LPS-induced clustering of CD14 triggers generation of PI(4,5)P₂

Lipopolysaccharide (LPS) is the main component of the outer membrane of Gram-negative bacteria and triggers strong pro-inflammatory responses upon recognition by Toll-like receptor 4 (TLR4). These reactions aim at eradication of the bacteria; however, an excessive reaction to LPS might lead to a systemic inflammatory reaction, termed sepsis (Poltorak et al., 1998; Salomao et al., 2012). In a typical scenario, activation of TLR4 is preceded by an interaction of LPS aggregates with LPS-binding protein (LBP) and subsequent extraction of LPS monomers by CD14, a glycosylphosphatidylinositol (GPI)-anchored glycoprotein that is fairly abundant on the surface of macrophages and other myeloid lineage cells (Simmons et al., 1989; Gioannini et al., 2005). CD14 transfers the LPS to MD-2 (also known as LY96) protein that is associated with the ectodomain of TLR4, and the binding of the LPS induces dimerization of TLR4–MD-2 complexes (Da Silva Correia et al., 2001; Park et al., 2009). This, in turn, triggers the assembly of signaling complexes at TLR4, which is initiated by two pairs of adaptor proteins – TIRAP–MyD88 and TRAM–TRIF (Kawai et al., 1999; Yamamoto et al., 2002, 2003). The first pair of adaptors triggers a signaling cascade that leads to early-phase activation of NF-κB and AP-1, which eventually control the production of pro-inflammatory cytokines, such as tumor necrosis factor-α (TNF-α). By contrast, the TRAM–TRIF adaptors initiate a signaling cascade that promotes expression of type-I interferons (IFN) and IFN-inducible genes; late-phase activation of NF-κB follows (Kawai and Akira, 2011). The activated TLR4 needs to be internalized in a dynamin-dependent manner and to reach endosomes in order to initiate the TRIF-dependent processes (Husebye et al., 2006; Kagan et al., 2008; Tanimura et al., 2008).

The internalization of activated TLR4, and thus the onset of the TRIF-dependent signaling, is controlled by CD14 (Perera et al., 1997; Jiang et al., 2005; Zanoni et al., 2011); CD14 also triggers activation of NFAT independently of TLR4 in LPS-stimulated dendritic cells. This latter process has been found to depend on the integrity of so-called plasma membrane rafts (Zanoni et al., 2009). Rafts are nanoscale (2–20 nm) assemblies with a lifetime of the order of nanoseconds and are enriched with cholesterol, saturated sphingolipids and proteins that have a GPI anchor, including CD14, as well as with palmitoylated proteins. After cell stimulation, such highly dynamic rafts can coalesce to form platforms that facilitate signal transduction (Lingwood and Simons, 2010; Kusumi et al., 2012). Ample data indicate that CD14-containing rafts are the sites of the assembly of multimolecular protein complexes that govern LPS recognition and the downstream signaling cascades of TLR4 (Pfeiffer et al., 2001; Triantafilou et al., 2004a,b; Dhungana et al., 2009; Zhu et al., 2010). Thus, the participation of CD14 in LPS-induced responses goes beyond LPS binding, and the localization of this protein to rafts can determine the signaling.

Despite the crucial role of lipid rafts in LPS-triggered signaling, the mechanisms that control the turnover of plasma membrane lipids in LPS-stimulated cells are relatively unknown. It has been shown that phosphatidylinositol 4,5-bisphosphate [PI(4,5)P₂] and its derivatives are vital for TLR4 signaling (Kagan and Medzhitov, 2006; Aksoy et al., 2012). PI(4,5)P₂ is confined almost exclusively to the inner leaflet of the plasma membrane and is generated mainly by phosphorylation of phosphatidylinositol 4-monophosphate [PI(4)P] catalyzed by type-I phosphatidylinositol 4-phosphate 5-kinase (PIP5K) isoforms Iα, Iβ and Iγ (encoded in humans by PIP5K1A, PIP5K1B and PIP5K1C, respectively) (Kunz et al., 2000). In LPS-stimulated cells, PI(4,5)P₂ ensures the binding of TIRAP to the plasma membrane and regulates MyD88-dependent signaling of TLR4 (Kagan and Medzhitov, 2006). Phosphorylation of PI(4,5)P₂ to PI(3,4,5)P₃ by the p110δ isoform of phosphatidylinositol 3-kinase serves as a switch from MyD88- to TRIF-dependent signaling (Aksoy et al., 2012), and PI(4,5)P₂ hydrolysis can also contribute to this switch (Chiang et al., 2012). It has recently been reported that LPS induces an increase of the PI(4,5)P₂ level in microglia by activating PIP5K Iα (Nguyen et al., 2013).

It has been proposed that the involvement of PI(4,5)P₂ in signaling pathways that are triggered by plasma membrane receptors relies on mechanisms that ensure PI(4,5)P₂ generation and/or PI(4,5)P₂ unmasking at discrete locations of the plasma membrane (Kwiatkowska, 2010). Because PI(4,5)P₂ can be enriched in rafts of the plasma membrane (Liu et al., 1998; Saito et al., 2003; Parmryd et al., 2003; Szymanska et al., 2009; Furt et al., 2010), we examined the involvement of CD14 in PI(4,5)P₂ generation in the response of J774 macrophage-like cells to LPS. We found that LPS induces rapid clustering of CD14 in the plane of the plasma membrane, which then triggers generation of PI(4,5)P₂ by PIP5K Iα and PIP5K Iγ in rafts required for the both signaling pathways of TLR4.

