Phosphatidylserine and RhoB connect PI4P and PA metabolism to maintain plasma membrane identity

Abstract

Proper functions of cellular organelles require tight control of membrane phospholipid composition, yet the mechanisms by which lipid imbalances are sensed and corrected remain largely unknown. Here, we present evidence of an unexpected metabolic connection between plasma membrane (PM) phosphoinositide metabolism and two key anionic lipids, phosphatidylserine (PS) and phosphatidic acid (PA). Prolonged depletion of PM phosphatidylinositol 4-phosphate (PI4P) by pharmacological inhibition of PI 4-kinase IIIα (PI4KIIIα/PI4KA) increases phospholipase D (PLD) activity and PA levels in the PM. Using lipidomics, RNA-seq, and proximity proteomics, we find that PI4P loss induces a concomitant decrease in PS, activating a reciprocal relationship between PS synthesis and PLD-mediated PA generation. These changes also drive transcriptional and translational upregulation of the small GTPase RhoB, which enhances PLD-mediated PA synthesis and actin cytoskeletal remodeling. Because reduced PI4KA activity underlies numerous hereditary diseases, our studies reveal how perturbation of PM phosphoinositide synthesis triggers an integrated response that maintains the anionic character and structural integrity of the PM.

Introduction

The phosphatidylinositol (PI) cycle, first described by Lowell and Mabel Hokin >70 years ago, encompasses the accelerated hydrolysis and resynthesis of phosphoinositides (PPIns) at the plasma membrane (PM) in cells stimulated by secretagogues (Hokin and Hokin, 1958a; Hokin and Hokin, 1958b; Hokin and Hokin, 1959). Subsequent work has shown that stimulated PLC-mediated hydrolysis of PI 4,5-bisphosphate [PI(4,5)P₂] and the resultant increased metabolic flux through the PI cycle intermediates generates important lipid and soluble intermediates, including diacylglycerol (DAG), phosphatidic acid (PA), and inositol trisphosphate. All of these molecules are critical components of the signal transduction machinery to alter a variety of cellular functions (Epand, 2017). Though the PI cycle was originally described in excitable, secretory tissues featuring high rates of exocytosis and compensatory endocytosis, it is a universal feature of all eukaryotic cell types, though rates of flux through the PI cycle can vary considerably depending on cell type and physiological state, e.g., activation of GPCR–Gq or receptor tyrosine kinase signaling, which leads to PLC-mediated PI(4,5)P₂ hydrolysis (Jensen et al., 2009; Jensen et al., 2022; Kim et al., 2011).

An underappreciated feature of the PI cycle is that it is not a closed loop. PI(4,5)P₂ can be siphoned off into PI(3,4,5)P₃ via Class I PI 3-kinases, and most prominently, the glycerolipid intermediates DAG and PA are important lipid biosynthetic precursors for triglycerides and all major phospholipid classes that exist outside the core PI cycle. These routes include phosphatidylcholine (PC) and phosphatidylethanolamine synthesis via the Kennedy pathway (and subsequent conversions to phosphatidylserine [PS]) and production of phosphatidylglycerol and cardiolipin, in addition to PI, via the CDP-DAG pathway (Vance, 2015).

In addition to the connection of these lipid intermediates, which are mostly linked to homeostatic or housekeeping functions, to the receptor-controlled PI cycle, recent findings also identified additional inositol lipid cycles that do not involve agonist action and PLC activation. Notably, PI4P gradients between membranes of adjacent organelles can be used for the countertransport of structural lipids by members of the oxysterol-binding protein family (Mesmin et al., 2013; Chung et al., 2015b; Moser von Filseck et al., 2015). These cycles that we refer to as “short-loop” inositol lipid cycles are evolutionarily more ancient mechanisms that use different PI 4-kinase isoforms dedicated to serve these cycles at specific ER-organelle contacts.

Because PI4P in the PM is also the precursor to PI(4,5)P₂, how the short-loop PI4P cycle and the classical, agonist-induced PI cycle are interconnected remains a fascinating open question. Though PI4P is present on several organelle membranes, the PI4P pool at the PM is synthesized primarily by one of four PI 4-kinase isoforms, PI 4-kinase type IIIα (PI4KIIIα, encoded by PI4KA) (Audhya and Emr, 2002; Balla et al., 2005; Balla et al., 2008). Indeed, PI4KIIIα activity is required for acute PI(4,5)P₂ signaling in cells due to its critical role in PI(4,5)P₂ resynthesis following its hydrolysis, e.g., by GPCR–Gq–PLC signaling (Bojjireddy et al., 2014a; Chung et al., 2015a). However, acute inhibition of PI4KIIIα causes the PM PI4P pool to decrease, notably without a decrease in resting PI(4,5)P₂ levels (Hammond et al., 2012; Lu et al., 2022; Kim et al., 2024). In contrast, the decrease in PM PI4P under these conditions has profound effects on PS levels, especially under prolonged inhibition of the PI4KIIIα enzyme (Chung et al., 2015b; Sohn et al., 2018). Therefore, PI4KIIIα plays crucial housekeeping roles that affect the lipid landscape and therefore regulates cellular physiology and organismal development.

Accordingly, PI4KIIIα is essential in all tested model organisms. In Saccharomyces cerevisiae, the PI4KIIIα ortholog Stt4 is necessary for cell wall integrity and actin organization (Audhya et al., 2000; Audhya and Emr, 2002). In Drosophila, PI4KIIIα is required for Hippo signaling and egg chamber polarity and morphology (Yan et al., 2011; Tan et al., 2014). In mice, knockout (KO) or chronic inhibition of PI4KIIIα leads to mucosal epithelial degeneration of the gastrointestinal tract, resulting in lethality (Vaillancourt et al., 2009; Bojjireddy et al., 2014a). Mutations in either PI4KA or genes encoding essential non-catalytic components of a PI4KIIIα-containing complex (Baird et al., 2008; Nakatsu et al., 2012; Bojjireddy et al., 2014b; Baskin et al., 2016; Lees et al., 2017) required for PI4P synthesis (EFR3A/B, TTC7A/B, and FAM126A/B) have been linked to schizophrenia (Jungerius et al., 2008; Vorstman et al., 2009), hypomyelinating leukodystrophy (Baskin et al., 2016; Verdura et al., 2021; Salter et al., 2021), inflammatory bowel disease (Avitzur et al., 2014), and immunodeficiency (Salter et al., 2021). Indeed, the importance of PI4KIIIα in cell physiology is further underscored by its essential role during hepatitis C viral infection and replication, where the enzyme is hijacked to produce PI4P on viral replication organelles, consequently antagonizing its vital housekeeping roles at the PM (Ahn et al., 2004; Trotard et al., 2009; Vaillancourt et al., 2009; Berger et al., 2009; Borawski et al., 2009).

Collectively, numerous KO studies in model organisms and human disease genetics point to chronic deficiency of PM PI4P as leading to catastrophic outcomes at the cellular and organismal levels. However, the downstream events that cause these end results and the compensatory changes that are initiated under conditions of perturbation or stress remain largely unknown. Pharmacological inhibition of PI4KIIIα would enable the identification of first-order mechanisms that precede the pleiotropic, long-term effects of PI4KA defects. For example, we recently exploited PI4KIIIα inhibition to reveal key mechanistic insights into how PI4P-dependent interorganelle transport of PS contributes to the complex pathology of Lenz-Majewski syndrome, which is caused by activating mutations in the PS-synthesizing enzyme phosphatidylserine synthase 1 (PSS1) (Sohn et al., 2016).

In this study, we undertook an investigation of the cellular response to disruption of PM PPIns metabolism by pharmacological inhibition of PI4KIIIα. We employed distinct PI4KIIIα inhibition time points to probe different stages of the cellular response. Short-term treatment (1 h) was used to examine immediate signaling and proteomic changes at the PM, intermediate treatment (4–6 h) to capture early transcriptional and translational responses, and longer treatments (18–24 h) to assess more extensive lipidomic remodeling and downstream phenotypic outcomes. We uncovered an integrated response to PM PI4P depletion that leads to an unexpected accumulation of PA in the PM in large part via activation of phospholipase D (PLD) enzymes, which produce PA via PC hydrolysis. Mechanistically, this response involves a metabolic rewiring wherein PI4P loss causes a loss of PS at the PM, which we find exhibits a reciprocal relationship with PA under conditions of limiting PI4P. These metabolic changes lead to transcriptional upregulation of the small GTPase RhoB, which under PI4P-limiting conditions enhances PLD-mediated PA synthesis and subsequent changes in F-actin organization, explaining perturbations to the actin cytoskeleton observed both by PI4KIIIα inhibition and in genetic models of PI4KA-associated disease. Overall, our findings define a long-range metabolic circuit connecting PI4KIIIα and PLD signaling, revealing unexpected metabolic inputs into PM PPIns metabolism.

Results

Inhibition of PI4P synthesis at the PM elevates PA levels

To assess the effect of prolonged PM PI4P depletion on cellular lipid compositions, we treated HEK293 cells overnight with the selective PI4KIIIα inhibitor GSK-A1 (Bojjireddy et al., 2014a). We used a bioluminescence resonance energy transfer (BRET) assay, which allows for monitoring of organelle-specific lipid changes, to assess phospholipid changes in the PM at the cell population scale (Tóth et al., 2016). This assay involves expression of a Renilla luciferase variant fused to a lipid-binding domain and an organelle-anchored mVenus construct. The presence of the target lipid on the organelle of interest recruits the luciferase-fused lipid sensor, and subsequent BRET leads to an increase in mVenus fluorescence that serves as a readout for lipid levels on the organelle of interest (Kim et al., 2024). BRET-based measurements of PM PI4P levels using the 2xP4M domain showed that acute inhibition of PI4KIIIα gradually decreases PM PI4P levels, as observed previously (Kim et al., 2024). Such low levels of PI4P were still observed after overnight treatment with the inhibitor, and a freshly added inhibitor exerted no further decrease (Fig. 1 B). These measurements showed that PM PI4P can be significantly depleted throughout prolonged GSK-A1 treatment to mimic the effects of defective PI4KIIIα function on the cellular lipidome. Total cellular lipidomics revealed not only decreased PS levels, consistent with earlier reports (Sohn et al., 2016), but also a prominent increase in total PA species upon prolonged GSK-A1 treatment (Fig. 1 C).

Figure 1. Depletion of PI4P at the PM leads to changes in PA.

(A) Schematic depicting the short and long PI cycles. (B) Live-cell BRET assay analysis of PM PI4P levels in HEK293 cells. Cells were pretreated with either DMSO control (gray symbols) or GSK-A1 (100 nM) overnight (blue symbols). The vertical dotted line represents the acute addition of GSK-A1 (100 nM) to both groups. (C) Lipidomics analysis of PA levels in HEK293 cells treated with DMSO control or GSK-A1 (100 nM, 24 h) (n = 3 biological replicates from one representative experiment). (D) Live-cell imaging of PA localization in HEK293 cells expressing the PA biosensor GFP-Nir2-LNS2 (816–1,181), treated with or without GSK-A1 (100 nM, 24 h) or with or without PLD inhibitor FIPI (1 μM). Scale bars: 10 μm. Quantification performed on 5 cells per condition. (E) Live-cell BRET analysis of PA levels at the PM in HEK293 cells treated with or without GSK-A1 (100 nM, 24 h) or with or without PLD inhibitor FIPI (1 μM, 24 h) using the Nir2-LNS2 (816–1,181) PA biosensor (n = 3 biological replicates). (F) Schematic of IMPACT labeling for generating fluorescent reporters of PLD activity. (G) Confocal microscopy imaging of the localization of PLD activity, assessed using RT-IMPACT labeling in HEK293 cells treated with GSK-A1 (100 nM, 2.5 h) or with or without FIPI (1 μM). (H) Normalized PLD activity by IMPACT on HEK293T cells treated with GSK-A1 (100 nM, 18 h) or DMSO with or without FIPI pretreatment (750 nM, 30 min) (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). (I) Normalized PLD activity by IMPACT on HEK293T PLD1/2 double KO cells transiently overexpressing mScarlet-i empty vector or mScarlet-i-PLD1 treated with GSK-A1 (100 nM, 18 h) with or without FIPI pretreatment (750 nM, 30 min). Gray indicates untransfected population, and red indicates transfected population (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). (J) Normalized PLD activity by IMPACT on HEK293T cells treated with GSK-A1 (100 nM, 1 h), VT01454 (100 nM, 1 h), or DMSO with or without FIPI pretreatment (750 nM, 30 min) (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). (K) Myo-[³H]-inositol incorporation into PI in HEK293 cells treated with GSK-A1 (100 nM, 24 h) with or without FIPI pretreatment. Different symbols on plot correspond to different biological replicates (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). For IMPACT flow cytometry data, fluorescence from FIPI treatment group was subtracted from no FIPI group for normalization unless otherwise indicated, with more details on normalization provided in Materials and methods, IMPACT labeling and flow cytometry section.

