Hippo and PI5P4K signaling intersect to control the transcriptional activation of YAP

Abstract

Phosphoinositides are essential signaling molecules. The PI5P4K family of phosphoinositide kinases and their substrates and products, PI5P and PI4,5P₂, respectively, are emerging as intracellular metabolic and stress sensors. We performed an unbiased screen to investigate the signals that these kinases relay and the specific upstream regulators controlling this signaling node. We found that the core Hippo pathway kinases MST1/2 phosphorylated PI5P4Ks and inhibited their signaling in vitro and in cells. We further showed that PI5P4K activity regulated several Hippo- and YAP-related phenotypes, specifically decreasing the interaction between the key Hippo proteins MOB1 and LATS and stimulating the YAP-mediated genetic program governing epithelial-to-mesenchymal transition. Mechanistically, we showed that PI5P interacted with MOB1 and enhanced its interaction with LATS, thereby providing a signaling connection between the Hippo pathway and PI5P4Ks. These findings reveal how these two important evolutionarily conserved signaling pathways are integrated to regulate metazoan development and human disease.

INTRODUCTION

Phosphoinositides are a group of seven distinct species of phospholipids generated by the phosphorylation of the third, fourth and fifth positions of the inositol headgroup of phosphatidylinositol(1). The seven resulting phosphoinositides are cycled and regulated by a complex network of kinases and phosphatases that add or remove phosphate groups at specific positions. Phosphoinositide signaling impacts many cellular processes, including proliferation, survival, and growth, and its dysregulation is common in cancer and other diseases. One of the least studied phosphoinositide species is the low abundance phosphatidylinositol 5-monophosphate (PI5P), which, 20 years after its discovery, has yet to be definitively linked to any specific cellular pathway. PI5P is the lipid substrate for the type II phosphatidylinositol-5-phosphate 4-kinases (PI5P4Ks) and was the last phosphoinositide to be discovered (2). PI5P4Ks phosphorylate PI5P at the 4-position to generate distinct pools of phosphatidylinositol 4,5-bisphosphate (PI4,5P₂). The importance of this evolutionarily conserved pathway for generating PI4,5P₂, as well as for cycling PI5P, is still poorly understood. To date only a few PI5P4K regulators have been identified, including the kinases p38 MAPK(3) and mTOR(4). Similarly, the list of characterized binders to PI5P remains short. These binders are ING2(5), which connects nuclear PI5P to p53-dependent apoptosis; TOM1(6), which is involved in endosomal maturation; and Tiam1(7), which is related to cell migration and invasion.

Several phospholipids regulate the Hippo pathway (8–12), which consists of four major kinases, mammalian Ste20-like kinases 1 and 2 (MST1/2) and large tumor suppressor LATS1 and LAT2 (LATS1/2), as well as the associated adaptor proteins SAV and MOB1. The main downstream target of Hippo pathway, Yes-associated protein (YAP), activates various transcriptional programs related to cell growth. YAP dysregulation has been linked to various pathologies including cancer and cancer metastasis (13). Here, we provide a previously unrecognized connection between PI5P4K activity and Hippo pathway regulation. We identified MST1 and MST2 as direct inhibitors of PI5P4Ks. PI5P4K knockdown or knockout led to enhanced activity downstream of MST1/2 and decreased activity of YAP. Thus, our study reveals insights into how PI5P4Ks relay metabolic information to the Hippo pathway. PI5P4Ks are druggable kinases (14–22) and therefore, these data can be leveraged in targeting diseases with dysregulated Hippo signaling, such as cancers.

RESULTS

The MST kinases control intracellular turnover of PI4,5P₂ by directly phosphorylating PI5P4Ks

To identify new direct regulators for PI5P4Ks, we conducted in vitro kinase assays in which we incubated the lipid kinase PI5P4Kβ with a panel of protein kinases and ATP. Then, we included the lipid substrate PI5P with radio-labeled ATP and measured the generation of PI4,5P₂ by thin layer chromatography (TLC) (Fig. S1A). Of the approximately 29 kinases assayed, the Hippo pathway core kinase MST1 had the most potent effect on PI5P4Kβ, completely inhibiting its activity in a concentration-dependent manner (Fig. 1A, Fig. S1B). In addition, we found that MST1 also inhibited the activity of the PI5P4K⍺ isoform (Fig. 1A). MST1 is a germinal center kinase II (GCKII) protein, part of the “STE20-like” family of proteins that play essential roles as regulators of cell fate and cell growth. Other germinal center kinase family members also inhibited PI5P4Kβ, including MST2 (Fig. 1B), HGK, GLK, MST3 and TAOK1 (Fig. S1C). Moreover, we found this effect to be evolutionarily conserved because the human MST1 inhibited the activity of the Caenorhabditis elegans PPK-2 orthologue for PI5P4Ks (Fig. S1D). Finally, this regulation was specific to type II PI5P4Ks because MST1 did not affect the activities of type I PI4P5Ks (Fig. S1E). MST1/2 are serine-threonine protein kinases (23). The best characterized substrates for MST1/2 are MOB1A, MOB1B (24, 25), LATS1, and LATS2 (26), the core members of the classical Hippo cascade. Other substrates include NDR1/NDR2 (27), FOXO1 (28), and FOXO3a (29). Our initial observations indicated that MST1/2 kinase function inhibited the in vitro activity of PI5P4K. Indeed, we showed by SDS-PAGE autoradiography direct phospho-transfer by MST1 to both PI5P4K⍺ and PI5P4Kβ in vitro (Fig. 1C).

Figure 1. MST and MST2 promote intracellular turnover of PI4,5P₂ by directly phosphorylating PI5P4Ks.

(A and B) Thin-layer chromatography (TLC) and autoradiography of in vitro kinase assays in which recombinant PI5P4Kα or PI5P4Kβ was incubated with MST1 (A) and MST2 (B) prior to testing for phosphorylation of PI5P to generate PI4,5P₂ in the presence of radiolabeled ATP. Data are representative of 3 independent experiments. (C) SDS-PAGE and autoradiography of in vitro kinase assays in which MST1 was incubated with PI5P4K⍺ or PI5P4Kβ in the presence of radiolabeled ATP. Data are representative of 3 independent experiments. (D) Intracellular pools of PI4,5P₂ assessed by immunofluorescence in WT HEK293A cells, cells with complete CRISPR KO of both MST1 and MST2 (MST KO), and MST KO cells expressing WT MST1 (MST1 WT) or kinase-dead MST1 (MST1 KD). Nuclei are stained with DAPI. Area occupied by PI4,5P₂/cell area was quantified in ImageJ and presented in the adjacent graph as area % of WT. A minimum of 20 cells/experiment/group obtained from n=3 independent experiments were quantified. One-way ANOVA followed by Bonferroni correction, *** p≤ 0.001, ** p≤ 0.01. Scale bar, 50μm. (E) TLC and autoradiography of in vitro kinase assays testing for generation of PI4,5P₂ from PI5P by human PI5P4Kβ WT, T386A (phospho-null) and T386D (phospho-mimetic) in the presence of increasing amounts of MST1. Data are representative of 3 independent experiments. (F) Western blotting for FLAG and PI5P4Ks phosphorylated at Thr³⁸⁶ (PI5P4Kβ numbering) in WT or MST KO HEK 293A cells expressing FLAG-tagged PI5P4Kα or PI5P4Kβ. Western Blots were quantified from n=3 independent experiments (right, Student’s t-test, **p<0.01).