To decipher the involvement of PI(4,5)P₂ in TLR4 signaling, we examined whether the level of the lipid changes in J744 macrophage-like cells under the influence of 10–100 ng/ml LPS. In unstimulated cells, the amount of PI(4,5)P₂ was estimated at 76.2±0.4 pmol/mg of protein, and during the LPS treatment it increased in a biphasic manner (Fig. 1A,B). In cells exposed to 10 ng/ml LPS, the first peak of PI(4,5)P₂ accumulation was detected at 10 min, when its level increased about 1.7-fold. Following a drop during the next 20 min, the PI(4,5)P₂ level rose again markedly, to about 2.4 times that in unstimulated cells, after 60–90 min of stimulation with LPS (Fig. 1A). A similar pattern was found in cells that had been stimulated with 100 ng/ml LPS, although the first PI(4,5)P₂ peak was detected as early as 5 min when the PI(4,5)P₂ level increased about 1.5-fold; in the second peak seen at 60 min of stimulation with LPS, the PI(4,5)P₂ level rose about 3-fold (Fig. 1B). These dynamics of PI(4,5)P₂ were confirmed by using a protein–lipid overlay assay with a probe containing the pleckstrin homology domain of phospholipase Cδ₁ (PLC-PH–GST) (Szymanska et al., 2008). This approach allowed us to detect PI(4,5)P₂ in a dose-dependent manner in lipids extracted from 0.2×10⁶–2×10⁶ cells (Fig. S1), and it also indicated that there was an elevation of the PI(4,5)P₂ level at 5 min and 60 min of stimulation with LPS (Fig. 1C, left). In samples that had been exposed to the PH-PLC–GST probe pre-incubated with PI(4,5)P₂-containing liposomes, negligible nonspecific staining was detected (Fig. 1C, right). The binding of PLC-PH–GST was also inhibited when lipids were extracted from cells that had been pre-treated with 10 μM ionomycin in the presence of external Ca²⁺ (Fig. S1), which leads to PLC activation and PI(4,5)P₂ hydrolysis (Varnai and Balla, 1998).

The LPS-induced generation of PI(4,5)P₂ was nearly abolished when the binding of LPS to CD14 was inhibited with an antibody specific to CD14 (4C1 clone) (Fig. 1A,B); this antibody is known to efficiently prevent the activation of the CD14-dependent TRIF-mediated signaling pathway of TLR4 (Borzecka et al., 2013).

We studied the relationship between CD14 and PI(4,5)P₂ by examining their localization in the plane of the plasma membrane at the ultrastructural level. For this, sheets of the dorsal portion of the membrane were prepared through mechanical cleavage of cells. Labeling of CD14 before cell cleavage allowed us to visualize its distribution on the plasma membrane surface. In unstimulated cells, the gold particles labeling CD14 were scattered on the cell surface mainly as singlets or doublets, accounting for about 85% of all particles (Fig. 2A,H). After 5 min of treatment with 100 ng/ml LPS, a distinct redistribution of CD14 occurred, and nearly 50% of the CD14 gold label was found in clusters comprising 3–5, to over 15, particles (Fig. 2B,H). The proportion of clustered CD14 did not change substantially after 30 min of cell stimulation (Fig. 2C,H) but was clearly reversible, and after another 30 min returned nearly to the level found in unstimulated cells (Fig. 2D,H). Furthermore, when the total number of gold particles marking CD14 in a 100-µm² area of the plasma membrane sheet was determined, substantial changes in the amount of CD14 on the cell surface were found. Thus, in unstimulated cells, 525±15 gold particles were found (mean±s.e.m.), and after 5 and 30 min of stimulation with LPS, their number increased to 1388±41 and 1545±39, respectively, and then decreased to 774±24 at 60 min. These data indicate that during the early stages of stimulation with LPS, an enrichment and clustering of CD14 on the surface of the plasma membrane occurs.

Taking into account the preferential location of CD14 in plasma membrane rafts, we examined whether LPS also affects the distribution of ganglioside GM1, another raft component. Unlike for CD14, no translocation of GM1 was detected; the glycolipid was decorated mainly by singlets or doublets of gold particles both before and during stimulation of cells with 100 ng/ml LPS (Fig. 2E–H).

Having established the pattern of LPS-induced CD14 redistribution, we analyzed the localization of PI(4,5)P₂ in the inner leaflet of the plasma membrane using double gold labeling. In unstimulated cells, most PI(4,5)P₂ was distant from CD14, and only about 23% of the overall PI(4,5)P₂ content was located within the vicinity of CD14 (Fig. 3A; Table S1). The relative distribution of CD14 and PI(4,5)P₂ did not change after 1 min of stimulation of cells with 100 ng/ml LPS (Fig. 3B); however, after 5 min, an increased colocalization of PI(4,5)P₂ and CD14 was detected (Fig. 3C). Quantification of gold particles revealed that, within 5 min of exposing the cells to 10 or 100 ng/ml LPS, as many as 42% and 46%, respectively, of the gold particles attributed to PI(4,5)P₂ colocalized with CD14 (Table S1). This colocalization tended to decline after 30 min of stimulation, and after 60 min only 30–35% of the gold particles attributed to PI(4,5)P₂ coincided with CD14 (Fig. 3D,E; Table S1). At this stage of cell stimulation, ∼100-nm vesicles decorated with PI(4,5)P₂ were often seen (Fig. 3F).

Both before and during stimulation of the cells with 10 or 100 ng/ml LPS, PI(4,5)P₂ was visualized mainly as assemblies of 3–5 gold particles (Table S1). However, the colocalization of PI(4,5)P₂ clusters, exhibited by assemblies of 3–5 as well as 6–10 gold particles, and CD14 increased 2- to 2.3-fold after 5 min of stimulation with LPS (Table S1). To underscore the positive relationship between the clustering of CD14 and of PI(4,5)P₂, we summed up all the clustered gold particles (assemblies of at least three gold particles), reflecting assemblies of CD14 with PI(4,5)P₂ nearby (Fig. 3G). This analysis confirmed that the prominent clustering of CD14 induced by 100 ng/ml LPS after 5 min was accompanied by co-clustering of PI(4,5)P₂. After 60 min of cell stimulation, when the clustering of CD14 reversed and the overall amount of cell-surface CD14 declined, markedly fewer PI(4,5)P₂ clusters coincided with the protein (Fig. 3G) despite the presence of substantial amounts of PI(4,5)P₂ at this stage of cell stimulation (see Fig. 1B). The accumulation of PI(4,5)P₂ around CD14 was inhibited when the binding of LPS to CD14 was blocked. Under these conditions, after 5 min of cell stimulation with 10 or 100 ng/ml LPS, the distribution of labeled PI(4,5)P₂ resembled that observed in unstimulated cells (Table S1).

