Research ResourceSystems Biology

Protein Interaction Network of the Mammalian Hippo Pathway Reveals Mechanisms of Kinase-Phosphatase Interactions

See allHide authors and affiliations

Science Signaling  19 Nov 2013:
Vol. 6, Issue 302, pp. rs15
DOI: 10.1126/scisignal.2004712


The Hippo pathway regulates organ size and tissue homeostasis in response to multiple stimuli, including cell density and mechanotransduction. Pharmacological inhibition of phosphatases can also stimulate Hippo signaling in cell culture. We defined the Hippo protein-protein interaction network with and without inhibition of serine and threonine phosphatases by okadaic acid. We identified 749 protein interactions, including 599 previously unrecognized interactions, and demonstrated that several interactions with serine and threonine phosphatases were phosphorylation-dependent. Mutation of the T-loop of MST2 (mammalian STE20-like protein kinase 2), which prevented autophosphorylation, disrupted its association with STRIPAK (striatin-interacting phosphatase and kinase complex). Deletion of the amino-terminal forkhead-associated domain of SLMAP (sarcolemmal membrane–associated protein), a component of the STRIPAK complex, prevented its association with MST1 and MST2. Phosphatase inhibition produced temporally distinct changes in proteins that interacted with MOB1A and MOB1B (Mps one binder kinase activator–like 1A and 1B) and promoted interactions with upstream Hippo pathway proteins, such as MST1 and MST2, and with the trimeric protein phosphatase 6 complex (PP6). Mutation of three basic amino acids that are part of a phospho-serine– and phospho-threonine–binding domain in human MOB1B prevented its interaction with MST1 and PP6 in cells treated with okadaic acid. Collectively, our results indicated that changes in phosphorylation orchestrate interactions between kinases and phosphatases in Hippo signaling, providing a putative mechanism for pathway regulation.


The Hippo pathway was first characterized in Drosophila melanogaster, in which genetic inhibition of pathway components leads to cellular overgrowth reminiscent of tumor hyperplasia (1, 2). The core components of the pathway are conserved in mammals and implicated in the regulation of organ size. Consistent with a role in maintaining tissue homeostasis, genetic inhibition of the Hippo pathway in mice leads to tumor development, and dysregulation of the Hippo pathway is associated with cancer in humans (2, 3).

The core Hippo pathway consists of a kinase cascade that promotes inhibition of transcriptional activity through cytosolic retention of transcriptional coactivators. High cell density or stimulation of mechanosensors activates the mammalian homologs of Hippo known as MST1 and MST2 [mammalian STE20-like protein kinase 1 and 2, which are encoded by the genes STE20-like protein kinase 4 and 3 (STK4 and STK3)] (3). MST1 and MST2 phosphorylate and activate LATS1 and LATS2 (large tumor suppressor kinase 1 and 2, Warts in flies). Activated LATS1 and LATS2 phosphorylate the transcriptional coactivators Yes-associated protein 1 (YAP1) and Tafazzin (TAZ) (known as Yorkie in flies) and thereby promote binding of these coactivators to members of the 14-3-3 family of scaffolding proteins, leading to cytosolic retention and, in some cases, proteasome-mediated degradation (4, 5). In the absence of phosphorylation by LATS1 or LATS2, YAP1 and TAZ localize to the nucleus and interact with Tea domain–containing transcription factors (TEAD1 to TEAD4, Scalloped in flies) to induce the transcription of genes that prevent apoptosis and promote proliferation (3).

Physical and genetic interaction screens in Drosophila and mammals have linked a number of additional proteins to Hippo signaling. For example, activation of LATS1 or LATS2 requires interaction with the small proteins MOB1A or MOB1B of the Mps one binder kinase activator–like (MOB) family, which are phosphorylated by MST1 and MST2 (6). In addition, MST1 and MST2 associate with the scaffolding protein salvador homolog 1 (SAV1), which activates the pathway (1, 7), or with proteins in the Ras association domain family (RASSF), which can inhibit the pathway, but in mammalian cells can also result in other effects, such as microtubule stabilization, depending on the RASSF family member (811). Apical polarity proteins including the Crumbs complex and Angiomotin (AMOT) bind to YAP1 and TAZ to promote apical localization and inhibit transcriptional activation downstream of these proteins in response to cell density (12, 13). Several additional components of the pathway have been identified in flies and humans and are reviewed elsewhere (3, 14, 15).

Whereas the roles of kinases in Hippo signaling have been extensively studied, the roles of protein phosphatases are also important. Increasing evidence suggests that dephosphorylation catalyzed by protein phosphatases plays a critical role in the activation of Hippo signaling. Knockdown of the protein phosphatase 2A (PP2A) catalytic subunit by small interfering RNA in mammalian cells leads to pathway activation by increasing the abundance of autophosphorylated MST1 and MST2 (16). Because serine and threonine phosphatases are multisubunit holoenzymes comprising catalytic subunits and regulatory subunits that confer substrate specificity [more than 60 different PP2A holoenzymes exist in mammalian cells (excluding splice variants) (17)], the identity of the precise PP2A holoenzyme responsible for inhibition of Hippo signaling is unclear. Additional complexity arises from the observations that protein phosphatase 1 (PP1) dephosphorylates TAZ (18), whereas PP2A dephosphorylates YAP1 (19). Thus, phosphatases act at multiple levels to promote Hippo signaling, and there is much to be discovered with respect to the particular enzymes involved in this process.

In Drosophila, there is precedent for inhibition of Hippo signaling by a PP2A subunit–containing holoenzyme known as STRIPAK (striatin-interacting phosphatase and kinase) (20). STRIPAK is a large evolutionarily conserved protein complex containing the PP2A catalytic and scaffolding subunits and the regulatory subunit striatin (STRN) that recruits other proteins, including all three Sterile 20–like kinases of the germinal center kinase III family (GCKIII) (2124). In mammals, STRIPAK comprises proteins from 10 different protein families encoded by 20 genes and forms at least two mutually exclusive complexes: one containing proteins of the cortactin-binding family (CTTNBP2 or CTTNBP2NL) and a second containing sarcolemmal membrane–associated protein (SLMAP) and proteins of the suppressor of IKKε (inhibitor of nuclear factor κB kinase ε) family (SIKE or FGFR1OP2) (21). Whether human STRIPAK associates with the human Hippo pathway and, if so, with which of the different STRIPAK complexes remain to be explored.