To gain more precise information about the subcellular localization of the PA increase, we used live-cell imaging with the Nir2-LNS2 domain (residues 816–1,181) to detect PA and found that it showed prominent increases in the PM following GSK-A1 treatment (Fig. 1 D). Using this Nir2-LNS2 (816–1,181) biosensor in the BRET configuration to measure PA at the PM (Kim et al., 2015) at the population scale, we found that PI4KIIIα inhibition induced an increase in PA levels at the PM (Fig. 1 E). PM PA levels usually show an increase during PLC activation, as the DAG product is converted to PA by DAG kinase (DGK) enzymes (Kim et al., 2022). However, it is important to note that prolonged GSK-A1 treatment did not cause PLC activation, and PI(4,5)P₂ levels showed a small increase rather than a decrease (Fig. S1 B). Therefore, it was unlikely that PI(4,5)P₂ hydrolytic product was the source of the elevated level of PA in the PM. Notably, PI3P levels measured by BRET analysis showed no significant change upon GSK-A1 treatment (Fig. S1 C), indicating that the lipid changes are specific to PI4P and does not induce a global perturbation of all PPIn pools.

Beyond DGKs, PA is produced by additional biosynthetic pathways in the cell, including the hydrolysis of PC by PLDs. Of the two PLD isoforms (PLD1/2) that produce PA via PC hydrolysis, PLD1 translocates to the PM to produce PA upon receptor activation (Jenkins and Frohman, 2005), and PLD2 is constitutively localized at the PM (Du et al., 2004). We found that PA increases caused by GSK-A1 treatment at the PM were greatly reduced in cells pretreated with a pan-PLD1/2 inhibitor, 5-fluoro-2-indolyl des-chlorohalopemide (FIPI) (Su et al., 2009), as judged by confocal microscopy (Fig. 1 D) or BRET analysis based on the Nir2-LNS2 (816–1,181) biosensor (Fig. 1 E).

To test the involvement of PLDs using an orthogonal method, we used a small-molecule chemical biology tagging approach termed Imaging PLD Activity with Clickable Alcohols via Transphosphatidylation (IMPACT), which we previously developed to quantify PLD activity in single cells (Bumpus and Baskin 2017). Briefly, IMPACT involves treatment of cells with an azide-containing primary alcohol, which is used as a substrate for PLD in a transphosphatidylation reaction that produces a phosphatidyl azidoalcohol lipid instead of PA. Next, cells are treated with a cyclooctyne-bearing fluorophore (e.g., BCN-BODIPY), which selectively tags the phosphatidyl azidoalcohols via a bioorthogonal reaction. The cells can then be analyzed by flow cytometry to quantify BODIPY-derived fluorescence as a readout of PLD activity (Fig. 1 F; see also Fig. S1 A for flow cytometry gating). Because IMPACT relies on the transphosphatidylation activity of PLDs with an exogenously supplied primary alcohol, it provides a direct readout of PLD-dependent PA biosynthetic activity. Importantly, the small-molecule alcohol substrate is membrane permeable and labels PLD activity at the membrane where the enzyme is active. IMPACT followed by flow cytometry revealed an increase in PLD activity of HEK293T cells treated with GSK-A1, and this increase was prevented by pretreatment with the PLD inhibitor FIPI (Fig. 1 H). We then used a real-time variant of IMPACT (RT-IMPACT) to visualize the subcellular localization of active PLDs (Liang et al., 2019). Here, a different alcohol probe containing a trans-cyclooctene (TCO) is used along with fluorogenic tetrazine detection reagents in a bioorthogonal reaction termed the tetrazine ligation, which exhibits rapid kinetics and no-rinse imaging such that localizations of IMPACT-derived fluorescent lipids can be visualized prior to subsequent interorganelle transport. Using this RT-IMPACT protocol, we found that the localization of PLD activity induced by GSK-A1 was at the PM (Fig. 1 G). These experiments revealed that PLDs are involved in the GSK-A1–induced PA increase.

To delineate the contributions of individual PLD isoforms, we employed the isoform-selective inhibitors: VU0359595 (250 nM) for PLD1 and VU0364739 (350 nM) for PLD2. In U2OS cells, inhibition of PLD1 produced a stronger reduction in the GSK-A1–induced IMPACT signal, whereas in HEK293T cells PLD2 inhibition had a greater effect (Fig. S1 D), indicating that the relative contributions of the two PLD isoforms are cell type dependent. Notably, however, U2OS cells exhibited a stronger overall PLD activation in response to GSK-A1 treatment (Fig. 3 A and Fig. S2 A), and immunoblotting showed that U2OS cells express higher levels of PLD1 protein than HEK293T cells (Fig. S1 E). Together, these results suggest that PLD1 likely makes a larger contribution to A1-induced PA generation, particularly in the more strongly responding U2OS cells.

Figure 3. Depletion of PI4P at the PM leads to transcriptional and translational changes.

(A) Normalized PLD activity by IMPACT on HEK293T cells treated with GSK-A1 (100 nM) for 0, 1, 4, or 18 h or with PMA (100 nM) for 30 min with or without FIPI pretreatment (750 nM, 30 min) (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). (B) Western blot analysis of HEK293T cells treated with GSK-A1 (100 nM) for 0, 1, 4, or 18 h, PMA (100 nM) for 30 min, or VT01454 (100 nM) for 1 h. Quantification performed by normalization of PLD1 or PKCα protein band intensity to total protein content (Ponceau S). Representative western blot shown on the left with quantification on the right (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). Note that α-tubulin blot is identical to that in Fig. 4 A because the same samples were probed for multiple markers shown in Figs. 3 B and 4 A. (C) Normalized PLD activity by IMPACT on HEK293T cells pretreated with cycloheximide (5 μg/ml, 1 h), followed by treatment with GSK-A1 (100 nM, 4 h) or PMA (100 nM, 30 min) with or without FIPI pretreatment (750 nM, 30 min) (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). (D) mRNA-seq results on HEK293 cells treated with DMSO control or GSK-A1 (100 nM, 24 h). Volcano plot highlighting differentially expressed genes in A1 vs. control. The plot summarizes the GSK-A1 inhibitor treatment effect where points represent genes (n = 14,038). The x-axis shows the estimated log₂ FC (A1—control) obtained with adaptive shrinkage, and the y-axis shows –log₁₀ of the Benjamini–Hochberg adjusted P value. Genes colored red passed the 10% FDR threshold and have positive log₂ FCs; genes colored blue passed the 10% FDR threshold and have negative log₂ FCs; gray points are nonsignificant. Positive log₂ FC values indicate higher expression in A1 treated cells, and negative values indicate lower expression. Point size is determined by log₁₀ expression across all samples (A1 and control) for the gene. The adjusted P value for RHOB rounds to zero, and its −log10 value is infinity, which cannot be plotted. Its location was manually and arbitrarily set to the y-axis limit for the purpose of visualization. The horizontal bar chart at right displays the 40 most strongly up- and downregulated genes by GSK-A1 inhibitor treatment that met the 10% FDR threshold. The x-axis shows the estimated log₂ FC (GSK-A1—control) derived with adaptive shrinkage; red bars extending right denote higher expression in A1-treated cells, while blue bars extending left denote lower expression. Gene symbols appear on the y-axis, ordered by descending effect size. Full dataset is provided in Table S1. (E) Normalized PLD activity by IMPACT on HEK293T cells treated with or without the PSS1 inhibitor DS55980254 (1 μM, 24 h), followed by treatment with or without cycloheximide (5 μg/ml, 1 h), followed by treatment with or without GSK-A1 (100 nM, 4 h) with or without FIPI pretreatment (750 nM, 30 min) (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). Different shapes on plot correspond to different biological replicates. For IMPACT flow cytometry data, fluorescence from FIPI treatment group was subtracted from no FIPI group for normalization unless otherwise indicated, with more details on normalization provided in Materials and methods, IMPACT labeling and flow cytometry section. Source data are available for this figure: SourceData F3.

Further, in PLD1/2 double KO HEK293T cells, IMPACT analysis did not detect increased PLD-mediated PA production following prolonged GSK-A1 treatment, but expression of mScarlet-i-PLD1 but not mScarlet-i alone restored the increased PA production, confirming the specificity of the IMPACT readout to PLDs (Fig. 1 I). Elevation of PLD1 levels using a CRISPRa approach further augmented the GSK-A1–induced PLD activity (Fig. S1, F and G). To further ensure that the PA increase was the result of PI4P loss rather than off-target effects of GSK-A1, we inhibited PM PI4P generation by interfering with PI transfer to the PM using the class I PITP inhibitor VT01454, which prevents PI transport from the ER to the PM, thus hindering PI4P synthesis (Kim et al., 2024). We found that VT01454 treatment phenocopied GSK-A1, eliciting a comparable elevation of PLD activity in HEK293T cells (Fig. 1 J), with similar results observed in U2OS cells (Fig. S2 B). Together, these results demonstrate that disruption of PI4P synthesis at the PM triggers a response in which PLD activation drives a PA increase at the PM.

We also determined whether the PA generated by PLD at the PM under PI4P depletion conditions could also be used to synthesize PI in the ER, as a compensatory response to increase PI synthesis. Indeed, prolonged A1 treatment caused an increased rate of PI synthesis, which was partially reversed by FIPI treatment, suggesting that PLD-derived PA can at least partially feed the PI cycle by promoting PI resynthesis, presumably through Nir2-mediated PA/PI exchange at ER–PM contact sites (Fig. 1 K).

PS counteracts PLD-mediated PA production under PI4P depletion

Because PLD-derived PA is not directly coupled to PI4P synthesis at the PM, we examined whether additional changes to lipid levels induced by PI4P depletion might be related to the increase of PA. Lipidomics analysis of HEK293 cells revealed a decrease in multiple PS species upon 24-h treatment with GSK-A1 (Fig. 2 A), suggesting that PI4P loss decreases the pool of anionic lipids at the PM, consistent with our earlier work (Sohn et al., 2016). To determine whether PS abundance regulates PLD activation under PI4P depletion conditions, we performed IMPACT on cells overexpressing either WT PSS1 or a gain-of-function mutant PSS1^P269S associated with Lenz-Majewski syndrome that is insensitive to product inhibition and therefore causes substantial accumulation of PS in the cell (Sohn et al., 2016). We found that HEK293T cells overexpressing wild-type PSS1 exhibited a small decrease in the GSK-A1–induced PLD response, whereas those overexpressing PSS1^P269S showed almost no PLD response to GSK-A1 treatment (Fig. 2 B). Conversely, pharmacological inhibition of PS synthesis with a PSS1 inhibitor (Yoshihama et al., 2022) significantly amplified the PLD response to GSK-A1 (Fig. 2 D), with similar trends observed in U2OS cells (Fig. S2 C). We also performed confocal imaging with a PM PA sensor (GFP-Nir2(816–1,181)) under PSS1i + A1 combination treatment conditions and observed increased PA signal on the PM (Fig. 2 E). It is important to note that PA is metabolized relatively quickly, and therefore the confocal imaging with PA sensor would be expected to show a smaller change than IMPACT, which is a readout of PLD activity. Notably, pharmacological inhibition of PSS1 alone, without concomitant PI4P depletion, was not sufficient to activate PLD (Fig. 2 D), indicating that PS reduction alone does not drive PLD activation but rather modulates the magnitude of the response when PI4P is depleted.

