PtdIns5P was discovered in 1997 [Rameh, Tolias, Duckworth and Cantley (1997) Nature 390, 192–196], but still very little is known about its regulation and function. Hitherto, studies of PtdIns5P regulation have been hindered by the inability to measure cellular PtdIns5P using conventional HPLC, owing to poor separation from PtdIns4P. In the present paper we describe a new HPLC method for resolving PtdIns5P from PtdIns4P, which makes possible accurate measurements of basal and inducible levels of cellular PtdIns5P in the context of other phosphoinositides. Using this new method, we found that PtdIns5P is constitutively present in all cells examined (epithelial cells, fibroblasts and myoblasts, among others) at levels typically 1–2% of PtdIns4P levels. In the β-pancreatic cell line BTC6, which is specialized in insulin secretion, PtdIns5P levels were higher than in most cells (2.5–4% of PtdIns4P). Using subcellular fractionation, we found that the majority of the basal PtdIns5P is present in the plasma membrane, but it is also enriched in intracellular membrane compartments, especially in SER (smooth endoplasmic reticulum) and/or Golgi, where high levels of PtdIns3P were also detected. Unlike PtdIns3P, PtdIns5P was also found in fractions containing very-low-density vesicles. Knockdown of PIP4K (PtdIns5P 4-kinase) leads to accumulation of PtdIns5P in light fractions and fractions enriched in SER/Golgi, whereas treatment with Brefeldin A results in a subtle, but reproducible, change in PtdIns5P distribution. These results indicate that basal PtdIns5P and the PtdIns5P pathway for PtdIns(4,5)P2 synthesis may play a role in Golgi-mediated vesicle trafficking.
PIs (phosphoinositides) have long been known to participate in basal cellular functions, such as vesicle transport and cytoskeleton dynamics, as well as responses triggered by extracellular cues including proliferation, differentiation and chemotaxis . Whereas PtdIns4P and PtdIns(4,5)P2 are abundant in cells, the PI3K (phosphoinositide 3-kinase) lipid products PtdIns(3,4)P2 and PtdIns(3,4,5)P3 are virtually absent from quiescent cells, but can be rapidly stimulated by extracellular factors. PtdIns3P, on the other hand, is constitutively present in eukaryotic cells, but is also regulated in specific subcellular locations.
PtdIns5P was the last member of the PI family to be discovered and very little is known about its regulation and function . Like PI3K lipid products, PtdIns5P levels are low in abundance, but can be up-regulated by extracellular stimuli. PtdIns5P levels increase in response to stress signals , insulin  or TCR (T-cell receptor) stimulation , after thrombin-stimulated platelet aggregation  or during cell-cycle progression .
Cellular PtdIns5P was also shown to increase during bacterial invasion owing to the catalytic activity of the virulence factors IpgD from Shigella flexneri  or SigD/SopB from Salmonella spp. , indicating that PtdIns5P may play a role in membrane and cytoskeleton events that facilitate pathogen invasion. Two new phosphatases capable of generating PtdIns5P have been identified; namely PtdIns(4,5)P2 4-phosphatase types I and II, which are mammalian analogues of IpgD that can generate PtdIns5P from the dephosphorylation of PtdIns(4,5)P2in vitro . PtdIns5P levels are negatively regulated by PIP4K (PtdIns5P 4-kinase; also known as PIPK type II), which are a family of 4-kinases that specifically use PtdIns5P as a substrate to generate PtdIns(4,5)P2 . PLIP [PTEN (phosphatase and tensin homologue deleted on chromosome 10)-like PI 5-phosphatase] is also a potential negative regulator of cellular PtdIns5P [12,13].
Despite the identification of several enzymes involved in the regulation of PtdIns5P, many questions remain regarding the synthesis and degradation of basal and inducible cellular PtdIns5P. It is not yet clear whether PtdIns5P can only be generated by phosphatases or whether a PtdIns-specific 5-kinase exists. The role of different PIP4K isoforms on the regulation of basal or stimulated PtdIns5P is also unclear. PIP4K type IIβ, for instance, is present in the nucleus and is phosphorylated and inactivated in response to stress signals, leading to an increase in nuclear PtdIns5P [3,14–17]. This isoform interacts with the EGF (epidermal growth factor) and TNFα (tumour necrosis factor α) receptors [18,19], and modulates early insulin responses , suggesting that PtdIns5P is also present at the plasma membrane. In addition, the type IIα isoform translocates to the cytoskeleton in response to platelet aggregation . Based on this evidence, many have suggested that different enzymes or cues regulate distinct subcellular pools of PtdIns5P . However, the subcellular distribution of this lipid has never been fully examined.
