Abstract
Activating mutations in the Wnt pathway drive a variety of cancers, but the specific targets and pathways activated by Wnt ligands are not fully understood. To bridge this knowledge gap, we performed a comprehensive time-course analysis of Wnt-dependent signaling pathways in an orthotopic model of Wnt-addicted pancreatic cancer, using a PORCN inhibitor currently in clinical trials, and validated key results in additional Wnt-addicted models. The analysis of temporal changes following Wnt withdrawal demonstrated direct and indirect regulation of >3,500 Wnt activated genes (23% of the transcriptome). Regulation was both transcriptional via Wnt/β-catenin, and through the modulation of protein abundance of important transcription factors including MYC via Wnt/STOP. Our study identifies a central role of Wnt /β-catenin and Wnt/STOP signaling in controlling ribosomal biogenesis, a key driver of cancer proliferation.
Introduction
Wnts are a family of 19 secreted proteins that play key roles in cell proliferation, cell-cell communication and differentiation and are essential during embryonic development and in adult tissue homeostasis (Nusse and Clevers, 2017). The binding of Wnts to their receptors and co-receptors results in the regulation of multiple downstream signaling pathways (Acebron and Niehrs, 2016). Our knowledge of the specific events and signals regulated by Wnts derives from a variety of genetic, molecular, and biochemical approaches that has generated a rich map of these downstream pathways (Ramakrishnan and Cadigan, 2017; Yu, 2014).
The Wnt/β-catenin pathway, also known as canonical WNT signaling, has been intensively studied. In the presence of Wnts, β-catenin is stabilized and translocates to the nucleus where it drives expression of target genes in a context specific manner via binding to TCF7L2 and other factors (Ju et al., 2014; Nakamura et al., 2016; Ramakrishnan and Cadigan, 2017). In addition to the Wnt/β-catenin pathway, Wnts regulate signaling through diverse β-catenin independent non-canonical pathways such as the PCP (planar cell polarity) pathway and the Wnt/STOP (Wnt-dependent stabilization of proteins) pathway that are less well characterized (Acebron et al., 2014; Chien et al., 2009; Koch et al., 2015; Taelman et al., 2010; Zhang et al., 2012).
While the downstream mutations that stabilize β-catenin (e.g. in the adenomatous polyposis coli (APC) gene) clearly cause human cancers, genetic lesions that cause Wnt over-expression have not been found (Nusse and Varmus, 2012). A subset of mutations that block Wnt receptor internalization and confer dependency on Wnt ligands have been identified in a range of carcinomas. These include loss of function mutations in RNF43, an E3-ligase, and translocations leading to increased R-spondin levels (Jiang et al., 2013; Ong et al., 2012; Seshagiri et al., 2012). RNF43 mutations(Cancer Genome Atlas Research Network, 2017) and translocations involving RSPO2 and RSPO3 are found in 7% of pancreatic adenocarcinoma (PDAC) and 10% of colorectal cancers respectively (Seshagiri et al., 2012). Cancers with RNF43 or RSPO3 mutations have a markedly increased abundance of Frizzled receptors on the cell surface and are uniquely Wnt-addicted (Madan et al., 2016; Madan and Virshup, 2015).
Wnts are palmitoleated by a membrane bound O-acyltransferase, PORCN. This modification is(Rios-Esteves and Resh, 2013) essential for binding to chaperone WLS and Frizzled receptors, and is therefore required for the activity of all Wnts. Pharmacological PORCN inhibitors such as ETC-159 and LGK-974 have progressed to Phase I clinical trials due to their efficacy in preclinical models of RNF43-mutant pancreatic and RSPO3-translocated colorectal cancers (Jiang et al., 2013; Madan et al., 2016; Proffitt et al., 2013). The recent development of these PORCN inhibitors that block all Wnt secretion provides an opportunity to investigate how Wnt regulated genes change over time following withdrawal of signaling (Chen et al., 2009; Coombs et al., 2010; Janda et al., 2012; Liu et al., 2013; Madan et al., 2016; Takada et al., 2006). In order to provide the most relevant data, it is important to use the most predictive pre-clinical models. A large body of literature demonstrates that cancers in vivo behave very differently than cancers in tissue culture (Killion et al., 1998). These differences have led to the development of orthotopic xenograft models and patient-derived xenografts that better reflect the behavior of Wnt-addicted cancers in a complex cancer-host environment (Byrne et al., 2017).