Figure 2. PS levels modulate the extent of PLD activation induced by PI4KIIIα inhibition.

(A) Lipidomics analysis of PS levels in HEK293 cells treated with DMSO control or GSK-A1 (100 nM, 24 h) (n = 3 biological replicates from one representative experiment). (B) Normalized PLD activity by IMPACT on HEK293T cells transiently overexpressing PSS1^WT-mCherry or PSS1^P269S-mCherry treated with GSK-A1 (100 nM, 18 h) with or without FIPI pretreatment (750 nM, 30 min). Gray indicates untransfected population, and red indicates transfected population (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). (C) Normalized PLD activity by IMPACT on HEK293T cells knocked down by siRNA against NC1 (nontargeting negative control #1) or ORP5 and ORP8, followed by transient overexpression of PSS1^WT-mCherry or PSS1^P269S-mCherry treated with GSK-A1 (100 nM, 18 h) with or without FIPI pretreatment (750 nM, 30 min) (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). (D) Normalized PLD activity by IMPACT on HEK293T cells pretreated with or without the PSS1 inhibitor DS55980254 (1 μM, 24 h), followed by treatment with or without GSK-A1 (100 nM, 4 h) with or without FIPI pretreatment (750 nM, 30 min) (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). Different symbols on plot correspond to different biological replicates (n = 3). Statistical significance was determined using one-way ANOVA with Tukey post hoc test, with P values indicated on the plot. (E) Live-cell imaging of PA localization in HEK293 cells. Cells expressing the PA biosensor GFP-Nir2-LNS2 (816–1,181) were analyzed following a 20-h treatment with GSK-A1 (100 nM), FIPI (1 μM), or DS55980254 (1 μM). Control cells were treated with DMSO for the same duration. Scale bars: 10 μm. For IMPACT flow cytometry data, fluorescence from FIPI treatment group was subtracted from no FIPI group for normalization unless otherwise indicated, with more details on normalization provided in Materials and methods, IMPACT labeling and flow cytometry section.

Because PS is delivered to the PM in exchange for PI4P via ORP5/8 transporters (Chung et al., 2015b; Sohn et al., 2018), we next investigated whether such PI4P/PS countertransport played a role in this phenomenon. Knockdown of ORP5/8 by siRNA caused only a minor change in PLD activity under control conditions and exerted no effect at all under conditions of excess PS synthesis in the ER, i.e., overexpression of PSS1^P269S (Fig. 2 C). These results suggest that ORP5/8 transport does not play a pivotal role in the PI4P–PS–PA regulation and that other PS-related regulators are likely also involved. Nevertheless, these data reveal a reciprocal relationship between two negatively charged phospholipids, PA and PS, under conditions when levels of PI4P at the PM are reduced by PI4KIIIα inhibition. That is, the decreased PS synthesis caused by PM PI4P depletion, as a likely result of decreased PI4P-driven PS transport, leads to elevation of PA at the PM, and maintenance of PS levels even under limited PI4P supply at the PM can prevent PLD-mediated PA synthesis.

GSK-A1–induced PLD activation is mechanistically distinct from PKC-mediated PLD activation

Given the strong response of PLD activation to PI4KIIIα inhibition, we sought to establish mechanisms underlying this phenomenon. We first compared PI4P depletion-driven PLD activation induced by GSK-A1 with classical PKC-dependent PLD stimulation by measuring PLD activity in cells treated with GSK-A1 over an 18-h time course or via acute (30 min) treatment with the PKC activator PMA. PMA, a chemical mimic of DAG, activates classical PKCs, including PKCα, which in turn acutely activate PLDs (Lee et al., 1997). PMA triggered a rapid, maximal PLD response within minutes, whereas GSK-A1 elicited a slowly developing increase that plateaued after 4 h (Fig. 3 A and Fig. S2 A). We therefore hypothesized that PLD activation by PI4KIIIα inhibition might have a component that involves gene transcription and protein expression, which would explain the slower timescale.

We found no increase in PLD1 protein levels upon GSK-A1 treatment; in fact, prolonged treatment caused somewhat reduced PLD1 protein levels (Fig. 3B and Fig. S2 D). We then treated cells with cycloheximide prior to GSK-A1 treatment to block protein synthesis and found that the GSK-A1–mediated effect on PLD activity was strongly reliant on de novo protein synthesis, in contrast to PMA treatment, whose effects on PLD activity were minimally changed by cycloheximide (Fig. 3 C). These results indicate that PI4P depletion engages a translation-dependent (and possibly transcription-dependent) pathway to activate PLD. Therefore, we then performed mRNA sequencing (RNA-seq) analysis of cells treated with GSK-A1 for 24 h (Fig. 3 D). These studies revealed upregulated mRNA levels of several transcription factors (TFs), including ZCCHC12, EGR1, ETV4, ETV5, KLF6, FOS, and JUN, as well as strong upregulation of the small GTPase RhoB following GSK-A1 treatment. To confirm that the upregulated mRNA levels were related to an increase in de novo protein synthesis, we additionally performed ribosome profiling (Ribo-seq) on cells treated with GSK-A1 for 0, 1, and 6 h and indeed found that RhoB exhibited upregulated translation upon GSK-A1 treatment (Table S4).

To dissect the interplay between PS-mediated suppression and transcriptional control of PLD activation, we combined cycloheximide treatment with pharmacological PSS1 and PI4KIIIα inhibition. Cycloheximide attenuated but did not abolish the enhanced PLD activation seen upon PS and PI4P depletion, and the overall PLD activity remained above levels induced by GSK-A1 alone (Fig. 3 E). Thus, these data collectively support a model wherein the response to PM PI4P depletion involves a transcriptional response resulting in elevated PLD-mediated PA synthesis.

RhoB enhances PLD activation in response to PI4P depletion

As described above, transcriptomics and Ribo-seq analyses revealed the striking upregulation of RhoB upon PI4KIIIα inhibition. RhoB shares 85% amino acid sequence identity with the known PLD activator RhoA (Wheeler and Ridley, 2004), but RhoB has not been previously implicated in PLD1/2 regulation. Consistent with the RNA-seq and Ribo-seq data, immunoblotting showed a substantial increase in RhoB protein levels after GSK-A1 treatment in both HEK293T and U2OS cells, with a more pronounced increase in U2OS cells, which might be related to their higher PLD activity response (Fig. S2 D and Fig. 4 A). To assess whether RhoB is required for PI4P depletion–driven PLD activation, we generated mixed cell populations exhibiting reduced levels of RhoB via transient delivery of sgRNAs for CRISPR KO, with downstream analysis performed on short timescales to avoid compensatory activities resulting from prolonged depletion of RhoB. These pools of RhoB “KO” cells showed a reduced RhoB response upon GSK-A1 treatment, and they also exhibited a substantial reduction in GSK-A1–induced PLD activity compared with WT controls (Fig. 4, B and C). Conversely, CRISPRa-mediated increase of RhoB expression enhanced the levels of PLD activity in response to GSK-A1 (Fig. 4 D). We noted that previous studies have reported that tagging RhoB with fluorescent proteins can affect its localization and function (Wherlock et al., 2004; Castillo et al., 2023). Therefore, we transiently overexpressed untagged RhoB via plasmid transfection with a bicistronic mScarlet-i-P2A-RhoB construct, resulting in co-expression of an untagged RhoB and free mScarlet-i, allowing for identification of transfected cells by flow cytometry. By performing IMPACT labeling on such cells overexpressing RhoB, we found an enhancement in GSK-A1–induced PLD activity compared with GSK-A1 alone (Fig. 4 E). It is noteworthy, though that RhoB alone, without GSK-A1 treatment did not increase PLD activity (Fig. 4 E).

Figure 4. RhoB enhances PLD activity under conditions of PM PI4P depletion.

(A) Western blot analysis of HEK293T cells treated with GSK-A1 (100 nM, 0–18 h), PMA (100 nM, 30 min), or VT01454 (100 nM, 1 h). Quantification performed by normalization of RhoB protein band intensity to total protein content (Ponceau S). Representative western blot shown on the left with quantification on the right (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). Note that α-tubulin blot is identical to that in Fig. 3 B because the same samples were probed for multiple markers shown in Figs. 3 B and 4 A. (B) Normalized PLD activity by IMPACT on mixed U2OS cell populations expressing Cas9 and the indicated sgRNA (nontargeting [NT1] or RhoB [KO1, KO2]) treated with GSK-A1 (100 nM, 18 h) with or without FIPI pretreatment (750 nM, 30 min) (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). (C) Western blot analysis of U2OS cells expressing Cas9 and the indicated sgRNA (nontargeting [NT1] or RhoB [KO1, KO2]) treated with or without GSK-A1 (100 nM, 18 h). Note the reduced RhoB response in the pool of KO cells, where because KO and subsequent analysis is performed transiently to avoid compensatory activities, and RhoB is unlikely to be eliminated from all cells. (D) Normalized PLD activity by IMPACT on K562 CRISPRa cells expressing the indicated sgRNA (nontargeting or RhoB) treated with GSK-A1 (100 nM, 18 h) with or without FIPI pretreatment (750 nM, 30 min) (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). See Fig. S4 A for validation of RhoB CRISPRa. (E) Normalized PLD activity by IMPACT on HEK293T cells transiently overexpressing mScarlet-i empty vector or mScarlet-i-P2A-RhoB treated with or without GSK-A1 (100 nM, 18 h) with or without FIPI pretreatment (750 nM, 30 min). Gray indicates untransfected population, and red indicates transfected population (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). (F–I) Normalized PLD activity by IMPACT on HEK293T cells transiently overexpressing mScarlet-i empty vector (mScarlet EV), mScarlet-i-P2A-RhoB, mScarlet-i-P2A-RhoB^Q63L, mScarlet-i-P2A-RhoB^T19N, mScarlet-i-P2A-RhoB^CAIM, mScarlet-i-P2A-RhoB^CLLL, mScarlet-i-P2A-RhoB^C193S, mScarlet-i-P2A-RhoB^C192S, or mScarlet-i-P2A-RhoB^C189/192S treated with GSK-A1 (100 nM, 18 h) with or without FIPI pretreatment (750 nM, 30 min). Gray indicates untransfected population, and red indicates transfected population (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). Different shapes on plot correspond to different biological replicates. For IMPACT flow cytometry data, fluorescence from FIPI treatment group was subtracted from no FIPI group for normalization unless otherwise indicated, with more details on normalization provided in Materials and methods, IMPACT labeling and flow cytometry section. Source data are available for this figure: SourceData F4.

Because Rho GTPases function via GTP/GDP cycling, we assessed the importance of nucleotide binding for the effects of RhoB on PLD. IMPACT labeling of cells expressing either the GTP-locked RhoB^Q63L or the GDP-locked RhoB^T19N mutant forms (Bery et al., 2019) resulted in no further increase to PLD activity by GSK-A1 compared with GSK-A1 alone, suggesting that dynamic GTPase cycling is important for the enhancing effect of RhoB on PLD activity (Fig. 4 F). RhoB undergoes prenylation at C193, which can influence its membrane localization and other downstream effects (Adamson et al., 1992). Mutation of C193 nearly completely abolished its ability to enhance PLD activity (Fig. 4 G), whereas mutations to the C-terminal CAAX motif to control the length of the prenylation tag (e.g., farnesylation using CAIM vs. geranylgeranylation using CLLL [Baron et al., 2000]) had equivalent stimulatory effects (Fig. 4 H). These results indicate that prenylation of RhoB was required for its PLD enhancement but that the form of prenylation did not impact this effect. Unlike RhoA, RhoB is also S-acylated (palmitoylated) adjacent to the prenylation site, and because palmitoylation can affect peripheral membrane protein localization and behavior within the bilayer (Linder and Deschenes, 2007), we also investigated whether RhoB palmitoylation was important for its PLD enhancement activity. These studies revealed that ablation of the two C-terminal palmitoylation sites (C189 or C192) caused only a modest reduction in RhoB enhancement of PLD activity (Fig. 4 I), suggesting that palmitoylation fine-tunes but is not absolutely required for the enhancing effect of RhoB on PLD activity upon PI4KIIIα inhibition.