Taken together, these data indicate that LPS induces rapid clustering of CD14 in the plane of the plasma membrane that is concomitant with the generation of PI(4,5)P₂ and its accumulation within the vicinity of the CD14 assemblies.

Taking into account the observed correlation between LPS-induced clustering of CD14 and PI(4,5)P₂ generation, we examined whether clustering of CD14 without an involvement of LPS could also trigger PI(4,5)P₂ production. For this purpose, J774 cells were exposed to an antibody against CD14 at 0°C in order to promote antibody binding and were subsequently incubated with a secondary antibody at 37°C, thereby inducing CD14 crosslinking. Following 5–10 min of such treatment, CD14 translocated from an apparently random distribution to form distinct large polar aggregates, called caps (Fig. 4A,B). At the same time, the cellular level of PI(4,5)P₂ rose by up to 2.7-fold, as analyzed by using an ELISA assay (Fig. 4C). A significant increase of PI(4,5)P₂ was also detected in these conditions by using a protein–lipid overlay assay with the PLC-PH–GST probe (Fig. 4D, left), but not when the probe had been pre-adsorbed with PI(4,5)P₂-containing liposomes (Fig. 4D, right). Immunofluorescence analysis indicated that the lipid accumulated at the CD14 caps (Fig. 4B). Neither exposure of the cells to the antibody against CD14 alone at 0°C or 37°C, nor to the secondary antibody alone, induced comparable PI(4,5)P₂ generation (Fig. 4C,D).

In contrast to crosslinking CD14, crosslinking of TLR4 with antibodies for 10 min induced only a slight (about 1.37-fold) elevation of the PI(4,5)P₂ level (Fig. 4E). Furthermore, crosslinking of GM1 with cholera toxin B subunit (CTX) and an antibody against CTX did not affect the cellular level of PI(4,5)P₂ (Fig. 4F). These results indicate that strong PI(4,5)P₂ generation is a specific response to CD14 clustering.

To verify the role of CD14 in PI(4,5)P₂ generation, we expressed the protein in HEK293 cells, which lack endogenous expression of CD14 and TLR4, and exposed the cells to 100 ng/ml LPS, as in J774 cells. The HEK-CD14 transfectants expressed CD14 at a high level on the cell surface. The protein and PI(4,5)P₂ were rather homogenously distributed in the plane of the plasma membrane (Fig. 5A). Following exposure to LPS, fine aggregates of CD14 in the membrane were detectable, especially at the dorsal surface of cells (Fig. 5B). In this time (5 min), colocalization of CD14 and PI(4,5)P₂, reflected by the Pearson's correlation coefficient, increased from 0.44±0.03 to 0.61±0.01 (in relation to 1 as the maximal value; mean±s.e.m.). LPS-treated HEK-CD14 cells reproduced the biphasic profile of PI(4,5)P₂ generation found in J774 cells (Fig. 5D, see Fig. 1), although in HEK-CD14 cells, the first peak of PI(4,5)P₂ was more prominent than the second. Thus, after 5 min of stimulation with LPS, the level of PI(4,5)P₂ in HEK-CD14 cells rose about 2.4-fold, returned to the basal level and then rose again by about 50% after 60 min. Pre-incubation of HEK-CD14 cells with the function-blocking antibody against CD14 nearly abolished the LPS-induced PI(4,5)P₂ increase, resembling the results obtained in J774 cells (Fig. 5D). CD14 expression was required for the PI(4,5)P₂ generation because no changes in the level of PI(4,5)P₂ were found in non-transfected HEK293 cells that had been exposed to LPS (Fig. 5D). However, exposure of HEK-CD14 cells to 100 ng/ml Pam₂CSK₄ diacylated lipopeptide, a ligand of the TLR2–TLR6 dimer – which to some extent depends on CD14 for activation (Schroder et al., 2004; Borzecka et al., 2013) – induced only a negligible elevation of PI(4,5)P₂ (Fig. 5E). By contrast, the crosslinking of CD14 with antibodies for 5–10 min without LPS evoked a prominent elevation of the amount of PI(4,5)P₂ in the HEK-CD14 transfectants that was not observed when either the antibody against CD14 or the secondary antibody were omitted. After 10 min of CD14 crosslinking, a gradual and steady decrease of the PI(4,5)P₂ level was detected (Fig. 5F). The crosslinking of CD14 with antibodies induced its translocation to distinct aggregates at the cell periphery, which colocalized with assemblies of PI(4,5)P₂ (Fig. 5C). Taken together, these data indicate that clustering of CD14 in the plane of the plasma membrane is sufficient to trigger PI(4,5)P₂ generation.

In an attempt to identify the type-I PIP5K isoform(s) that contribute to the generation of PI(4,5)P₂ in LPS-stimulated cells, we first estimated the relative abundance of PIP5K isoforms Iα, Iβ and Iγ in J774 cells using reverse transcriptase quantitative (RT-q)PCR to analyze their mRNA, and found that mRNAs encoding PIP5K Iγ and PIP5K Iα were most abundant (Fig. 6A). Transfection of cells with small interfering (si)RNAs targeting PIP5K Iα or PIP5K Iγ led to a profound depletion of the respective kinase at the mRNA and protein levels (Fig. 6B,C). However, the downregulation of PIP5K Iα with the specific siRNA correlated with a moderate increase of the amount of PIP5K Iγ (by up to 18%) and vice versa (Fig. 6C). As expected, targeting the two kinases downregulated both of them (Fig. 6C). The level of mRNA encoding PIP5K Iβ was unaffected by silencing of the other kinase isoforms (Fig. 6B).