To explore how protein interactions in the Hippo pathway are affected by phosphatases and identify proteins that interact with known components of the Hippo pathway in a phosphorylation-dependent manner, we used affinity purification mass spectrometry (AP-MS) to systematically characterize protein interactions among Hippo pathway proteins in the presence or absence of phosphatase inhibition by okadaic acid (OA). OA is a potent inhibitor of serine and threonine phosphatases and is most effective against PP2A and its closest relatives PP4 and PP6 (2528). Treatment with OA results in activation of MST1 in cell culture, as shown by in-gel kinase assays (29, 30), and increases phosphorylation of MOB1A and MOB1B, LATS1 and LATS2, YAP1, and TAZ (1922). Whether increased phosphorylation of these proteins results directly from inhibition of dephosphorylation or indirectly from de novo phosphorylation induced by the activation of MST1 and MST2 is not known. We developed a high-confidence Hippo protein interaction network (interactome) using both FLAG-tagged protein–based affinity purification (FLAG AP) and in vivo BirA fusion protein–based (proximity-dependent) protein biotinylation followed by streptavidin-based affinity purification (BioID) coupled to MS. This dual approach revealed protein interactions among several phosphatases and known Hippo pathway proteins. Comparison of cells treated with and without OA identified several protein interactions that were modulated by phosphatase inhibition, leading us to perform a focused investigation of candidate protein interactions, which demonstrated that recognition of phosphorylated amino acids is a defining feature of kinase-phosphatase interactions in the Hippo pathway.


FLAG AP-MS and BioID are complementary approaches used to define the mammalian Hippo interactome

To systematically profile protein interactions in the Hippo pathway, we performed FLAG AP-MS using several bait proteins in the Hippo pathway. We generated human embryonic kidney (HEK) 293 Flp-In T-REx cells stably expressing 3×FLAG epitope–tagged versions of murine Lats2 or human MST1, MST2, SAV1, MOB1A, or MOB1B (Fig. 1A, table S1, and fig. S1A). We analyzed the data from two independent biological replicates and 15 negative controls (table S2) using SAINT (Significance Analysis of INTeractome) to generate a probability score for each protein interaction (3133). In additional experiments, we performed FLAG AP-MS with proteins identified in the first round of experiments to validate the interaction network and categorize Hippo pathway proteins into complexes. We identified 182 high-confidence interactions using 21 bait proteins (Fig. 1B, figs. S2 and S4, and table S5;

Fig. 1 Comparison of FLAG AP-MS and BioID used to define the Hippo protein interactome.

(A) List of the baits used for FLAG AP-MS and BioID. (B) Heatmap of the abundance of protein interactions discovered by FLAG AP-MS. (C) Heatmap of the abundance of protein interactions discovered by BioID. (D) Venn diagram showing the overlap of protein interactions discovered by FLAG AP-MS and BioID for the 13 bait proteins that were used in both approaches. In (C) and (D), data represent the sum of spectral counts for each protein from two biological replicates.

Some proteins are poorly solubilized by the lysis conditions used for FLAG AP-MS, and some protein interactions are too weak to withstand the FLAG AP procedure. To overcome these limitations, we used an approach known as BioID (34). We created stable HeLa and HEK293 Flp-In T-REx cell lines expressing 19 bait proteins (Fig. 1A and fig. S1B) fused in-frame with a promiscuous biotin ligase (BirA) that activates biotin to covalently bind to proximal proteins in living cells. We lysed the cells and isolated biotinylated proteins using streptavidin affinity purification with stringent lysis and wash conditions and analyzed the resultant mixture by MS. We analyzed data from two independent biological replicates with SAINT using a wide range of negative controls (table S3) and identified 487 high-confidence bait-prey relationships (Fig. 1C, fig. S3, and table S5). Comparison of FLAG AP-MS and BioID identified 53 proteins in common between the two data sets, whereas BioID revealed 284 and FLAG AP-MS revealed 63 unique proteins not found in the complementary data set (Fig. 1D).

Several examples show the utility of BioID for the isolation of proteins from diverse cellular compartments. Using human LATS1 or murine Lats2 as bait, we found that BioID, but not FLAG AP-MS, revealed interactions with proteins from the plasma membrane, cell-cell junctions, and the centrosome (figs. S5 to S7), consistent with reports that Lats2 is a centrosome-associated protein implicated in mitotic progression (35, 36). Similarly, chromatin-associated protein interactions, which are difficult to identify by AP-MS because of poor protein solubility (37), were detected with BioID. We found that BioID, but not FLAG AP-MS, identified the association of YAP1 with the Hippo pathway DNA binding transcription factors TEAD1 and TEAD3 (38) and the transcriptional coactivator B cell lymphoma 9–like protein (BCL9L) (fig. S8). The most abundant chromatin-associated proteins detected with BioID using YAP1 as bait were members of the SWI-SNF chromatin remodeling complex, including ARID1A and ARID1B (AT-rich interactive domains 1A and 1B; also known as BAF250a and BAF250b), SMARCD1 (SWI-SNF–related, matrix-associated, actin-dependent regulator of chromatin, subfamily d, member 1), and the transcription factor DPF2 (D4, zinc and double PHD fingers family 2; also known as Requiem or BAF45d) (fig. S8). These results suggest that, as in Drosophila (39), YAP1 may have additional roles in transcriptional control outside of Hippo signaling.

Although we detected fewer protein interactions unique to the FLAG AP-MS data set as compared to BioID, we identified potential functionally relevant protein interactions. For example, Lats2 interacted with scaffolding (PPP2R1A) and regulatory (PPP2R2A) subunits of PP2A (figs. S5 and S7). Collectively, these data validate the use of BioID and FLAG AP-MS as orthogonal approaches to define the Hippo interactome and identify several previously uncharacterized interactions that may be functionally relevant to Hippo signaling.