We next sought to validate that MST1/2 phosphorylated and inhibited PI5P4Ks in cells. To measure the activity of PI5P4K, we visualized PI4,5P₂ by immunofluorescence using a commercially available antibody as an indicator of PI5P4K activity. To distinguish from PI4,5P₂ generated by the activity of the largely plasma membrane-localized type I PI4P5Ks, we specifically examined intracellular PI4,5P₂, which is predominantly generated by the activity of type II PI5P4Ks. Using this approach we measured intracellular PI4,5P₂ in HEK 293A cells with CRISPR/Cas9-mediated knockout (KO) for both MST1 and MST2 (MST KO, Fig. S1F). In support of the in vitro data, we observed a significant increase in intracellular PI4,5P₂ in the cytoplasm of MST KO cells compared to WT cells (Fig. 1D). This effect was rescued by expression of functional WT MST1 (MST1 WT(30)), but not by that of a kinase dead mutant MST1 (MST KD (30)) (Fig. 1D). Further, we observed a similar pattern in Mst1^−/−Mst2^+/− mouse embryonic fibroblasts (MEFs)(31) (Fig. S1G). We measured a 2.6-fold increased accumulation of intracellular PI4,5P₂ in Mst1^−/−Mst2^+/− MEFs as compared to WT MEFs (Fig. S1H).

We next identified and validated the MST1 phosphorylation site on the PI5P4Ks. MST1 is threonine-specific and prefers aliphatic residues at the +1 position and basic residues, especially lysine, at the +2 position (T-[aliphatic]-K)(32). Thr³⁷⁶ in PI5P4K⍺ and Thr³⁸⁶ in PI5P4Kβ fit MST1’s motif criteria and are located in the substrate binding sites (Fig. S1D). This site and the surrounding sequence are mostly conserved in the C. elegans orthologue (Fig. S1D) and is not present in type I PI4P5Ks (Fig. S1E). To determine if this was the regulatory phosphorylation site, we repeated the in vitro kinase assay for PI5P4Kβ, comparing the activity of PI5P4Kβ WT with that of either a phospho-null T386A PI5P4Kβ or a phospho-mimetic T386D PI5P4Kβ mutant. First, the phospho-mimetic mutant showed no basal activity, consistent with the inhibitory effects of phosphorylation (Fig. 1E). Second, the phospho-null mutant was insensitive to MST1 treatment, consistent with Thr³⁸⁶ being the targeted site (Fig. 1E). Finally, to explore this regulatory mechanism in cells, we raised polyclonal antibodies against the MST1 threonine phosphorylation sites on the PI5P4Ks. To validate these antibodies (PI5P4K-P(K-T-K) #1, #2, #3, Fig. S1I), we expressed either human WT PI5P4K⍺ and PI5P4Kβ or the phospho-null T376A PI5P4K⍺ and T386A PI5P4Kβ in Pip4k2a^−/−Pip4k2b^−/− (⍺βDKO) MEFs. We confirmed that the antibodies recognized PI5P4K⍺ and PI5P4Kβ WT but not their respective phospho-null mutants (Fig. S1I). Using this tool, we assessed the phosphorylation of human PI5P4K ⍺_FLAG and β_FLAG in WT and MST KO HEK 293A cells and measured a significant decrease in phosphorylation at this site in MST KO cells as compared to WT cells (Fig. 1F). This suggests that the phosphorylation of PI5P4K⍺ and PI5P4Kβ at Thr³⁷⁶ and Thr³⁸⁶, respectively, depends on the activity of cellular MST1/2. Based on these data, we propose that MST1/2 potently inhibits PI5P4K activity by directly phosphorylating Thr³⁷⁶ in PI5P4K⍺ and Thr³⁸⁶ in PI5P4Kβ.

PI5P links PI5P4K signaling to the Hippo pathway

Because MST1/2 can inhibit PI5P4Ks, we asked if the resulting change in PI5P4K activity and PI5P/PI4,5P₂ ratio influenced the functional output of Hippo signaling downstream of MST1/2. We tested this hypothesis with MEFs isolated from WT or Pip4k2a^flx/flxPip4k2b^−/− (⍺β DKO) mice. First, we measured the phosphorylation of the adaptor protein MOB1 and found a significant increase in MOB1 phosphorylation in ⍺β DKO cells compared to WT cells, indicating activated Hippo signaling (Fig. 2A). A similar effect was also observed in HEK293A cells with CRISPR/Cas9-generated KO for both PI5P4K⍺ and PI5P4Kβ (⍺β DKO, Fig. S2A). High confluency activates MST1/2 and increase MOB1 phosphorylation. As expected, MOB1 phosphorylation was enhanced with increasing confluency in WT MEFs, whereas ⍺β DKO MEFs showed no further increase in MOB1 phosphorylation (Fig. S2B). Because MST1/2 inhibited PI5P4K activity, we further investigated if the observed increase in MOB1 phosphorylation depended on PI5P4K catalytic activity by re-expressing mouse WT PI5P4K⍺ (PI5P4K⍺ WT) or a kinase dead D359N PI5P4K⍺ (PI5P4K⍺ KD) mutant in ⍺β DKO MEFs. MOB1 phosphorylation was partially restored by re-expression of PI5P4K⍺ WT but remained unchanged by re-expression of PI5P4K⍺ KD (Fig. 2A). We next performed streptavidin bead pull-downs for MOB1 tagged with a streptavidin binding protein tag (SBP) transiently transfected into WT and ⍺β DKO HEK 293A cells. MOB1-LATS binding significantly increased upon PI5P4K loss, suggesting an increase in MOB1-LATS activity (Fig. 2B).

Figure 2. PI5P4K activity inhibits MOB1 phosphorylation and activation of the MOB1-LATS regulatory complex.

(A) MOB1 phosphorylation in WT and ⍺β DKO with or without re-expression of PI5P4K⍺ (⍺ WT) or kinase-dead PI5P4K⍺ (⍺ KD) MEFs. Quantified Western blot results are presented as P-MOB1/MOB1 ratio (right), average of 3 independent experiments (one way ANOVA, followed by Welch’s correction). (B) Co-immunoprecipitation of over-expressed MOB1^SFB and endogenous LATS in WT and ⍺β DKO HEK 293A cells. Quantified Western blot results are represented as LATS1/MOB1^SFB ratio averaged over 3 independent experiments (right, Student’s t-test, **p<0.01). (C) PIP array spotted with eight phosphoinositides at decreasing concentrations and probed with purified human MOB1^MYC. Representative of 3 independent experiments. (D) PIP strip probed with purified human MOB1^WT or a truncated ΔN33–216 MOB1 (MOB1^∆N-term). MOB proteins were immunoblotted with an antibody against His-Tag. Representative of 3 independent experiments. (E) Computed binding energies of 9 positions from PyRx blind docking of PI5P and PI4,5P₂ (16:0/16:0) to MOB1 (p=0.0001 by unpaired t-test). (F) RMSD over 50ns Amber17 MD simulations of MOB1 bound to either PI5P (green) or PI4,5P₂ (blue) using the lowest-energy docked structures as the initial state. (G) RMSF of the indicated N-terminal MOB1 residues over the entire MD simulation. (H) RMSF of the phosphoinositide atoms over the MD simulation. (I) Representative structure from PyRx docking showing interaction of Arg⁸ in MOB1 with the PI5P headgroup. The N-terminus of MOB1 is indicated by a yellow sphere.