We next explored how PI4P depletion triggers RhoB upregulation. Among the TFs induced by GSK-A1 were several MAPK pathway regulators (Fig. 3 D and Table S4) and c-Jun (Fig. 3 D). Based on previous studies reporting an association of RhoB with MAPK (Ahn et al., 2011; Delmas et al., 2015), we used the MEK inhibitor AZD6244 to assess the involvement of MAPK pathways in the RhoB enhancement of PLD activity upon PI4P depletion. We found that AZD6244 treatment led to a small but significant increase in GSK-A1–induced PLD activity, accompanied by increased levels of RhoB, and phosphorylated forms of c-Jun and its kinase JNK (Fig. 5, A and B). Moreover, PLD activity did not correlate with levels of phosphorylation of ERK1/2, suggesting that the p42/44 MAPK pathway is not involved in activation of PLD activity (Fig. 5 B). Interestingly, a prior study found that RhoB transcription is activated by c-Jun (Delmas et al., 2015). Therefore, we treated cells with the JNK inhibitor, JNK-IN-8, and found that it significantly blunted the GSK-A1–mediated activation of PLDs as well as levels of RhoB and phospho-c-Jun (Fig. 5, C and D), suggesting that c-Jun and JNK are involved in the RhoB response.

Figure 5. JNK–c-Jun signaling promotes RhoB-dependent PLD activation after PI4P depletion.

(A) Normalized PLD activity by IMPACT on U2OS cells pretreated with or without the MEK inhibitor AZD6244 (2 μM, 24 h), followed by treatment with or without GSK-A1 (100 nM, 4 h) with or without FIPI pretreatment (750 nM, 30 min) (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). (B) Western blot analysis of U2OS cells pretreated with or without the MEK inhibitor AZD6244 (2 μM, 24 h), followed by treatment with or without GSK-A1 (100 nM, 4 h). Quantification performed by normalization of target protein band intensity to total protein content (Ponceau S). Representative western blot shown on the left with quantification on the right (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). (C) Normalized PLD activity by IMPACT on U2OS cells pretreated with or without the JNK inhibitor JNK-IN-8 (1 μM, 24 h), followed by treatment with or without GSK-A1 (100 nM, 4 h) with or without FIPI pretreatment (750 nM, 30 min) (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). (D) Western blot analysis of U2OS cells pretreated with or without the JNK inhibitor JNK-IN-8 (1 μM, 24 h), followed by treatment with or without GSK-A1 (100 nM, 4 h). Representative western blot shown on the left with quantification on the right (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). (E) Normalized PLD activity by IMPACT on U2OS cells pretreated with or without the p38 MAPK inhibitor SB203580 (5 μM, 19 h), followed by treatment with or without GSK-A1 (100 nM, 18 h) with or without FIPI pretreatment (750 nM, 30 min) (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). (F) Western blot analysis of U2OS cells pretreated with or without the p38 MAPK inhibitor SB203580 (5 μM, 19 h), followed by treatment with or without GSK-A1 (100 nM, 18 h). Quantification performed by normalization of target protein band intensity to total protein content (Ponceau S). Representative western blot shown on the left with quantification on the right (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). Different shapes on plot correspond to different biological replicates. For IMPACT flow cytometry data, fluorescence from FIPI treatment group was subtracted from no FIPI group for normalization unless otherwise indicated, with more details on normalization provided in Materials and methods, IMPACT labeling and flow cytometry section. Source data are available for this figure: SourceData F5.

Additionally, we looked into the mammalian p38 kinases, a MAPK family that responds to general stress and the activation of which has been reported to increase RhoB mRNA stability (Li et al., 2011). To evaluate whether RhoB upregulation is simply part of a general stress response rather than a response to prolonged PM PI4P depletion, we pretreated HEK293T and U2OS cells with the p38 MAPK inhibitor SB203580, followed by GSK-A1 treatment. Western blot analysis showed that inhibition of the p38 MAPKs did not alter the increased RhoB protein levels following GSK-A1 treatment, indicating that RhoB induction is not mediated through the canonical p38 stress-activated pathway (Fig. 5 F and Fig. S2 F). Interestingly, in parallel studies, p38 MAPK inhibition slightly increased GSK-A1–induced PLD activity (Fig. 5 E and Fig. S2 E), and we also found increased levels of phospho-JNK in cells treated with the p38 MAPK inhibitor (Fig. 5 F). This could explain the increased PLD response observed after inhibition of p38 MAPK. Additionally, and quite notably, we did not observe upregulated translation of RhoA, which also responds to cellular stress and is a well-characterized PLD1 activator, upon GSK-A1 treatment (Table S4). Overall, these data support a model wherein depletion of PI4P at the PM activates a c-Jun–dependent transcriptional program that contributes to RhoB upregulation, which amplifies PLD activity by a mechanism that requires its GTPase activity and membrane targeting.

GSK-A1–stimulated PLD activity increases actin fiber formation

To gain additional information on how cells respond to PI4P depletion from the PM, including possible downstream mechanistic changes, we leveraged proximity labeling with a PM-anchored TurboID to assess proteomic changes in the local membrane environment upon PI4KIIIα inhibition (Fig. 6 A). To identify proteins recruited to the PM upon PI4P depletion, we performed such proximity labeling in HeLa and HEK293T cells treated with GSK-A1 (Fig. S3) or the PITP inhibitor VT01454 (Fig. 6 B), respectively. Collectively, these studies revealed enrichment of multiple actin cytoskeleton-associated factors, including the Rho GTPase–binding protein Rhophilin-2 (RHPN2), which was previously implicated in RhoA and RhoB signaling and actin organization (Peck et al., 2002). Supporting the specificity of this approach, we found that the protein that was most depleted from the PM upon PI4P inhibition was ORP8 (OSPBL8), consistent with its requirement of binding to PI4P in the PM for lipid transport functions (Chung et al., 2015b; Sohn et al., 2018).

Figure 6. Acute PI4P depletion remodels actin cytoskeleton through PLD activity.

(A) Schematic of PM-tagged TurboID proximity labeling proteomics. (B) TurboID proteomics results on HEK293T Lyn₁₀-TurboID-V5 cells treated with or without VT01454 (100 nM, 1 h). Green indicates hits enriched, and red indicated hits depleted from the PM after VT01454 treatment. FC in abundance is indicated on the x axis (in log₂), and significance was determined using adjusted P values, shown on the y axis (in 1og₁₀). Full dataset is provided in Table S3. (C) Normalized PLD activity by IMPACT on K562 CRISPRi cells with no sgRNA or RHPN2 sgRNAs treated with GSK-A1 (100 nM, 4 h) with or without FIPI pretreatment (750 nM, 30 min) (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). See Fig. S4 B for validation of RHPN2 CRISPRi cell lines using qPCR. (D) Normalized PLD activity by IMPACT on HEK293T cells transiently overexpressing mScarlet-i empty vector or mScarlet-i-RHPN2 treated with GSK-A1 (100 nM, 4 h) with or without FIPI pretreatment (750 nM, 30 min). Gray indicates untransfected population, and red indicates transfected population (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). (E) Phalloidin staining on U2OS cells pretreated with or without the PLD inhibitor FIPI (750 nM, 30 min), followed by treatment with or without GSK-A1 (100 nM, 4 h). Scale bar: 10 μm. Quantification performed by selecting 10 areas per condition over three frames per biological replicate (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). (F) Phalloidin staining on U2OS cells pretreated with or without LatB (2.5 μM, 1 h) or cytoD (2.5 μM, 1 h), followed by treatment with or without GSK-A1 (100 nM, 4 h). Scale bar: 10 μm. Quantification performed by selecting 10 areas per condition over three frames per biological replicate (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). (G) Normalized PLD activity by IMPACT on HEK293T cells pretreated with or without LatB (2.5 μM, 1 h) or cytoD (2.5 μM, 1 h), followed by treatment with or without GSK-A1 (100 nM, 4 h) with or without FIPI pretreatment (750 nM, 30 min) (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). (H) Western blot analysis on HEK293T cells pretreated with or without LatB (2.5 μM, 1 h) or cytoD (2.5 μM, 1 h), followed by treatment with or without GSK-A1 (100 nM, 4 h). Quantification performed by normalizing RhoB protein band intensity over Ponceau S intensity. Representative western blot shown on the left with quantification on the right (n = 3 biological replicates, one-way ANOVA with Tukey post hoc test). (I) Working model. Different shapes on plot correspond to different biological replicates. For IMPACT flow cytometry data, fluorescence from FIPI treatment group was subtracted from no FIPI group for normalization unless otherwise indicated, with more details on normalization provided in Materials and methods, IMPACT labeling and flow cytometry section. Source data are available for this figure: SourceData F6.

We next tested the role of RHPN2 in PLD regulation by using CRISPR interference. CRISPRi-mediated RHPN2 knockdown modestly but significantly increased GSK-A1–induced PLD activity, whereas RHPN2 overexpression had the opposite effect on PLD activity (Fig. 6, C and D), suggesting that RHPN2 alone can negatively regulate the PLD response to PI4KIIIα inhibition. Because RHPN2 can disassemble actin fibers when overexpressed (Peck et al., 2002), we examined actin cytoskeleton organization after PI4P depletion. Using phalloidin staining, we found that GSK-A1 treatment induced a substantial increase in actin fiber levels, and importantly, this increase could be blocked by PLD inhibition in both U2OS and HEK293T cells (Fig. 6 E and Fig. S2 G). Together with the enrichment of several actin-binding proteins at the PM upon PI4P depletion in our proximity proteomics datasets (Fig. 6 B and Fig. S3), these data suggest a causal relationship between PI4P depletion, PLD activation, and increased actin fiber formation.

Indeed, even under conditions that induce actin depolymerization, i.e., treatment with the actin-depolymerizing agents latrunculin B (LatB) or cytochalasin D (cytoD), PI4KIIIα inhibition partially rescued defects in actin fiber levels (Fig. 6 F). A further connection between the actin cytoskeleton and PLD was indicated by the finding that treatment with LatB or CytoD under conditions of PI4KIIIα-inhibition, a larger increase in PLD activity was observed. Notably, similar to the effects of PI4KIIIα inhibition, the actin-depolymerizing agents also induced increases in RhoB protein levels yet did not activate PLD on their own (Fig. 6 H). This result was consistent with our earlier findings that RhoB activation alone is not sufficient to increase PLD activity (Fig. 4 E) but greatly enhances the effect of PI4KIIIα antagonism (Kusner et al., 2002) (Fig. 6 G). Overall, these data support a model (Fig. 6 I) wherein cells respond to lipidomic changes induced by PI4P depletion by activating a cascade culminating in RhoB expression, PLD-mediated PA synthesis, and F-actin assembly as a consequence to disruptions to the short, PI4P-centric PI cycle (Fig. 1 A).

PI4P depletion impacts endocytosis, ER homeostasis, and cell viability

All the effects described so far suggested that prolonged inhibition of PI4KIIIα triggers an integrated cell response that attempts to maintain the anionic character of the inner leaflet of the PM and increase actin polymerization to increase PM stability. To assess the broader biological significance of these changes, we examined their impact on endocytosis, ER homeostasis, and cell viability. Using a fluorescent transferrin uptake assay to measure clathrin-dependent endocytosis, we found that GSK-A1 treatment slightly increased transferrin uptake in both HEK293T and U2OS cells. This effect was independent of PLD activity, as FIPI co-treatment did not alter the endocytic response (Fig. S4 C). These results suggest that PI4P depletion modestly enhances clathrin-dependent endocytosis through a mechanism that does not involve the PLD-mediated PA increase, possibly reflecting the slight increases in PI(4,5)P₂ at the PM. This result was also consistent with the finding that the class II PI-3 kinase, PIK3C2A, was the top enriched hit in our PM VT01454 TurboID proteomics (Fig. 6 B), as this enzyme has established roles in clathrin-mediated endocytosis and endocytic pit maturation (Gaidarov et al., 2001).

