Wnt proteins synergize to activate β-catenin signaling

Wnts are secreted morphogens that control myriad biological processes during development and adult tissue homeostasis in animals (Chien et al., 2009; Nusse and Varmus, 2012; Yu and Virshup, 2014). Aberrant Wnt signaling is associated with many pathological conditions (Clevers and Nusse, 2012). Nineteen distinct genes in the human genome encode Wnt ligands, which bind to a variety of receptors including ten frizzled proteins (FZDs) and the co-receptors low-density lipoprotein receptor-related protein 5 and 6 (LRP5/6), as well as an increasing number of alternative receptors and co-receptors including the single transmembrane receptors RYK, ROR1/2, PTK7 and GPR124 (also known as ADGRA2) to trigger various downstream signaling pathways (MacDonald et al., 2009; Niehrs, 2012; Posokhova et al., 2015; Zhou and Nathans, 2014). Many Wnt ligands can stimulate β-catenin-induced gene transcription. In some settings, distinct Wnt genes utilize unique promoters and enhancers to drive expression with distinct developmental timing and tissue specificity. However, in both normal and disease states, multiple Wnt genes are often expressed in combination (Akiri et al., 2009; Bafico et al., 2004; Benhaj et al., 2006; Suzuki et al., 2004). For example, stromal cells that support the intestinal stem cell niche express at least six different Wnts at the same time (Kabiri et al., 2014). While in isolated instances, specific Wnt pairs have been shown to combine to enhance β-catenin signaling during embryonic development, whether this is a general phenomenon remains unclear (Cha et al., 2008; Cohen et al., 2012; Miller et al., 2012). Notably, increased expression of multiple Wnt ligands has been described in a number of cancers (Akiri et al., 2009; Bafico et al., 2004; Benhaj et al., 2006; Suzuki et al., 2004). The source of these cancer-associated Wnts is often from the cancer cells themselves, although stromal cells (including fibroblasts and hematopoietic cells) also produce multiple Wnts (Luga et al., 2012; Macheda and Stacker, 2008). Hence, Wnt ligands from multiple sources can converge on target cells in both physiological and pathological settings. How these Wnt ligands interact with each other to regulate β-catenin signaling is not well understood.

In the absence of Wnt ligand–receptor interaction, β-catenin released from membrane complexes is sequentially phosphorylated and targeted for ubiquitylation and proteasomal degradation by a cytosolic ‘destruction complex’ (reviewed in Anastas and Moon, 2013; MacDonald et al., 2009; Yu and Virshup, 2014). Wnt ligands bind to both FZD proteins and LRP5/6 receptors, with different Wnts interacting with two distinct interaction domains on the LRP5/6 receptors (Bourhis et al., 2010; Ettenberg et al., 2010; Gong et al., 2010). This initiates formation of multimeric signalosomes that suppress β-catenin phosphorylation by GSK3 (Gammons et al., 2016; reviewed in Kim et al., 2013; MacDonald et al., 2009). The stabilized β-catenin can then translocate to the nucleus to act as a transcriptional coactivator in conjunction with T cell factor (TCF) and lymphoid enhancer-binding factor (LEF) family transcription factors (reviewed in Cadigan and Waterman, 2012). β-catenin acts as a scaffold to recruit additional proteins to Wnt target gene promoters (reviewed in Mosimann et al., 2009; Valenta et al., 2012). The armadillo repeats of β-catenin recruit activators such as BCL9 and PYGO proteins, while the C-terminus interacts with many proteins involved in chromatin remodeling and transcription initiation including histone acetyltransferases (HATs) such as P300 (also known as EP300), CBP (also known as CREBBP) and KAT5 (also known as TIP60), histone methyltransferases (MLL1 and MLL2, also known as KMT2A and KMT2B), SWI/SNF factors (BRG1 and ISWI, also known as SMARCA4 and SMARCA5, respectively) and members of the PAF complex (reviewed in Mosimann et al., 2009; Valenta et al., 2012).

Using a recently constructed uniform Wnt expression library (Najdi et al., 2012), we screened 18 human Wnts in pairwise combinations for enhanced ability to activate a TCF/LEF reporter. Unexpectedly, multiple Wnt pairs were identified that, when co-expressed, synergize to potentiate β-catenin signaling. The mechanism of synergistic signaling from WNT1 and WNT7B was examined in depth. Synergy required both FZD5 and FZD8, as well as the recently identified WNT7A and WNT7B co-receptors GPR124 and RECK (Posokhova et al., 2015; Zhou and Nathans, 2014). The WNT7B–GPR124 interaction markedly stimulated K49 acetylation of β-catenin. Wnt synergy has important biological consequences, as co-expression of WNT1 and WNT7B regulated the expression of multiple genes and markedly enhanced the tumorigenicity of YCC11 gastric cancer cells. Co-expression of Wnts that synergize through novel pathways may have important combinatorial consequences in β-catenin-driven gene expression in diverse developmental, homeostatic and pathological processes.