Because PI5P4Ks control intracellular pools of PI5P, we hypothesized that the increased MOB-LATS binding in the absence of the most active isoforms of PI5P4Ks may be mediated by PI5P, the substrate of these kinases. Phospholipids regulate components of the Hippo pathway (8–12), although LATS does not appear to bind to PI5P or PI4,5P₂ (9). Therefore, we focused our attention on MOB1. We first used a commercially available PIP array blotted with decreasing amounts of phosphoinositides and observed that Myc-tagged MOB1 immunoprecipitated with PI5P to the greatest degree among all mono-phosphoinositides tested and only showed weak interaction to diphosphoinositide (PI4,5P₂) and triphosphoinositide (PI3,4,5P₃) (Fig. 2C).

Positively charged surfaces within protein structures are important for phospholipid-protein interactions (33). We identified basic residues clustered within the N-terminus of MOB1 that could be responsible for binding to PI5P (Fig. 2D). To determine if altering these residues influences MOB1 interaction with PI5P, we purified recombinant forms of both full-length MOB1 and a N-terminal truncated MOB1 lacking the first 32 amino-acids (MOB1 ∆N-term), which also harbors Thr¹², one of the MST1/2 phosphorylation residue. We used these proteins to blot commercially available PIP strips spotted with various phospholipids, including PI5P and PI4,5P₂. Consistent with the PIP array results, full-length MOB1 interacted with the monophosphoinositides PI3P and PI5P to the greatest extent and less so with PI4,5P₂ (Fig. 2D). In contrast, the truncated version of MOB1 did not bind to monophosphoinositides (Fig. 2D). We mutated the positive residues within this region (Arg⁸, Lys¹¹, Lys¹⁴, Lys¹⁶, and Lys¹⁷) to alanine and examined two mutated residues at a time (K11/14A and K16/17A) as well as all the 5 positively charged residues together (5A mutant, MOB1 5A). Two mutations simultaneously (K11/14A and K16/17A) reduced the PI5P binding on PIP strips but mutating all 5 positively charged residues had a more pronounced effect with the MOB1 5A mutant showing little detectable binding to PI5P (Fig. S2C). Together, these studies are consistent with MOB1 binding to solid-phase PI5P and suggest the binding depends on the presence of basic residues in the N-terminus.

Attempts to co-crystalize the MOB1-PI5P complex failed; thus, computational docking and molecular dynamics (MD) simulations were used to examine these interactions with an AlphaFold2-based model of full-length MOB1 (Fig. S2D). Blind, rigid-body computational docking to MOB1 predicts that PI5P binds to MOB1 with lower energy than PI4,5P₂ (Fig. 2E), with the acyl chains located in a hydrophobic pocket of MOB1 (Fig. S2D). Interaction of PI5P with MOB1 remained stable throughout a 50ns MD simulation (Movie S1). In contrast, PI4,5P₂ interaction with MOB1 became unstable after just 23ns into the simulation, with the acyl chains adopting a rigid conformation and the headgroup moving away from the MOB1 protein (Movie S2). PI5P induced less root mean square deviation (RMSD) in MOB1 structure than did PI4,5P₂ over the course of the simulation (Fig. 2F). Fluctuation of individual MOB1 residues was similar in the PI5P and PI4,5P₂ simulations (Fig. 2G, S2E), except at the N-terminus of MOB1 where PI5P induced less fluctuation (Fig. 2G. S2E). The PI5P ligand itself also fluctuated to a lesser extent than the PI4,5P₂ ligand over the simulation, with largest differences in atoms of the inositol headgroup (atoms 62–91) and the sn2 acyl chains (atoms 92–139, Fig. 2H). All computationally docked structures of PI5P to MOB1 predicted the acyl chains to be buried in a hydrophobic cleft of MOB1, with residues of the N-terminus protecting the acyl chains from solvent (Fig. 2I). Docked models of the MOB1-PI5P interaction predict that Arg⁸ of MOB1 interacts with the 5-position phosphate of PI5P headgroup, whereas Lys¹¹, Lys¹⁴, and Lys¹⁶ do not (Fig. 2I). PyRx docking to multiple crystal structures of the MOB1-LATS complex (PDB:5B5W and PDB:5BRK) predicts that PI5P binds to MOB1-LATS with lower computed energy than does PI4,5P₂, regardless of which crystal structure was used for docking (Fig. S2F-G). Moreover, according to these docked models, PI5P is predicted to bind at the interface between MOB1 and LATS, establishing contact with both proteins. This alignment is consistent with the stabilization of the MOB1-LATS interaction in PI5P4K null cells by exogenous PI5P (Fig. 2B). Together, these data suggest that PI5P interacts with MOB1 in a manner dependent on positively charged residues in the N-terminus of MOB1, an activity we did not observe for PI4,5P₂ in silico. Thus, our data suggest MOB1 selectively interacts with the PI5P4K reaction substrate over the PI5P4K reaction product, consistent with PI5P acting as a second messenger that connects PI5P4K signaling to the Hippo pathway.

Downregulation of YAP activation upon PI5P4K inhibition

The activation of the MOB1-LATS complex leads to phosphorylation of YAP(34), which prevents YAP from translocating to the nucleus and promotes its proteasomal degradation(35–37). Therefore, we next tested whether PI5P4K silencing resulted in increased YAP phosphorylation and decreased nuclear translocation. Inhibition of proteasomal degradation with MG132 significantly increased YAP phosphorylation in ⍺β DKO MEFs compared to WT MEFs (Fig. 3A). Further, re-expressing PI5P4K⍺ WT in ⍺β DKO MEFs decreased YAP phosphorylation, whereas YAP phosphorylation was restored to WT levels in ⍺β DKO MEFs expressing PI5P4K⍺ KD, supporting that the kinase activity of PI5P4Ks are necessary for YAP activation (Fig. 3A). YAP is the effector of the Hippo pathway and when dephosphorylated, YAP translocates to the nucleus where it activates various target genes through its interaction with transcription factors such as TEADs(38, 39). The YAP transcriptional program is essential for development(37) and has been linked to various human diseases such as cancer. To measure the activation of the TEAD transcription factor(40), we expressed the TEAD promotor region fused with luciferase (TEAD^luc), which revealed that TEAD activation was decreased in ⍺β DKO as compared to WT HEK 293A cells (Fig. 3B). Moreover, increasing cell density conditions in WT cells correlated with decreasing luciferase intensity, suggesting a decrease in TEAD activity. However, that decrease was less marked in ⍺β DKO HEK293A cells (Fig. 3B). To confirm these observations, we analyzed YAP localization. By immunofluorescence analysis, we measured an approximately 30% decrease in the nuclear/cytoplasmic YAP ratio in ⍺β DKO as compared to WT MEFs (Fig. S3A). As observed above with MOB1 phosphorylation, higher confluency decreased YAP nuclear localization in WT cells but did so to a lesser extent in ⍺β DKO cells (Fig. S3B). Similarly, we generated MCF10A human breast epithelial lines in which we deleted PI5P4K⍺ by CRISPR/Cas9 gene editing and PI5P4Kβ by a doxycycline (DOX)-inducible shRNA (⍺^−/−;β sh_in, Fig. S3C). Consistent with our results in MEFs, ⍺^−/−;β sh_in MCF10A cells also showed a decrease in the nuclear:cytoplasmic YAP ratio, which was partially rescued upon re-expression of human PI5P4K⍺ WT, but not by that of PI5P4K⍺ KD (Fig. 3C). This observation further validates that the kinase activity of PI5P4Ks has an important role in modulating the Hippo pathway. We next asked whether silencing PI5P4Ks was sufficient to inhibit YAP activity in human cancer cell lines. We stably suppressed PIP4K2B with shRNA and silenced PI5P4K2A with two different PIP4K2A hairpins (α sh_1/β sh, α sh_2/β sh) in both HCC1806 breast cancer cells and in SKOV3 ovarian cancer cells (Fig. S3D). As suggested by our observations in MEFs and in MCF10A cells, YAP nuclear protein amounts were reduced and the nuclear:cytoplasmic ratios were lower in the α sh_1/β sh, α sh_2/β sh lines in both cancer lines when compared to controls (Fig 3D and Fig. S3E).