To determine whether PI4P depletion and the associated lipid changes induce ER stress, we probed for established ER stress markers by western blot analysis across a treatment time course (0, 1, 4, and 18 h GSK-A1) with and without FIPI in both HEK293T and U2OS cells. We only observed a slight increase in CHOP upon prolonged GSK-A1 treatment in U2OS cells (Fig. S4 D), suggesting that ER stress is not a major factor during these treatment regimes. To assess the impact on overall cell viability, we examined apoptosis markers in U2OS cells. GSK-A1 treatment induced a time-dependent increase in apoptosis, measured by cleaved PARP and cleaved caspase-3 levels, with significant increases observed at 18 h (Fig. S4 E). Importantly, co-treatment with FIPI further increased apoptotic markers at 18 h, indicating that PLD-mediated PA production serves a protective role under these conditions (Fig. S4 E). In HEK293T cells, levels of cleaved PARP and caspase-3 were very low under all conditions, suggesting cell type–dependent sensitivity to PI4P depletion–induced apoptosis. Collectively, these findings established that the lipid changes induced by PI4KIIIα inhibition have functional consequences that may manifest in different forms in different tissues and cell lines.

Discussion

In this study, we describe a multicomponent integrated response that is activated by prolonged depletion of PI4P at the PM leading to stimulation of PLD enzymes for production of PA in this membrane. An important effect of the loss of PM PI4P is the concomitant decrease in total cellular PS levels as shown by several earlier studies and documented here as well (Nakatsu et al., 2012; Sohn et al., 2016). Under such conditions of low PM PI4P, levels of the anionic lipid PA are sensitive and reciprocally related to those of PS, the most abundant anionic lipid in the PM at steady state. These lipidomic changes are the result of a series of molecular events that involve a complex set of transcriptional changes, including upregulation of the small GTPase RhoB, which enhances PLD activity and subsequently also promotes cortical actin fiber formation. Collectively, these findings reveal how cells integrate lipid metabolism and gene regulatory pathways to adjust membrane lipid levels in attempts to achieve homeostasis.

The negative charge of the PM inner leaflet, conferred by lipids such as PI4P, PI(4,5)P₂, PS, and PA, is critical for recruitment and function of a number of peripheral membrane proteins (Yeung et al., 2008; Zhou et al., 2024; Hammond et al., 2012; Simon et al., 2016). PI4P depletion is expected to compromise this electrostatic landscape, primarily through also decreasing PS levels in the inner leaflet of the PM, potentially disrupting downstream signaling (Hammond et al., 2012). In addition to relying on PI(4,5)P₂, cells appear to enlist PA synthesis as an alternative anionic lipid produced at the PM. PS and PA coexist as major anionic lipids at the PM, yet their relative abundances and distributions are dynamically interconverted via lipid-exchange proteins (Leventis and Grinstein, 2010). By manipulating PS synthesis and interorganelle transport, we reveal that PS and PA, two negatively charged lipids, may function as part of a rheostat to maintain PM inner leaflet charge and signaling competence under conditions of PI4P depletion. In this model, PS occupies a gatekeeping function where its abundance is inversely correlated with PLD-mediated production of PA. Even beyond maintaining the anionic character of the PM, cells also use the PLD-generated PA to increase PI synthesis in the ER, in an attempt to correct the impaired PI4P gradient at the ER-PM contact sites during prolonged PI4KIIIα inhibition.

The delayed kinetics of PLD-mediated PA production following PI4P inhibition suggested that de novo protein synthesis was at least partially involved in this regulatory connection. Whereas the transcriptional and translational responses to PI4KIIIα inhibition were indeed complex and multifaceted, important themes emerged that were bolstered by targeted mechanistic studies. One was the involvement of de novo gene expression via activation of the JNK/c-Jun pathway. A second was production of the small GTPase RhoB via this pathway and changes to a RhoB interactor RHPN2. A third was changes to organization of the actin cytoskeleton organization. Notably, overexpression of RhoB has been associated with increased actin fiber formation (Fernandez-Borja et al., 2005), whereas overexpression of RHPN2 resulted in loss of actin fibers and cell contraction (Peck et al., 2002). Our findings that increased actin fiber formation caused by PI4KIIIα inhibition was preventable by co-treatment with a PLD inhibitor indicate that PA formation by PLD directs increased actin fiber formation. These results are consistent with earlier work establishing that PLD-derived PA can stimulate F-actin fiber formation (Ha and Exton, 1993). Conversely, free G-actin monomers have been shown to negatively regulate PLD activity (Kusner et al., 2002). Our findings that perturbation with LatB or cytoD, both of which antagonize F-actin formation (Spector et al., 1983; Shoji et al., 2012), while having no effects on their own, further increased PLD activity upon PI4KIIIα inhibition, also suggests a feedback loop wherein lipid perturbations can alter the actin cytoskeleton, which, in turn, can also act by changing the lipidome.

Actin cytoskeleton remodeling was one of the earliest phenotypes associated with perturbations to PI4P at the PM. In yeast, the PI4P pool produced by the PI4KIIIα ortholog Stt4 was shown to be crucial for vacuole morphology and actin cytoskeleton organization (Foti et al., 2001). Further, the major PI4P-degrading enzyme is named suppressor of actin 1 because its mutation rescues a defective actin allele; its inactivation leads to an 8–10-fold increase in PI4P levels and can rescue defects in actin organization and growth caused by Stt4 mutants (Guo et al., 1999; Nemoto et al., 2000; Foti et al., 2001; Cleves et al., 1989; Novick et al., 1989; Tahirovic et al., 2005). Our studies also provide some clues to better understand the developmental and functional defects described in human subjects with documented mutations in the PI4KA gene or in the genes encoding the proteins that are critical for anchoring the enzyme to the PM. Conditional mouse models of PI4KA deficiency in Schwann cells show major myelination defects in peripheral nerves, with massive changes in the lipidome and actin remodeling (Alvarez-Prats et al., 2018). Defects in PI4KA function in humans lead to a congenital hypomyelinating leukodystrophy, characterized by impaired myelin sheath formation in the central nervous system (Verdura et al., 2021). Myelination by highly specialized oligodendrocyte cells depends on extensive actin remodeling to extend and wrap PM processes around axons (Nawaz et al., 2015). Our mechanistic studies here raise the possibility that for diseases associated with PI4KA dysfunction, insufficient PI4P production activates a RhoB–PLD–PA pathway, leading to changes in PA levels and a reorganization of the actin cytoskeleton. Some of these changes appear to preserve the anionic character of the PM against decreased PS levels, and whereas increased actin polymerization may help some cells to support their PM, it may also impair processes where continuous actin remodeling is critical for function, such as in myelinating Schwann cells.

In conclusion, in these studies, we reveal how cells mount an integrated response to perturbations to PM PI4P pools. Lipidomic changes to PI4P and PS depletion rely on a reciprocal relationship between PS and PA in apparent efforts to bolster PM inner leaflet negative charge and boost the synthesis of PI, the main PI cycle intermediate, via stimulation of PA-synthesizing PLD enzymes. A key component of this response is de novo transcription of RhoB, whose activation of PLD-mediated PA synthesis not only causes lipidomic changes but also induces changes to F-actin organization. Our findings highlight the intricate crosstalk among lipid metabolic enzymes, small GTPases, and the actin cytoskeleton, providing a framework for understanding how PM lipids orchestrate both signaling and structural programs critical for cellular homeostasis. Beyond the effects on actin dynamics, we demonstrate that the lipid rheostat has broader functional consequences. PI4P depletion modestly enhanced clathrin-mediated endocytosis through a PLD-independent mechanism, potentially reflecting compensatory changes in other anionic lipids at the PM. Furthermore, prolonged PI4P depletion induced apoptosis and PLD inhibition exacerbated this situation, revealing a protective role for PLD-mediated PA production in buffering the stress of PI4P depletion. These findings position the PI4P–PS–PA rheostat as a central node in maintaining PM homeostasis with consequences extending to ER function, PI cycle maintenance, and cell survival.

Materials and methods

Antibodies

Antibodies are shown in Table 1.

Table 1.

Antibodies

Antibody	Source	Cat no	RRID
Mouse monoclonal anti-PLD1 (F-12)	Santa Cruz Biotechnology	sc-28314	AB_677324
Mouse monoclonal anti-PKCα (H-7)	Santa Cruz Biotechnology	sc-8393	AB_628142
Mouse monoclonal anti-α-tubulin (B-5-1-2)	Sigma-Aldrich	T5168	AB_477579
Mouse monoclonal anti-actin (AC-40)	Sigma-Aldrich	A4700	AB_476730
Mouse monoclonal anti-p-JNK (G-7)	Santa Cruz Biotechnology	sc-6254	AB_628232
Mouse monoclonal anti-JNK (D-2)	Santa Cruz Biotechnology	sc-7345	AB_675864
Mouse monoclonal anti-p-c-Jun (KM-1)	Santa Cruz Biotechnology	sc-822	AB_627262
Mouse monoclonal anti-c-Jun (G-4)	Santa Cruz Biotechnology	sc-74543	AB_1121646
Rabbit polyclonal anti-RhoB	Proteintech	14326-1-AP	AB_2179092
Mouse monoclonal anti-p-p44/42 MAPK (p-ERK1/2) (E10)	Cell Signaling Technology	9106	AB_331768
Rabbit polyclonal anti-p44/42 MAPK (ERK1/2)	Cell Signaling Technology	9102	AB_330744
Rabbit polyclonal anti-p38 MAPK	Cell Signaling Technology	9212	AB_330713
Rabbit polyclonal anti-p-p38 MAPK	Cell Signaling Technology	9211	AB_331641
Mouse monoclonal CHOP (L63F7)	Cell Signaling Technology	2895	AB_2089254

Reagents

Lipofectamine 2000, Invitrogen (11668019); Polybrene Infection/Transfection Reagent, Sigma-Aldrich (TR-1003); Trypsin Gold Mass Spectrometry Grade, Promega (V5280); D-biotin, Chem-Impex (00033, Cas #: 58-85-5); (+)-S-Trityl-L-cysteine, Alfa Aesar (L14384, CAS #: 2799-07-7); Puromycin dihydrochloride, Sigma-Aldrich (P8833, CAS # 58-58-2); Doxycycline hyclate, Acros (446061000, CAS #: 24390-14-5); cOmplete Protease Inhibitor Cocktail, Roche (5056489001); GSK-A1, SYNkinase, (SYN-1219, CAS #: 1416334-69-4); Phorbol-12-Myristate-13-Acetate, Santa Cruz Biotechnology (sc-3576, CAS #: 16561-29-8); FIPI, Cayman Chemical (13563, CAS #: 939055-18-2); JNK-IN-8, Ambeed Inc. (A206353, CAS #: 1410880-22-6); AZD6244, BioVision (2234–5, CAS #: 606143-52-6); Latrunculin B, Sigma-Aldrich (L5288, CAS #: 76343-94-7); Cytochalasin D, MedChemExpress (HY-N6682, CAS #: 22144-77-0); DS55980254, ProbeChem (PC-49548, CAS #: 2488609-41-0); 32% Paraformaldehyde, Electron Microscopy Sciences (15714); Phalloidin-AF488, Invitrogen (A12379); Phalloidin-AF594, Invitrogen (A12381); Pierce High Capacity Streptavidin Agarose, Thermo Fisher Scientific (20359); Streptavidin-conjugated HRP, Gene-Tex (GTX85912); ProLong Diamond Antifade Mountant with DAPI, Thermo Fisher Scientific (P36971); Clarity Western ECL Substrate, Bio-Rad (1705061); BCA Protein Assay Kit, Thermo Fisher Scientific (23225); Human Transferrin, CF568 Conjugate, Biotium (50-196-4092).