We investigated the consequences of Wnt gene co-expression on β-catenin signaling, starting with HEK293 cells with a stably integrated SuperTopFlash reporter (STF, a β-catenin-activated TCF/LEF transcriptional reporter, denoted HEK293-STF; Veeman et al., 2003). Several pairs of Wnts showed synergistic interactions when nanogram quantities of Wnt expression plasmids were transfected (Table S1). For example, WNT7B expression alone did not activate the STF reporter, but it significantly potentiated the activity of both WNT1 and WNT3A (Fig. 1A,B). WNT7A similarly synergized with WNT1 (Fig. S1A). Another Wnt ligand, WNT10B, was found to potentiate the activity of WNT3A (Fig. 1B). Synergy was not simply a consequence of transfection, as it was also observed with purified proteins. Purified recombinant WNT10B, while inactive alone, potentiated signaling stimulated by recombinant WNT3A protein (Fig. 1C).

We found that the choice of Wnt pairs was important. For example, WNT10B did not potentiate the activity of WNT1 (Fig. 1A). This correlates well with the finding that WNT1 and WNT10B interact with the same domain on LRP6, while WNT1 and WNT7B interact with different LRP6 domains (Gong et al., 2010). This effect is general, as Wnt synergy was also seen in YCC11 (gastric cancer) and HeLa (cervical cancer) cells (Fig. 1D,E; Fig. S1B,C). Demonstrating specificity for the Wnt pathway, neither WNT1 nor WNT7B alone nor in combination activated either an nuclear factor (NF)-κB or an AP-1 reporter construct (Fig. S1D,E). Interestingly, WNT10B, which potentiated the activity of WNT3A on the STF reporter in HEK293 cells, did not have this effect in YCC11 cells (Fig. 1E). We speculate that this is due to differences in the Wnt receptors and co-receptors expressed in different cell types. Synergy was not due to increase in Wnt ligand expression upon co-expression of two Wnts (Fig. S6).

To quantitatively assess synergy, we performed Wnt titrations and calculated combination indices (CI) (Chou and Talalay, 1983, 1984) for the WNT combinations tested (WNT1 and WNT7B; WNT3A and WNT10B) (described in detail in the Materials and Methods). A combination is defined as additive when the CI=1, synergistic when the CI<1 and antagonistic if the CI>1. Both the combination of WNT1 and 7B and the combination of WNT3A and 10B had a CI of <<1, indicating they are highly synergistic (Fig. 1F).

In testing various pairwise combinations, again using nanogram quantities of Wnt expression plasmids (Najdi et al., 2012), we made several other intriguing observations. We found that two ‘non-canonical’ Wnts that did not activate Wnt/β-catenin signaling on their own in HEK293 cells could do so when co-expressed. WNT2, WNT4, WNT9A and WNT9B did not activate the STF reporter by themselves in HEK293 cells, but WNT9B, and to lesser extent, WNT9A, synergistically activated signaling when expressed in combination with WNT2 or WNT4 (Fig. 1F,G). In addition, WNT6 and WNT8A, which interact with the same LRP6 domain as WNT10B (Gong et al., 2010), were found to inhibit the activity of WNT10B in HEK293 cells (Fig. S1F,G). Therefore, multiple sets of Wnts can interact in different ways to regulate β-catenin signaling activity in cells.

We examined the consequences and requirement of the WNT1 and WNT7B interaction in more detail. We tested whether WNT1 and WNT7B co-operate in regulating the expression of endogenous Wnt/β-catenin target genes. The expression of AXIN2, LEF1, NKD1 and TCF7 was assessed in YCC11 cells by quantitative real-time RT-PCR (qRT-PCR) (Fig. 2A–D). WNT1 and WNT7B alone each increased expression of these endogenous genes. In all cases, combined transfection of 100 ng of each Wnt plasmid produced at least twice the expression of the endogenous target genes as 200 ng of a single Wnt. Combination indices were calculated for these pairs as described above and a CI of <<1 was obtained for all the target genes (AXIN2, LEF1, NKD1 and TCF7). Thus, Wnt synergy occurs on both a model reporter and on bona fide endogenous Wnt target genes.

To identify additional Wnt target genes that are synergistically regulated by combined WNT1 and WNT7B signaling, we performed a global transcriptome analysis by RNA-seq on YCC11 cells transfected with plasmids expressing WNT1 (200 ng), WNT7B (200 ng) or both WNT1 and WNT7B (100 ng each). This identified multiple additional genes synergistically up- and down-regulated by the WNT1 and WNT7B combination (Tables S2, S3; Fig. S1H,I,J). Taken together with the finding that multiple Wnt pairs can synergize, this result suggests that Wnt combinations have the potential to regulate the expression of multiple downstream pathways.