Figure 3. PI5P4K activity is relevant for YAP activation.

(A) WT and ⍺β DKO MEFs reconstituted or not with ⍺ WT and ⍺ KD were treated with the proteasomal inhibitor MG132 at 5μM or 4 hours and subjected to western blot analysis for phosphorylated YAP. Quantified Western blot results are presented as P-YAP/YAP ratio (right) obtained from 3 independent experiments (Student’s t-test, *p≤0.05). (B) TEAD luciferase assay performed in WT and ⍺β DKO HEK 293A cells under increasing confluency conditions. Data represents random luminescence intensities [RLU] obtained from 5 independent experiments (Students t-test, **p≤0.01). (C) YAP immunofluorescence in SCR and DKO MCF10A cells reconstituted with either ⍺ WT or ⍺ KD MCF10A cells. (scale bar=50μM). Average nuclear: cytoplasmic intensity ratio was quantified from n=3 independent experiments with minimum 20 cells/experiment. Data are presented as a YAP nuclear/YAP cytoplasmic ratio (right). Statistical significance determined by one-way ANOVA followed by Bonferroni, **p≤0.01, *p≤0.05). (D) YAP immunofluorescence in ⍺ sh_1/β sh, ⍺ sh_2/β sh and control (WT CT) HCC1806 breast cancer cells. Average nuclear: cytoplasmic intensity ratio was quantified from 3 independent experiments. Data is represented as a YAP nuclear/YAP cytoplasmic ratio (right). Statistical significance determined by Student’s t-test, **p≤0.01, *p≤0.05. (E) GSEA RNA-seq analysis for the YAP signature in SCR and DKO MCF10A cells (n=3 biological replicates per group). (F) GSEA RNA-seq analysis for the YAP signature in control (WT CT), ⍺ sh1/β sh, and ⍺ sh2/β sh HCC1806 breast cancer cells (n=3 biological replicates per group). (G) qPCR analysis for the relevant top genes and relative values presented in RNA-seq analysis of MCF10A cells (n=3 biological replicates per group). Statistical significance determined by Student T-test, **p≤0.01, *p≤0.05. (H) qPCR analysis for the relevant and relative values presented in RNA-seq analysis of HCC1806 cells (n=3 biological replicates per group). Statistical significance determined by Student’s t-test, **p≤0.01, *p≤0.05.

To confirm the inhibition of YAP activity by PI5P4Ks, we measured the gene expression of two classical YAP target genes Ccn2 (which encodes CTGF) and Ccn1 (which encodes CYR61) (38), which were significantly decreased in ⍺β DKO MEFs (Fig. S3F). To further understand the downstream effects of silencing PI5P4Ks on YAP activity, we performed RNA sequencing (RNA-seq) analysis to compare gene expression. Gene set enrichment analysis (GSEA) of the genome-wide dataset revealed a significant decrease in the expression of the Cordenonsi YAP signature gene list (41) in ⍺^−/−;β sh_in MCF10A cells (Fig. 3E) and in α sh_1/β sh, α sh_2/β sh relative to control cells (Fig. 3F). In MCF10A cells, some of the top hits were genes encoding for growth factors such as TGFB2 and FGF2 and were confirmed by qPCR (Fig. 3G). To validate the importance of PI5P4Ks in regulating YAP transcriptional activity, we overexpressed either PI5P4K⍺ or PI5P4Kβ (Fig. S3G) in MCF10A cells, both of which increased the expression of these genes (Fig. S3H). Moreover, in HCC1806 cells, one of the top hits identified was the gene encoding the tyrosine kinase receptor AXL, a metastasis driver in human cancers (42, 43), and the gene encoding SLIT2, a modulator of cancer cell migration (44, 45). We confirmed this strong decrease in gene expression of the genes encoding CCN2, SLIT2 and AXL by qPCR analysis (Fig. 3H). In ovarian cancer cells, we also measured reduced expression of the genes encoding CCN2, CCN1 and SLIT2 (Fig. S3I). Together, these observations demonstrate the regulation of YAP transcriptional activity by PI5P4Ks in cancer cells.

PI5P4Ks regulate the functional output of the Hippo pathway

Because several hits identified in the RNA-seq datasets have been implicated in various aspects of epithelial to mesenchymal transition (EMT)(46), a YAP-driven program, we performed GSEA analyses for the standard EMT signature (Hallmark Epithelial to Mesenchymal Transition). Indeed, in both MCF10A and the HCC1806 breast cancer cell line, the EMT signature was enriched in control groups compared to groups lacking or with reduced PI5P4K expression (Fig. S4A). In MCF10A DKO cells, the expression of the gene encoding the EMT driver TWIST1 was decreased in our RNA-seq data set, which was confirmed by qPCR analysis (Fig. 3G). YAP activates TWIST1 transcription, along with other EMT drivers (47). The top hits for decreased expression in HCC1806 cells with decreased PI5P4K expression included genes encoding vimentin, FN1, and CDH2 (N-cadherin), known markers of EMT (Fig. S4A). A standard indicator of EMT is loss of the cell adhesion protein E-cadherin. YAP and E-cadherin are tightly linked: YAP has been proposed to be a target of E-cadherin and conversely, YAP can alter the activity of E-cadherin through its transcriptional targets as well as through protein-protein interaction with the small GTPase RAC1. ⍺^−/−;β sh_in MCF10A cells showed a 1.5-fold increase in E-cadherin immunofluorescence, an increase that was partially reversed by expression of WT PI5P4K⍺, but not kinase-dead PI5P4K⍺, again emphasizing the importance of kinase activity (Fig. 4A). Next, we assessed EMT phenotypes that are transcriptionally regulated by YAP(46, 48), specifically migration and invasion. Migration was examined by wound healing assays, which showed that ⍺^−/−;β sh_in cells had a significantly decreased migration speed compared to SCR cells after wounding (Fig. 4B). We also examined migration capacity in HCC1806 breast cancer cells. Cells with double knock-down of PIP4K2A and PIP4K2B migrated approximately 50% less than control cells (Fig. 4C). Next, we analyzed the in vitro invasiveness of MCF10A cells stimulated with a serum gradient, which revealed that a significantly lower percentage of ⍺^−/−;β sh_in cells invaded through the matrigel membrane when compared to SCR MCF10A cells (Fig. 4D). A similar difference was observed in α^sh_1/β^sh and α^sh_2/β^sh HCC1806 cells when compared to controls (Fig. 5E). These data support the decrease in the expression of EMT-related genes as shown by RNA-seq analyses and collectively demonstrate that PI5P4K ablation inhibits the YAP pathway as well as YAP-related phenotypes.

Figure 4. Silencing of PI5P4Ks triggers a decrease in key YAP-related phenotypes.