The FHA domain of SLMAP of the STRIPAK complex binds to MST1 and MST2

In Drosophila, the STRIPAK complex binds Hippo kinase and inhibits Hippo activation through dephosphorylation of the kinase activation loop (20). In mammalian cells, STRIPAK exists as two mutually exclusive complexes defined by the presence of alternative subunits, including either CTTNBP2-like adaptors or SLMAP in association with SIKE or FGFR1OP2 (21). BioID with MST1 and MST2 as baits revealed interactions with STRIPAK proteins, including several STRNs, SLMAP, and FGFR1OP2, but not CTTNBP2 or CTTNBP2NL (Fig. 2, A and B, and figs. S9 and S10), suggesting that MST1 and MST2 interact with the STRIPAK-SLMAP complex. Using SLMAP or CTTNBP2NL as bait protein for BioID, we discovered that SLMAP, but not CTTNBP2NL, could recover MST1 (Fig. 2B and fig. S3), suggesting that MST1 specifically interacts with the SLMAP containing STRIPAK complex.

Fig. 2 Interaction of the FHA domain of SLMAP of the STRIPAK complex with MST1 and MST2.

(A) Diagram of protein interactions for MST1 discovered by FLAG AP-MS and BioID. (B) Heatmap of the abundance of protein interactions discovered by BioID for the indicated baits and cell types. A black dot in the middle of a cell indicates that the SAINT cutoff was not met in that condition. Data represent the sum of spectral counts from two biological replicates. (C) Western blot of lysates and FLAG affinity purification fractions from HEK293T cells expressing GFP-MST1 or GFP-MST2 and the indicated FLAG-SLMAP constructs. (D) Western blot of lysates and FLAG affinity purification fractions from HEK293T cells expressing the indicated GFP-MST2 constructs and FLAG-SLMAP. In (C) and (D), blots are representative of three biological replicates.

To confirm the interactions among SLMAP and MST1 or MST2 and to identify the domains and amino acids required for these interactions, we used FLAG AP coupled to Western blot. We coexpressed green fluorescent protein (GFP)–tagged MST1 or MST2 and FLAG-tagged full-length SLMAP, variants of SLMAP, or negative control proteins in HEK293T cells. FLAG affinity purification of full-length SLMAP but not of empty vector or FLAG-CTTNBP2NL coimmunoprecipitated GFP-MST1 and GFP-MST2 (Fig. 2C), validating the BioID MS data. SLMAP is composed of a central coiled-coil region with two leucine zipper motifs, a variable C-terminal region with two alternatively spliced transmembrane domains, and an alternatively spliced N-terminal forkhead-associated (FHA) domain, which recognizes phospho-threonine (4042). Deletion of the FHA domain (amino acids 1 to 86) of FLAG-SLMAP (SLMAPΔFHA) reduced its interaction with GFP-MST1 and GFP-MST2 (Fig. 2C). Moreover, mutation of residues in the FHA domain that promote the interaction between SLMAP and threonine-phosphorylated proteins [R32A or H55A (43)] reduced the ability of FLAG-SLMAP to coimmunoprecipitate GFP-MST1 and GFP-MST2 (Fig. 2C). MST1 and MST2 are activated by autophosphorylation of Thr183 (MST1) or Thr180 (MST2) in the kinase activation loop (8). We found that mutation of MST2 Thr180 to Ala decreased the ability of FLAG-SLMAP to coimmunoprecipitate GFP-MST2 (Fig. 2D). Thus, these data suggest that the FHA domain of SLMAP binds to MST1 and MST2 in a phosphorylation-dependent manner and recruits these proteins to the STRIPAK complex.

Phosphatase inhibition alters the Hippo interactome

Our data suggested that phosphorylation regulates protein interactions in the Hippo pathway. Phosphatase inhibition by OA activates Hippo signaling (28). We validated that treating HeLa cells for 2.5 hours with OA increased the Ser127 phosphorylation and cytosolic localization of YAP1 (Fig. 3A), indicative of Hippo pathway activation. To identify changes in protein interactions caused by inhibition of dephosphorylation, we treated HEK293 Flp-In T-REx cells expressing FLAG-tagged bait proteins with OA and performed FLAG AP-MS. These experiments were done in parallel with those in Fig. 1B. In cells treated with OA, we observed increases in more than 100 protein interactions with 14-3-3β (encoded by YWHAB) as compared to cells treated with dimethyl sulfoxide (DMSO) (Fig. 3, B and C, and fig. S11), consistent with the fact that 14-3-3 proteins bind phosphorylated proteins (44, 45) and verifying that we were able to detect changes in protein interactions due to phosphatase inhibition by OA.

Fig. 3 Comparison of YAP1 phosphorylation and protein interactions in cells treated with the phosphatase inhibitor OA.

(A) Western blot of cytosolic and nuclear lysates from HeLa cells treated with DMSO or OA. P-YAP indicates YAP1 phosphorylated at Ser127. α4 and PARP [poly(adenosine diphosphate–ribose) polymerase] are control proteins for cell fractionation. C, cytosolic; N, nuclear. Blot is representative of four biological replicates. (B) Venn diagram showing the overlap of protein interactions discovered by FLAG AP-MS in cells treated with DMSO and OA. (C) Histogram showing the number of protein interactions discovered by FLAG AP-MS for the indicated bait proteins in cells treated with DMSO and OA. Data represent the sum from two biological replicates.

Comparison of the Hippo interactome in the presence and absence of OA revealed that OA had minimal effects on most interactions among established Hippo pathway proteins, including YAP1 (Fig. 4A and fig. S8). Association of phosphorylated YAP1 with 14-3-3 proteins is one mechanism by which the activation of Hippo signaling leads to inhibition of TEAD-dependent transcription (46). Using YAP1 as bait, we were unable to detect changes in interaction with 14-3-3 proteins in cells treated with OA compared to those treated with DMSO (Fig. 4A), which could be because 14-3-3 proteins were abundant in YAP1 purifications in the absence of OA (Fig. 4A). Moreover, OA did not alter the interaction of YAP1 with AMOT; junctional and apical polarity proteins, including INADL (InaD-like), PARD3 (par-3 family cell polarity regulator), and MPDZ (multiple PDZ domain protein); and other known interaction partners, including LATS1, PTPN14 (protein tyrosine phosphatase, nonreceptor type 14), WBP2 (WW domain–binding protein 2), SLC9A3R2 (solute carrier family 9, subfamily A member 3 regulator 2), and TP53BP2 [tumor protein p53–binding protein 2; also known as ASPP2 (47)] (Fig. 4A and fig. S8).