The silencing of PIP5K Iα and/or PIP5K Iγ did not significantly reduce the basal level of PI(4,5)P₂ in comparison to that found in cells transfected with control scrambled siRNA, but it substantially inhibited the generation of PI(4,5)P₂ during a subsequent stimulation of the cells with LPS (Fig. 6D). The silencing of PIP5K Iα was less efficient in preventing PI(4,5)P₂ generation because, under these conditions, its level rose by about 29% after 60 min of stimulation with LPS (Fig. 6D). A joint downregulation of PIP5K isoforms Iα and Iγ led to strong depletion of PI(4,5)P₂ in LPS-stimulated cells (Fig. 6D).

The depletion of PI(4,5)P₂ after silencing of PIP5K Iα and PIP5K Iγ correlated with a significant inhibition of NF-κB, the activation of which is a hallmark of the pro-inflammatory signaling that is triggered by LPS (Fig. 6E). Notably, LPS-induced activation of NF-κB, indicated by IκB phosphorylation, followed the biphasic profile described above for PI(4,5)P₂, with the first peak of IκB phosphorylation seen at 5–10 min and the second after 60 min of stimulation with LPS (Fig. S2A). After silencing of PIP5K Iα and PIP5K Iγ, both peaks of IκB phosphorylation were diminished by 27–46% (Fig. 6E). The inhibition of IκB phosphorylation was followed by a reduced production of TNF-α and RANTES (also known as CCL5), the cytokines used to gauge the MyD88- and TRIF-dependent pathways of TLR4, respectively (Bjorkbacka et al., 2004). The TNF-α and RANTES production stimulated by 10 ng/ml LPS was reduced by about 38% and 55%, and that induced by 100 ng/ml by about 26% and 37%, respectively, when PIP5K Iα and PIP5K Iγ were jointly silenced (Fig. 6F,G). The effect of silencing either of the PIP5KI isoforms alone on cytokine production depended on the LPS concentration – depletion of PIP5K Iα inhibited TNF-α and RANTES release only in cells that had been exposed to 10 ng/ml LPS, whereas the depletion of PIP5K Iγ inhibited the cytokine production induced by 10 or 100 ng/ml LPS by 21–41% (Fig. 6F,G). In either case, the production of RANTES was more sensitive to the downregulation of the kinase(s).

Taken together, the data indicate that both PIP5K Iα and PIP5K Iγ contribute to the production of PI(4,5)P₂ in cells stimulated with LPS, although PIP5K Iα seems to be dispensable for the generation of the cytokines at the higher dose of LPS.

Further data on the involvement of PI(4,5)P₂ in the signaling pathways of TLR4 has come from studies that have used drugs affecting PI(4,5)P₂ generation and availability in J774 cells. LiCl, an inhibitor of inositol monophosphatase (Ohnishi et al., 2007), lowered the basal level of PI(4,5)P₂ by about 14% only. However, stimulation of the LiCl-treated cells with 100 ng/ml LPS decreased the amount of PI(4,5)P₂ further by an additional 50% after 60 min, whereas in control cells, it brought about a 2.4-fold increase (Fig. 7A). The depletion of PI(4,5)P₂ upon treatment with 10 mM LiCl correlated with an inhibition of both peaks of LPS-induced IκB phosphorylation (Fig. 7D), and of production of TNF-α and RANTES (Fig. 7E,F). The inhibition of cytokine production by LiCl was dose dependent (in the range of 0.1–10 mM LiCl) (Fig. 7E,F). At 10 ng/ml LPS, 10 mM LiCl reduced the production of TNF-α by over 50%, whereas with 100 ng/ml and 1000 ng/ml, production was reduced by 38% and 23%, respectively (Fig. 7E). The TRIF-dependent production of RANTES was in fact more sensitive to LiCl than that of TNF-α – at 10 mM LiCl, the release of RANTES was reduced by 66%, 64% and 43% in cells stimulated with 10, 100 and 1000 ng/ml LPS, respectively (Fig. 7F). In another approach aimed at interfering with PI(4,5)P₂ generation in LPS-treated cells, we used 2-bromohexadecanoic acid (BPA). BPA is an inhibitor of protein palmitoylation (Davda et al., 2013), a post-translational modification that is typical for numerous raft-associated proteins and that is also required for the activity and membrane association of three out of four phosphatidylinositol 4-kinases generating PI(4)P, a PI(4,5)P₂ precursor (Jung et al., 2008; Barylko et al., 2009; Nakatsu et al., 2012). BPA reduced PI(4,5)P₂ production in cells that had been stimulated with 100 ng/ml LPS by about 47% (Fig. 7B). In parallel, phosphorylation of IκB was also inhibited, and production of TNF-α and RANTES was reduced by about 49% and 40%, respectively (Fig. 7D–F).

In accordance with the effects of the drugs that interfere with PI(4,5)P₂ generation, application of 2 mM neomycin, which sequesters PI(4,5)P₂ (Laux et al., 2000), also reduced the LPS-induced IκB phosphorylation (Fig. S2B) and TNF-α and RANTES production, although less efficiently than LiCl or BPA did (Fig. 7E,F). As it has been found previously (Szymanska et al., 2008), pre-treatment of cells with 2 mM neomycin abrogates the binding of PLC-PH–GST to cells that have been permeabilized with Triton X-100 (Fig. 7C), illustrating that neomycin can preclude the interaction of PI(4,5)P₂ with effector proteins in living cells, thereby affecting both signaling pathways of TLR4. For the same reason, neomycin interfered with both assays used in the present study for quantification of PI(4,5)P₂ (not shown).

Based on the electron microscopy data that pointed to the generation of PI(4,5)P₂ within the vicinity of CD14, we examined the presence of PI(4,5)P₂ and PIP5K isoforms Iα and Iγ in the Triton-X-100-insoluble and octyl-β-glucoside-soluble [detergent-resistant membrane (DRM)] membrane fraction, which is enriched with raft constituents (Kusumi et al., 2012). In unstimulated cells, the DRM fraction contained about 22% of the total cellular pool of PI(4,5)P₂, only about 10% of PI(4,5)P₂ was soluble in Triton X-100, whereas the majority of the lipid was recovered in the residual SDS-soluble fraction (Fig. 8A). Notably, upon stimulation of cells with 10 or 100 ng/ml LPS, the level of PI(4,5)P₂ rose significantly only in the DRM fraction, to about 2.1–3.0 times that in unstimulated cells, after 5–10 and 60 min of stimulation with LPS (Fig. 8A). By contrast, the content of PI(4,5)P₂ in the Triton-X-100-soluble fraction remained fairly stable, whereas that in the SDS-soluble fraction increased only by 24–29% (Fig. 8A).