PtdIns5P studies have been hindered by the inability to measure PtdIns5P levels using conventional HPLC, owing to poor separation from PtdIns4P. For this reason, most studies thus far have used an enzymatic assay based on the ability of PIP4K to use PtdIns5P as a substrate . This approach, however, does not allow for measurements of PtdIns5P in the context of the other cellular PIs and is susceptible to interference by PIP4K inhibitors in the assay, such as its own product PtdIns(4,5)P2.
In the present study we describe a new HPLC method to measure cellular PtdIns5P levels in the context of the other PIs. This allows sensitive and accurate detection of basal PtdIns5P levels and changes in response to extracellular factors. Using this method, we found that all cells examined thus far have detectable basal levels of PtdIns5P, which are typically around 1% of that of PtdIns4P. The insulinoma cell line BTC6 had higher levels of PtdIns5P than other cells. Using cellular fractionation combined with HPLC measurements of PIs, we defined the subcellular localization of basal PtdIns5P in HeLa and BTC6 cells, which was previously impossible due to the lack of PtdIns5P-specific probes. This approach revealed that the majority of PtdIns5P resides in various intracellular vesicles and plasma membrane, but are particularly enriched in light microsomal and SER [smooth ER (endoplasmic reticulum)]/Golgi-containing fractions. PtdIns3P was also found to be specifically concentrated in SER/Golgi-enriched fractions, but in contrast with PtdIns5P, it was completely absent from light microsomal fractions. Knockdown of PIP4Ks resulted in accumulation of PtdIns5P in the Golgi-enriched fractions and Brefeldin A treatment resulted in the redistribution of PtdIns5P, indicating that PtdIns5P may play a role in Golgi-mediated intracellular trafficking.
MATERIALS AND METHODS
Cell lines, maintenance and manipulations
HeLa and BTC6 cells (A.T.C.C.) were cultured in DMEM (Dulbecco's modified Eagle's medium) supplemented with 10% FBS (fetal bovine serum; Gibco). Retroviruses carrying the pSuper.retro.puro shRNA (small-hairpin RNA) vectors (OligoEngine) were generated by transiently transfecting HEK (human embryonic kidney)-293T cells using Lipofectamine™ Plus (Invitrogen). Stable knockdown cells were generated by infection with pSuper.retro.puro-derived retrovirus carrying the target sequence for PIP4K IIα, PIP4K IIβ, PIP4K IIγ or a control sequence, which consisted of the target sequences with four to six bases mismatched (C1). Infected cells were selected by puromycin treatment.
Metabolic labelling of PIs
Cells were metabolically labelled with 200 μCi/ml [32P]Pi for 1.5–4 h in phosphate-free DMEM or with 10 μCi/ml [3H]inositol for 24–72 h in inositol-free DMEM supplemented with dialysed FBS and 200 mM L-glutamine.
Cells plated in 150-mm-diameter tissue culture dishes and labelled or not for the times indicated were washed in PBS, rinsed with cytosol buffer [0.2 M sucrose, 25 mM Hepes (pH 7), 125 mM potassium acetate, 1 mM dithiothreitol, 1 mM sodium orthovanadate, 2 mg/ml sodium fluoride, 2 mg/ml 2-glycerophosphate, 1 mM phenanthroline, 1 mM benzamidine and protease inhibitor cocktail from Sigma) and scraped from the dish. The cell suspension was passed through a 29 1/2 gauge needle 12 times and centrifuged at 100 g for 10 min. The supernatant (microsomal fraction) was loaded on top of a discontinuous sucrose gradient (20–65% sucrose in cytosol buffer) and centrifuged for 4.5 h at 55000 rev./min in a TL100 centrifuge, using a TLS55 rotor (Beckman). Six fractions were collected from the top of the gradient and designated microsomal fractions 1 (lightest) through to 6 (densest). The pellet containing unbroken cells, nuclei and associated membranes was resuspended in cytosol buffer, frozen and thawed three times to break any remaining intact cells, and centrifuged at 100 g for 10 min. The supernatant was designated fraction X. The pellet was resuspended in cytosol buffer containing 1% Triton X-100 and centrifuged at 16000 g for 10 min to separate the pellet containing the nuclear fraction from the supernatant containing the Triton-soluble membranes. For lipid analysis, 2 M HCl was added to each fraction to a final concentration of 1 M and the lipids were extracted with 1:1 (v/v) methanol/chloroform. For protein analysis, each fraction was mixed with SDS-containing loading buffer and analysed by Western blotting.