The following compounds were prepared as described previously: 3-azido-1-propanol (Yuan et al., 2012); BCN-BODIPY (Alamudi et al., 2016); VT01454 (Li et al., 2022).

Plasmids and cloning

mScarlet-i-tagged PLD1, PSS1-mCherry, and PSS1(P269S)-mCherry were generated previously (Sohn et al., 2016; Tei and Baskin, 2020). The pSBtet-Lyn₁₀-TurboID-V5-puro plasmid used for making a HeLa cell line stably expressing Lyn₁₀-TurboID-V5 under doxycycline induction was generated previously (Cao et al., 2022).

The PM-specific PI(4,5)P₂ BRET biosensor, L₁₀-mVenus-T2A-sLuc-PLCδ1_PH, and the PI4P BRET sensor, L₁₀-mVenus-T2A-sLuc-P4M(2x) was described previously (Tóth et al., 2016). The PI3P BRET sensor sLuc2xFYVE^Hrs-T2A-mVenus-Rab5 was described previously (Pemberton et al., 2020). To construct PM-specific PA BRET sensor, PLCδ1_PH domain was replaced with Nir2 LNS domain (residues 816–1181) from GFP-Nir2 (816–1181) (Kim et al., 2015).

To construct the mScarlet-i-P2A-RhoB plasmid, the RhoB insert was amplified from pcDNA-GFP10-RhoB (182237; Addgene) and the mScarlet-i insert was amplified from the mScarlet-i-C1 vector (85044; Addgene). The mScarlet-i insert was subcloned into EcoRI/XhoI-digested pCAGGS-Lyn11-FRB*-dGFP-PLDs48-P2A-PLDs48-mCh-iFKBP (Masaaki Uematsu, Baskin Lab), replacing Lyn11-FRB*-dGFP-PLDs48. The RhoB insert was subcloned into KpnI/SalI-digested pCAGGS-mScarlet-P2A-PLDs48-mCh-iFKBP, replacing PLDs48-mCh-iFKBP. pCAGGS-mScarlet-i was made by inserting mScarlet-i into EcoRI/XmaI-digested M269. For amino acid mutagenesis, mutations were incorporated into the mScarlet-i-P2A-RhoB plasmid using the QuikChange XL Site-directed mutagenesis kit (Agilent).

To construct the mScarlet-i-RHPN2 plasmid, the RHPN2 insert was amplified from pDONR221-RHPN2 (DNASU 43557) and subcloned into the XhoI/KpnI-digested mScarlet-i-C1 vector.

For CRISPRa/i, sgRNAs against target genes were selected from the Weissman lab published dataset and cloned into the pCRISPRia-V2 vector using the BstXI and BlpI cut sites. For CRISPR KO, guides were designed with Benchling CRISPR tool and cloned into the BsmBI-digested pLentiCRISPR v2 vector.

Primers

Primers are shown in Table 2.

Table 2.

Primers

Gene	sgRNA forward primer	sgRNA reverse primer	Note
RhoB	5′-TTGGGCAGGGCCAGCTTGATTGAGTTTAAGAGC-3′	5′-TTAGCTCTTAAACTCAATCAAGCTGGCCCTGCCCAACAAG-3′	CRISPRa guide 1
	5′-TTGGACCCCGCGCGCAAACGCTGGTTTAAGAGC-3′	5′-TTAGCTCTTAAACCAGCGTTTGCGCGCGGGGTCCAACAAG-3′	CRISPRa guide 2
	5′-CACCGCGGTGGGCACGTACACCTCG-3′	5′-AAACCGAGGTGTACGTGCCCACCGC-3′	CRISPR KO guide 1
	5′-CACCGCACATAGTTCTCGAAGACGG-3′	5′-AAACCCGTCTTCGAGAACTATGTGC-3′	CRISPR KO guide 2
RHPN2	5′-TTGGCGCTGGAGAAGGAGAACGAGTTTAAGAGC-3′	5′-TTAGCTCTTAAACTCGTTCTCCTTCTCCAGCGCCAACAAG-3′	CRISPRi guide 1
	5′-TTGGAGCGCGTCGGTCATGCTAGGTTTAAGAGC-3′	5′-TTAGCTCTTAAACCTAGCATGACCGACGCGCTCCAACAAG-3′	CRISPRi guide 2

Gene	qPCR forward primer	qPCR reverse primer
RhoB	5′-GTTTGACACTTAATGCACTCGTC-3′	5′-TGAGGTTACAGCGTACAAGTG-3′
RHPN2	5′-GCTGGAGAGTGCCATAGATG-3′	5′-GTATGGGTAAATGTGTCTTTCAGG-3′
GAPDH	5′-TCTCCTCTGACTTCAACAGCGAC-3′	5′-CCCTGTTGCTGTAGCCAAATTC-3′

Cell culture

All cells were maintained in a 5% CO₂, water-saturated atmosphere at 37°C. Flp-In T-Rex HeLa cells (Thermo Fisher Scientific) were maintained in DMEM (Corning) supplemented with 10% FBS (Corning) and 1% penicillin and streptomycin (P/S, Corning). HEK293T (ATCC) and HEK293TN (gift from Anthony Bretscher, Cornell University, Ithaca, NY, USA) cells were maintained in DMEM supplemented with 10% FBS, 1% P/S, and an additional 1 mM sodium pyruvate (Corning). K562-dCas9-KRAB (K562i) and K562-dCas9-VP64 cells (gifts from Martin Kampmann, UCSF, San Francisco, CA, USA) were maintained in RPMI 1650 medium (Corning) supplemented with 10% FBS, 1% P/S, and an additional 2 mM L-glutamine (Corning). U2OS cells were maintained in McCoy’s 5A medium (Corning) supplemented with 10% FBS and 1% P/S. Doxycycline-inducible lyn10-TurboID-V5 stable HeLa and HEK293T cell lines were generated in our previous studies (Cao et al., 2022). HEK293T PLD1/2 double KO cell line was previously generated (Tei and Baskin, 2020).

Plasmid transfection

Plasmids were transfected into mammalian cells using 1 μl Lipofectamine 2000:2 μg DNA or 1 μl PEI:3 μg DNA ratio following the manufacturer’s protocol. Cells were incubated with the transfection mix in Opti-MEM with 10% FBS for 6–8 h and then cultured in regular growth medium until harvest.

Lentivirus production

HEK293TN cells were seeded in 6-well plates with 2 ml media. Cells were transfected at 40–50% confluency. The transfection mixture was created by adding Lipofectamine 2000–containing Opti-MEM in 1:1 vol/vol ratio to DNA-containing Opti-MEM (2:1: 3 Pax2:VSVg:lentiviral plasmid) and incubating for 20 min. Cell media was replaced with fresh Opti-MEM + 10% FBS, and transfection mixture was added dropwise to HEK293TN cells. Opti-MEM was replaced with fresh growth medium after 6–8 h. Viral production was allowed to continue for 48 h with media harvested every 16 h. Virus-containing media was filtered through a 0.45-μm filter and store at 4°C to be used within 1 wk or aliquoted and stored in −80°C for long-term storage.

Lentivirus transduction

For adherent cell lines, cells were seeded in 6-well plate with 2 ml media. One “kill control” well was included in which the cells would not be transduced with virus media. Cell media was replaced with 0.5 ml fresh media + 1.5 ml virus-containing media + 8 μg/ml polybrene at ~60% cell confluency. Virus-containing media was replaced every 12 h for total of three times, and cells were allowed to recover in fresh media for 12 h before antibiotic selection begins. For K562 suspension cells, 1 million cells were resuspended with 2 ml virus-containing media + 8 μg/ml polybrene and centrifuged at 1,000 g for 2 h at 33°C. After centrifugation, cells were recollected and centrifuged to remove virus-containing media, then resuspended with fresh RPMI media. Cells were allowed to recover overnight before antibiotic selection begins.

Validation

CRISPRa/i cell lines were validated by assessing mRNA levels by qPCR. CRISPR KO cell lines were validated by assessing protein levels by western blot analysis.

Confocal microscopy

For live-cell imaging, cells were seeded on 35-mm glass bottom dishes (14 mm diameter, #1.5 thickness, Matsunami Glass). For immunofluorescence, cells were seeded on 12-mm cover glass (#1.5 thickness, Fisherbrand) in 12-well plates (Corning). Glass bottom dish or cover glass was first incubated with 50 μg/ml poly-L-lysine in PBS for 2–5 h and rinsed three times with PBS before seeding of HEK293T cells. Cells were imaged live or fixed by immunofluorescence 24 h after transfection. Images were acquired via the Zeiss Zen Blue 2.3 software on Zeiss LSM 800 confocal laser scanning microscope equipped with Plan Apochromat objectives (40× 1.4 NA) and two GaAsP PMT detectors. Solid-state lasers (405, 488, and 561) were used to excite DAPI/tagBFP, EGFP/Alexa Fluor 488, and mCherry/mScarlet-i/Alexa Fluor 568, respectively. Acquired images were analyzed using FIJI.

Live-cell imaging

Cells were rinsed twice with PBS and imaged in Tyrode’s-HEPES buffer (T/H buffer: 135 mM NaCl, 5 mM KCl, 1.8 mM CaCl₂, 1 mM MgCl₂, 5 mg/ml glucose, 5 mg/ml BSA, and 20 mM HEPES, pH 7.4) at room temperature.

Immunofluorescence

Cells were rinsed twice with PBS, fixed in 4% paraformaldehyde in PBS for 10 min at room temperature, permeabilized with 0.5% Triton X-100 in PBS for 5 min, and blocked in 1% BSA and 0.1% Tween-20 in PBS (blocking buffer) for 30 min. Cells were incubated with primary antibody in blocking buffer for 1 h at room temperature and rinsed three times with 0.1% Tween-20 in PBS. Cells were then incubated with secondary antibody in blocking buffer for 1 h at room temperature in dark. Cells were rinsed three times with 0.1% Tween-20 in PBS before being mounted on slides in ProLong Diamond Antifade with DAPI and incubated overnight in dark before imaging. Slides were stored at 4°C.

Lipidomics analyses

Passage 6–9 of 600,000 HEK293 cells were plated onto 60-mm culture dishes and cultured overnight. The following day, GSK-A1 (100 nM) was added and cells were incubated for 24 h in DMEM medium with high glucose and 10% serum. The harvested cell pellets were resuspended in PBS at a concentration of 6,000 cells/μl and then frozen and sent to Lipotype GmBH for mass spectrometry–based lipid analysis as described previously (Surma et al., 2021). Briefly, lipids were extracted using a chloroform/methanol procedure (Ejsing et al., 2009). Samples were spiked with internal lipid standard mixture. After extraction, the organic phase was transferred to an infusion plate and dried in a speed vacuum concentrator. The dry extract was resuspended in 7.5 mM ammonium formate in chloroform/methanol/propanol (1:2:4; vol:vol:vol). All liquid handling steps were performed using Hamilton Robotics STARlet robotic platform with the Anti Droplet Control feature for organic solvents pipetting.

MS data acquisition

Samples were analyzed by direct infusion on a QExactive mass spectrometer (Thermo Fisher Scientific) equipped with a TriV-ersa NanoMate ion source (Advion Biosciences). Samples were analyzed in both positive and negative ion modes with a resolution of Rm/z = 200 = 280,000 for MS and Rm/z = 200 = 17,500 for MSMS experiments in a single acquisition. MS/MS was triggered by an inclusion list encompassing corresponding MS mass ranges scanned in 1 Da increments (Surma et al., 2015).

