Application of an integrated physical and functional screening approach to identify inhibitors of the Wnt pathway

Bryan W Miller^1,2, Garnet Lau^1,2, Chris Grouios³, Emanuela Mollica¹, Miriam Barrios-Rodiles^5,6, Yongmei Liu⁵, Alessandro Datti⁶, Quaid Morris^2,3,4,7, Jeffrey L Wrana^5,6,7 & Liliana Attisano^1,2

Article highlights

Synopsis

Large-scale proteomic approaches have been used successfully to study signalling pathways (Kabuyama et al, 2004; Moffat and Sabatini, 2006). However, identification of biologically relevant hits from a single screen remains challenging due to limitations inherent in each individual approach (Hakes et al, 2008; Stelzl and Wanker, 2006). To overcome these limitations, we implemented an integrated, multi-dimensional approach and used it to identify Wnt pathway modulators.

For the physical mapping of a Wnt interactome, we used the LUMIER protein-protein interaction method (Barrios-Rodiles et al, 2005), which measures the interaction of a Firefly luciferase-tagged bait with Flag-tagged preys in mammalian cells (Figure 2A). Eleven cytoplasmic mediators of Wnt signals, -catenin, Dishevelled (Dvl) 1, 2 and 3, Axin1 and 2, glycogen synthase kinase (GSK) 3, casein kinase (CK) 1 and 1 and Naked (Nkd) 1 and 2, were tagged with Firefly luciferase for use as baits and screened for interactions with a library of 640 3XFlag-tagged cDNAs. The data were used to generate a network map comprised of 829 protein-protein interactions between 11 baits and 265 Flag-tagged preys (Figure 2B and C).

Figure 2

High-throughput screening of protein–protein interactions of the Wnt pathway. (A) The LUMIER assay involves coexpression of a Flag-tagged prey and Firefly luciferase (FF-Luc)-tagged bait in mammalian cells. An interaction is detected by -Flag immunoprecipitation followed by measurement of luciferase activity. (B) Results of the LUMIER screen of the Wnt pathway. Luciferase-tagged baits (left) were screened against 640 Flag-tagged cDNAs (numbered across top) in the absence (-) or presence (+) of Wnt3A-conditioned media. The mLIR score for each interaction is represented by colors according to the indicated scale. (C) The Wnt interactome as defined by LUMIER. Interactions with an mLIR >2 in at least 2 of 3 runs or in the single Wnt3A run, and with a total average mLIR of >2 are visualized with Cytoscape (Shannon et al, 2003). The complete network is shown as a circular representation with baits represented by large yellow nodes, preys as small nodes and interactions by edges.

Full figure and legend (647K)Figures & Tables index

In order to complement this interaction map, we carried out two functional screens to determine the effect of altering the expression of each of the proteins encoded by the Flag-tagged cDNA library on Wnt-dependent transcription. As a functional readout of the canonical Wnt pathway, we used the Wnt3A-responsive transcriptional reporter, TOPflash, which is comprised of three LEF/TCF binding sites located upstream of the Firefly luciferase gene. We first examined the effect on transcription upon protein overexpression by transfecting cells with plasmids encoding each individual Flag-tagged cDNA. Next, we examined the effect on signalling upon abrogation of the expression of each of these genes using siRNAs. Each screen was run in the presence and absence of Wnt3A ligand and analysis of normalised data revealed both known and novel Wnt pathway regulators.