FZDs are the primary Wnt receptors on the cell surface and trigger various downstream signaling cascades based on the co-receptors involved (Niehrs, 2012). To test their involvement in WNT1 and WNT7B synergistic signaling, we individually knocked down each FZD receptor that was expressed in both HEK293 and YCC11 cells (FZD2, FZD3, FZD4, FZD5, FZD6, FZD7 and FZD8) using two independent siRNAs. Knockdown of either FZD5 or FZD8 reduced both the WNT1 signal as well as WNT1 and WNT7B synergy in HEK293-STF cells (Fig. 3A). There was no effect of the knockdown of other FZDs. Synergistic activation of endogenous AXIN2 expression in YCC11 cells was likewise dependent on FZD5 and FZD8 expression (Fig. S1K). Consistent with a central role for FZD5 and FZD8 in combined Wnt signaling, their overexpression increased both the basal WNT7B signal and the synergistic signal (Fig. 3B). Thus, both FZD5 and FZD8 are required and rate limiting for WNT1 and WNT7B signaling.

GPR124 was recently identified as a co-receptor for WNT7A and WNT7B involved in Wnt/β-catenin signaling during central nervous system (CNS) angiogenesis (Posokhova et al., 2015; Zhou and Nathans, 2014). We therefore examined the involvement of GPR124 in WNT7B signaling and WNT1 and WNT7B synergy. Consistent with published reports, siRNA-mediated knockdown of GPR124 in HEK293-STF and YCC11 cells abrogated the WNT7B-induced STF reporter activity (Fig. 4A,B). Similar results were seen with multiple independent siRNAs (Fig. S2A–C). GPR124 knockdown also abrogated the induction of AXIN2 mRNA by WNT7B in YCC11 cells (Fig. 4C). Importantly, knockdown of GPR124 had no effect on WNT1 signaling, but markedly reduced WNT1 and WNT7B synergy (Fig. 4A–C; Fig. S2A). Conversely, overexpression of GPR124 increased WNT7B signaling in both HEK293-STF and YCC11 cells (Fig. S2E,F) and increased WNT1 and WNT7B synergy (Fig. S2G). We conclude that WNT1 and WNT7B synergy requires engagement of GPR124 by WNT7B.

The glycophosphatidylinositol (GPI)-anchored membrane protein RECK has also been shown to be a part of the WNT7–GPR124 complex at the membrane required for β-catenin signaling (Vanhollebeke et al., 2015). Consistent with this, siRNA-mediated knockdown of RECK also decreased WNT7B-induced luciferase reporter activity and synergy in both HEK293-STF and YCC11 cells (Fig. 4D,E; Fig. S2D) and reduced WNT7B-induced AXIN2 mRNA expression in YCC11 cells (Fig. 4F). Therefore, WNT7B signaling at the membrane requires GPR124 and RECK along with FZD5 and FZD8 to signal alone and to potentiate the WNT1-induced signaling.

To investigate the mechanism of WNT1 and WNT7B cooperation, events downstream of the Wnt–receptor interaction were examined. Prior to the identification of GPR124 as a WNT7A and WNT7B co-receptor, Miller et al. reported that WNT7A and WNT7B cooperate in foregut development, and this cooperation was mediated through the PDGF pathway (Miller et al., 2012). However, while we found that GPR124 knockdown reduced synergy, in our system inhibition of PDGFR (and other growth factor receptor tyrosine kinases) had no effect on WNT1 and WNT7B synergy (Fig. S2H,I).

Binding of Wnt ligands to their FZD receptors and LRP6 co-receptor leads to LRP6 phosphorylation followed by inactivation of the β-catenin destruction complex (Clevers and Nusse, 2012; MacDonald et al., 2009; Yu and Virshup, 2014). To test the role of LRP6 in WNT1 and WNT7B synergy, we used DKK1, which binds to and blocks LRP6 function. Recombinant DKK1 decreased total Wnt/β-catenin signaling but did not reduce synergy of the residual signal (Fig. 5A). LRP6 S1490 phosphorylation was increased by WNT1 but not WNT7B; however, the combination of Wnts did not lead to additional LRP6 phosphorylation (Fig. 5B; Fig. S3A, Fig. S8). Thus, synergy does not appear to occur at the level of LRP6 activation. The WNT3A and WNT10B synergy on the other hand was inhibited by DKK1 (Fig. S3B) indicating that the WNT3A and WNT10B synergy may require LRP6 engagement via different propeller domains (Ettenberg et al., 2010; Gong et al., 2010). Supporting a central role for LRP6 in Wnt synergy, DKK1 also inhibited WNT2 or WNT4 synergistic signaling with WNT9B (Fig. S3C).