HPLC method for PI analysis
Cellular PIs were extracted and deacylated as described previously . Deacylated lipids were separated by anion-exchange HPLC (Agilent 1200) using two partisphere SAX columns (Whatman) in tandem and a four-step gradient of ammonium phosphate, pH 6.0 (10–40 mM over 60 min; 40–150 mM over 5 min; 150 mM isocratic for 20 min and 150–650 mM over 25 min). Radiolabelled eluate was detected by an online flow scintillation analyser (PerkinElmer) and quantified using ProFSA software (PerkinElmer).
Western blot analysis
Protein lysates were prepared from either intact cells or subcellular fractions. Total protein lysates were obtained by lysis in buffer containing 50 mM Tris/HCl (pH 7.5), 100 mM NaCl, 1 mM EDTA, 1% Triton X-100 and 10% glycerol as well as protease and phosphatase inhibitors. Lysates were centrifuged at 14000 g to remove the insoluble fraction and were normalized based on total protein content measured using the Bradford assay (Bio-Rad). Total cellular protein or protein from various subcellular fractions were mixed with SDS-loading buffer, boiled for 5 min and separated by SDS/PAGE (10% gels). The proteins were transferred on to a nitrocellulose membrane, which was blocked with 5% (w/v) non-fat dried skimmed milk dissolved in TBS (Tris-buffered saline; 20 mM Tris/HCl, pH 7.5, and 137.5 mM NaCl) plus 1 mM sodium orthovanadate. Membranes were probed overnight with the appropriate primary antibody. After washing, membranes were incubated for 60 min with secondary antibodies conjugated to IR680 (Rockland and Molecular Probes) or IR800 (Rockland). The membranes were washed in TBS-Tween (TBS plus 0.1% Tween 20) and bound antibodies were detected and quantified using the Odyssey Infrared Imaging System (LI-COR). Antibodies against β1-integrin, GM130 (cis-Golgi matrix protein of 130 kDa), ATP synthase, caveolin and tubulin were from BD Transduction Laboratories. Antibodies against S6 ribosomal protein, calnexin, phospho-histone and EEA1 (early endosome antigen 1) were from Cell Signaling Technology. Antibodies against lamin A/C and ERK (extracellular-signal-regulated kinase) were from Santa Cruz Biotechnology, against βCOP was from ABR, and against 58K Golgi was from Abcam. Antibodies against Rab5 were from BD Transduction Laboratories and from Cell Signaling Technology. An antibody against PIP4K IIβ was a gift from Dr M. Chao (Department of Cell Biology, New York University School of Medicine, New York, U.S.A.) and an antibody against PIP4K IIγ was generated in rabbits by injection of a GST (glutathione transferase)–PIP4K IIγ N-terminal fusion protein (Pocono Rabbit Faru and Laboratories).
Separation and detection of cellular PtdIns5P using a novel HPLC method
In order to study PtdIns5P levels in cells, we developed a new HPLC method for separation of PtdIns5P from cellular PtdIns4P. Deacylated PIs from [3H]inositol-labelled HeLa cells expressing the bacterial phosphatase IpgD were used to test various protocols, since they have elevated levels of PtdIns5P. Using a conventional anion-exchange column and shallow ammonium phosphate gradient for the elution of PIs, the PtdIns5P peak was eluted 0.5–1.0 min after the PtdIns4P peak, which was not sufficient for a complete separation of the PtdIns5P peak from the declining base of the much larger cellular PtdIns4P peak. Since decreasing the slope of the gradient had little or no effect on the separation of these two isomers, we decided to test whether increasing the bed volume of the column would improve separation. For this purpose, we attached two 250 mm Whatman Partisphere SAX columns in tandem, which improved the separation of PtdIns5P from PtdIns4P by 1 min. We tested various pHs for ammonium phosphate and determined pH 6.0 to be the optimal pH for separation. The combination of the larger bed volume and higher pH resulted in a 4 min separation between PtdIns4P and PtdIns5P, and complete separation of the bases of the two peaks in cells expressing IpgD (Figure 1A). Using this new method we were also able to detect basal levels of PtdIns5P in control HeLa cells, without interference from the PtdIns4P peak (Figure 1B).