The DRM fraction of unstimulated cells contained about 43% and 60% of PIP5K Iα and PIP5K Iγ, respectively, and also about 89% of the cellular pool of CD14 (Fig. 8B,C). About 27% of PIP5K Iα was present in the SDS-soluble fraction of cells, which contained the cytoskeleton judging from the presence of actin (Fig. 8B,C). Stimulation of cells with 10 ng/ml LPS for 5 min induced a reproducible moderate enrichment of PIP5K Iα in the DRM fraction at the expense of that in the Triton-X-100-soluble fraction (Fig. 8B,C). This redistribution of PIP5K Iα was transient as it was absent after 60 min of stimulation with 10 ng/ml LPS (Fig. 8B,C). It was not detected in cells that had been stimulated with 100 ng/ml LPS (Fig. S3A,B). The distribution of PIP5K Iγ between the fractions did not change substantially during stimulation of cells with LPS (Fig. 8B,C). We were unable to detect TLR4 in the DRM fraction, regardless of stimulation of the cells with 10 or 100 ng/ml LPS (Fig. 8B,C; Fig. S3A,B), nor following application of various antibodies for its detection (results not shown).

To further assess the role of rafts in the LPS-induced generation of PI(4,5)P₂, we exposed cells to 10 mM methyl-β-cyclodextrin (CDX). The drug reduced the cellular level of cholesterol by about 46±5%, and caused a concomitant redistribution of CD14 from the DRM to the Triton-X-100-soluble fraction and a reduction of the basal level of PI(4,5)P₂ in the DRM fraction by about 32% and by 17% in the SDS-soluble fraction (Fig. 8D). Notably, no increase of PI(4,5)P₂ was found in CDX-treated cells after their stimulation with 100 ng/ml LPS for 5 or 60 min (Fig. 8D).

The results of the cell fractionation analysis prompted us to study the localization of PIP5K Iα and PIP5K Iγ in the plane of the plasma membrane by using immunoelectron microscopy (Fig. 8E–H). Stimulation of cells with 10 ng/ml LPS induced an accumulation of PIP5K Iα in the vicinity of CD14 clusters that formed after 5 min of cell stimulation. No such colocalization was seen before cell stimulation (Fig. 8E,F). PIP5K Iγ tended to localize in the vicinity of small clusters of CD14 in LPS-stimulated cells and, to a lesser extent, also in the vicinity of non-clustered CD14 in unstimulated cells (Fig. 8G,H).

Taken together, biochemical and ultrastructural data indicate that, at the onset of stimulation of cells with LPS, PI(4,5)P₂ is generated by PIP5K isoforms Iα and Iγ in proximity of CD14 clusters.

The aim of this study was to reveal the mechanism governing the participation of PI(4,5)P₂ in TLR4 signaling that is triggered by LPS. We found that activation of J744 cells with LPS leads to a biphasic increase of the PI(4,5)P₂ level. The first peak of PI(4,5)P₂ generation could be detected after only 5–10 min of exposure of cells to LPS, when electron microscopy analysis indicated that CD14 underwent prominent clustering in the plane of the plasma membrane and that those CD14 clusters were accompanied by PI(4,5)P₂ and PIP5K isoforms Iα and Iγ. Several lines of data indicate that the LPS-induced PI(4,5)P₂ generation is governed by CD14 rather than TLR4 – (i) the elevation of PI(4,5)P₂ in J774 cells was inhibited by a function-blocking antibody against CD14; (ii) LPS induced biphasic PI(4,5)P₂ generation in HEK-CD14 transfectants that lacked TLR4, and this process was also inhibited by an antibody that interfered with CD14 function; (iii) crosslinking of CD14 with antibodies, without LPS or TLR4 participation, induced the generation of PI(4,5)P₂ in J774 cells and in HEK-CD14 transfectants. In both cell lines, the lipid colocalized with conglomerates of crosslinked CD14.

It has been reported previously that crosslinking of CD14 with antibodies induces an increase of intracellular Ca²⁺ in THP-1 cells (Pugin et al., 1998). A similar signaling effect is exerted by crosslinking of CD59 or other GPI-anchored proteins (Shenoy-Scaria et al., 1992; Morgan et al., 1993). More recently, single-molecule-tracking studies have revealed that antibody-induced clusters of CD59 recruit PLCγ, which cleaves PI(4,5)P₂ leading to the release of Ca²⁺ from intracellular stores (Suzuki et al., 2007). Those data reinforce our results and indicate that clustering of CD14 in the plasma membrane can initiate signal transduction events, such as PI(4,5)P₂ generation.