As expected, we found that β-catenin is required for signaling and synergy. siRNA-mediated knockdown of β-catenin abrogated the WNT1 signal (Fig. 5C). However, the residual β-catenin still functions better in the presence of combined WNT1 and WNT7B expression. To test whether the synergistic signal is due to increased stabilized β-catenin, we probed for total β-catenin by western blotting. However, WNT7B co-expression did not cause any further increase in stabilized β-catenin beyond that caused by expression of WNT1 alone (Fig. 5B).

After cytoplasmic stabilization, nuclear translocation of β-catenin is the next step in signaling. Surprisingly, there was no additional increase in nuclear β-catenin with combined WNT1 and WNT7B co-expression in HEK293 cells (Fig. 5D). We also examined the levels of non-phosphorylated ‘active’ β-catenin and found that there was no increase in ‘active’ β-catenin upon WNT1 and WNT7B co-expression (Fig. 5E). We therefore considered whether WNT7B directly enhanced the activity of nuclear β-catenin. WNT7B expression, which alone produces no STF activity in HEK293 cells, enhanced the activity of a stabilized β-catenin with an S45A mutation (Fig. 5F), albeit to a lesser extent (<1.5-fold) than the combination of WNT1 and WNT7B (>3-fold). Calculation of combination indices for WNT7B and β-catenin-S45A also yielded a CI of <<1, indicative of synergy. Interestingly, this effect is not unique to WNT7B as several other ‘non-canonical’ Wnts (WNT2, WNT9B and WNT10B) that did not signal alone could also modestly activate β-catenin-S45A signaling (Fig. 5G). These results suggest that the ‘non-signaling’ Wnts (WNT2, WNT7B, WNT9B and WNT10B) can signal directly to stabilized β-catenin, but that formation of a two-Wnt signaling complex at the membrane is a much more potent means of generating synergy in the nucleus.

We next examined events downstream of β-catenin stabilization and nuclear entry. Recruitment of CBP/p300 to β-catenin is one of the key events that occurs after β-catenin translocates to the nucleus and has been proposed to be important for the transcriptional activity of the β-catenin–TCF complex (Levy et al., 2004). The acetylation status of β-catenin was therefore examined by immunoblotting with a β-catenin K49Ac-specific antibody. WNT1 and WNT7B individually each induced acetylation of β-catenin to a greater or lesser extent in YCC11 and HEK293 cells (Fig. 6A,B; Fig. S7). However, β-catenin K49Ac was markedly increased when WNT1 and WNT7B were co-expressed. This suggests an increase in specific nuclear acetyltransferase activity or a decrease in deacetylase activity in response to WNT7B signaling.

We tested whether CBP was responsible for the synergistic signaling. CBP acetylates β-catenin on K49 (Wolf et al., 2002). Treatment with the β-catenin–CBP interaction inhibitor ICG-001 (Emami et al., 2004) indeed globally decreased β-catenin signaling. However, the IC₅₀ of inhibition of signaling by ICG-001 was the same for WNT1 alone as for the WNT1 plus WNT7B synergistic signal (Fig. S3D). Since ICG-001 is a binding inhibitor, this suggests there is no increase in the CBP bound to the complex during synergistic signaling.

GPR124 is the co-receptor responsible for the WNT7B signaling in HEK293 and YCC11 cells. GPR124 knockdown decreased both the WNT7B-induced β-catenin K49Ac, and the synergistic β-catenin acetylation induced by WNT1 and WNT7B (Fig. 6C; Fig. S3E). Thus, the WNT7B interaction with GPR124 activates β-catenin acetylation and synergistic β-catenin signaling. Ongoing studies aim to define the signaling pathway from GPR124 to nuclear protein acetylation.

The functional importance of β-catenin acetylation is not well established. To test whether modification of β-catenin on K49 is required for synergy, endogenous β-catenin was knocked down using siRNA and rescued with either siRNA-resistant wild-type β-catenin or the K49R mutant. We found that the K49R acetylation mutant of β-catenin could signal and synergize with WNT7B as well as with the wild type (Fig. S5A). Similarly, using a second approach, we found that WNT7B could synergize with the K49R mutant of stabilized β-catenin S45A to the same extent as it does with the wild-type β-catenin S45A (Fig. S5C,D). In both instances, the K49R mutation had no effect on β-catenin signaling. These findings suggest that Wnt synergy enhances nuclear protein acetylation, and that acetylation of β-catenin K49 correlates with, but is not essential for, Wnt synergy.

