Figure 1. TWIST1 interactome in cranial NCCs revealed using BioID and network propagation.

(A) BioID procedure to identify TWIST1-interacting partners in neural crest stem cells (NCCs). TWIST1-BirA* (TWIST1 fused to the BirA* biotin ligase) labeled the proteins partners within the 10 nm proximity in live cells. Following cell lysis and sonication, streptavidin beads were used to capture denatured biotin-labeled proteins, which were purified and processed for mass spectrometry analysis. (B) TWIST1-specific interaction candidates identified by BioID mass-spectrometry analysis in NCC cell line (p<0.05; Fold-change >3; PSM#>2) overlap with all reported TWIST1 interactions on the Agile Protein Interactomes DataServer (APID) (Alonso-López et al., 2019). (C) Networks constructed from stringent TWIST1-specific interaction at a significant threshold of adjusted p-value (adjp) <0.05 and Fold-change >3. Unconnected nodes were removed. Top GO terms for proteins from three different clusters are shown. Node size = -Log10 (adjp). Genes associated with human and mouse facial malformation (HP:0001999, MP:0000428) were used as seeds (dark red) for heat diffusion through network neighbors. Node color represents the heat diffusion score. (D) Expression of candidate interactor genes in cranial neural crest from E9.5 mouse embryos; data were derived from published transcriptome dataset (Fan et al., 2016). Each bar represents mean expression ± SE of three biological replicates. All genes shown are expressed at level above the microarray detection threshold (2⁷, red dashed line).

Figure 1—figure supplement 1. Nuclear localization of TWIST1-BirA* biotinylated proteins and the endogenous TWIST1.

(A) Immunofluorescence analysis revealed co-localization of TWIST1-BirA* (HA tagged) and biotinylated proteins (labeled with streptavidin-GFP) in the nucleus in NCCs. Bar = 20 μm. (B) Expression and localization of endogenous TWIST1 in untransfected cells detected by immunostaining with α-TWIST1. Bar = 50 μm. (C) Immunofluorescence detection of proteins in cells co-expressing FLAG-TWIST1 (α-TWIST1) and 10 HA-tagged proteins (α-HA). Nuclei were stained by DAPI.

Figure 1—figure supplement 2. Identification of core NCC regulators within the TWIST1-CRM.

(A) Profile of streptavidin-purified proteins in GFP-BirA* and TWIST1-BirA*-expressing 3T3 cells visualized by Coomassie staining (left panel). Box: Gel bands sampled for mass spectrometry analysis. Expression of the TWIST1-BirA*HA encoded by the transgene and TCF4, a known TWIST1 interactor, by western blot analysis of the streptavidin-beads purified proteins (right panel). (B) Mean peptide spectrum match (PSM) across samples, normalized by total PSM of the peptide library. (C) PCA plot of normalized PSM data. G: GFP-BirA* Green dots, T: TWIST1-BirA* Red dots. (D) Complete network of 140 BioID candidates (p<0.05; Fold-change >3; PSM#>2) interacting physically with TWIST1 in the O9-1 neural crest stem cells. Functional interactions (edges) of these candidates, based on prior evidences of co-expression, protein-protein interaction, evolutionary conservation and text mining, were retrieved from STRING database (Szklarczyk et al., 2015). Medium confidence (combined score >0.4) was used as the cut-off for interactions. The MCL algorithm was used to generate protein interaction hubs with strongest connection (dark edges). Result from previous protein interaction survey of 56 TFs (Li et al., 2015) was referenced to annotate putative specific (red), non-specific (gray) or promiscuous TF interactors (green) among the BioID candidates. Blue nodes are putative-specific TWIST1 partners not annotated in Li et al. study. Node size = -Log10 (P-value). (i and ii) Example clusters. (E) Pairwise correlation of TWIST1 BioID data from O9-1 NCCs (x-axis) versus 3T3 fibroblasts (y-axis). Each data point represents one protein, plotted with their log2 fold change of PSM in TWIST1-BirA* versus GFP group (Log2FC). Point density were represented by rug plot next to the axis. Adjusted p-values (adjp) were generated with EdgeR package using negative binomial model: blue, adjp <0.05 (significant) for O9-1 but not 3T3, red, adjp <0.05 for both cell lines, black, adjp >0.05 for O9-1. (F) Plots generated by NetworkAnalyzer (Assenov et al., 2008) for Diffusion Rank of nodes against Degree of connection (number of edges), and cluster where peaks of highly connected nodes were labeled (left panel), and Between-ness centrality, a measures of how fast information spreads to other nodes (right panel). Putative neural crest disease-causing factors are found in the shaded region.