The signaling abilities of CD14 are most likely related to its location in plasma membrane rafts. This claim is supported by the results of cell fractionation analysis, which revealed that PI(4,5)P₂ produced in LPS-stimulated cells associated with the DRM fraction, accommodating the vast majority of CD14 and also PIP5K Iα and PIP5K Iγ. Notably, depletion of plasma membrane cholesterol with CDX abolished the LPS-induced accumulation of PI(4,5)P₂ in the DRM and induced a shift of CD14 from the DRM to the Triton X-100-soluble fraction. Such behavior is characteristic of components of sphingolipid- and cholesterol-based rafts, although the DRM fraction is not equivalent to native rafts of the plasma membrane (Kusumi et al., 2012; Plociennikowska et al., 2015). Accordingly, some reports have shown an association of LPS-activated TLR4 receptor with the DRM (Triantafilou et al., 2004a), whereas others, including our studies, fail to do so (Dhungana et al., 2009). This suggests that the association of TLR4 with rafts in the plasma membrane can be weak and lost upon solubilization of some cells with Triton X-100. Nevertheless, rafts are indicated as sites of the CD14–TLR4 interaction based on several methodological approaches, and many raft-anchored proteins collaborate with TLR4 in the recognition of LPS or endogenous ligands, and in signal transduction (Plociennikowska et al., 2015). Earlier studies have shown that the signaling events triggered by crosslinking of CD14, CD55 or CD59 are abolished when the GPI anchor of these proteins is replaced by a transmembrane protein sequence, suggesting that, specifically, the localization to rafts endows those proteins with signaling abilities (Shenoy-Scaria et al., 1992; Pugin et al., 1998; Suzuki et al., 2007). A direct link between the signaling activity of CD14 and its raft localization has been revealed in dendritic cells, where binding of LPS to CD14 leads to calcineurin-induced NFAT activation. The onset of this signaling is independent of TLR4 but requires raft integrity, which is most likely to support the interaction of CD14 and Src-family kinases (Zanoni et al., 2009). The clustering of rafts into larger signaling platforms has been described for CD59 bound to its ligand and is typical for receptors that associate with rafts (Suzuki et al., 2007, 2012). It is tempting to speculate that the clustering of CD14 is triggered by the LPS molecule itself through a fatty acid residue that lies outside of the LPS-binding pocket of CD14 and facilitates the subsequent association of CD14 with MD-2 in the TLR4–MD-2 complex (Kelley et al., 2013). An involvement of locally generated PI(4,5)P₂ in the feedback loop to control the dynamics of CD14-bearing rafts through interaction with the submembraneous actin cytoskeleton (Chaudhuri et al., 2011; Kusumi et al., 2012; Zhang et al., 2012) is an intriguing possibility. Notably, Pam₂CSK₄ failed to induce substantial PI(4,5)P₂ generation in HEK-CD14 cells. During binding to TLR2–TLR6, the bridging function of Pam₂CSK₄ needs to be enhanced by the strong intermolecular bonding of the receptors (Kang et al., 2009), which suggests that the CD14-clustering potential of Pam₂CSK₄ is also rather poor. The mechanism of the second phase of PI(4,5)P₂ generation observed here in LPS-stimulated cells is unknown but can be linked with PI(4,5)P₂ replenishment, which has been described in hormone-stimulated cells (Balla et al., 2008).

The rapid LPS-induced clustering of CD14 in the plasma membrane of J774 cells was accompanied by an overall increase of its content, followed by loss of CD14 from the plasma membrane. Although internalization of CD14–LPS is widely accepted (Zanoni et al., 2011), few reports to date indicate a rapid recruitment of a cytoplasmic pool of CD14 to the plasma membrane shortly after exposure of cells to LPS (Antal-Szalmas et al., 2000; Vasselon et al., 1999). The fairly high increase of CD14 in the plasma membrane of J774 cells found in our electron microscopy studies suggests that the dorsal surface of cells examined in these studies is the preferred site of CD14 recruitment. Notably, stimulation of cells with LPS did not affect the distribution of GM1 in the plasma membrane, indicating that LPS triggers clustering of CD14-bearing rafts rather than an overall raft rearrangement; this also argues against the possibility that the clustering of CD14 was induced during immunolabeling of LPS-stimulated cells.

It is well established that PI(4,5)P₂ is involved in the TIRAP–MyD88-dependent signaling pathway of TLR4 (Kagan and Medzhitov, 2006). Our data are in agreement with those findings, but they also underscore the role of PI(4,5)P₂ in the TRAM–TRIF-dependent pathway of TLR4. The two peaks of PI(4,5)P₂ generation coincided with two peaks of NF-κB activation found in LPS-stimulated J774 cells, which is likely to reflect the MyD88- and TRIF-dependent signaling cascades of TLR4 (Kawai and Akira, 2011). The interference of PI(4,5)P₂ generation resulting from siRNA or drug application inhibited both peaks of NF-κB activation and affected production of TNF-α and, to an even greater degree, the TRAM–TRIF-dependent production of RANTES. The inhibition of the latter could follow the inhibition of TRAM–MyD88; however, in the TRAM–TRIF-dependent pathway of TLR4, PI(4,5)P₂ can serve as a source of PI(3,4,5)P₃ and inositol trisphosphate, governing the release of intracellular Ca²⁺, both of which are required for the uptake of activated TLR4 (Zanoni et al., 2011; Aksoy et al., 2012; Chiang et al., 2012). Another possibility is a direct interaction of PI(4,5)P₂ with the proteins that mediate the uptake of LPS-activated TLR4 through macropinocytosis (Jiang et al., 2005; Plociennikowska et al., 2015) or clathrin-mediated endocytosis (Kitchens et al., 1998; Husebye et al., 2006; Wang et al., 2012; Czerkies et al., 2013). Distinct isoforms of PIP5K type I have been identified as controlling clathrin-mediated endocytosis in various cell types, and several components of the clathrin coat, including AP-2, are recruited to clathrin-coated pits through binding to PI(4,5)P₂ (Kwiatkowska, 2010). PI(4,5)P₂ was seen here at vesicles that had formed following prolonged stimulation of cells with LPS (Fig. 3F).

In conclusion, our data underscore the signaling role of CD14 in LPS-stimulated cells. Binding of LPS to CD14 induces rapid clustering of the protein, which triggers local generation of PI(4,5)P₂ by the PIP5K isoforms Iα and Iγ, which is required for the maximal production of pro-inflammatory mediators in both the MyD88- and TRIF-dependent signaling pathways of TLR4. How the clustering of CD14 leads to generation of PI(4,5)P₂ through type-I PIP5K isoforms remains unknown, but it is likely that it involves the raft-based compartmentalization of signaling molecules.