Here we investigated the temporal impact of acute withdrawal of Wnt ligands on the perturbed transcriptome of a Wnt-addicted human pancreatic cancer in an orthotopic mouse model. The time-series analysis identified direct and indirect Wnt targets based on their dynamics, distinguishing immediate early, early and late response genes. This analysis led to identification of an important role of the Wnt/STOP pathway in regulating tumor growth. We further disentangled the WNT vs MYC dependencies in these cancers. This comprehensive time course analysis in an in vivo Wnt-addicted cancer provides both a valuable resource and new insights into the central role of Wnt/β-catenin and Wnt/STOP signaling in regulation of ribosomal biogenesis pathway.
Results
Time-dependent global transcriptional changes follow PORCN inhibition in a Wnt-addicted pancreatic cancer model
We aimed to identify the genes and biological processes that are directly or indirectly regulated by Wnt ligands in a Wnt-addicted cancer in vivo. To mirror the tumor microenvironment and recapitulate tumor stromal interactions, we established an orthotopic mouse model using a highly WNT-dependent HPAF-II cell line with RNF43 inactivating mutation (Jiang et al., 2013; Madan et al., 2016). As expected, ETC-159 significantly inhibited the growth of HPAF-II orthotopic xenografts (Fig. 1A), and led to pronounced histomorphological changes (Fig. 1B) (Madan et al., 2016). Tumors from the control group were characterized by the presence of neoplastic cells with poorly defined acini and cell boundaries. Treatment with ETC-159 induced changes in cellular organization and by 7 days the tissue appeared more differentiated with a decreased nuclear cytoplasmic ratio, and diminished anisocytosis and anisokaryosis.
To identify immediate early, early and late responses to Wnt inhibition, mice with established HPAF-II orthotopic xenografts were treated with ETC-159 and tumors were collected at 3, 8, 16, 32, 56 hours and at 7 days after treatment (Fig. 1C). Comprehensive gene expression analysis was performed using RNA-seq of 4-7 independent tumors at each time point (Fig. S1A). Inhibition of WNT signaling led to a marked change in the transcriptome, with the expression of 11,673 genes (75% of all expressed genes) changing over time (false discovery rate (FDR) < 10%) (Table S1). Expression of 773 genes changed as early as 8 h after the first dose of ETC-159. 1,578 and 1,883 genes were upregulated or downregulated, respectively after 56 h (FDR < 10%, absolute fold-change > 1.5) (Figs 1D, S1B-C). The majority of genes that exhibited significant differences at 56 h were also differentially expressed at 7 days, suggesting that the effect of Wnt inhibition is primarily established within 3 days.
To better understand how the withdrawal of Wnt signaling affected gene expression over time we performed time-series clustering (Hensman et al., 2015) of differentially expressed genes. Genes with significant changes in response to treatment were grouped into 64 clusters, with each cluster consisting of genes exhibiting similar dynamic responses following PORCN inhibition (Fig. S1D, Table S2). Further analysis of these clusters to identify consistent global patterns of transcriptional response identified two major robust patterns (Fig. S1E-F), a supercluster comprising of genes consistently down-regulated (Wnt-activated genes) and a supercluster containing genes consistently upregulated following PORCN inhibition (Wnt-repressed genes) (Fig. 1E). Here, to demonstrate the usefulness of this data resource, we focus on Wnt-activated genes, and we provide the analysis of Wnt-ligand regulated genes in Tables S1-S2.
Analysis of Wnt-activated genes
The Wnt-activated genes supercluster contained 11 clusters with distinct dynamics, consisting of 3,549 genes (23% of the transcriptome). For each of these clusters, we calculated the time to 50% inhibition (TI50) based on its mean profile. We operatively classified these clusters into four waves with TI50 ranging from 5.3 to 46.6 hours (Fig. 1E) (first wave, 5-3-9.8 h; second, 12.9-15.1; third, 20.3-26.5; fourth, 40.8-46.6). Well-established Wnt target genes had distinct time courses and were present throughout the first three waves with TI50 ranging from 5.3 to 26.5 hrs. For example, Cluster 9 (TI50 = 8.5 h) included well-known β-catenin targets (e.g. AXIN2, NKD1, RNF43, BMP4 and LGR5) and was significantly enriched for pathways and processes relating to Wnt signaling and development (Fig. 1F). C5 and C12 in the third wave (TI50 20.6 and 22.9) similarly contained known Wnt target genes e.g. NOTUM (Giraldez et al., 2002). We speculate that these complex dynamics and broad range of response times of the various β-catenin targets relates to the cell-type specific context of the co-regulatory elements of these genes and the stability of the specific mRNAs.