Multiple Wnt genes appear in the genomes of all metazoans including sponges, suggesting synergy between Wnts may be a central feature of this pathway (Nichols et al., 2006). To test whether increased co-expression of Wnt genes might play a role in cancer, we assessed the expression of Wnt genes in a cohort of 201 primary gastric cancer samples previously stratified by gene expression signature into mesenchymal (Fig. 7, orange), proliferative (Fig. 7, cyan) and metabolic (Fig. 7, purple) groups (Lei et al., 2013). The data suggest that there is increased expression of distinct combinations of specific Wnt ligands in different subsets of gastric cancers (Fig. 7A). For example, WNT2B and WNT9A expression clusters in the mesenchymal group, WNT5A is expressed primarily in the proliferative group, and multiple Wnts including WNT1, WNT3 and WNT7 are expressed in the metabolic group. Wnt gene expression can be in either cancer cells themselves, and/or from cells in the stroma. Additionally, Wnt gene expression in cancers and stroma may be induced by the interactions in the local microenvironment. Consistent with a key role for stroma, expression profiling of 37 gastric cancer cell lines (Ooi et al., 2009) grown in plastic in the absence of stroma did not show similar Wnt-high subsets (Fig. S4). We next asked whether significant Wnt gene expression occurs in tumor stroma. We examined six gastric cancer patient-derived xenografts propagated in NSG mice, where the cancer cells are human but the stroma is murine. Using human- or mouse-specific PCR primers (Table S4), we found multiple but different mouse (i.e. stromal) Wnts were upregulated, with samples GC38, GC84 and GC47 having high human and mouse Wnt expression, and GC72, GC66 and GC45 having low expression (Fig. 7B,C). Taken together, the data suggest that Wnt synergy can occur in a subset of gastric cancers that have coordinate upregulation of multiple Wnts produced both in cancer and stromal cells.

As multiple Wnts are co-expressed in a subset of gastric cancers, we tested the biological consequences of WNT1 and WNT7B synergy in the gastric cancer cell line YCC11. WNT1 alone, WNT7B alone or WNT1 plus WNT7B were stably expressed in YCC11 cells by lentiviral transduction and the ability of these cells to form colonies in soft agar was assessed (Fig. 8). Cells expressing WNT1 alone were hindered in their ability to form colonies, while WNT7B expression alone had no effect. The combination of WNT1 and WNT7B doubled the number of colonies and also increased their size. The data are consistent with Wnt co-expression promoting tumorigenesis by synergistic regulation of downstream target genes.

Multiple Wnts potently regulate diverse biological responses. Wnt genes are often coordinately expressed, but they are most often studied individually. Here, using a library of 19 Wnts, we assessed interactions between multiple Wnts and find synergy to be a general phenomenon. We find that specific combinations of Wnts signal synergistically through both the well-known FZD5 and FZD8 receptors and the recently described WNT7 receptor GPR124 and its co-receptor RECK. One surprising consequence of synergy between WNT1 and WNT7A occurs at the level of GPR124-dependent increased β-catenin acetylation. The data suggest a model whereby ‘non-canonical’ Wnts such as WNT7B interact with specific FZDs and LRP5/6 in the presence of a third co-receptor, in this case GPR124, to signal through currently unknown pathways to enhance β-catenin activity, nuclear lysine acetylation and, hence, gene expression.

GPR124 is a putative G-protein-coupled receptor required for angiogenesis in the brain, and was recently shown to be a co-receptor for WNT7 family members (Posokhova et al., 2015; Zhou and Nathans, 2014). GPR124 is likely to be important in vasculature outside of the brain, as it was also identified as tumor endothelial marker 5 (TEM5), a transcript enriched in the vasculature of human colorectal cancer and murine tumors (Carson-Walter et al., 2001) and which is itself upregulated by TGF-β (Anderson et al., 2011) and Rac signaling (Vallon et al., 2010). Whether activated GPR124 couples to a G-protein, or whether it signals to the nucleus via other mechanisms is not currently known.

β-catenin acetylation by CBP/p300 has been frequently observed during canonical Wnt/β-catenin pathway activation (Hecht et al., 2000; Levy et al., 2004) and may play a role in signaling at β-catenin-responsive promoters. Our data suggest that WNT7B and GPR124 signaling either increases the activity of the β-catenin K49 acetyltransferase (CBP) (Wolf et al., 2002), or decreases the activity of the deacetylase. Of course, changes in the activity of these enzymes may also regulate the acetylation of other important lysine residues leading to changes in promoter activation. We have not examined the acetylation status of other lysine residues on β-catenin (e.g. K345) due to lack of good antibodies. Involvement of other HATs that may acetylate additional lysine residues on β-catenin or other proteins cannot be ruled out based on our current data. Additional work is needed to clarify the landscape of lysine acetylation changes in response to WNT7B and GPR124 signaling. We found the acetylation of β-catenin on lysine 49 to be a consequence of increased acetyltransferase (or reduced deacetylase) activity and that it was not required for synergy as the K49R mutant of β-catenin could signal and synergize with WNT7B as well as did the wild type (Fig. S5).