Fig. 4 Comparison of YAP1 and Lats2 protein interactions in cells treated with OA.

(A) Heatmap of the abundance of protein interactions with YAP1 discovered by FLAG AP-MS. (B) Western blot of lysates and FLAG affinity purification fractions from HEK293T cells expressing GFP-YAP1 and FLAG-PPP1CA. (C) Histogram of TEAD reporter activation in HEK293T cells by the indicated overexpression constructs. Data are means ± SD of three biological replicates. * fold decrease P < 0.05, # fold increase P < 0.05, compared to control (t test). (D) Heatmap of the abundance of protein interactions with Lats2 discovered by FLAG AP-MS. In (A) and (D), a black dot in the middle of a cell indicates that the SAINT cutoff was not met in that condition. Data represent the sum of spectral counts from two biological replicates.

We did observe modest changes in some protein interactions with YAP1 in OA-treated cells. For example, YAP1 coimmunoprecipitated the catalytic subunit of PP1 (encoded by PPP1CA) slightly more in cells treated with OA as compared to DMSO (Fig. 4A). Moreover, affinity purification of FLAG-PPP1CA coimmunoprecipitated GFP-tagged YAP1 when coexpressed in HEK293T cells in the absence of OA as assessed by Western blot (Fig. 4B). PP1 interacts with ASPP2 to promote activation of TAZ by dephosphorylation of Ser89 and Ser311 (18). ASPP2 also binds YAP1 and promotes its nuclear accumulation (48). Thus, we hypothesized that PP1 promotes YAP1-dependent transcription. Overexpression of FLAG-PPP1CA in HEK293T cells increased the activity of a multimerized TEAD-binding element luciferase reporter by 2.3-fold compared to cells transfected with the reporter alone (Fig. 4C), consistent with a functional role for PP1 in the regulation of YAP1 nuclear localization and transcriptional activation by dephosphorylation.

In contrast to YAP1, treating cells with OA altered several protein interactions with Lats2 (Fig. 4D and fig. S5). For example, treating cells with OA decreased interactions among Lats2 and proteins involved in proteostasis, such as LONP1 (mitochondrial lon peptidase 1), ANAPC1 and ANAPC5 (anaphase-promoting complex subunit 1 and 5), and CUL7 (cullin 7); junctional and apical polarity proteins, including INADL, PARD3, and MPDZ; as well as scaffolding and regulatory subunits of PP2A (Fig. 4D and fig. S5). Other protein interactions with Lats2 increased in the cells treated with OA, including those with MST1, MST2, and AMOT (Fig. 4D and fig. S5). These data suggest that increased phosphorylation of LATS kinases or associated proteins in cells treated with OA induced a shift from interactions that could repress LATS activity, such as ubiquitin ligases and phosphatases, toward interactions that promote LATS activity, including MST1 and MST2 (49).

MOB1 proteins recognize phosphorylated kinases and phosphatases in the Hippo pathway

Analysis of changes in protein interactions in cells treated with OA revealed a phosphorylation-dependent interaction network involving MST- and MOB1-family proteins. Treating cells with OA promoted interactions among MST1 and MST2 and proteins including Aurora kinase B, kinesin family member 2C, and PRKRIR (repressor of interferon-inducible double-stranded RNA–dependent inhibitor of protein kinase) and between MST2 and polo-like kinase 1, which was previously reported to interact with MST2 in the absence of OA (50) (figs. S9 and S10). In addition, treating cells with OA induced the interaction of MST1 and MST2 with MOB1A and MOB1B (Fig. 5A and figs. S9, S10, S12, and S13). Using MOB1A and MOB1B as bait proteins, we confirmed that MOB1A and MOB1B interacted with both MST1 and MST2 as well as other upstream Hippo components, including RASSF2 and SAV1 (Fig. 5A and figs. S12 and S13). As a control for the specificity of MOB1-MST protein interactions, we performed FLAG AP-MS using MOB4, a distant member of the MOB family that is a component of the STRIPAK complex (51) (table S4). Unlike MOB1A and MOB1B, MOB4 did not interact with proteins in the Hippo pathway in the presence or absence of OA (table S7). Because MST1 and MST2 can phosphorylate MOB1A and MOB1B (6), these results suggest that the phosphorylation-dependent interactions among MST- and MOB1-family proteins could be a mechanism of Hippo pathway feedback.

Fig. 5 Comparison of MOB1A and MOB1B protein interactions in cells treated with OA.

(A) Heatmap of the abundance of protein interactions with MOB1A and MOB1B discovered by FLAG AP-MS. Data represent the sum of spectral counts from two biological replicates. (B) Diagram of protein interactions with MOB1A and MOB1B discovered by FLAG AP-MS and BioID. (C) Western blot of lysates and FLAG affinity purification fractions from HEK293T-REx cells stably expressing FLAG-MOB1A or FLAG-MOB1B and treated with DMSO or OA. (D) Heatmap of the abundance of protein interactions with MOB1A discovered by FLAG AP-MS from cells treated over a time course with OA. Heatmap is representative of the sum of spectral counts for two biological replicates. (E) Western blot of lysates and FLAG affinity purification fractions from HEK293T-REx cells treated with DMSO or OA and stably expressing FLAG-MOB1B or a mutant FLAG-MOB1B with alanine substitutions at three basic amino acids in the putative phospho-serine– and phospho-threonine–binding pocket. In (C) and (E), blots are representative of three biological replicates.