Interestingly, we observed that the early-changing clusters such as C9 contained genes that are not known to be direct β-catenin target genes (Fig. S2) and thus may be β-catenin independent and rely on mechanisms such as Wnt/STOP. These included well-studied regulators of ribosomal biogenesis (e.g. NPM1, DKC1, NOL6, RRS1) and nucleocytoplasmic transport (e.g. XPO5, NUP37) (van Riggelen et al., 2010). Other smaller rapidly responding clusters in the first and second waves similarly contained genes not known to be β-catenin target genes. These clusters were also enriched for processes associated with ribosome biogenesis (e.g. C20: POLR1A, POLR1B; C25: NOP14, RRP9). The slowest responding genes, found in the fourth wave clusters 7 and 21 (TI50 40.8-46.6 h) are likely to be regulated by processes downstream of initial Wnt signaling events and were also enriched for processes relating to ribosomal biogenesis (Fig. 1E & F). In addition to ribosome biogenesis, there was a broad enrichment throughout the clusters for genes involved in nucleic acid metabolism and cell cycle, especially in the third wave, C1, C5 and C12 (Fig. 1F). Selected examples of Wnt-activated genes with differences in their pattern of response to Wnt inhibition are depicted in Fig. 1G.
Our dataset provides a comprehensive resource of genes whose expression is highly dependent on Wnt signaling in vivo (Table S1). As several of the early changing genes are not known to be direct targets of β-catenin, this analysis identified genes that may depend on additional pathways such as Wnt/STOP. Importantly, the dataset highlights that in addition to its recognized role in cell cycle regulation, an early consequence of blocking Wnt signaling is the down-regulation of genes involved in ribosome biogenesis and its associated processes.
A common core of Wnt-regulated gene expression changes are more robust in orthotopic xenografts
To test if the gene expression changes seen in the HPAF-II pancreatic cancer were generalizable to other Wnt-addicted cancers, we compared our data to our previously published dataset of a Wnt-addicted RSPO3-translocation patient-derived colorectal cancer xenograft (CRC PDX) (Table S3) (Madan et al., 2016). The strength of the correlation between the expression changes induced by PORCN inhibition at 56 h in the two experimental systems (r2 = 0.36, Fig. 2A) indicated that regardless of the upstream mutation and tissue of origin, the downstream effect of Wnt inhibition on tumor gene expression was similar. In keeping with the central role of Wnts in regulating the differentially expressed genes, the majority of Wnt-activated genes (62%, FDR < 10%, absolute fold-change > 1.5) were also down regulated at 56 h in the CRC PDX (Fig. 2B). Within C9, 69% of the genes expressed in both systems were down-regulated in CRC PDX, suggesting that these genes may be direct targets of WNT signaling in both of these tumor models. This highlights that the core processes and mechanisms responsible for the Wnt-activated genes are shared between the CRC PDX and HPAF-II orthotopic xenografts. Notably, the overlapping set of Wnt-activated genes was enriched for genes involved in cell cycle regulation and ribosome biogenesis (Fig. S3A), again suggesting the centrality of these pathways in Wnt-addicted cancers.
The shared response of the Wnt-addicted pancreatic and colorectal cancers led us to evaluate the importance of the stromal microenvironment in regulating the response to Wnt inhibition. We compared the effects of ETC-159 on the transcriptome of HPAF-II cells implanted as orthotopic xenografts, subcutaneous xenografts or cultured in vitro. The response to Wnt inhibitor treatment in orthotopic xenografts was correlated to the response in both the subcutaneous model (r2 = 0.64, Fig. 2C) and the in vitro model (r2 = 0.46, Fig. 2D). However, the magnitude of the responses varied widely between the models. Differential expression analysis of the three HPAF-II models identified 4,409 genes whose response to ETC-159 was significantly different (interaction test, FDR < 10%) between models. These genes were enriched for processes including cell cycle, ribosome biogenesis (Fig. S3B). 51% of all Wnt-activated genes responded differently to ETC-159 across the three systems, with many more genes showing robust fold changes in the orthotopic model compared with the other systems (Fig. 2E). In particular, cell cycle associated genes changed most robustly in the orthotopic model (Figs 2C-D and F). For example, AURKA decreased by 3.7 fold at 56 hours after the start of PORCN inhibition in the orthotopic system, 2.1 fold in the subcutaneous model and only 1.7 fold in vitro. Cyclin E1 did not change in vitro but decreased by 1.7 fold and 3.8 fold in the subcutaneous and orthotopic model, respectively (Fig. 2F). Negative regulators of the cell cycle such as CDKN2B also exhibited differences in their magnitude of response to PORCN inhibition. In addition to cell cycle associated genes we identified several other genes that did not respond to Wnt inhibition in vitro but behaved as WNT targets in vivo(i.e. EPHB3 and TGFBI, Fig. S3C). Thus, the overall response to Wnt inhibition was reduced in the subcutaneous model and further blunted in vitro (Figs 2C-F).