(A) Immunofluorescence of E-cadherin in SCR, DKO, and DKO cells reconstituted with PI5P4K⍺ (⍺ WT) or kinase dead mutant (⍺ KD) MCF10A cells. Quantification of images are resented as normalized average intensity. E-cadherin in yellow, nuclei (DAPI) in blue. Scale bar=50μm. Statistical significance determined by one-way ANOVA followed by Bonferroni post, ***p≤ 0.0001, **p ≤ 0.001, *p≤0.05. (B and C) Wound healing assays performed in a confluent monolayer of SCR and DKO MCF10A epithelial cells (B) and WT CT, ⍺ sh1/β sh and ⍺ sh2/β sh HCC1806 breast cancer lines (C). Images were taken every two hours for 24 hours after scratching (T24). Wound areas were quantified from 3 independent experiments with 3–4 technical replicates per experiment and presented as percent area migrated (right). Statistical significance was determined by two-tailed Student’s t-test (** p ≤ 0.01), scale bar=250μm. (D and E) SCR and DKO MCF10A cells (D) and WT CT, ⍺ sh1/β sh and ⍺ sh2/β sh HCC1806 breast cancer cells (E) were serum starved overnight and transferred to invasion chambers on top of a matrigel membrane in serum-depleted media. Bottom chamber contained complete media, creating a gradient. 24 hours after transfer, cells were fixed and stained with 0.05% crystal violet. Scale bar=500μm. The total number of cells that migrated over n=3 independent experiments were counted and raw values are presented in the graph. Statistical significance determined by Student’s t-test (**p ≤ 0.01). (F) Schematic depicting the interaction between PI5P4Ks and the Hippo pathway. The proposed model shows that through regulation of PI5P pools, the lipid kinase activity of PI5P4Ks modulates the formation of the MOB1-LATS complex, triggering the activation of the downstream target YAP. MST1/2, the protein kinases of the Hippo pathway upstream of LATS1, regulate the activity of PI5P4Ks, thus controlling PI4,5P₂/PI5P turnover.

To assess the potential relevance of this PI5P4K-Hippo pathway interaction to human cancers, we correlated the two signatures (Cordenonsi YAP conserved signature and hallmark EMT) and PIP4K2A expression in human patients with invasive breast carcinoma from the TCGA Pan-Cancer Atlas data base. Higher expression of PIP4K2A correlated with significantly higher expression of YAP and EMT signatures (Fig. S4B). Using the same data set, we further confirmed that CCN2, CCN1, AXL and SLIT2 expression positively correlated with PIP4K2A expression (Fig. S4C). Finally, we extended our correlative analysis between PIP4K2A expression and the Cordenonsi YAP conserved signature (Fig. S4D) and the hallmark EMT signature (Fig. S4E) to other human cancers and again observed significant and strong positive associations in ovarian, colorectal adenocarcinoma, prostate and head and neck cancers. These correlative analyses hint at the utility of future studies regarding the potential role of PI5P4Ks in YAP-driven cancers.

DISCUSSION

The present work highlighted the core Hippo pathway kinases MST1/2 as direct regulators of PI5P4Ks and revealed a role for PI5P4Ks in regulating the Hippo pathway. Inhibition of the PI5P4Ks stimulated MOB1 phosphorylation and formation of the MOB1-LATS complex. This correlated with increased YAP phosphorylation and decreased nuclear localization. Finally, we found that YAP-related phenotypes were reduced by PI5P4K deficiency, thus demonstrating that these phosphoinositide kinases play a role in regulating YAP function.

Finding upstream regulators for PI5P4Ks will enhance our understanding of the cellular roles PI5P4Ks and phosphoinositide regulation. PI5P4Ks are regulated by only a small number of protein kinases, with cellular stress being a common denominator. The stress MAP4K p38 is the only known kinase that directly phosphorylates Thr³²⁶ in PI5P4Kβ, which inhibits its activity in response to UV irradiation stress, leading to an increase in nuclear PI5P(3). Similarly, the isomerase Pin1 decreases PI5P4K activity(49). Our efforts led to the identification of the core Hippo pathway kinases MST1/2 and other GCKs as specific inhibitors of the activity of PI5P4Ks. Both enzymatically active isoforms PI5P4K⍺ and PI5P4Kβ seem to ultimately be inhibited by cellular stressors, which leads to increased PI5P(49, 50). This body of previous work makes the case for PI5P as a secondary messenger for stress signaling. Another regulator of PI5P4Ks highlighted here is HPK/GCK-like kinase (HGK/MAP4K4), which directly interacts with and activates LATS1(51). In fact, MST1/2 is dispensable for LATS1/2 activation and can be replaced by other regulators such as HGK/MAP4K4, suggesting a degree of redundancy in the regulation of the Hippo pathway. Thus, at least two kinases that impinge on the Hippo pathway can also regulate PI5P4Ks. Further research will be necessary to determine the degree to which HGK/MAP4k4 utilizes PI5P to enhance LATS1/2 activation and inhibit YAP activity.

MST1/2 are protein kinases central to cellular stress responses(29, 52–54). They can be regulated by cell-cell contact, oxidative stress(29), energy(54) and growth factors(55, 56). Thus, we propose that, by activating MST1/2, stress can stimulate an increase in PI5P pools by attenuating PI5P4K activity. To support this hypothesis, we showed that knocking down PI5P4Ks also promoted MOB1-LATS complex formation. Activation of MST1/2 leads to inhibition of YAP nuclear translocation and activity. Indeed, knocking down PI5P4Ks led to reduction of YAP nuclear translocation and YAP gene expression signatures, which could not be rescued by a kinase dead PI5P4K. Together, our data support a role for PI5P4K activity in regulating the functional output of the classical Hippo pathway by PI5P signaling. Lastly, we demonstrate that PI5P4Ks affected YAP-related phenotypes, more specifically in regulating the EMT program. Indeed, YAP regulates EMT(57, 58) and various EMT drivers(47), along with phenotypes such as migration and invasion(48). YAP overexpression induces EMT(58), whereas YAP inhibition can stop EMT progression(57). In the same way, ablation of PI5P4Ks in MCF10A epithelial cells and breast cancer cells reduced EMT gene signatures and invasion and migration.

PI5P is the least abundant phosphoinositide in the cell. There are several lipid kinases and phosphatases that directly contribute to PI5P synthesis. Three 3-phosphatases (MTM1, MTMR2 and MTMR3) likely produce PI5P from PI3,5P₂. PIKfyve is a 5-kinase that contributes to PI5P biosynthesis by phosphorylating PI3P to PI3,5P₂. Finally, 4-phosphatases I and II recycle PI5P from PI4,5P₂. Some of these enzymes stimulate EMT events such as invasion and migration. For instance, PIKfyve regulates the invasion of cells through TIAM-1-dependent cytoskeletal regulation(7). However, TIAM1 also antagonizes YAP signaling and inhibits invasion of epithelial cells(59). Further, both PIKfyve and MTMR3 may promote invasion by activating Rac1(60) presumably through PI5P production. On the other hand, PI5P4Ks are 4-lipid kinases that consume PI5P and have been suggested to affect cellular processes by turning over PI5P rather than through their product, PI4,5P₂, due to the lower abundance of PI5P compared to that of PI4,5P₂. Our data propose one such mechanism for PI5P4Ks in regulating PI5P pools available for the Hippo pathway activation which, to a certain extent, does not fully agree with more established roles for PI5P. These observations highlight the complexity of phospholipid regulation. Although phospholipid dynamics is still not fully understood, the functions of various phospholipids are predicted to be context-dependent in terms of subcellular localization or physiological condition. It remains unclear how MOB1 might acquire phospholipid ligand. As an initial hypothesis, MOB1 might be expected to interact with membrane lipids not unlike how phosphatidylinositol-transfer proteins (PITP)(61–64) or phosphoinositide-binding nuclear receptors (SF-1)(65–68) acquire phospholipids. Although the details of how PITPs acquire phospholipids is better defined(63), how proteins like SF-1 or MOB1 might accomplish this task are almost completely uninvestigated.