Mst1-mediated phosphorylation of Mob1 promotes its interaction with dedicator of cytokinesis 8 (DOCK8) in mouse thymocytes (52). We found that treating HEK293 Flp-In T-REx cells with OA increased the interactions among MOB1A and MOB1B and DOCK6 and DOCK7 (Fig. 5, A and B). DOCK6 to DOCK8 are atypical guanine exchange factors for the cytoskeletal-associated guanosine triphosphatase Rac (53). FLAG AP-MS using MOB1A, MOB1B, DOCK7, or DOCK8 as bait revealed interactions with the cytoskeletal-associated, leucine-rich repeats and calponin homology domain–containing proteins (LRCH1 to LRCH3) (54) and with a protein of unknown function, CRLF3 (cytokine receptor–like factor 3) (55). FLAG AP-MS experiments using LRCH3, LRCH4, and CRLF3 revealed reciprocal interactions among each other and with DOCK6 to DOCK8 (Fig. 5B and table S5), suggesting that these proteins form a complex that associates with MOB1A and MOB1B in a phosphorylation-dependent manner.

FLAG AP-MS of MOB1 proteins revealed interactions with the serine and threonine protein phosphatase PP6 in cells treated with OA (Fig. 5, A and B). MOB1A or MOB1B interacted with PPP6R1 to PPP6R3 and the ankyrin repeat–containing protein ANKRD28 (Fig. 5, A and B), which are regulatory subunits that confer substrate specificity to the PP6 holoenzyme (56, 57). Using FLAG AP coupled to Western blot, we confirmed that OA enhanced the interaction of the PP6 catalytic (PPP6C) and regulatory (PPP6R3) subunits with MOB1A and MOB1B (Fig. 5C).

We performed a time-course experiment to characterize the order in which proteins were recruited to MOB1A after phosphatase inhibition. We treated HEK293 Flp-In T-REx FLAG-MOB1A cells with OA for 30, 90, or 150 min and performed FLAG AP-MS. We observed that LATS1 interacted with MOB1A at all time points, whereas MOB1A interacted maximally with the DOCK-LRCH-CRLF3 and the MST-SAV-RASSF complexes at 90 min and with PP6 at 150 min (Fig. 5D and table S6), suggesting that PP6 may be recruited after LATS and MST kinases to attenuate MOB1A phosphorylation and Hippo pathway activation.

In yeast, Mob1 interacts with Nud1, a protein homologous to the mammalian centrosome and midbody component Centriolin (58). This interaction requires three basic amino acids that form part of a phospho-serine– and phospho-threonine–binding pocket in Mob1 (59). To determine if the homologous residues in human MOB1 were required for its interaction with MST and PP6, we mutated Lys153, Arg154, and Arg157 to Ala in MOB1B and generated stable HEK293 Flp-In T-REx cell lines. We treated mutant and wild-type MOB1B-expressing cells with OA and performed FLAG AP coupled to Western blot. We found that OA induced interactions between MST1 or PPP6R3 and wild-type MOB1B, but not mutant MOB1B (Fig. 5E). Together, these results suggest that MOB1 recognizes phosphorylated proteins in both yeast and mammalian cells and that human MOB1A and MOB1B interact with phosphorylated kinases and phosphatases in the Hippo pathway to promote activation of signaling.


Here, we used FLAG AP-MS and BioID to define the Hippo protein interactome and showed that combining these strategies produced complementary data sets with both common and unique sets of protein interactions. BioID enabled us to identify interactions among proteins localized in cellular compartments that are difficult to analyze by AP-MS, such as membrane- and chromatin-associated proteins, and to recover potentially transient interactions that we were only able to detect by FLAG AP-MS in cells treated with OA. We were not able to perform BioID on cells treated with OA because of the length of incubation time required for biotin labeling. Thus, we could not perform a comparative analysis of BioID and FLAG AP-MS under this condition. However, our study highlights the utility of BioID and potentially similar methods, such as engineered ascorbate peroxidase (APEX) (60), to identify proximal proteins, thereby providing information about the physiological context of protein interactions in living cells.

We developed a high-confidence interactome revealing multiple interactions of Hippo pathway proteins with protein phosphatases. We found that the interaction between the STRIPAK phosphatase complex and Hippo, previously reported in Drosophila (20), is conserved in mammalian cells. MST1 and MST2 interacted with the SLMAP containing STRIPAK complex and required the phospho-threonine–binding FHA domain of SLMAP. The MST-related family of GCKIII kinases bind to STRIPAK independent of SLMAP through a protein known as cerebral cavernous malformation 3 (22, 23). Thus, it is formally possible for a STRIPAK complex to contain both MST1 or MST2 and a GCKIII kinase, which would permit kinase crosstalk; however, in this study, we did not detect an interaction among MST1 or MST2 and GCKIIIs (fig. S14).

We discovered that phosphatase inhibition by OA induced protein interactions with MOB1A and MOB1B, consistent with a recent report that defines a phospho-serine– and phospho-threonine–binding domain in yeast Mob1 (59). Mutation of the basic residues that form the phospho-serine– and phospho-threonine–binding domain (59) abrogated OA-induced interaction of MOB1B with MST1 and the PP6 subunit PPP6R3, suggesting that PP6 and MST1 may compete for interaction with MOB1 through a common binding mechanism. The phosphorylation sites recognized by MOB1 on MST1 and PP6 are not known, and there is no obvious sequence homology among known phosphorylation sites on MST1 or PP6 with that of the yeast Mob1-binding protein Nud1. In addition, whether components of the PP6 holoenzyme are substrates for MOB1-associated kinases (LATS1, LATS2, MST1, and MST2) remains to be elucidated. Nevertheless, our results argue that MOB1 could play a role in the integration of key phosphorylation and dephosphorylation events to fine-tune the timing or amplitude of Hippo pathway activation.

Our study provides a comprehensive view of interactions for core Hippo pathway proteins combining traditional AP-MS with proximity-dependent in vivo biotinylation by BioID. We used OA to demonstrate that phosphorylation-dependent changes in protein interactions are a potential mechanism for Hippo pathway activation, and we identified several uncharacterized protein interactions in the Hippo network. Thus, our data set serves as an important resource for the field of Hippo signaling research and opens exciting, new avenues to be explored.