Data analysis and post-processing

Data were analyzed with in-house developed lipid identification software based on LipidXplorer (Herzog et al., 2011; Herzog et al., 2012). Data after processing and normalization were performed using an in-house developed data management system. Only lipid identifications with a signal-to-noise ratio >5 and a signal intensity fivefold higher than in corresponding blank samples were considered for further data analysis. Lipidomic data are presented as mol%, defined as the molar percentage of each lipid species relative to total measured lipids in each sample.

Measurement of individual phospholipids at specific compartments using BRET-based sensors in live cells

HEK293 cells were seeded in white-bottom 96-well plates that had been pre-coated with 0.01% poly-L-lysine solution (Sigma-Al-drich) and then cultured overnight. Subsequently, the cells were transfected with 0.1 μg of one of the following lipid sensors: PM-targeted PI4P biosensor (L₁₀-mVenus-T2A-sLuc-P4MSidMx2), PM-targeted PI(4,5)P₂ biosensor (L₁₀-mVenus-T2A-sLuc- PLCd1_PH), PM-targeted PA biosensor (L₁₀-mVenus-T2A-sLuc-Nir2-LNS2, residues 816–1181), or Rab5-targeted PI3P biosensor (sLuc-2xFY-VE^Hrs-T2A-mVenus-Rab5). The transfection was performed using Lipofectamine 2000 according to the manufacturer’s protocol. After 4–6 h of Lipofectamine incubation, the media was replaced with fresh media containing DMSO or other reagents. After 24 h, the cells were quickly washed before being incubated for 30 min in 50 μl of modified Krebs-Ringer buffer at 37°C in ambient air. After the preincubation period, the cell-permeable luciferase substrate, coelenterazine h (50 μl, final concentration 5 μM), was added, and the signal from the mVenus fluorescence and sLuc luminescence were recorded using 485- and 530-nm emission filters over a 4 min baseline BRET measurement (15 s/cycle). The indicated inhibitors were maintained throughout all procedures. All measurements were performed in triplicate wells. BRET ratios (mVenus/Luciferase) were calculated for each well by dividing the 530-nm with the 485-nm intensity values. The processed BRET ratios obtained from drug-treated wells were then normalized to DMSO controls.

Analysis of PI synthetic rate

HEK293 cells were seeded in 12-well plates, which had been pre-coated with 0.01% poly-L-lysine solution and were cultured overnight. The cells were then incubated for 24 h with GSK-A1, either with or without FIPI. Following the initial incubation, the cells were labeled for 1 h with 10 μCi/ml myo-³[H]inositol (PerkinElmer) in an inositol- and serum-free medium. The GSK-A1 and FIPI treatments were maintained throughout this labeling period. The labeling reaction was terminated by adding ice-cold perchloric acid to a final concentration of 5%, followed by a 30-min incubation on ice. After scraping and centrifuging the cells, lipids were extracted from the resulting pellet using an acidic chloroform/methanol procedure, as described by Nakanishi et al. (1995). The extracted lipids were then evaporated, resuspended in a scintillation cocktail, and quantified using scintillation spectrometry.

mRNA-seq

HEK293 cells were treated with DMSO control or GSK-A1 (100 nM) for 24 h. RNA was purified from frozen cells using RNeasy kits (Qiagen). Total RNA integrity was assessed on an Agilent 2100 Bioanalyzer RNA Nano chip and quantified by Qubit prior to library construction with the Illumina TruSeq Stranded RNA kit. 12 libraries (three biological replicates, each with two technical replicates) were sequenced as 2 × 100 bp paired-end reads on an Illumina HiSeq 2500. lcdb-wf (https://github.com/lcdb/lcdb-wf) was used to perform all analysis steps up to and including differential expression analysis. Specifically, raw reads were adapter- and quality-trimmed with cutadapt v5.1 (Martin, 2011), then both pre- and post-trim QC metrics were collected with FastQC v0.12.1 (Andrews, 2010) and aggregated using MultiQC v1.30 (Ewels et al., 2016). Trimmed reads were aligned to GRCh38.p13 (GENCODE v46) using STAR v2.7.11b (Dobin et al., 2013), and gene-level counts were generated by featureCounts (Subread v2.1.1) using the GENCODE v46 GTF (Liao et al., 2014; Frankish et al., 2019). Differential expression analysis between A1 and control samples was conducted in R v4.2.3 using the DESeq2 v1.38.0 package (Love et al., 2014). Raw counts from technical replicates were summed for each biological replicate. A negative-binomial generalized linear model was fitted with the design formula ~ treatment to contrast A1 versus control. Log2 fold-change (FC) estimates were shrunk using the adaptive shrinkage (ashr) method for interpretable effect-size estimation. To flag known TFs, we cross-referenced each gene against the curated set of 1,435 human DNA-binding TFs by Lovering et al. (2021) and added the binary TF indicator to the results (Table S1). Enrichment of TFs among DEGs was assessed by Fisher’s exact test on a 2 × 2 contingency table comparing TF versus non-TF status and significant versus nonsignificant differential expression.

Polysome profiling

Four plates (10-cm) of HEK293T cells/condition was grown to 80% confluency and treated with GSK-A1 (100 nM) for 0, 1, or 6 h. Cells were washed with cold PBS and lysed in the polysome lysis buffer (10 mM HEPES, pH 7.4, 100 mM KCl, 5 mM MgCl₂, and 100 μg/ml cycloheximide with 1% Triton X-100). The nuclei were pelleted by spinning at 21,130 g for 10 min at 4°C. 500 μl of lysates were loaded onto a 15–45% (wt/vol) sucrose density gradients freshly prepared in a SW41 ultracentrifuge tube (Beckman) using a Gradient Master (Biocomp Instruments). Samples were centrifuged at 180,000 g for 2.5 h at 4°C in a Beckman SW41 rotor. Polysome profiles were recorded at A254 using the Brandel Gradient Fractionation System and an ISCO UA-6 UV/Vis detector.

Ribo-seq library construction

The cDNA library construction follows the Ezra-seq method described previously with minor modifications (Mao et al., 2023). In brief, an aliquot of ribosome fractions representing monosome and polysome were collected, followed by digestion with Escherichia coli RNase I (Ambion, 750 U per 100 A260 units) by incubation at 4°C for 1 h. RNA was extracted using Trizol LS reagent (Invitrogen), followed by ethanol precipitation. The ribosome-protected mRNA fragments were separated on a 15% polyacrylamide TBE-urea gel (Invitrogen) and visualized using SYBR Gold (Invitrogen). Selected regions in the gel corresponding to 25–35 nt were excised and dissolved by soaking in 400 μl RNA elution buffer (300 mM NaOAc pH 5.2, 1 mM EDTA, and 0.1 U/μl SUPERase·In) at 4°C for overnight. The gel debris was removed using a Spin-X column (Corning), followed by ethanol precipitation. 10–200 ng RNAs were mixed with 10 U T4 PNK (NEB), 1 μl homemade Ezra enzyme, 5 U Poly(A) Polymerase (NEB), and 20 U SUPERase·In in Ezra buffer and incubated at 37°C for 30 min followed by 70°C for 10 min. Ligation was performed for 60 min at 25°C by adding a 10 μl reaction mixture (0.5 μM biotinylated 5′ end adapter, 1 × T4 Rnl2 reaction buffer, 20 U SUPERase·In, 15% PEG8000, and 10 U T4 RNA ligase 2 truncated KQ [NEB]). The ligated RNA sample was mixed with 10 μl of pre-washed streptavidin beads (NEB) and incubated at room temperature for 10 min. After washing once with 2 × SSC, beads were resuspended in 12 μl nuclease-free water and mixed with 8 μl cDNA synthesis mixture (5 μM reverse transcription primer, 5 × first strand buffer, 0.1 M DTT, 10 mM dNTP, and SuperScript III), followed by incubation at 50°C for 30 min. After washing once with 2 × SSC, the cDNA was amplified by PCR using barcoded sequencing primers. PCR was performed by mixing 1 × HF buffer, 0.5 mM dNTP, 0.25 μM PCR primers, and 0.025 U Phusion polymerase. PCR was carried out under the following conditions: 98°C, 30 s; (98°C, 5 s; 68°C, 15 s; 72°C, 20 s) for 14 cycles; 72°C, 2 min. PCR products were separated on an 8% polyacrylamide TBE gel (Invitrogen). DNA products with the expected size around 180 bp were excised and recovered from DNA elution buffer (300 mM NaCl and 1 mM EDTA). After quantification by Agilent BioAnalyzer DNA 1000 assay, equal amounts of barcoded samples were pooled and sequenced using NextSeq 500 (Illumina).

Ribo-seq analysis

To align sequencing reads, the 5′ and-3′ adapters of the reads were trimmed by Cutadapt (version 2.8). The trimmed reads with length shorter than 15 nucleotides were excluded from the analysis. To keep accurate reading frame of Ribo-seq, low-quality bases at both ends of the reads were not subject to clip. The trimmed reads were first aligned to rRNAs using Bowtie (version 1.2.3) (Langmead et al., 2009). The rRNA sequences were downloaded from the nucleotide database of NCBI and RNAcentral. The reads unaligned to rRNAs were then mapped to the custom human transcriptome using STAR (version 2.7.10a) (Dobin et al., 2013). To avoid ambiguity, reads mapped to multiple positions or with >2 mismatches were disregarded for further analysis. The custom transcriptome was generated based on the reference genome and annotations obtained from Ensembl using human release 109 (GRCh38.p14). Protein-coding genes were extracted, and a single transcript was selected for each gene on the following procedure. For each gene, the transcript with the longest coding sequence (CDS) was initially selected. If the selected transcripts have equal CDS length, the longest transcript was included in the custom transcriptome. Mapping to ribosomal P sites was performed by shifting the read position from the 5′ ends to the position by 12 nt.

Data and code availability

The sequencing data reported in this manuscript have been deposited in NCBI’s Gene Expression Omnibus under accession number GSE308521. Custom Python scripts used to analyze the sequencing data are available at GitHub and Zotero at https://github.com/usa0ri/Huang2025 and https://doi.org/10.5281/zenodo.17144598.

Proximity biotinylation with Lyn₁₀-TurboID

pSBtet-Lyn10-TurboID-V5 expression were induced in HeLa or HEK293T cells with 2.5 μg/ml doxycycline for 48 h, treated with 0, 1, 6 h GSK-A1 or 1 h VT01454, respectively, and then incubated with 500 μM biotin for 10 min at 37°C under 5% CO₂. Cells were rinsed five times with PBS. Cell pellet was collected by trypsinizing cells and rinsing three times with PBS at 500 g, 4°C for 3 min. Cell pellet was resuspended with RIPA buffer (150 mM NaCl, 25 mM Tris, pH 8.0, 1 mM EDTA, 0.5% sodium deoxycholate, 0.1% SDS, and 1% Triton X-100) supplemented with cOmplete protease inhibitor cocktail and sonicated using a tip sonicator at 20% amplitude, 1 s on and 1 s off, for 4 s. The cell mixture was centrifuged at 13,000 g for 5 min at 4°C to clarify lysates. Protein concentration was quantified using the BCA assay (Thermo Fisher Scientific), and a small fraction of the clarified cell lysates was saved and normalized as input. The remaining cell lysate was subjected to pulldown using streptavidin-agarose with rotation at 4°C overnight (12–16 h). The resin was then centrifuged for 5 min at 1,000 g, washed two times with RIPA buffer, one time with 1 M KCl, once with 0.1 M Na₂CO₃, once with 2 M urea in 10 mM Tris, pH 8.0, and twice with RIPA buffer to reduce nonspecific binding. Samples were then denatured and analyzed by SDS-PAGE and western blot with detection by chemiluminescence using Clarity Western ECL substrate.