Furthermore, whether lipids are localized at the plasma membrane, internal membranes and potentially lipid rafts, and the other lipid kinases and phosphatases in the vicinity will most likely dictate phospholipid dynamics and function. Cellular localization could prove essential to the cellular roles for PI5P. Likewise, studying the different lipid kinases involved in regulating PI5P and their dynamics and interactions in specific cellular contexts could further our understanding of PI5P biology. Overall, our study uncovered putative PI5P4Ks regulators as well as a previously unknown signaling cascade. Although PI5P4Ks are relevant in various disease contexts such as cancer, the mechanisms and pathways are still not fully understood. Our study highlights a connection between PI5P4Ks and the Hippo pathway, with potential implications for development and cancer biology.

MATERIALS AND METHODS

Cell lines

All cells were incubated in a 37°C humidified incubator with 5% CO₂. Mouse embryonic fibroblasts (MEFs) and human epithelial kidney (HEK) 293A cells were cultured in Dulbecco’s modified Eagle medium (Corning 10–013-CV). Media was supplemented with 10% fetal bovine serum (Life Technologies), 100 U/ml penicillin/streptomycin (Gibco; 15140122). MCF10A human breast epithelial cells were grown in DMEM/F-12 medium (Gibco; 11330032) supplemented with 5% horse serum (Gibco 16050122), 20ng/ml EGF (Sigma-Aldrich; E9644), 10μg/ml insulin (Sigma-Aldrich; I1882), 100ng/ml cholera toxin (Sigma-Aldrich; C8052), 0.5mg/ml hydrocortisone (Sigma-Aldrich; H0888) and 100 U/ml penicillin/streptomycin.

MEFs were isolated from E13.5 embryos and genotyped as described previously(69). Pip4k2a^flx/flxPip4k2b^−/− MEFs were infected at a MOI=10 of FIVCMVCre VSVG to knockout Pip4k2a from the lines and generate Pip4k2a^−/−Pip4k2b^−/− MEFs (αβ DKO MEFs). Rescue lines expressing mouse WT PI5P4Kα (α WT) and kinase-dead PI5P4Kα D359N (α KD) were also generated by retroviral infection. For MCF10A cells, PIP4K2A was initially deleted using CRISPR and cells were lentivirally infected with a doxycycline-inducible shRNA targeting PIP4K2B and selected in 2ug/ml puromycin. To generate MCF10A lines lacking both isoforms of the kinases, cells were treated with 3ug/ml doxycycline for 3 consecutive days before carrying out experiments in the presence of doxycycline. The PIP4K2A^−/−;PIP4K2B^−/− HEK 293A cell line was generated using CRISPR as previously described (70). MCF10A breast epithelial mCherry CT, mCherryPI5P4Kα (α OE) and mCherryPI5P4Kβ (β OE) fusion overexpression lines were generated by lentiviral infection of MCF10A cells with pRRL mCherry, pRRL mCherry-PIP4K2A and pRRL mCherry-PIP4K2B. Infected cells were selected using mCherry fluorescence by fluorescence activated cell sorting (FACS).

The cancer cell lines HCC1806, SKOV3 and HCC116 were purchased from ATCC. PIP4K2B was stably knocked down by lentiviral transduction of shRNA. HEK293T cells were cotransfected with 12 μg of each shRNA containing pLKO.1 plasmid, 8 μg of psPAX2 packaging plasmid, and 4 μg of pMD2.G envelope plasmid using Lipofectamine 2000 (Invitrogen) in 10-cm plates. After 16 h, medium was replaced with fresh 10% FBS-DMEM without antibiotics, and 48 h later, the viral supernatant was collected and filtered through a 0.45-μm filter. HCC1806, SKOV3 and HCC116 cells were transduced with viral supernatant and 8 μg/ml polybrene, and stably infected cells were selected with 1 μg/ml puromycin. Knockdown of PIP4K2A was performed prior to each experiment. One day after seeding, cells were transduced with either control or αsh1 or αsh2 shRNA (Day 0). After 24 hours puromycin was added to the growth at a concentration of 1 ug/ml (Day 1). After 48 hours of selection cells were seeded for the respective experiments on day 3 after infection.

The 293T packaging cell line was used for retroviral and lentiviral amplification. In brief, viruses were collected 48 hours after transfection, filtered, and used for infecting cells in the presence of 10μg/ml polybrene (Santa Cruz Biotechnology; NC9840454) prior to puromycin selection or upon cell sorting for mCherry. pCL-Ampho and pCL-Eco were used for retroviral packaging for human and mouse cells respectively. pVSVg and pPax were used for lentiviral packaging. MST1/ MST2 CRISPR double knock-out (MST KO) HEK293A cells were a kind gift from Dr. Kun Lian Guan(51).

Coimmunoprecipitation

Cells were transfected with either MOB1B or LATS1 with a C-terminal SFB tag (S-tag;Flag-tag; Streptavidin Binding Protein-tag). Plasmids were transfected using Lipofectamine™2000 reagent (Thermo Fisher Scientific 11668019) as per manufacturers recommendations. In brief, cells at 70–80% confluency were incubated for 4 hours at 37°C with Lipofectamine-DNA complex (2:1). Fresh media was added and cell were incubated for 24 hours prior to treatment and processing. Cells were lysed in IP buffer (25mM Tris 7.4, 150mM salt, 5% glycerol and 1% NP40) with Halt protease and phosphatase inhibitor cocktail (Thermo Fisher Scientific, 78440). Equal amounts of protein of each sample were incubated with pre-washed streptavidin magnetic beads (New England BioLab^® S1420S) overnight at 4°C. Beads were thoroughly washed and the proteins were eluted under denaturing conditions in SDS-containing buffer and immunoblotted. MOB1B and LATS1 with a C-terminal SFB tag (S-tag;Flag-tag; Streptavidin Binding Protein-tag) or SFB alone kindly provided by Dr. Wenqi Wang.

In vitro kinase assay

For phosphorylation reactions, predetermined concentrations of protein kinase and lipid kinases were added to master mix containing 50 μM of ATP (Sigma), 0.1 mg/mL of BSA and 1x Assay Buffer I (SignalChem) in 40 μl total volumes. Reactions were carried out at 30°C for 30 minutes. Next, 0.01 mCi/mL ³²P-labeled ATP (Perkin Elmer) and 25 μM PI5P:PS or PI4P:PS liposomes (Avanti) were included in the kinase reaction, bringing total volume to 100 μL. Additional cold ATP was included to maintain its concentration at 50 μM. Lipid phosphorylation reactions were carried out at room temperature for 20 min. 50 μl of 4N HCl was added to quench the reaction and the lipid was extracted with 100 μl of CHCl3: MeOH (1:1) followed by vortexing for 1 minute and centrifugation at 14,000XG for 2 minutes. 10 μL of the lower hydrophobic phase was extracted with gel loading pipet tips and spotted onto a silica plate (EMD Millipore #M116487001) for thin layer chromatography. Plates were placed in a sealed chamber with 1-propanol: 2M acetic acid (65:35) mobile phase. After completion, the plates were removed, briefly dried, exposed to a storage phosphor screen BAS-IP (GE Life Sciences) and imaged with a Typhoon FLA 7000 phosphorimager (GE) and quantified by ImageQuant (GE). PI5P:PS or PI4P:PS liposomes were prepared by mixing lipid substrate with 2-fold mass excess PS in chloroform. The chloroform was evaporating with N₂ gas, and the lipid mixture was hydrated in 30 mM HEPES pH 7.4, 1 mM EGTA and dispersed by bath sonication.