Cell line generation

3×FLAG and BirA-FLAG constructs were generated via Gateway cloning into pDEST 5′ Triple FLAG pcDNA5 FRT TO or pDEST5 BirA FLAG pcDNA5 FRT TO, respectively. Accession numbers for the starting clones are shown in table S1. Point mutations were performed by polymerase chain reaction–directed mutagenesis (the position of the mutated amino acids is indicated on the basis of the reference sequence). All constructs were validated by sequencing. Stable cell lines were generated as Flp-In 293 T-REx or Flp-In HeLa T-REx cell pools as described (61), and expression was induced for 24 hours with tetracycline (1 μg/ml). OA was added at a concentration of 150 nM for 2.5 hours unless otherwise indicated, and DMSO was used as a negative control. For BioID experiments, the cells were treated with 50 μM biotin for 24 hours in combination with tetracycline. For the experiments presented in tables S6 and S7, complementary DNAs (cDNAs) were subcloned into pcDNA3-FLAG, and pools of stable transfectants were selected with G418 as previously described (21).


Cell pellets from two 150-mm plates were lysed in 50 mM Hepes-KOH (pH 8.0), 100 mM KCl, 2 mM EDTA, 0.1% NP-40, and 10% glycerol and affinity-purified with M2-FLAG magnetic beads and on-bead digest as described (59). Spectra were acquired on an LTQ (Thermo Fisher) mass spectrometer placed in line with an Agilent 1100 pump with split flow as described (62). The raw files are deposited in the MassIVE repository housed at the Center for Computational Mass Spectrometry at University of California, San Diego (UCSD) ( The main FLAG AP-MS data set consisting of 99 raw files has been assigned the MassIVE ID MSV000078449. The data set used for the time-course and specificity experiments presented in tables S6 and S7 (20 raw files) has been assigned the MassIVE ID MSV000078450.

BioID purification and data acquisition

Cell pellets from four 150-mm plates were incubated at 4°C in 2 ml of radioimmunoprecipitation assay (RIPA) buffer [50 mM tris-HCl (pH 7.5), 150 mM NaCl, 1% NP-40, 1 mM EDTA, 1 mM EGTA, 0.1% SDS, Sigma protease inhibitors P8340 1:500, and 0.5% sodium deoxycholate], supplemented with 250 U of benzonase. After incubation with agitation for 1 hour, the lysates were sonicated using three 10-s bursts with 2-s rest in between on ice at 30% amplitude. Lysates were centrifuged for 30 min at 20,817g at 4°C. Cleared supernatants were transferred to 2-ml microcentrifuge tubes, and a 30-μl bed volume of prewashed streptavidin-agarose beads (GE Healthcare, catalog no. 17-5113-01) was added to each sample. Affinity purification was performed at 4°C on a nutator for 3 hours, and the beads were washed twice in RIPA buffer (1 ml per wash), twice in TAP lysis buffer [50 mM Hepes-KOH (pH 8.0), 100 mM KCl, 10% glycerol, 2 mM EDTA, 0.1% NP-40], and three times in 50 mM ammonium bicarbonate (ABC; pH 8.0). After affinity purification and removal of all washing buffer, beads were resuspended in 200 μl of 50 mM ABC (pH 8) with 1 μg of trypsin (Sigma, no. T6567) added and incubated at 37°C overnight with agitation. The next day, an additional 1 μg of trypsin was added to each sample (in 10 μl of 50 mM ABC), and the samples were incubated for 2 hours at 37°C. Beads were pelleted (400g, 2 min), and the supernatant was transferred to a fresh 1.5-ml tube. The beads were then rinsed two times with 150 μl of MS-grade H2O, and these rinses were combined with the original supernatant. The pooled fractions were centrifuged at 16,100g for 10 min, and the supernatant was transferred to a new 1.5-ml tube and dried in a vacuum centrifuge. Tryptic peptides were resuspended in 10 μl of 5% formic acid, and 3 μl was used per analysis.

Peptides were analyzed by nano-HPLC (high-pressure liquid chromatography) coupled to MS. A spray tip was formed on a fused silica capillary column (0.75 μm internal diameter, 350 μm outer diameter) using a laser puller (Sutter Instrument Co., model P-2000, program = 4; heat = 280, FIL = 0, VEL = 18, DEL = 200). C18 reversed-phase material (Reprosil-Pur 120 C18-AQ, 3 μm) in methanol was packed [10 (±1) cm] into the column using a pressure injection cell. The column was equilibrated in 6 μl of 0.1% formic acid in water and connected to a NanoLC-Ultra 2D plus HPLC system (Eksigent) coupled to an LTQ Orbitrap Velos (Thermo Electron) equipped with a nanoelectrospray ion source (Proxeon Biosystems). The HPLC program ran the following ratios of buffer B (0.1% formic acid in acetonitrile) to buffer A (0.1% formic acid in water): 20 min at 400 μl/min with 2% buffer B, 95.5 min at 200 μl/min using a linear gradient from 2 to 35% buffer B, 5 min at 200 μl/min using a linear gradient from 35 to 80% buffer B, 6.5 min at 200 μl/min with 80% buffer B, and 18 min at 200 μl/min with 2% buffer B. The LTQ Orbitrap Velos was operated with Xcalibur 2.0 in data-dependent acquisition mode with the following parameters: one centroid MS (mass range, 400 to 2000) followed by MS-MS on the 10 most abundant ions. General parameters were as follows: activation type = CID, isolation width = 1 mass/charge ratio (m/z), normalized collision energy = 35, activation Q = 0.25, activation time = 10 ms. For data-dependent acquisition, the minimum threshold was 500, the repeat count = 1, repeat duration = 30 s, exclusion size list = 500, exclusion duration = 30 s, exclusion mass width (by mass) = low 0.6, high 1.2. The raw files are deposited in the MassIVE repository housed at the Center for Computational Mass Spectrometry at UCSD ( and assigned the MassIVE ID MSV000078460