Initial comparison of the two functional screens and the protein interaction screen suggested that a number of hits were common in multiple screens (Figure 4A). To integrate the three data sets, we developed the combined pathway (CP) score, a value that reflects the likelihood that a tested gene is a component of the signalling pathway of interest. For this, P-values were derived from each data set and converted to a condition-specific confidence score for each individual gene, ranging from 0 to 15. For each gene, this resulted in 15 individual scores; one for each bait in LUMIER, and two each for the functional screens run with and without Wnt3A. To ensure equal weighting of the three screens, we selected the maximum score achieved in each screen, and the sum of the three values was used to calculate the CP score. Analysis of the ranked scores revealed that while only 8.6% of all genes achieved a score of 15 or greater, over 89% of well-established key Wnt pathway components present in the screen scored 15 or higher, demonstrating that CP scoring is highly predictive for Wnt pathway components (Figure 4B). Additional analysis of performance revealed that the CPS outperformed each of the individual screens in identifying true positives while maintaining low false positive rates (Figure 4C). Of the top ranked genes, we selected Ube2m and Nkd1 for further characterisation.

Figure 4

Integration of data from multiple screens. (A) A proportional Venn diagram (prepared with www.cs.kent.ac.uk/people/staff
/pjr/EulerVennCircles/EulerVennApplet.html), indicating the number of hits in the LUMIER (as defined in Figure 2), RNAi and cDNA overexpression (where overexpression or knockdown gave an average fold over median >1.5 or <0.6 in the presence or absence of Wnt3A) and the degree of overlap. (B) Distribution of combined pathway scores (CPSs) of integrated screen data. The percentage of core (Group 1) and known (Group 2) components of the Wnt pathway and all tested genes (All) corresponding to the indicated CPS thresholds are shown. (C) Receiver operating characteristic (ROC) curves assessing screen and CPS performance. The sensitivity (true positives) versus 1-specificity (false positives) is plotted for the integrated (CPS) and individual screens (cDNA, LUMIER and RNAi).

Full figure and legend (200K)Figures & Tables index

Ube2m is a member of the E2 ubiquitin-conjugating enzyme family that transfer Nedd8, a ubiquitin-like moiety, to target proteins (Petroski and Deshaies, 2005). Consistent with the screen data, we found that Ube2m interacts with -catenin and that Ube2m overexpression activates TOPflash. Ube2m overexpression increased both total -catenin and phospho--catenin levels. As phospho--catenin is an intrinsically unstable form that is targetted for degradation, the results suggested that the increase in -catenin levels may be due to a disruption in the SCF complex responsible for -catenin ubiquitination. As function of this complex is dependent on neddylation, we tested the effect of increasing Nedd8 levels in the presence of Ube2m overexpression and found that Nedd8 overexpression reversed the effect of Ube2m on TOPflash. These data suggest that Ube2m regulates -catenin expression by recruiting -catenin to the SCF complex, and that increasing Ube2mexpression impairs the ability of this complex to degrade -catenin.

We also characterised the role of Nkd1 as an inhibitor of Wnt signalling. Nkd1 was previously reported to be a Wnt inhibitor that interacts with Dvl (Yan et al, 2001a, b), and consistent with this, our screening data showed that Nkd1 inhibits TOPflash and binds all three Dvls. In addition, we detected a novel interaction between Nkd1 and Axin. The histidine rich tail of Nkd1 was found to be crucial for association with Axin, and for inhibition of Wnt-dependent TOPflash activity. To test whether the association of Nkd1 with Axin was required for the inhibitory activity of Nkd1 on Wnt signalling, we tested the effect of Nkd1 overexpression in the presence and absence of Axin. Knockdown of Axin impaired the ability of Nkd1 to repress TOPflash, suggesting that this interaction is crucial for the repressive effect of Nkd1 on Wnt transcriptional activity. As both Nkd1 and Axin are Wnt responsive genes this suggests a model whereby Wnt induces expression of both inhibitors, which then cooperate to downregulate Wnt pathway signals.

In summary, determination of a Combined Pathway Score (CPS) allows integration of high throughput screens that examine physical and functional behaviors of individual proteins, and thereby enhances the selection of physiologically relevant pathway components. The novel insights gained into Wnt signalling using CP scoring demonstrates the power of our integrated mammalian-cell based physical and functional mapping method. In addition to Wnt, this approach can be applied to diverse signalling pathways to facilitate the identification of novel regulators and gain fresh insight into pathway function.