Further highlighting the importance of the orthotopic model was the considerable difference in baseline gene expression between models ~30% of all expressed genes (4,992 genes) were differentially expressed even before treatment when comparing in vitro and orthotopic xenografts. This included greater than two-fold increases (moving from cultured cells to the orthotopic model) in Wnt pathway genes WNT11, WNT2B, FZD8, FZD4, and LRP5, as well as the Wnt target genes NKD1, SP5, LGR5 and AXIN2 (Table S4). Additionally, 3,515 genes were differentially expressed at baseline when comparing subcutaneous xenografts with the orthotopic model (absolute fold change > 1.5, FDR < 10%) (Figs S3D-E). Not surprisingly, a number of genes more highly expressed in the orthotopic xenografts compared to in vitro were associated with multicellularity, innate immunity, extracellular matrix organization and cell adhesion (Fig. S3D).
The much less pronounced effect of ETC-159 on the expression of cell cycle genes in cell culture is consistent with our previous observations that PORCN inhibitors do not inhibit the growth of Wnt-addicted cancer cells in short-term 2D cell culture (Proffitt et al., 2013). This data illustrates that the orthotopic and PDX mouse models more accurately recapitulate the tissue-specific tumor microenvironment and highlights the value of the orthotopic model in identifying core Wnt regulated genes.
PORCN inhibition leads to early downregulation of MYC and its targets
The time-series clustering analysis (Fig. 1E) identified sets of genes (clusters) having similar dynamics of response to PORCN inhibition, suggesting that each cluster may be regulated by distinct mechanisms. To investigate the differences in the transcriptional regulation of these genes, we performed a Transcription Factor Binding Site (TFBS) motif analysis on the promoters of the Wnt-activated genes (Fig. 3A).
Unexpectedly, the promoters of genes downregulated immediately following Wnt withdrawal (e.g. C9, TI50 = 8.5 h) did not show significant enrichment for TCF7L2 binding sites (p-value=0.21). The majority of TCF7L2 binding events are found to be intergenic rather than promoter-associated (Fig. S4) (Stevens et al., 2017). The promoters of genes in the most rapidly responding clusters (i.e. C9, C10, C20, C24 and C25, TI50 < 20 h) were rather significantly enriched for canonical E-box motifs, bound by transcription factors including MYC, HEY1, CLOCK and ID2 (Fig. 3A). The genes in clusters that responded later (TI50 > 20 h) were enriched for E2F, NRF and NFYB binding sites(Dolfini and Mantovani, 2013).
The enrichment of E-box binding motifs occurred in some early responding Wnt-activated genes (e.g. C20, C9) whose expression fell even before MYC mRNA decreased, suggesting additional levels of regulation (Fig. 3A). Upon PORCN inhibition, MYC mRNA responded like a direct WNT target gene with an early and sustained decrease (C9), albeit only to ~50% of its initial mRNA abundance (Fig. 3B). This is consistent with the well-established role of β-catenin signaling in the regulation of MYC expression(Myant and Sansom, 2011). In addition to being a transcriptional target of Wnt signaling, MYC protein abundance can be directly regulated by GSK3 by phosphorylating it at Threonine 58, thus priming it for ubiquitylation and proteasomal degradation (Acebron et al., 2014; Arnold et al., 2009; Sears et al., 2000; Taelman et al., 2010). As Wnt signaling inhibits AXIN-associated GSK3, blocking Wnt signaling increases the activity of GSK3 and promotes MYC degradation. Indeed, we observed a more pronounced change in MYC protein (2.8-14.5 fold reduction) than MYC mRNA (2.0-2.2 fold reduction) in both HPAF-II orthotopic tumors and the colorectal PDX models 56 h after PORCN inhibitor treatment (Figs 3B-C). These results suggest that the Wnt-dependent decrease in MYC transcripts was coupled with post-transcriptional regulation of MYC protein abundance, i.e. a Wnt/STOP effect in Wnt addicted tumors.
Wnt signaling regulates MYC via WNT/STOP and Wnt/β-catenin pathway
To assess the relative contributions of Wnt-regulated MYC mRNA expression (Wnt/β-catenin) and Wnt/GSK3-regulated MYC protein stability, we generated HPAF-II cell lines stably overexpressing either Myc (MYC OE) or GSK3-resistant Myc (MYC T58A) under the control of the Wnt-independent CMV promoter. The CMV promoter drove 10-15 fold higher Myc mRNA expression in orthotopic tumors (Fig. 4A). We then compared the effect of PORCN inhibition on the growth of HPAF-II, HPAF-II (MYC OE) and HPAF-II (MYC T58A) orthotopic xenografts.