J774A.1 (J774) macrophage-like and HEK293 cells were cultured in Dulbecco's modified Eagle's medium (DMEM) with 10% fetal bovine serum (FBS). For stimulation, the medium was exchanged for 2 h and supplemented with 10–1000 ng/ml LPS (ultrapure smooth LPS from E. coli O111:B4, List Biological Laboratories, Campbell, CA). Stimulation of cells was also performed with 100 ng/ml Pam₂CSK₄ (InvivoGen, Toulouse, France). In some experiments, cells were pre-incubated with 10 µg/ml of rat IgG2b against CD14 (clone 4C1), or rat IgG2b (BD Biosciences) (30 min, 37°C), or 0.1 mM BPA (1 h, 37°C in DMEM with 2% FBS) in BSA or 0.07% BSA, prepared as described previously (Kwiatkowska et al., 2003), or 10 mM CDX (1 h, 37°C in DMEM with 2% FBS) or 10 μM ionomycin (10 min, 37°C in DMEM with 10% FBS and 1.5 mM CaCl₂) according to Varnai and Balla (1998). The level of cellular cholesterol was estimated, as described previously (Kwiatkowska et al., 2014). When indicated, cells were cultured for 18 h in DMEM with 2% FBS containing 0.1–10 mM LiCl (Sigma-Aldrich) or 2 mM neomycin (AppliChem, Darmstadt, Germany). The antibodies or drugs were present during subsequent stimulation of cells with LPS.

To obtain HEK293 cells expressing CD14, cells (1×10⁵/well in 12-well plate) were exposed to 0.3 µg of pUNO-mCD14 (InvivoGen) and 1 µl of FuGENE (Promega, Madison, WI) in 1 ml of DMEM with 10% FBS for 24 h before experiments.

For siRNA transfections, J774 cells (4×10⁵) were suspended in 800 μl of RPMI-1640 with 5% FBS and mixed with 800 μl of serum-free RPMI-1640 containing 160 pmol of siRNA (Qiagen) and 16 μl TrueFect-Lipo (United BioSystems, Herndon, VA). Cells were seeded (1×10⁵/well in 48-well plates), after 6 h, the medium was changed to DMEM with 10% FBS, and cells were further cultured for 18 h.

J774 (5×10⁵/well) or HEK-CD14 (1×10⁵/well) cells in 12-well plates were washed with ice-cold PD buffer (125 mM NaCl, 4 mM KCl, 10 mM NaHCO₃, 1 mM KH₂PO₄, 10 mM glucose, 20 mM Hepes, 1 mM MgCl₂, 1 mM CaCl₂, pH 7.4) and incubated with 1.6 μg/ml of rat IgG2a against CD14, clone Sa14-2 (BioLegend) or control rat IgG2a in serum-free DMEM (30 min, 4°C). After washing with ice-cold PD buffer, cells were either exposed to the F(ab′)₂ fragment of donkey anti-rat IgG (Jackson ImmunoResearch) to crosslink CD14 or with DMEM alone for 5–10 min at 37°C. Crosslinking of TLR4 in J774 cells was induced using 2 μg/ml mouse anti-TLR4 IgG2b clone 76B357.1 (Imgenex) followed by the F(ab′)₂ fragment of goat anti-mouse IgG (Jackson ImmunoResearch). To crosslink GM1 ganglioside, J774 cells were exposed to 5 μg/ml CTX–FITC (30 min, 4°C) followed by rabbit anti-CTX serum (both obtained from Sigma-Aldrich) for 10 min at 37°C. The anti-CTX serum was passed through a PD-10 desalting column to remove NaN₃.

Cells were seeded onto coverslips (3×10⁴ per 15×15 mm coverslip) and fixed with 3% paraformaldehyde, either before or after labeling of CD14 with rat anti-CD14 IgG2a clone Sa14-2 (30 min, 4°C) and the F(ab′)₂ fragment of donkey anti-rat IgG conjugated to Alexa Fluor 488 (5–10 min, 37°C). After fixing, the cells were incubated with TBS containing 10% heat-inactivated mouse serum (30 min, 20°C) and permeabilized with 0.05% digitonin in TBS (10 min, 20°C). After washing and blocking (3% gelatin and 1% polyvinylpirrolidone with 2% BSA), cells were processed to visualize PI(4,5)P₂ through application of 2 μg/ml PLC-PH–GST followed by rabbit anti-GST IgG (Sigma) and donkey anti-rabbit IgG–TRITC (Santa Cruz Biotechnology) (Szymanska et al., 2009). Samples were examined under a Leica confocal microscope (TCS SP8 SMD) and analyzed, as described previously (Kwiatkowska et al., 2014).

J774 cells were fractionated, as described previously (Brown, 2002). Briefly, cells (1×10⁶/sample) were lysed for 30 min at 4°C with 70 μl of 0.05% Triton X-100 in buffer A (100 mM NaCl, 2 mM EDTA, 2 mM EDTA, 30 mM Hepes, pH 7.4, 10 μg/ml leupeptin, 1 μg/ml pepstatin A, 10 μg/ml aprotinin, 1 mM PMSF, 1 mM Na₃VO₄, 10 mM p-nitrophenyl phosphate). Following centrifugation (5 min, 10,000 g), supernatants were collected, whereas pellets were washed once with ice-cold PBS buffer, resuspended in 70 μl of 1.8% octyl-β-D-glucoside in buffer A (30 min, 4°C) and centrifuged as above. Supernatants were collected, whereas remaining pellets were resuspended in 70 μl of 4% SDS in 150 mM NaCl, 50 mM triethylamine, 2 mM EDTA, 2 mM EDTA, 250 units of benzonase, and protease and phosphatase inhibitors.

Two approaches were used to estimate the PI(4,5)P₂ levels in cells – either a protein–lipid overlay assay or an ELISA test. Both required extraction of phosphoinositides from cells (5×10⁵ J774 cells or 1×10⁵ HEK-CD14 cells per sample) or cell fractions (100 µl). The procedure was performed according to Gray et al. (2003), omitting the neutral lipid extraction. Our analysis showed that the presence of neutral lipids in samples did not affect the final estimation of the PI(4,5)P₂ level.