MS data analysis

Samples analyzed on the LTQ were converted to mzXML using ProteoWizard 3.0.4468 (63) and analyzed using the iProphet pipeline (64) implemented within ProHits (65) as follows. The database consisted of the human and adenovirus sequences in the RefSeq protein database (version 57) supplemented with “common contaminants” from the Max Planck Institute ( and the Global Proteome Machine (GPM; The search database consisted of forward and reverse sequences (labeled “gi|9999” or “DECOY”); in total, 72,226 entries were searched. Spectra were analyzed separately using Mascot (2.3.02; Matrix Science) and Comet [2012.01 rev.3 (66)] for trypsin specificity with up to two missed cleavages; deamidation (Asn or Gln) or oxidation (Met) as variable modifications; single-, double-, and triple-charged ions allowed, mass tolerance of the parent ion at 3.00 atomic mass units (amu); and the fragment bin tolerance at 0.36 amu. The resulting Comet and Mascot results were individually processed by PeptideProphet (67) and combined into a final iProphet output using the Trans-Proteomic Pipeline (TPP; Linux version, v0.0 Development trunk rev 0, Build 201303061711). TPP options were as follows: general options were -p0.05 -x20 -d“gi|9999,” iProphet options were –ipPRIME, and PeptideProphet options were –OpdP. Samples analyzed on the LTQ Orbitrap Velos were analyzed with the following differences: the mass tolerance of the precursor ion was set at 50 parts per million (ppm), the fragment bin tolerance at 0.6 amu, and the general TPP options were -p0.05 -x20 -PPM -d“DECOY.” All proteins with a minimal iProphet probability of 0.05 were parsed to the relational module of ProHits. For analysis with SAINT, only proteins with an iProphet protein probability of >0.95 were considered. This corresponds to an estimated false discovery rate (FDR) of about 0.5%.

Interaction scoring for FLAG AP-MS

For each bait protein, two plates per cell line were grown in parallel and treated with 150 nM OA in DMSO or with DMSO alone for 2.5 hours before harvesting. Samples were prepared for MS analysis, as described above, in biological duplicate from cells grown, treated, and processed at different times to maximize the variability and increase the robustness in the detection of true interactors. Cells expressing the FLAG tag alone or a fusion of GFP protein to the FLAG tag were used for negative controls and processed in parallel to mitigate “batch effect” artifacts (68). Each peptide sample was loaded onto a single-use reversed-phase column with pressure bomb loading to prevent carry-over. The quality of each sample was assessed by manually aligning the runs for the biological replicates in ProHits. Samples of low quality were discarded, and additional biological replicates were acquired.

SAINT was implemented similar to (31, 32) but using a more computationally efficient process ( (33). The default SAINT options were lowMode = 1, minFold = 0, norm = 0. SAINT probabilities computed independently for each biological replicate were averaged (AvgP) and reported as the final SAINT score. Fifteen negative control experiments (eight treated with DMSO and seven treated with OA) were compressed to seven “virtual” controls using the seven highest spectral counts for each protein. For the FLAG AP-MS experiments, proteins with AvgP ≥0.88 were considered true positive interactions with a calculated FDR of ≤1%. Fold change was calculated for each prey protein as the ratio of average spectral counts from replicate bait purifications over the average spectral counts across all negative controls (peptide spectral counts were summed for each protein) (69). A background factor of 0.1 was added to the average spectral counts of negative controls to prevent division by zero. Proteins included in the final interactome list had an AvgP ≥0.88 or fold change ≥35 relative to negative control experiments and were detected in both of the biological duplicates with ≥2 spectra per peptide in each replicate. The complete list of the interactions with scores is available at MassIVE and in a searchable format on our Web site: (see the Supplementary Materials for a tutorial).

Interaction scoring for the BioID

The BioID interaction data set comprises samples generated in HeLa Flp-In T-REx cells as well as in HEK293 Flp-In T-REx cells. Like the FLAG AP-MS data set, biological replicates for each bait protein were generated on separate days. Nonspecific contamination in BioID could result from endogenously biotinylated proteins in the absence of the BirA* fusion protein and from abundant proteins whose biotinylation depends on BirA*. Several negative controls were used, including parental HeLa and HEK293 cells and those expressing BirA* alone or a BirA*-GFP fusion protein. We performed 28 negative control experiments that were compressed into four virtual controls for SAINT analysis by selecting the four highest spectral counts for each protein. Selection of the threshold for SAINT scores was based on receiver operating curve analysis performed using publicly available protein interaction data and the FLAG AP-MS data set as a list of true positive interactions. A SAINT score of AvgP ≥0.87 was considered a true positive BioID protein with an estimated FDR of ≤1%.

Pathway annotation, functional enrichment, and data visualization

Interactions in the iRefIndex (70) repository (v11, 9606.mitab.05192013.txt) were processed to only include those curated from IntAct (71), BioGRID (72), MINT (73), and DIP (74). These resulting interactions were collapsed at the gene level, directionality was removed, and the data were integrated with the FLAG AP-MS and BioID networks. Functional enrichment analysis was performed using DAVID Bioinformatics Resources 6.7 (75, 76). Unless otherwise indicated, we performed Functional Annotation Clustering and reported Cellular Component (CC FAT) enrichment probabilities adjusted for Benjamini-Hochberg FDR correction. Heatmaps were generated with MultiExperiment Viewer version 4.8.1 (MeV; (77); hierarchical clustering was performed using Pearson correlation and average linkage. For the heatmaps displaying the comparison of the interactions in the DMSO- and OA-treated samples, the following rules were applied: any protein that has AvgP ≥0.88 or fold change ≥35 relative to negative control experiments in either OA or DMSO treatment condition was considered an interactor, and its spectral counts were displayed on the heatmap regardless of the SAINT score in the particular treatment. If the protein score fell below the SAINT score threshold in one of the treatment conditions, this was indicated in the figure by a black dot on the heatmap. Network diagrams were generated using Cytoscape version 3.0 (78).