The MYC OE orthotopic tumors had a marked increase in MYC protein but there was no overall increase in tumor growth and remarkably they still responded significantly to PORCN inhibition (Figs 4B-C). Thus, restoration of MYC by overexpression does not rescue tumors from the effects of Wnt inhibition. This could be either because MYC protein is not rate-limiting for tumor growth in this setting, or that PORCN inhibition was able to drive MYC degradation. Consistent with this Wnt/STOP effect, we found that PORCN inhibition caused a decrease in MYC protein abundance despite no change in ectopic MYC mRNA levels (Figs 4A, 4C).
We next examined if blocking the Wnt/STOP effect on MYC protein altered the response to PORCN inhibition. The abundance of MYC T58A did not change upon PORCN inhibition (Fig. 4C) and tumors with stabilized MYC grew larger and showed a partial response to PORCN inhibition (Fig. 4B). Taken together, these findings indicate that in addition to its transcriptional regulation, inhibiting Wnt signaling regulates the growth of the tumors by directly regulating MYC protein abundance via a GSK-dependent mechanism. Further, the finding that tumors with stabilized and over-expressed MYC still partially respond to PORCN inhibition shows that Wnts regulate the growth of the pancreatic tumors via both Myc-dependent and Myc-independent pathways. This is similar to findings in the murine intestine, where MYC is essential for the oncogenic effects of APC deletion (Sansom et al., 2007) but alone is insufficient to drive tumorigenesis (Finch et al., 2009).
Distinguishing between Myc-dependent and Myc-independent regulation of Wnt target genes
To directly identify MYC-independent and MYC-dependent WNT target genes, we performed an additional set of RNA-seq experiments to examine gene expression changes in orthotopic tumors generated from HPAF-II WT, MYC OE or MYC T58A cells. We selected as the time points 0, 8 and 56 hours after the start of therapy to allow us to examine early direct targets of both Wnt/β-catenin and Wnt/GSK3/MYC signaling. Consistent with the previous experiment, ETC-159 treatment reduced MYC protein by ~70% in 56 h in HPAF-II WT tumors, while protein levels in MYC T58A tumors showed no decrease (Fig. 5A).
We identified 2,131 genes whose transcriptional response to PORCN inhibition in vivo was dependent on MYC status (FDR < 10%, Table S5). These genes, whose response to PORCN inhibition was different between WT, MYC OE or MYC T58A tumors, were classified as MYC-dependent Wnt target genes. Of these genes, 827 (23%) were found in our set of Wnt-activated genes. In each of the clusters of Wnt-activated genes we determined the fraction that exhibited MYC-dependent or MYC-independent responses (Fig. 5B). The majority of genes that were downregulated most rapidly upon Wnt inhibition (i.e C9, C10, C20, and C25, TI50 < 20 h) (Fig. 1E) were MYC-independent. Selected examples of well-established Wnt-regulated MYC-independent genes such as AXIN2 and NKD1 are illustrated in Fig. 5D, examining both the relative transcript abundance (top panel) and the log fold changes (bottom panel). Not surprisingly, the MYC-independent Wnt target genes in C9 were associated with Wnt signaling pathways and embryonic patterning (Fig. 5C).
Only third wave clusters C5 and C1, changing with TI50 of 20.6 and 26.5 h, contained a sizeable fraction (> 25%) of MYC-dependent genes (Fig. 5B) that were enriched in ribosome biogenesis and cell cycle processes discussed below. We did not observe enrichment for E-boxes in the clusters C1 and C5 (Fig. 3A). Interestingly, the subset of genes in these clusters that were MYC-dependent were also not specifically enriched for MYC TFBS, suggesting either they are indirect targets of MYC or that TFBS analysis is not powerful enough to detect a clear enrichment for these MYC motifs.
Regulation of cell cycle by Wnts in Wnt-addicted cancers is Wnt/GSK3 dependent
Our initial analysis (Fig. 1) demonstrated that cell cycle and ribosomal biogenesis are two key pathways that are transcriptionally regulated by Wnt inhibition, with multiple genes regulating cell cycle changing in a time-dependent manner, including CDK1, E2F2, E2F1, CDKN2B and CDKN2A (Fig. 6A). Consistent with this robust regulation of cell cycle genes, Ki67 positive cells were significantly reduced in the tumors as early as 56 hours after starting ETC-159 and were further reduced at 7 days of treatment (Fig. 6B).