TurboID proteomic sample preparation

Cells were grown in five to six 15-cm dishes per condition. Biotinylation and streptavidin IP were carried out as described above. Following the last RIPA rinse, the GFP-Trap magnetic resin was incubated with elution buffer (100 mM Tris, pH 8.0, 1% SDS) at 95°C for 5 min, and the resin was spun at a tabletop centrifuge for 5 s, and the supernatant was collected. This step was repeated two more times, and supernatants were collected and combined for TMT labeling sample preparation. IP eluates were reduced by using 200 mM TCEP for 1 h at 55°C. After that, samples were alkylated for 30 min at room temperature and in the dark using 375 mM iodoacetamide. Trypsin Gold, mass spectrometry grade (catalog no. V5280; Promega), was used to digest the samples at a ratio of 1:100 enzyme to substrate. The samples were then incubated to overnight at 37°C. The Pierce Quantitative Colorimetric Peptide Assay (catalog no. 23275; Thermo Fisher Scientific) was utilized to quantify the concentrations of peptides. For TMT tests, samples were resuspended and normalized using 1 M triethylammonium bicarbonate (catalog no. 90114; Thermo Fisher Scientific). Samples were labeled using TMT 16plex label reagent sets (catalog no. A44520; Thermo Fisher Scientific) at a (wt/wt) label-to-peptide ratio of 20:1 for 1 h at room temperature. Labeling reactions were quenched by the addition of 5% hydroxylamine for 15 min and pooled and dried using a SpeedVac. Labeled peptides were enriched and fractionated using Pierce High pH Reversed-Phase Peptide Fractionation Kit according to the manufacturer’s protocol (catalog no. 84868; Thermo Fisher Scientific). Liquid chromatography–tandem mass spectrometry fractions were analyzed using an EASY-nLC 1200 System (catalog no. LC140; Thermo Fisher Scientific) equipped with an in-house 3 μm C18 resin-(Michrom Bioresources) packed capillary column (125 μm × 25 cm) coupled to an Orbitrap Fusion Lumos Tribrid Mass Spectrometer (catalog no. IQLAAEGAAPFADBMBHQ; Thermo Fisher Scientific). The mobile phase and elution gradient used for peptide separation were as follows: 0.1% formic acid in water as buffer A and 0.1% formic acid in 80% acetonitrile as buffer B; 0–5 min, 5%–8% B; 5–65 min, 8–45% B; 65–66 min, 45%–95% B; 66–80 min, 95% B; with a flow rate set to 300 nl min–1. MS1 precursors were detected at m/z = 375–1500 and resolution = 120,000. A CID-MS2-HCD-MS3 method was used for MSn data acquisition. Precursor ions with charge of 2+ to 7+ were selected for MS2 analysis at resolution = 50,000, isolation width = 0.7 m/z, maximum injection time = 50 ms, and CID collision energy at 35%. 6 SPS precursors were selected for MS3 analysis, and ions were fragmented using HCD collision energy at 65%. Spectra were recorded using Thermo Xcalibur Software v.4.4 (catalog no. OPTON-30965; Thermo Fisher Scientific) and Tune application v.3.4 (Thermo Fisher Scientific). Raw data and the V5 sequence (ggcaagcccatccccaaccccctgctgggcctggacagcacc) were searched using Proteome Discoverer Software 2.5 (Thermo Fisher Scientific) against an UniProtKB human database.

Downstream proteomic analysis

The computational tool Magma was used to analyze mass spectrometry proteomics data. Magma quantifies the differences in protein abundance between different experimental conditions by calculating FC and P values using a customized linear mixed-effects model inspired by the MSStats TMT package (Huang et al., 2020; Gupta et al., 2024, Preprint). Custom normalization was performed using the V5 sequence, assumed to remain constant between conditions. By comparing each bait protein against untransfected HeLa or HEK293T cells, the bait protein’s interactors were identified using criteria of FC > 2, adjusted P value < 0.05, and peptide-spectrum matches > 5. Analysis was then narrowed to the combined set of interactors for both DMSO control and GSK-A1/VT01454-treated samples. To elucidate the specific effects of the GSK-A1/VT01454 treatment, we generated a volcano plot using the FC and adjusted P values derived from the comparison between the GSK-A1/VT01454-treated samples and the DMSO control samples. Known contaminants in AP-MS experiments, including keratin, myosins, small ribosomal subunit proteins, heat shock–related 70 kDa proteins (HSPA), and large ribosomal subunit proteins, were excluded from the analysis.

Western blot

Cells were lysed with RIPA buffer (150 mM NaCl, 25 mM Tris, pH 8.0, 1 mM EDTA, 1% Triton X-100, 0.5% sodium deoxycholate, and 0.1% SDS) supplemented with cOmplete protease inhibitor cocktail on ice. Cell lysate was sonicated with a tip sonicator at 20% intensity, 1 s on and 1 s off, for 4 s and centrifuged at 13,000 g for 5 min at 4°C to clear lysate. Protein concentrations of the clarified cell lysates were quantified using the BCA assay. Cell lysates were normalized to 1–5 mg/ml and denatured with 6× Laemmli buffer (13.3% SDS, 0.067% bromophenol blue, 52.2% glycerol, 67 mM Tris, pH 6.8, and 11.1% β-mercaptoethanol) at 95°C for 5 min. Samples were analyzed by SDS-PAGE and western blot with detection by chemiluminescence using Clarity Western ECL substrate (Bio-Rad).

Transferrin uptake assay

Cells were seeded in 24-well plates and allowed to grow for 2 days before drug treatments. Cells were serum starved for 1 h at 37°C, with drug treatment continuing. Subsequently, cells were incubated with 25 μg/ml Transferrin-CF568 in PBS + 0.1% BSA for 30 min at 37°C. After 2× cold PBS + 0.1% BSA rinses, cells were trypsinized and transferred to a 96-well V-bottom plate and centrifuged at 1,000 g, 4°C to remove supernatant. Cells were resuspended in 175 μl PBS + 0.1% BSA/well and centrifuged at 1,000 g, 4°C for rinsing. After two more rinses, cells were resuspended in 4% paraformaldehyde in PBS and incubated at room temperature in the dark for 10 min. After two more rinses with FACS buffer (PBS + 0.1% FBS), cells were resuspended with 120–150 μl of FACS buffer and subjected to flow cytometry analysis.

Agonists and inhibitors treatment summary

The following compounds were added to cells in conditions as summarized:

PLD activators: PMA, 100 nM, 30 min; GSK-A1, 100 nM, 1–24 h; VT01454, 100 nM, 1 h.

PLD inhibitor: FIPI, 750 nM, 30 min prior to PLD stimulation.

PSS1 inhibitor: DS, 1 μM, 24 h.

MEK inhibitor: AZD6244, 2 μM, 24 h.

JNK inhibitor: JNK-IN-8, 1 μM, 24 h.

p38 MAPK inhibitor: SB203580, 5, 1 h pretreatment prior to PLD stimulation.

Actin polymerization inhibitors: LatB, 2.5 μM, 1 h prior to PLD stimulation; cytoD, 2 μM, 1 h prior to PLD stimulation.

Protein expression inhibitor: cycloheximide, 50 μg/ml, 1 h prior to PLD stimulation.

IMPACT labeling and flow cytometry

For adherent cell lines, cells were seeded in 24-well plate and allowed to grow for 2 days before transfection or drug treatment. After transfection or any drug pretreatment, negative control cells were treated with 750 nM FIPI for 30 min for subtraction of IMPACT background. Cells were subsequently treated with PLD stimulus for indicated time. Cells were then treated with 1 mM azidopropanol for 30 min at 37°C to generate azido phosphatidylalcohols. Media was then aspirated, and cells were rinsed with three times with PBS. Cells were then incubated with prewarmed 1 μM BCN-BODIPY in T/H buffer (135 mM NaCl, 5 mM KCl, 1.8 mM CaCl₂, 1 mM MgCl₂, 5 mg/ml glucose, 5 mg/ml BSA, and 20 mM HEPES, pH 7.4) at 37°C for 30 min, rinsed two times with PBS, and incubated with prewarmed T/H buffer at 37°C for 15 min to rinse out unreacted BCN-BODIPY. After two more PBS rinses, cells were trypsinized and transferred to a 96-well V-bottom plate and centrifuged at 1,000 g, 4°C to remove supernatant. Cells were resuspended in 175 μl PBS/well and centrifuged at 1,000 g, 4°C for rinsing. After two more rinses, cells were resuspended in 4% paraformaldehyde in PBS and incubated at room temperature in the dark for 10 min. After two more rinses with FACS buffer (PBS + 0.1% FBS), cells were resuspended with 120–150 μl of FACS buffer and subjected to flow cytometry analysis. A 488 nm laser was used to excite BODIPY-derived IMPACT fluorescence as a measure of PLD activity, and a 561 nm laser was used to excite the mScarlet-i- or mCherry-containing constructs. For experiments involving two colors, compensation was performed using single color control cells and unstained cells, and IMPACT fluorescence from cells with similar mScarlet-i/mCherry expression levels were extracted for data analysis. A BD Accuri instrument was used for 1-color flow cytometry, and a Thermo Fisher Scientific Attune instrument was used for 2-color flow cytometry analysis. Cells were gated to exclude cell debris, followed by singlet isolation, and then FL2-A or FL1-A values were recorded. For each experimental condition, the average of median fluorescence intensities of the negative control FIPI-treated samples were subtracted from the respective no FIPI samples, and the differences of median fluorescence were normalized and plotted.

RT-IMPACT labeling and confocal microscopy

HEK293 cells were labeled as previously described (Liang et al., 2019). Briefly, cells (500,000) were seeded onto 29-mm glass-bottom dishes (Cellvis, Cat# D29-20-1.5-N) pre-coated with 0.01% poly-L-lysine solution 1 day prior to imaging. All media used for labeling was pre-warmed to 37°C. Cells were pretreated with the PLD inhibitor FIPI or DMSO vehicle control in culture media for 30 min at 37°C. Following pretreatment, the media was completely aspirated, especially media within the glass well at the center of the dish. A 200 μl solution of 3 mM trans-5-oxocene (oxoTCO) was prepared in media, with or without FIPI, and was then added to cover only the central glass area of the dish and incubated for 10 min at 37°C. After incubation, the oxoTCO solution was completely aspirated, followed by a brief rinse with 1 ml of fresh media and a final long rinse (5 min) with an additional 1 ml of fresh media at 37°C. After rinsing, 1 ml of pre-warmed T/H buffer was added, and the dish was quickly placed on the microscopy stage, and cells were put into focus. The buffer was then carefully removed while the dish was on the stage, and 100 μl of pre-warmed T/H buffer was added to the central glass well. Time-lapse imaging was initiated, followed by the addition of 100 μl of a 2 μM (2×) tetrazine-BODIPY solution in pre-warmed T/H buffer, resulting in a final concentration of tetrazine-BODIPY of 1 μM. Time-lapse imaging was commenced at 37°C, with frames acquired every 3–4 s, followed by addition of Tz-BODIPY (1 μM). Images were acquired for the subsequent 1–2 min. Shown are images of the earliest time points with detectable fluorescent signal above background (~10 s).

Statistical analysis

For all experiments involving quantification, statistical significance was calculated in GraphPad Prism using tests as indicated in the figure legend. P values for each comparison are reported, and the number of biological replicates or cells analyzed is stated in the legend. All the raw data were plotted into graphs using GraphPad Prism. For all scatter plots, the black line indicates the mean.

Supplementary Material

Fig. S1 shows flow cytometry sample gating strategy and effects of GSK-A1 on PPIn levels beyond PI4P and PLD activity. Fig. S2 shows that GSK-A1 effects are recapitulated across U2OS and HEK293T cells. Fig. S3 shows that TurboID proximity labeling proteomics results for GSK-A1–treated HeLa cells. Fig. S4 provides qPCR validation for CRISPRa/i cell lines and shows the impact of PI4P depletion on endocytosis, ER homeostasis, and cell viability. Table S1 contains the full dataset from RNA-seq experiments. Table S2 contains the full dataset from ribosome profiling experiments. Table S3 contains the full dataset from proximity labeling experiments. Table S4 contains a list of increased hits from ribosome profiling experiments.

Data availability

Raw FASTQ files and DESeq2 library–size normalized counts (counts.tsv) have been deposited in the Gene Expression Omnibus under accession no. GSE303723.