Miller et al. (2012) reported that WNT2 and WNT7B cooperate in foregut development. They found that WNT2 and WNT7B cooperation was specific to the mesenchymal cell lineage and did not occur in epithelial cells. We confirmed that WNT2 and WNT7B did not synergize in HEK293 cells. Conversely, they did not see synergy between WNT1 and 7B in mesenchymal cells, while we found robust synergy between these Wnts in multiple cell types of epithelial origin. The differences may well be related to expression patterns of known and novel Wnt receptors in various cell types. Further synergistic partners may be discovered by conducting similar screens in different cell types or model systems.

One clue to mechanism is the correlation between synergy pairs and the ability of different Wnts to bind to different domains on LRP6 (Bourhis et al., 2010; Ettenberg et al., 2010; Gong et al., 2010). We speculate that the synergy of the Wnt pairs may be due to the formation of a multimeric complex comprising multiple LRP6/WNT/FZD and alternative receptors such as GPR124/RECK at the membrane. A speculative model (based on Bourhis et al., 2010) is presented in Fig. 8D.

Wnt ligand synergy in activating the Wnt/β-catenin (and possibly other) pathways has substantial physiological relevance as many of these ligands are co-expressed at low individual levels in various organs (Farin et al., 2012) and during development (Witte et al., 2009). WNT7B has been shown to be involved in the development of lung (Rajagopal et al., 2008), kidney (Yu et al., 2009), nervous system (Stamatakou and Salinas, 2014) and pancreas (Afelik et al., 2015), and is implicated in the pathogenesis of breast (Yeo et al., 2014), pancreatic (Arensman et al., 2014), prostate (Zheng et al., 2013) and bladder cancer (Bui et al., 1998). The ability of WNT7B to potentiate the activity of canonical Wnts may play a significant part in its contribution towards these processes of development and disease progression. The synergy with other Wnts may be needed for proper pathway activation in vivo. This provides a further regulatory mechanism for fine-tuning Wnt signaling activity in time and space. Intersecting fields of Wnt expression can focus signaling into discrete areas. Wnt synergy may also explain how high β-catenin signaling activity can be maintained without concurrent high expression of any single ligand. Since WNT7B can further activate a stabilized mutant of β-catenin (S45A), this provides an additional route for Wnt ligands to function in cancers with stabilized β-catenin and perhaps even adenomatous polyposis coli (APC) mutations. The study of Wnt interactions will provide additional insights into the complex role that Wnts play in development and disease.

HEK 293-STF cells (with the stably integrated Super8×TOPFlash reporter) were a kind gift from Kang Zhang (Institute for Genomic Medicine, University of California, San Diego, La Jolla, CA) (Xu et al., 2004). HeLa cells were obtained from American Type Culture Collection (ATCC). YCC11 cells were a kind gift from Dr Patrick Tan (Programme in Cancer and Stem Cell Biology, Duke-NUS Medical School, Singapore). Authenticity of the cell lines was not validated. The cells were routinely tested and regularly confirmed to be mycoplasma free. All cells were grown in Dulbecco's modified Eagle's medium (DMEM; Nacalai Tesque) containing 4.5 g/l glucose, penicillin-streptomycin, 10% fetal bovine serum (FBS) and 1 mM sodium pyruvate in a humidified incubator with 5% CO₂. Recombinant DKK1, human WNT3A and human WNT10B were purchased from R&D Systems (Minneapolis, MN). Super8×TOPFlash reporter (Addgene Plasmid #12456; deposited by Randall Moon; Veeman et al., 2003). Expression plasmids for FZD5 and FZD8 were a kind gift from Jeffrey Rubin (Laboratory of Cellular and Molecular Biology, NCI, NIH, Bethesda, MD). The expression plasmid for GPR124 was a kind gift from Bradley St Croix [Tumor Angiogenesis Section, Mouse Cancer Genetics Program (MCGP), NCI at Frederick, NIH, Frederick, MD] (Posokhova et al., 2015). Wnt expression constructs were prepared as part of the Open Source Wnt project (Addgene Kit #1000000022) (Najdi et al., 2012). The plasmid for β-catenin expression, pCS2+Myc:β-catenin, was mutated to be resistant to β-catenin siRNA #11 (see below) and K49R using site-directed mutagenesis. The plasmid for stabilized β-catenin, pCS2+Myc:β-catenin S45A, was mutated to K49R by using site-directed mutagenesis. All patient samples were collected with informed patient consent from National University Hospital Singapore according to the National Healthcare Group Domain Specific Review Board (DSRB) guidelines (DSRB-B/07/367). All animal experiments were conducted with the approval of Institutional Animal Care and Use Committee (IACUC) in the National University of Singapore.