In vitro phosphorylation

Predetermined concentrations of MST1 and PIP4K2A and PIP4K2B were incubated in 50 μM of ATP, 0.01 mCi/mL ³²P-labeled ATP (Perkin Elmer), and 1x Assay Buffer I (SignalChem) at 30°C for 30 minutes. Reactions were run on SDS-PAGE and stained with Coomassie. Bands corresponding to PIP4K2A and PIP4K2B were examined for radioactivity incorporation by imaging with a Typhoon FLA 7000 phosphorimager (GE).

MOB1 expression and purification

Full length wild type MOB1B (MOB1 WT, residues 1–223) and alanine mutant MOB1B (MOB1 5A, residues 1–223) were subcloned into a pET21a vector with a C-terminal 6XHIS tag with ampicillin selection. For MOB1 5A, mutated residues were all contained in the N-term (R8A, K16A, K19A, K21A, K22A). All constructs were transformed into BL21 DE3 bacteria and purified by affinity chromatography. Briefly, single clones were grown in LB media under ampicillin selection at 37°C until OD600 reached ~0.5. Cultures were chilled on ice, induced at OD600 = 0.6 with 1mM IPTG for 18hr at 14°C, shaking at 100RPM. Cells were harvested by centrifugation for 15min at 3,000xg and resuspended in lysis buffer (50mM Tris HCl (pH 8.0), 150mM NaCl, cOmplete Protease tablets, and 1mg/mL lysozyme). Lysates were incubated on ice for 1hr with gentle rocking prior to sonication at 70% power with Branson Sonicator equipped with a titanium microtip on ice at 4°C for 10 minutes. Lysates were centrifuged at 14,000xg for 60 minutes. Supernatants were incubated for 1hr with HisPur Ni-NTA metal affinity resin (Thermo Scientific). The beads were washed 2X with 10 column volumes (CV) of Buffer A (50mM Tris HCl (pH 8.0), 150mM NaCl, and 25mM imidazole) and loaded onto a gravity column, washed with 10CV of Buffer A and eluted with 2CV Buffer B (50mM Tris HCl (pH 8.0), 150mM NaCl, and 300mM imidazole). The eluate was dialyzed in QBuffer A (50mM Tris HCl (pH 8.0)) overnight at 4C and loaded onto a HiTrap Capto Q column (Cytiva) pre-equilibrated with 50mM Tris HCl (pH 8.0) and eluted with QBuffer A + 1M NaCl. Eluted fractions corresponding to MOB1 WT or MOB1 5A mutant were pooled and concentrated using a 4mL Amicon® 10kDa filter prior to being aliquoted and stored at −80C. All samples analyzed were from aliquoted, frozen samples, and native gel electrophoresis showed no change in native gel migration after 1 freeze thaw cycle. MOB1B sample protein sequence was confirmed to be full-length MOB1B by native mass spectrometry.

Structural modeling of disordered regions in full-length MOB1

The crystal structure of MOB1B (PDB:5B5V) has disordered regions at both the N- and C-termini and at residues 100–105. To repair these regions missing from the crystal structures for computational docking studies to the full-length MOB1B, the MOB1B crystal structure (PDB:5B5V) was modeled by several independent approaches, including DMPfold(71), RaptorX(72, 73), InteractiveROSETTA(74) and AlphaFold2(75). The structure derived from AlphaFold2(75) had the lowest overall energy and was most similar to 5B5V and other structures of MOB1B in the protein database (Alphafold2 model RMSD with PDB:4JIZ = 0.306Å; PDB:5B5W = 0.701Å; PDB:5BRK = 0.448Å; PDB:5YF4 = 1.208Å). The full length AlphaFold2 model was therefore used for computational docking studies.

Computational phosphoinositide docking

Prior to docking, all phosphoinositide ligands were prepared by completing 200 steps of an energy minimization protocol using a universal force field (UFF). Global, undirected docking of dipalmitoyl PI5P (C16:0/C16:0) (1-(1,2-dioctanoylphosphatidyl)inositol-5-phosphate) or dipalmitoyl PI4,5P₂ (C16:0/16:0) (1-(1,2-dihexadecanoylphosphatidyl)inositol-4,5-bisphosphate) to the AlphaFold2 predicted structure of full-length MOB1B was performed using python-based PyRx(76) in a box defined as 55.9Å X 54.9Å X 57.0Å centered on the midpoint of the MOB1 protein (x = 0.68277982839, y = −3.4843207781, z = 0.510424592091). This global docking approach minimized investigator-driven bias toward any specific binding region within MOB1. For phosphoinositide docking to the MOB1/LATS complex, similar unbiased global docking of UFF-minimized PI5P (C16:0/C16:0) and PI4,5P₂ (C16:0/C16:0) to crystal structures of the MOB1-LATS complex (PDB:5B5W and PDB:5BRK) were performed, as above. For 5B5W, the box was defined as 43.8Å X 68.0Å X 69.2Å centered on the midpoint of the 5B5W crystal structure (x = 11.1273943743, y = 28.4483213519, z = 45.9137039229). For 5BRK, the box was defined as 50.3Å X 66.3Å X 64.6Å centered on the midpoint of the 5BRK crystal structure (x = 24.7530485596, y = 40.3357547512, z = 7.6352332862). All solvent-exposed regions of the MOB1 protein were accessible to ligand during all docking runs.

Molecular dynamics simulations

To compare dynamics of PI5P and PI4,5P₂ binding to MOB1 WT and computationally test the stability of these interactions, we performed 50ns molecular dynamics (MD) simulations on MOB1 WT-PI5P and MOB1 WT-PI4,5P₂ complexes created by docking described above. Complexes were prepped for MD simulation using the CHARMM-GUI(77). Briefly, the complex was solvated in an octahedral box of explicit TIP3P water with a 10-Å buffer around the protein complex. A total of 26 potassium (K⁺) and 22 Cl^- ions were added to neutralize the system charge and simulate physiologic conditions. SHAKE(78) bond length constraints were applied to all bonds involving hydrogen. Non-bonded interactions were evaluated with a 9-Å cutoff and calculated with a particle-mesh Ewald summation method(79, 80). The MD system was first minimized for 100 steps using steepest descent followed by 5,000 steps of conjugate gradient minimization. The system was heated to 300K over 125ps in the NVT ensemble using a step size of 1fs. After switching to the NPT ensemble, positional restraints on protein and ligand were gradually removed and Langevin dynamics with a collision frequency of 1.0ps⁻¹ were used for temperature regulation. Production MD was conducted for 50ns using a step size of 2fs with constant pressure periodic boundary conditions, isotropic pressure scaling and Langevin dynamics. MD trajectories printed every 10ps and 5000 frames were created for further analysis. All minimizations and MD simulations were performed with Amber17(81) employing the ff14SB force field for proteins, GAFF2 force field for the phosphoinositide ligands, and the TIP3P force field for water molecules(82). Analysis of MD trajectories was performed using the CPPTRAJ module(83) of AmberTools. Root mean square deviations (RMSD) and root mean square fluctuations (RMSF) were calculated on all non-hydrogen atoms of protein residues for each frame in the trajectory using the initial structure as the reference. RMSF was also calculated for all atoms of PI5P and PI4,5P₂ for each frame in the trajectory, also using the initial state as the reference.