Coimmunoprecipitation and cell fractionation

Cells were lysed in ice-cold CHAPS lysis buffer [40 mM Hepes (pH 7.5), 120 mM NaCl, 1 mM EDTA, 0.3% CHAPS] with 1× protease inhibitors (Sigma). Lysates were immunoprecipitated with M2-FLAG beads for 2 hours at 4°C. M2-FLAG beads and the bound proteins were washed three times with CHAPS lysis buffer. Samples were resolved on 10% SDS–polyacrylamide gel electrophoresis and transferred onto a nitrocellulose membrane, probed with the primary antibodies indicated in table S8 and fluorescent secondary antibodies [mouse (CW800) and rabbit (CW680), LI-COR], and directly imaged on an Odyssey scanner (LI-COR). The antibody recognizing PPP6C was generated by immunizing rabbits with a KLH (keyhole limpet hemocyanin)–conjugated peptide corresponding to the sequence 280-REPKLFRAVPDSERVIPPR-298 in human PPP6C. The antibody does not cross-react with the two closest relatives of the PP6 catalytic subunit, PP4 (79) and PP2A (fig. S15). In Fig. 3A, HeLa cells were lysed with the NE-PER Nuclear and Cytoplasmic Extraction Reagents (Thermo Scientific) as per the manufacturer’s instructions.

Luciferase reporter assays

HEK293T cells were grown in 24-well plates and transfected at 75 to 80% confluence with 200 ng of cytomegalovirus-driven yellow fluorescent protein (YFP; Addgene, 22780), 200 ng of pGL3-Basic firefly luciferase reporter (Promega) fused to a minimal Hsp70 promoter and a 10× multimerized TEAD-binding site (TGGAATGT), 10 ng of hemagglutinin-tagged TAZ, and 100 ng of the experimental plasmid using jetPRIME (PolyPlus) according to the manufacturer’s protocol. Twenty-four hours after transfection, cells were lysed in TNT buffer [50 mM tris (pH 7.6), 15 mM NaCl, 0.5% Triton X-100, 1 mM EDTA], and 30 μl of clarified lysate was transferred to a black round-bottomed 96-well plate. YFP fluorescence (emission at 530 nm) and luciferase activity were monitored after addition of luciferase assay reagent (Promega) with a Perkin-Elmer Envision 2104 Multilabel Reader. Background fluorescence was subtracted from the YFP and luciferase readings, and luciferase values were then normalized to YFP values. P values were calculated with StatPlus (AnalystSoft).


Fig. S1. Bait protein expression in inducible cell lines.

Fig. S2. High-resolution heatmap for the FLAG AP-MS data.

Fig. S3. High-resolution heatmap for the BioID data.

Fig. S4. Network diagram depicting the complete Hippo interactome.

Fig. S5. Detailed Lats2 interactome.

Fig. S6. Detailed LATS1 interactome.

Fig. S7. Comparison of Lats2 FLAG AP-MS and BioID data sets.

Fig. S8. Detailed YAP1 interactome.

Fig. S9. Detailed MST1 interactome.

Fig. S10. Detailed MST2 interactome.

Fig. S11. Detailed YWHAB interactome.

Fig. S12. Detailed MOB1A interactome.

Fig. S13. Detailed MOB1B interactome.

Fig. S14. Model for MST1 and MST2 recruitment to STRIPAK.

Fig. S15. Validation of specificity of PPP6C antibody.

Table S1. Proteins used as baits.

Table S2. Composition of the FLAG AP-MS data set.

Table S3. Composition of the in vivo biotinylation data set.

Table S4. Composition of the additional FLAG AP-MS data set.

Table S5. High-confidence data set reported in this study.

Table S6. Time course of OA treatment.

Table S7. Comparison of the MOB-family protein interactions.

Table S8. List of antibodies used in this study.

Website_tutorial.pdf. Tutorial to the Web site.

Hippo_interactome.cys. Cytoscape file depicting the complete Hippo interactome.


Acknowledgments: We would like to thank M. Goudreault and L. Llano for assistance with construct generation and cell lines, G. Liu and J. P. Zhang for the design of the Web site, and B. Larsen and M. Tucholska for expert advice on MS. We thank S. Guettler for the LATS1 cDNA, C. Badouel for the YAP1 cDNA, L. McBroom-Cerajewski for Strn and STRN3 entry vectors, C. Go for the TJP2 expression vector, B. Gonzalez Badillo for the PPP2R1A and PPP2R2A expression constructs, and A. Gregorieff for cloning of the TEAD reporter plasmids. We thank N. Bandeira and J. Carver for help with the MassIVE submission, S. Orchard and P. Porras for IntAct, and M. Tyers and L. Boucher for BioGRID. We thank H. McNeill, L. Attisano, Z. Steinhart, and S. Angers for helpful discussions. We thank B. Raught, P. Samavarchi-Tehrani, and L. Enderle for critical reading of the manuscript. Funding: This work was supported by the Canadian Institutes for Health Research (MOP-84314 and MOP-123433 to A.-C.G.), the Government of Ontario via a Global Leadership Round in Genomics and Life Sciences (T.P., J.L.W., and A.-C.G.). A.-C.G. is the Canada Research Chair in Functional Proteomics and the Lea Reichmann Chair in Cancer Proteomics. J.D.R.K. was supported by a postdoctoral fellowship from the Canadian Heart and Stroke Foundation. M.J.K. was supported by a Banting and Best Canadian Graduate Scholarship. Author contributions: A.L.C. generated 3×FLAG and BirA-FLAG samples, performed MS and validation experiments, designed the experiments, and analyzed the data. J.D.R.K. contributed reagents, discussions, and BirA-FLAG control runs. M.J.K. performed initial interaction mapping for MOB1 proteins. G.T. implemented the new SAINT tools and performed data analysis. A.W. provided TEAD reporter constructs and helped with assay development. W.H.D. contributed BirA-FLAG control runs and assisted with sample preparation. Z.-Y.L. helped with mass spectrometric analysis. R.D.B. contributed reagents and discussions. F.S. co-supervised M.J.K.; T.P. co-supervised J.D.R.K. and supervised R.D.B.; J.L.W. supervised A.W. and contributed to the functional validation. H.C. and A.-C.G. developed the scoring scheme for MS data. A.-C.G. directed the project. A.L.C. and A-C.G. wrote the manuscript with input from all authors. Competing interests: The authors declare that they have no competing interests. Data availability: All raw MS data are available through the MassIVE Web portal (, and the results can be searched at (see Materials and Methods and Supplementary Materials for a tutorial). High-confidence interactions have been deposited in BioGRID (64) and the International Molecular Exchange (IMEx) consortium ( through IntAct (58) and assigned the identifier IM-21541.
View Abstract

Stay Connected to Science Signaling

Navigate This Article