We utilized previously reported gene expression data for 201 primary gastric tumors (Gene Expression Omnibus GSE15459 and GSE34942) (Lei et al., 2013). Based on the gene expression profiles, these 201 gastric tumors were classified into three molecular subtypes: mesenchymal (orange), proliferative (cyan), and metabolic (purple) (Fig. 7). We extracted WNT genes (probesets) from the gene expression profiles and generated a heatmap on the gene expression values (robust multi-array average) with clustering on WNT genes. We obtained previously reported gene expression data for 37 gastric cancer cell lines (Gene Expression Omnibus GSE22183) (Ooi et al., 2009). All the gastric cancer cell lines except MKN7 were classified into one of the three molecular subtypes using the gastric cancer classifier GC-Class developed in Lei et al. (2013), while MKN7 was unclassifiable. Similarly, we generated the heatmap using the WNT gene expression values with clustering on WNT genes (Fig. S4).

SuperTOPFlash (β-catenin-activated TCF/LEF transcriptional reporter; STF) assays were performed in 24-well plates, transfecting 400 ng of total plasmid/well, composed of indicated amounts of WNT, 100 ng of mCherry expression plasmid and 100 ng of SuperTOPFlash (where needed) by using Lipofectamine 2000 (11668019, ThermoFisher Scientific). Lysates were prepared in PBS with 0.6% NP-40 with complete protease inhibitor cocktail without EDTA (Sigma-Aldrich, St. Louis, MO) and firefly luciferase activity was measured with a luciferase assay kit according to the manufacturer's recommendations (Promega, Madison, WI) by using a Tecan Infinite M200 plate reader (Tecan Trading AG, Switzerland). All assays unless otherwise indicated were performed in triplicate and graphed as mean±s.d. Each experiment was repeated at least three times with similar results.

For siRNA-mediated knockdown experiments, cells were plated in 12- or 24-well plates and transfected with the indicated siRNAs (at indicated concentration) using Lipofectamine 2000 (ThermoFisher Scientific, Waltham, MA). Wnt expression plasmids were transfected 48 h later and cells were harvested after a further 24 h incubation and analyzed for luciferase activity or by qRT-PCR. The siRNAs used are as follows: control/non-targeting (#D-001810-01-05), β-catenin (#J-093415-11, 5′-GCGTTTGGCTGAACCATCA-3′), FZD8 (#L-003962, #1: 5′-AGACAGGCCAGATCGCTAA-3′, #3: 5′-TCACCGTGCCGCTGTGTAA), GPR124 (#L-005540, #5: 5′-GAGCGAAACTACCGTCTAA-3′, #8: 5′-CGACTAAACATATCTGGAA-3′) and RECK (#L-011474) were from GE-Dharmacon (Lafayette, CO). The targeting sequences for the other FZD siRNAs are as follows: FZD2 (5′-CGGTCTACATGATCAAATA), FZD5 (#1: 5′-TCCTCTGCATGGATTACAA-3′, #2: 5′-AGACGGACAAGCTGGAGAA-3′).

The extent of synergy between two factors (WNT1 and WNT7B, or WNT3A and WNT10B) was quantified using by determining the combination index (Chou and Talalay, 1983, 1984). The combination index, CI, was originally conceptualized in combination therapy to study the extent of interaction between two drugs, where CI=1 indicates purely additive effects. The CI is derived from the median-effect equation which is a unified theory of multiple mass-action equations (Chou, 2006). A combination is classified as synergistic when CI<1, or as antagonistic when CI>1.

To compute the CI for pairs of canonical and non-canonical Wnts, such as WNT1 and WNT7B, or WNT3A and WNT10B, we first obtained dose–response curves for each Wnt individually. We then stimulated the cells with pairs of Wnt ligands, using a matrix of dose combinations, and we measured the effects with the same SuperTOPFlash assay. As the CI for a pair of drugs is not necessarily constant across different doses, we used the matrix of dose combinations to compute the combination index (CI) for each treatment (Fig. 1F). An alternative method would have been to use serial dilutions of combination at a constant ratio, such as D_can:D_non-can=(IC₅₀)_can:(IC₅₀)_non-can (where ‘can’ is the ‘canonical Wnt’). Constant ratio experiments permit simulation of estimated CIs for a generalized range of dose combinations. However, they require that different batches of cells be used for determining the IC₅₀ and for quantifying the combination effects. This introduces batch effects, which are problematic to analyze. More importantly, the non-canonical Wnts have a very weak SuperTOPFlash response by themselves, causing their IC₅₀ to be far beyond the physiological range, and producing a poor signal-to-noise ratio in the non-canonical single-Wnt curve. In contrast, our matrix design (non-constant ratio combinations) allowed us to use dose combinations that are of practical relevance. We then used the Chou–Talalay method to quantify the CI for those experimentally verified combination data points.