Immunofluorescence

PI4,5P₂ staining for intracellular membranes was done using the protocol described(84). In brief, cells were fixed in 2% pre-warmed methanol free PFA followed by permeabilization with 20 μM digitonin. Samples were blocked in 5% normal goat serum in buffer A (20mM PIPES pH 6.8, 2.7mM KCl, 137 mM NaCl) with 50mM NH₄Cl before being incubated for 1hour at room temperature in primary antibody against PI4,5P₂ (1:100) (Echelon Biosciences; Z-P045–2). Samples were washed twice before incubation with secondary antibody for 45 min at room temperature, followed by post-fixation with 2% PFA in PBS. During the final washes, DAPI was added to the wash buffer. The samples were imaged with 40x oil objectives on a Zeiss LSM710 microscope. Total cell area and % area represented by PI4,5P₂ was quantified in ImageJ. For YAP (Cell Signaling Technology^®, S14074 YAP (D8H1X) XP^® rabbit monoclonal antibody) and E-cadherin (Proteintech, 60335–1 E-Cadherin rabbit monoclonal antibody) imaging, cells were fixed with 4% PFA and blocked in 0.3% Triton-X with 5% normal goat serum in PBS for 1hour at room temperature. Antibodies were diluted 1:100 in antibody dilution buffer (0.3% Triton-X with 1% BSA in PBS) and incubated overnight at 4C. Secondary antibodies were diluted 1:1000 in antibody dilution buffer and incubated for 1hour at room temperature. During the final washes, DAPI was added to the wash buffer. The samples were imaged with a 20X objective using the Biotech Imager. Average intensity was measured in Image J.

Western blotting

Cells were lysed with RIPA lysis buffer containing 1% Triton X-100, 0.1% SDS, 150 mM NaCl, 50 mM Tris-HCl pH 7.5 supplemented with Halt protease and phosphatase inhibitor cocktail (Thermo Fisher Scientific, 78440). Protein concentrations were quantified using the BCA method (Thermo Scientific; 23225) and 20μg of total cell lysates were run on an SDS–polyacrylamide gel electrophoresis. Proteins were transferred on to a nitrocellulose membrane, and membranes were probed overnight at 4°C with the appropriate primary antibody. See Table S1 for a complete list of antibodies used. Membranes were incubated with appropriate secondary antibodies and visualized using a LI-COR Odyssey® Imager. All quantifications were done using ImageJ.

PIP Array

PIP Array (P6100, Echelon) were incubated with recombinant human MOB1B C-MYC/DDK (TP306337, OriGene) and processed per the manufacturer’s instruction. In brief, strips were blocked in 3% free fatty acid (FFA) BSA for 1H at room temperature and incubated with 0.5ug/ml of protein in 3% FFA BSA overnight at 4ºC. Strips were thoroughly rinsed in TBST and incubated for 2 hours at room temperature with Myc antibody, then with HRP-conjugated secondary antibody for 45 min. Standard Western blot developing procedures were followed.

PIP strips

PIP strips were incubated with recombinant MOB1B (XM-005265709) protein with a C-terminal 6X His tag (MOB1 WT construct made in Emerling lab; the three mutant MOB1 constructs, MOB1 K11/14A, MOB1 K16/17A and MOB1 5A, were synthesized by GENEWIZ). Proteins were expressed in competent E. coli BL21(D3) (New England BioLab^® C2527). E. coli single colonies housing the appropriate plasmids were grown in LB cultures until O.D. = 600nm. Protein synthesis was induced with 0.5 mM IPTG at 18°C overnight. Protein extraction was done with B-PER™ Bacterial Protein Extraction Reagent (Thermo Scientific™ 78243) containing lysozyme (100ug/ml) and DNase I (5U/ml). Protein was isolated using Ni-NITA magnetic beads (New England Biolabs). Synthesized proteins were visualized by SDS-PAGE and proteins were quantified using Quick Start™ Bradford Reagent (Bio-Rad 5000205). PIP strips (P6002, Echelon) were processed per the manufacturer’s instructions. In brief, strips were blocked in 3% FFA BSA for 1H at room temperature. Following blocking, they were incubated with 10μg total protein 3% FFA BSA overnight at 4ºC. Strips were thoroughly rinsed in TBST and incubated for 2 hours at room temperature with His (Proteintech) antibody, then with secondary antibody (Li-COR) for 45 min. Standard Western blot developing procedures were followed.

RNA extraction and qPCR

Total RNA was prepared using Directzol RNA MiniPrep (Zymo Research Corporation; 50–444-628). cDNA was synthesized using High-capacity cDNA Reverse Transcription kit (ThermoFisher Scientific; 4368814) and qRT-PCR performed utilizing PowerUp SYBR green (ThermoFisher Scientific; A25742) and the LightCycler 96 (Roche). See Table S2 for a complete list of primers.

RNA sequencing and alignment

RNA was extracted from cell pellets washed with PBS using Directzol RNA MiniPrep (Zymo Research Corporation; 50–444-628). RNA yield and purity was tested using the Qubit 4 Fluorometer prior to analysis. Sequencing libraries were prepared using 250ng RNA using standard Illumina Tru-seq single indexing protocols and sequenced using the Illumina NextSeq 500 instrument. Adapter remnants of sequencing reads were removed using cutadapt v1.18(85). Read alignment was performed using human genome v.38 and Ensembl gene annotation v.84 using STAR aligner v.2.7(86). DESeq2 was used for differential gene expression analysis(87).

Gene set enrichment

Gene set enrichment analysis (GSEA) was performed using GSEA v4.0.3. Normalized enrichment scores (NES) and p values were used to determine the significance of findings.

Wound healing assay

Scratch assays were performed in 24 well plates. In brief, 3.5X10⁵ cells were plated in each well, three-four technical replicates per condition. 24 hours post-plating, the scratch was performed using an AutoScratch Wound Making Tool (BioTek). Cells were thoroughly washed and fresh media was added. Wound closure was monitored for 24 hours on a Cytation5 imager (BioTek) using the Scratch Assay App (BioTek). Wound area was measured using ImageJ.

Invasion assays

Invasion assays were performed in BioCoat™ Matrigel® Invasion Chambers (Corning; 354480) following the manufacturer’s instructions. MCF10A cells were FBS-deprived overnight. Prior to plating, invasion inserts were rehydrated with FBS free media for 2 hours in a humidified tissue culture incubator at 37°C and 5% CO₂. Cells were resuspended in FBS free media at 1X10⁵ cell/ml. Chemoattractant media (media containing FBS) was added to the companion 24 well plates, and rehydrated inserts placed in the plate. Cells were added to the inserts and incubated overnight. Cells were fixed with 70% ethanol for 20 minutes at room temperature, thoroughly washed with diH₂O and stained in a solution of 0.05% crystal violet for 30 minutes at room temperature. After thoroughly washing the excess stain, cells in the upper chamber were carefully removed with a cotton swab and the membranes were allowed to dry for at least 24 hours prior to mounting on microscopy slides. For each group, at least 2–3 technical replicates were assessed per experiment. Images were acquired using an Olympus microscope.

Statistics

Data are expressed as means SD. Data was verified for normality using Shapiro-Wilk normality test. Statistical analyses for all data, including microscopy quantification and qPCR were performed by Student’s two-tailed t-test or ANOVA with Bonferroni’s method used to correct for multiple comparisons, as indicated, using GraphPad Prism. Statistical significance is indicated in the figures. *p<0.05, **p<0.001, and ***p<0.0001 unless specified otherwise.

Data and material availability:

The RNA-seq data have been deposited into the Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/) with the accession ID GSE 221217. All other data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials. All materials are available from B. M. E. or R. D. B. upon reasonable request under a materials transfer agreement from Sanford Burnham Prebys and Vanderbilt University.