Genes associated with mitotic cell cycle processes and DNA replication were enriched in the third wave of clusters (C1, C5, C12; TI50 20.6-26.5 hrs) (Fig. 1F). These clusters were enriched for binding sites for the E2F and NFY families of transcription factors (Fig. 3A) that cooperatively regulate cell cycle genes (Dolfini and Mantovani, 2013; Ly et al., 2013; van den Heuvel and Dyson, 2008).
Interestingly, E2F1 and E2F2 gene expression decreased at the same rate as the other cell cycle genes, suggesting that the early decrease in the expression of cell-cycle related genes was not due to changes in these E2F mRNAs (Fig. 6A). E2F activity is also regulated by cyclin dependent kinase signaling through p105/Rb. Indeed, the expression of the CDK inhibitors increased as early as 8 h after PORCN inhibition in orthotopic tumors, associated with a subsequent decrease in Rb phosphorylation (Figs 6A and C). Thus, CDK inhibition and decreased Rb phosphorylation is likely to be a major mechanism driving the decrease in the transcription of E2F target cell cycle genes.
Notably, the abundance of cell cycle regulators such as Cyclin D1 and E1, amongst others, is also regulated by Wnt/STOP signaling (Acebron et al., 2014). Similar to MYC, after 56 h of ETC-159 treatment the protein abundance of CCND1 was reduced by ~3 fold (Fig. 6C) while CCND1 transcript levels were only reduced by ~1.5 fold in both HPAF-II xenografts and colorectal PDX (Figs 6A and D). Thus, our in vivo data in Wnt driven cancers support the data from in vitro studies (Acebron et al., 2014) that Wnt signaling regulates CCND1 and MYC by both transcriptional and post-transcriptional mechanisms.
We further examined the role of MYC in the regulation of cell cycle genes. Notably, MYC overexpression had no effect on baseline expression of the cell cycle genes and Wnt inhibition reduced their expression, albeit to differing extents even in the presence of stabilized MYC (Figs 6E-F). Stabilized MYC blunted the effect of PORCN inhibition on the expression of a subset of the cell cycle genes, e.g. CDK1and MKI67 (Fig. 6E). However, a number of other cell cycle genes (e.g. CDKN2B, CDC7, RBBP8 and RPA3) were MYC-independent and responded to Wnt inhibition even in MYC-stabilized tumors (Fig. 6F). Consistent with the observed transcriptional response, there was a partial reduction of Ki67 staining in ETC-159 treated MYC-stabilized tumors (Fig. 6G). Taken together these findings indicate that Wnt regulates the cell cycle in cancers via multiple pathways, both dependently and independently of MYC, and through both transcriptional and Wnt/STOP mechanisms.
Wnt signaling regulates Ribosomal Biogenesis
The enrichment for rRNA processing and ribosomal biogenesis in the Wnt activated gene clusters (Fig. 1F) suggested that PORCN inhibition would lead to a reduction in ribosome formation and protein synthesis. Indeed, nearly all genes encoding ribosomal protein subunits (RPSs and RPLs) were downregulated (Fig. 7A), with 94% of differentially regulated RPSs and RPLs being present in late responding clusters, C7 (TI50 46.6 h) or C1 (TI50 26.5 h). Although the expression of RPS and RPL genes was reduced by only ~30-40 %, the changes were largely coherent albeit with some outliers of unknown significance. The changes were more apparent following 32 h of treatment, suggesting that these genes are indirectly regulated by Wnt signaling. We next examined if this was reflected in the abundance of ribosomal subunit proteins. In a parallel mass spectrometry experiment that only detected high abundance proteins (see Methods), we confirmed that the RPS and RPL proteins were also coherently downregulated at 56 h (Fig. 7B, Table S6). Given the high abundance of ribosomal proteins, this suggests a dramatic shift in ribosome biogenesis.
Ribosomal biosynthesis requires multiple processes including nucleocytoplasmic transport and rRNA expression and processing (van Riggelen et al., 2010). We found that genes required for nucleocytoplasmic export including exportins and nucleoporins were similarly coherently downregulated implicating the regulation of ribosome assembly by Wnt signaling (Fig. 7C). Multiple components of the machinery required for rRNA transcription, including several subunits of RNA polymerases POLR1 and POLR3 (Fig. 7D) and rRNA processing factors (e.g. NPM1, DKC1) were also downregulated (Fig. 7E). Finally, consistent with the changes in gene and protein expression, the size of nucleolar organizer regions was reduced by ETC-159 treatment (Fig. 7F). Taken together, these data indicate that ribosomal biogenesis is globally regulated by Wnt signaling. A global decrease in protein synthesis coupled with a halt in the cell cycle likely explains how PORCN inhibition blocks tumor progression in Wnt-addicted cancers (Ruggero and Pandolfi, 2003).