Responses to combination treatments (referred to as f_A) were normalized such that 1.0 would represent the maximum response to a saturating dose. The theoretical maximum response to each individual Wnt was estimated by assuming that the maximum measured effect was 80% of the saturated maximum effect. To verify the impact of the 80% assumption, we repeated the CI analysis over a range of assumed percentages (10, 20, 30…100%) and found that the calculated CI values and CI trends were insensitive to the assumption.

The computation of CI requires computing a linear fit between log(dose) versus log[f_A/(1−f_A)] for each of the individual stimuli. Lines with higher slope (higher efficacy of each dose) yield lower synergism when computing the CI. Since the exact linear relationship is uncertain (e.g. due to experimental error) and the CI values might be sensitive to the slope of the fitted line, we fitted an ensemble of lines to the points (for each Wnt) and selected the one with maximum slope (which would give the lowest synergy), to be the most conservative linear fit for the relationship between log(dose) and log[f_A/(1−f_A)].

RNA was isolated using an RNeasy kit (cat. #74106, Qiagen, Hilden, Germany), reverse transcribed with the i-script RT kit (cat. #170-8891, Bio-Rad Hercules, CA), and quantified on a Bio-Rad CFX96 real-time cycling machine using the SsoFast EvaGreen PCR assay (cat #172-5200, Bio-Rad). Sequences for the human- and mouse-specific Wnt and housekeeping primers and other primers used are given in Table S4.

Cells (HEK293 or YCC11) were transfected with the indicated Wnt expression plasmids and lysates were made 24 h post transfection. Nuclear and cytoplasmic extracts were prepared using the NE-PER Nuclear and Cytoplasmic Extraction Kit (PI-78835, ThermoFisher Scientific, Waltham, MA) following the manufacturer’s recommendations. Whole-cell lysates were made in 50 mM Tris-HCl, 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS, 1 mM DTT and protease inhibitor cocktail (Sigma-Aldrich, St. Louis, MO). The proteins were separated on 8 or 10% SDS polyacrylamide gels and transferred onto PVDF membrane (Hybond, GE Healthcare, Pittsburgh, PA). Membranes were blocked in 5% BSA in 1× TBS with 0.1% Tween-20 for 1 h and incubated overnight at 4°C with primary antibodies. Antibody against β-catenin (cat. #610154) was from BD Biosciences (Franklin Lakes, NJ) and was used at 1:1000 dilution. Antibodies against acetylated β-catenin (K49, cat. #9030), non-phosphorylated (active) β-catenin (cat. #8814), LRP6 (cat. #2560), phospho-LRP6 (Ser1490, cat. #2568) were from Cell Signaling Technology (Danvers, MA) and were used at 1:1000 dilution. Antibody against Eg5 (4H3-IF12) was from Cell Signaling (cat. #4203) and was used at 1:2000 dilution. Antibodies against actin (cat. #3280) and tubulin (cat. #52623) were from Abcam (Cambridge, UK) and were at 1:1000 dilution, and antibody against lamin b (cat. # SC-6213), used at 1:1000 dilution, was from Santa Cruz Biotechnology (Dallas, TX). Antibody against V5 was from Invitrogen (cat. #R96025) and was used at 1:5000 dilution. Antibody against WNT1 was from Genetex (cat #GTX111182) and was used at 1:1000 dilution and anti-WNT3A antibody was a generous gift from Shinji Takada (Okazaki Institute for Integrative Bioscience and National Institute for Basic Biology, Okazaki, Japan) (culture supernatant used at 1:100 dilution).

After incubation with primary antibodies, blots were washed and incubated with horseradish peroxidase-conjugated (Bio-Rad; dilution 1:5000) or IR dye-conjugated secondary antibodies (ThermoFisher Scientific; dilution 1:15,000). The blots were detected using SuperSignal West Dura substrate for chemiluminescence (cat. #34075, ThermoFisher Scientific, Waltham, MA). The membranes were imaged for chemiluminescence (ImageQuant LAS 4000, GE Healthcare, Pittsburgh, PA) or fluorescence (Odyssey, LI-COR, Lincoln, NE).

YCC11 cells were infected with lentiviral constructs expressing human WNT1, WNT7B or both and stable cell lines generated using antibiotic selection (puromycin and/or blasticidin). YCC11 cells stably expressing human WNT1, WNT7B or both were seeded in soft agar (bottom 0.6%, top 0.36%) in 24-well plates at a density of 5000 cells per well. The medium was replenished twice a week and colonies were quantified after 4 weeks using Crystal Violet staining and manual counting.

Statistical analysis was performed using GraphPad Prism version 5 for Mac (GraphPad Software, La Jolla, CA) using one-way or two-way ANOVA, correcting for multiple comparisons using Tukey's test. Significance for all tests was set at P<0.05 unless otherwise stated.

We thank Chan Shing Leng of Cancer Science Institute, Singapore for assistance with the gastric cancer patient-derived xenograft tumor tissues, and Patrick Tan for assistance with gastric cancer gene expression studies.