelial cells and pericytes were maintained in endothelial and pericyte maintenance medium respectively. The cells were used from passages 2-5. Both the differentiated endothelial cells and the pericytes were seeded and cultured at low confluency and experiments were performed within 48 h of the passage. For the tube formation assay, endothelial cells and pericytes were trypsinized, counted and resuspended in human endothelial/serum-free growth medium (EC-SFM). The cells were diluted to a concentration of 1×10⁵ cells/ml and were combined in a ratio of 1:4 endothelial cells to pericytes. They were placed in an angiogenesis μ-slide (ibidi) that was coated with ColIV. The cells were placed in the incubator at 37°C and 5% CO₂. They were fed at day 3 with EC-SFM and they were fixed and stained for imaging at day 8. The transwell migration assay of induced endothelial cells was performed as previously described (Chen, 2005). Briefly, the migration assay was performed in Transwell inserts in 24-well plates (Corning). We placed 1×10⁵ cells in 500 μl EC-SFM in the upper chamber, and 750 μl of the same medium containing 20 ng/ml of VEGF was placed in the lower chamber. The plates were incubated for 4 h at 37°C in 5% CO₂. Cells were washed with 600 μl PBS twice and fixed in 100% methanol for 5 min. Non-migrating cells in the upper chambers were removed with a cotton swab; migrated cells were stained with 0.5% crystal violet for 20 min and photographed under an optical microscope (Stemi 2000c, Zeiss). Five random fields were chosen in each well and the cell number was quantified manually.

Approximately 4×10⁵ cells for each sample were harvested by trypsinization, washed in PBS, centrifuged at 500 g for 5 min and pellets were resuspended in PBS. To each sample 100% ice cold ethanol was added dropwise while vortexing, and samples were incubated overnight at −20°C. Cells were harvested by centrifugation at 400 g for 2 min, supernatant was discarded and pellets resuspended in PBS. Cells were treated with 20 µg/ml RNase A (Roche) for 30 min at 37°C and the cell suspension was moved to flow cytometry tubes, passing the suspension through cell strainer caps. Propidium iodide was then added to each sample (20 µg/ml final concentration) and samples were analyzed using a BD LSRFortessa cell analyzer (BD Biosciences).

For confocal microscopy imaging of cultured cells and heart cryosections, cells or embryos were washed in PBS and fixed in 2% PFA for 5-15 min at 4°C, then washed briefly with methanol followed by incubation in pre-chilled (at −20°C) methanol for 5 min at −20°C. Alternatively, phalloidin-stained cells or embryos were fixed in 4% PFA for 15 min at room temperature followed by permeabilization with 1% Triton X-100. Cells were washed with PBS 3× for 5 min each. Nonspecific binding was blocked by adding blocking solution (0.5% gelatin in PBS) for 30 min at room temperature. The blocked cells were incubated with primary antibodies (Table S6) diluted in washing media (0.1% gelatin, 1% goat or donkey serum in PBS) overnight at 4°C. After five washes in washing buffer, cells were incubated for 40 min with secondary antibodies. DAPI staining was performed after washing off the secondary antibodies. Cells on coverslips were mounted using Prolong Gold (Life Technologies) and cells in wells were mounted with Slowfade Gold (Life Technologies). Laser scanning confocal microscopy was performed using a Zeiss LSM880 and 20× (0.8 NA) or 40× oil (1.4 NA) Plan/Apo objectives and with a pinhole aperture of 1-1.5 AU. For multiphoton imaging, whole embryos were permeabilized with 0.5% Triton and stained using the same antibodies and conditions except primary antibody staining was performed over 1-2 days in PBS with 0.5% Triton X-100, and secondary antibody and DAPI staining were performed overnight. Embryos were mounted in a viewing dish in a minimal volume of melted and cooled (45°C) 0.5% Ultrapure agarose (Invitrogen) prepared in sterile ddH₂O. Two-photon images of embryonic hearts were acquired using an upright Leica TCS SP8 DIVE multiphoton microscope with Insight X3 (Spectra-Physics) infrared laser and a 25×/0.95 NA FLUOTAR water immersion objective. For acquisition of images to encompass the lateral and axial dimensions of the E10 heart (∼1200×800 μm), mosaic tiling with a 10% overlap was employed over the designated region of interest with a z-stack depth range from 414-280 μm and a z-interval of 1.7 μm. Sequential scanning was performed using the 1040 nm fixed line to excite AF568 (phalloidin) and the tunable laser line for AF488 (PECAM1) and DAPI using 950 nm and 750 nm, respectively. The emission was captured separately on non-descanned hybrid (HyD) detectors. Stitching to create a seamless mosaic was performed using Leica LASX software, and Imaris 9.1 (Bitplane) was used to render 3D volumes and visualize vasculature and heart tissue using the green (PECAM1) and red (phalloidin) channels, respectively. Wild-type and Usp22 mutants were processed in parallel using the same parameters and the Imaris Blend and Orthoslicer features.

Total histones from the induced endothelial cells were prepared using a Histone Purification Mini Kit (Active Motif). Laemmli buffer was added and the lysates were analyzed using western blotting.

We thank Dr Jan Parker-Thornburg and Debra Hollowel for assistance with mouse line generation. We thank Amanda Martin for animal husbandry, and Tim Macatee and Pam Whitney for excellent technical support. We also thank Dr Ondine Cleaver for critical reading of the manuscript. Sequencing was done by the University of Texas MD Anderson Cancer Center (MDACC) Science Park NGS Core, supported by Cancer Prevention and Research Institute of Texas (CPRIT) Core Facility Support Grants RP120348 and RP170002. Bioinformatics support was made possible by CPRIT Core Facility Support Grants RP120348 and RP170002 and by support from the MDACC Center for Cancer Epigenetics. Animal services were supported by the MDACC Support Grant CA16672 from the National Cancer Institute. Histological analysis was performed at the Science Park histology core, partially supported by the MDACC Center for Environmental and Molecular Carcinogenesis. Confocal and multiphoton imaging and flow cytometry were performed at the Science Park Flow Cytometry and Cell Imaging Core and supported by CPRIT Core Facility Grant RP170628.