We asked if Wnt regulation of ribosome biogenesis was explained by its effect on MYC, a recognized regulator of ribosome biogenesis (van Riggelen et al., 2010). In contrast to the cell cycle genes, the baseline expression of ribosome subunit and biogenesis genes was increased by stabilized MYC (Figs 7G-H). However, a number of these genes remained sensitive to PORCN inhibition and decreased after 56 h of ETC-159 treatment even in cells with MYC T58A (Fig. 7G). These Wnt-regulated, Myc-independent ribosome genes includes virtually all of the RPLs and RPSs (C1 and C7) (Figs. 7G and S5). Another subset of genes involved in rRNA synthesis and processing (e.g., NPM1, DKC1, POLR1B) were MYC-dependent WNT target genes. These genes were both highly MYC responsive at baseline, and consistent with their regulation by Wnt/STOP regulation of MYC protein abundance, did not respond to Wnt inhibition if MYC T58A was present (Fig. 7H). These MYC-dependent genes are enriched for E-boxes in their promoters.
Our analysis thus establishes a key role of Wnt signaling in ribosome biogenesis via two routes. One route, via MYC, is regulated both through Wnt-driven MYC expression and via the Wnt/STOP pathway. The other route is MYC-independent and is a downstream effect of WNT signaling on the transcription of ribosomal genes.
Discussion
The development of targeted drugs that rapidly and robustly inhibit PORCN provides a unique opportunity to examine in real time the consequences of Wnt withdrawal in Wnt-addicted human cancers. This time-based analysis of Wnt signaling and its interaction with MYC, provides a comprehensive assessment of the role played by Wnt ligands in driving Wnt-addicted cancer. Importantly, the high concordance of the transcriptional changes in Wnt-addicted RSPO3-mutant colorectal and RNF43-mutant pancreatic cancers reveals core shared pathways regulated by Wnt signaling in cancer. Previous studies examining the targets of Wnt signaling in cancer have focused on models that are driven by loss of function mutations in APC. Here, the use of Wnt ligand driven cancer mouse models casts a broader net, identifying a unexpectedly large number of genes whose expression depends on continued presence of Wnt ligand many of which are independent of β-catenin.
The genes whose expression changes most rapidly after PORCN inhibition, the early wave clusters, were predictably enriched for well-established β-catenin target genes (Moon and Gough, 2016). However, our analysis revealed a large number of co-regulated genes that were not known β-catenin targets. DNA sequence-based analysis of enrichment for TCF/LEF binding sites was not a useful approach to discriminate if these early changing genes could be additional β-catenin targets or they could be regulated by multiple non-canonical pathways. Indeed, while many individual studies find TCF/LEF sites in the promoters of selected genes, our findings support the results from genome-wide analyses showing that functional TCF/LEF sites are often present at large distances from transcriptional start sites (Ramakrishnan and Cadigan, 2017). Additionally, recent studies have established that even β-catenin promoter binding is not sufficient to identify β-catenin transcriptionally regulated genes5, 40.
Interestingly our analysis revealed that E-box transcription factor binding sites are enriched in the early changing genes, followed at later time points by enrichment for E2F-binding sites. Finally, the fourth wave of genes was enriched for a broader set of TFBS that are likely to be regulated as secondary, downstream events. The enrichment for E-boxes strongly suggested a role for MYC. MYC is a potent oncogene and its activation is a hallmark of cancer initiation and maintenance (Dang, 2010; Gabay et al., 2014). MYC is required for tumorigenesis following β-catenin activation by APC loss in the gut but not in the liver (Reed et al., 2008; Sansom et al., 2007). Hence, it was an open question if MYC would be important downstream of RNF43 mutations in pancreatic cancers, where many additional pathways are activated by the Wnt addiction (Sansom et al., 2007; Wilkins and Sansom, 2008). Using a model of Wnt-addicted human cancer with stabilized MYC we were able to disentangle the interaction of Wnt and MYC and stratify the role of Wnts and MYC in regulating cell cycle and ribosomal biogenesis. One notable difference was that stabilization of MYC did not enhance the expression of cell cycle genes. However, stabilized MYC could partially overcome the effect of Wnt inhibition on expression of a subset of cell cycle genes (Annibali et al., 2014). Whereas, Myc overexpression and stabilization more profoundly affected genes regulating various processes associated with ribosomal biogenesis. Here too, the response to Wnt inhibition was variable as a large subset of genes, including ribosomal proteins, responded to PORCN inhibitors with similar fold changes, while others were “immune” to Wnt inhibition in the presence of stabilized MYC. This suggests a complex interaction of MYC and Wnt-regulated pathways driving these processes.