For detection of PI(4,5)P₂ using the protein–lipid overlay assay, the extracted lipids were resuspended through sonication in 60 μl of 50 mM NaCl, 5 mM DTT, 0.05% CHAPS, 1.5% sodium cholate, 50 mM Hepes, pH 7.4 (Gray et al., 2003). Aliquots of 2 μl were spotted onto a nitrocellulose membrane, which was blocked with 1% polyvinylpirrolidone and 0.1% gelatin and incubated with 0.1 μg/ml of PH-PLC–GST and chicken anti-GST Iggy conjugated with peroxidase (Gallus Immunotech, Cary, NC). Immunoreactive spots were detected by using chemiluminescence, and densitometry analysis was performed using ImageJ. In a series of experiments, the PH-PLC–GST probe was pre-incubated twice with liposomes that either contained dioleoyl phosphatidylcholine and PI(4,5)P₂ (0.05:0.95 molar ratio; Echelon) or did not contain PI(4,5)P₂, as described by Gao et al. (2009).

A competitive 96-well ELISA assay was performed to estimate the amount of PI(4,5)P₂ in samples, according to the manufacturer's instructions (Echelon, Salt Lake City, UT). To calculate the amount of PI(4,5)P₂ as pmol/mg of protein, the protein content in a corresponding number of cells was determined using the Bradford ULTRA kit (Novexin, Babraham, UK).

Cell lysates (10 μg of protein per lane) or equal volumes of cell fractions (10 μl) were subjected to 10% SDS-PAGE analysis. Separated proteins were transferred onto nitrocellulose and immunoblotted with goat antibody against PIP5K Iα (Santa Cruz Biotechnology), rabbit antibody against PIP5K Iγ (Cell Signaling, Leiden, The Netherlands), rat antibody against CD14, mouse antibody against β-actin (BD Biosciences) or rabbit antibody against TLR4 (Santa Cruz Biotechnology), followed by anti-goat, anti-rabbit or anti-mouse IgG conjugated with peroxidase (Sigma-Aldrich and Jackson ImmunoResearch). Immunoreactive bands were visualized and analyzed as described above.

Cytokine levels were determined in supernatants of J774 cells (5×10⁵/sample) with the use of murine ELISA kits according to the manufacturer's instructions (BioLegend, R&D Systems, Abingdon, UK).

Cells were seeded onto coverslips, fixed with 3% paraformaldehyde and incubated with a blocking solution of 10% heat-inactivated mouse serum (30 min, 20°C). To label CD14, cells were incubated with rat anti-CD14 IgG2a clone Sa14-2 (1.5 h, 20°C) followed by goat anti-rat IgG conjugated to 10-nm gold (Aurion, Wageningen, The Netherlands; 2 h, 20°C). Alternatively, rabbit anti-CD14 IgG (Santa Cruz Biotechnology) and donkey anti-rabbit IgG conjugated to 10-nm gold (Aurion) were applied to allow subsequent labeling of PIP5K Iα. In another series of experiments, cells were exposed to 2 μg/ml CTX–FITC, followed by rabbit anti-CTX and the F(ab′)₂ fragment of goat anti-rabbit IgG conjugated with 6-nm gold particles (Aurion). After surface labeling, sheets of the dorsal portion of plasma membrane were obtained through mechanical cleavage of the cells, according to Sanan and Anderson (1991) and Strzelecka-Kiliszek et al. (2004). Samples were incubated overnight in 10% heat-inactivated mouse serum and 50 mM l-lysine in TBS, followed by either 0.5 μg/ml PLC-PH–GST (1.5 h, 20°C) or goat anti-PIP5K Iα IgG or rabbit anti-PIP5K Iγ IgG. To visualize PI(4,5)P₂, rabbit anti-GST IgG and the F(ab′)₂ fragment of goat anti-rabbit IgG conjugated to 6-nm gold (2 h, 20°C) were applied, and PIP5K Iα was labeled with donkey anti-goat IgG conjugated to biotin (Rockland, Limerick, PA) and streptavidin-conjugated 6-nm gold (Aurion). PIP5K Iγ was labeled with the use of sheep anti-rabbit IgG conjugated to biotin (Rockland) and streptavidin conjugated to 6-nm gold. The concentrations of secondary antibodies were adjusted to exclude non-specific staining of control samples, in which incubation with primary antibodies was omitted. Samples were postfixed, counterstained (Szymanska et al., 2009) and examined under a JEM 1400 (Jeol) electron microscope.

Gold particles marking CD14, PI(4,5)P₂ and GM1 were scored from 20–25 micrographs corresponding to a 100-µm² area of the plasma membrane surface and classified as singlets or doublets, or clusters of 3–5, 6–10, 11–15 and ≥15 particles. The particles were considered to form clusters or colocalize if located within a radius ≤10 nm. For studies, translucent membrane areas of low electron density were used. The analyses were performed on data from three or four independent experiments.

Total RNA was extracted with TRI reagent (Sigma-Aldrich), and cDNA was synthesized using High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems). RT-qPCR was performed in a StepOnePlus instrument using fast SYBR Green Master Mix (Applied Biosystems). The following primers were used: sense 5′-CGGTTCCAGCGTTTCATGTG-3′ and antisense 5′-TGATCGCCGAGAGAAAGACG-3′ for PIP5K Iα; sense 5′-CGGCGAACACTATCCACAT-3′ and antisense 5′-TCAGGTAGATGGTCCCCTCT-3′ for PIP5K Iβ; sense 5′-CACGCTTGAGGATGAAGGC-3′and antisense 5′-GGCTGCTCCGATGTATCTGA-3′for PIP5K Iγ; sense 5′-CAGTCCCAGCGTCGTGA-3′ and antisense 5′-GCCTCCCATCTCCTTCAT-3′ for HPRT. The relative mRNA expression levels for the investigated genes (in comparison to the mRNA level of HPRT) were calculated by using the ΔΔC_T method.

The significance of differences between groups was calculated by using Student's t-test or two-way ANOVA without interactions, when indicated. P-values ≤0.05 were considered to be statistically significant.

We are grateful to Professor Andrzej Sobota for critical discussion of the results. ANOVA analysis was performed with the help of the Laboratory of Bioinformatics of the Nencki Institute, microscopic analyses with the help of the Laboratory of Electron Microscopy and the Laboratory of Imaging Tissue Structure and Function.