Ribosomes are overexpressed in cancer and have become novel targets for anticancer therapies, for instance, by triggering nucleolar stress (Pelletier et al., 2018; Quin et al., 2014; Ruggero and Pandolfi, 2003; Sulima et al., 2017). While MYC is known to regulate ribosome biogenesis (van Riggelen et al., 2010), the role of Wnts has been less clear (Kraushar et al., 2015; Pfister and Kühl, 2018). Here we show for the first time that Wnt signaling globally impacts multiple steps in ribosomal biogenesis both directly and by regulating the transcription and protein abundance of Myc via the Wnt/STOP pathway and this is shared in both Wnt-addicted pancreatic and colorectal cancers.
Comparing the effect of PORCN inhibition across different models confirmed the value of studying Wnt signaling in an orthotopic microenvironment or in the present of native stroma (CRC PDX). The experimental value of the orthotopic model using a cell line is that it is more amenable to genetic manipulation such as the introduction of stabilized MYC, allowing a more detailed analysis of the role of downstream drivers.
The stabilization of MYC via the Wnt/GSK3 signaling axis highlights how this mechanism can target MYC and other cell cycle proteins in cancer, impacting aberrant cell growth (Dang et al., 2017). The Wnt/STOP pathway is likely to have multiple additional targets (Koch et al., 2015; Taelman et al., 2010) that may also play a role in these pancreatic and colorectal cancers. Future studies with high-resolution mass spectrometry at early time points after Wnt inhibition may facilitate their identification. The data provided in this study can facilitate biomarker discovery for patients suffering from Wnt addicted cancers and provides a significant resource for the Wnt and cancer community.
Material and Methods
Tumor growth and mice treatment
Mouse xenograft models from HPAF-II cells were established by orthotopic injection of HPAF-II cells in NOD scid gamma mice as described in the supplemental methods.
Western Blot analysis
Tumors were homogenized in 4% SDS buffer and proteins were resolved on 10% SDS-polyacrylamide gel. Western blots were performed according to standard methods.
Immunohistochemistry and AgNOR staining
Formalin fixed and paraffin embedded tissue sections were then stained with hematoxylin and eosin, Ki67 or nucleolar organizer regions using standard protocol. Images were acquired using Nikon E microscope.
RNA isolation and Data analysis
Tumors were homogenized in RLT buffer and total RNA was isolated using RNAeasy kit (Qiagen) according to manufacturer’s protocol. The RNA-seq libraries were prepared using the Illumina TruSeq stranded Total RNA protocol with subsequent PolyA enrichment. Details for QC and data processing for RNA-seq, transcription factor binding site analysis, time series clustering and ChIP-seq analysis are provided in the supplemental methods.
Proteomics
Tumors were homogenized on dry ice and solubilized with 8M urea and 20 mM HEPES. Trypsin digested Peptides in 0.1% TFA were separated using an Ultimate 3000 RSLC nano liquid chromatography system coupled to a Q-Exactive mass spectrometer. Following data-dependent acquisition, raw data files were loaded and analyzed using Progenesis QI.
Author Contributions
Babita Madan, Nathan Harmston, Enrico Petretto and David M. Virshup designed the study. Babita Madan and Gahyathiri Nallan performed the animal studies and biochemical analysis. Nathan Harmston designed and performed the bioinformatics analysis. Alex Montoya and Peter Faull performed the mass spectrometry. Enrico Petretto and David M. Virshup supervised the study. Babita Madan, Nathan Harmston, Enrico Petretto and David M. Virshup wrote the manuscript.
Declaration of Interests
Babita Madan and David M. Virshup have a financial interest in ETC-159. The authors have no other competing interests.
Acknowledgements
We acknowledge the assistance of Yunka Wong and other members of the Virshup lab and members of Experimental Therapeutics Centre. We acknowledge Ralph Bunte, D.V.M., for his expert advice with histological analysis, and the assistance of the vivarium staff including Hock Lee.
This research is supported in part by the National Research Foundation Singapore and administered by the Singapore Ministry of Health’s National Medical Research Council under the STAR Award Program to D.M.V. E.P. acknowledges the support of the MRC London Institute of Medical Sciences, Imperial College, London.
Footnotes
Lead Contact: David M. Virshup: david.virshup{at}duke-nus.edu.sg