PtdIns5P measurements using a novel HPLC method
To validate this new HPLC method, we measured the levels of PtdIns5P in cells treated with insulin or hydrogen peroxide, which increase PtdIns5P levels as measured by enzymatic assay [4,24]. In CHO (Chinese-hamster ovary)-IR cells labelled with 32P, insulin induced PtdIns(3,4,5)P3 and PtdIns(3,4)P2 (results not shown) as expected, and also increased PtdIns5P levels 2.5-fold (Figure 1C) without affecting the levels of PtdIns4P or PtdIns(4,5)P2 (results not shown). In HeLa cells labelled with [3H]inositol, basal PtdIns5P levels were approx. 1% of the levels of PtdIns4P (Figure 1B and Table 1) and approx. 30% of the levels of PtdIns3P. Hydrogen peroxide treatment of HeLa cells significantly increased PtdIns5P levels and, to a lesser extent, PtdIns4P and PtdIns(4,5)P2 levels (Figure 1D). PtdIns5P levels in hydrogen-peroxide-treated cells were equal to or higher than PtdIns3P levels. After normalization with PtdIns4P, hydrogen peroxide treatment increased PtdIns5P 2.5-fold (Figure 1D, inset). These results are comparable with the results obtained by other groups using the enzymatic assay for PtdIns5P detection [4,24] and demonstrate that our new HPLC method allows us to consistently and accurately measure small levels of cellular PtdIns5P in the context of other PIs.
|Cell line||FBS in growth medium (%)||Labeling time (h)||PtdIns5P (% of PtdIns4P±S.E.M.)|
|Cell line||FBS in growth medium (%)||Labeling time (h)||PtdIns5P (% of PtdIns4P±S.E.M.)|
Basal PtdIns5P levels are present in various cell lines
Using this new HPLC method, we compared the levels of basal PtdIns5P in various cell lines labelled with [3H]inositol. As PtdIns4P is one of the most abundant PIs and is constitutively present in all cells, we chose to normalize the data using PtdIns4P, in order to compare the levels of PtdIns5P between cell lines with different rates of PI metabolism. Table 1 shows that HeLa (human cervical cancer), CHO, NIH-3T3 (mouse fibroblast) and 3T3-L6 (rat myoblast) cells have similar levels of PtdIns5P (approx. 1% of PtdIns4P), whether they were labelled in high- or low-serum medium. In fact, from a panel of more than 20 cell lines including cancer cells and normal cell lines, PtdIns5P was detected in all cells examined at levels ranging from 0.5 to 2.0% of PtdIns4P (Table 1 and D. Sarkes, K. Vasudevan, L. Garraway and L. E. Rameh, unpublished work). Therefore, unlike the PI3K lipid products PtdIns(3,4)P2 and PtdIns(3,4,5)P3, which are absent from or below detection levels in most cells, PtdIns5P is constitutively present in all cells examined. Notably, the mouse insulinoma cell line BTC6 had higher levels of PtdIns5P than other cell lines tested (2.5–4% of PtdIns4P). In these cells, the level of incorporation of [3H]inositol into PtdIns was similar to the level in HeLa cells, but the levels of [3H]inositol incorporation into PtdIns4P and PtdIns(4,5)P2 were significantly lower, probably owing to a slower PI turnover. For example, after 48 h labelling, the levels of [3H]PtdIns4P in BTC6 cells were half that of the levels in HeLa cells, and the levels of [3H]PtdIns(4,5)P2 were 5–10-fold lower, whereas [3H]PtdIns(3,5)P2 was completely undetectable. Notwithstanding, the levels of [3H]PtdIns5P in BTC6 cells were higher than in HeLa cells and did not correlate with the levels of [3H]inositol labelling of PtdIns(4,5)P2. This raises the possibility that basal PtdIns5P is generated from phosphorylation of PtdIns rather than dephosphorylation of PtdIns(4,5)P2 or PtdIns(3,5)P2 as previously suggested .
Subcellular distribution of PIs and protein markers
In order to investigate the distribution of PIs within the cell, we used this new HPLC method combined with biochemical fractionation of cellular organelles. HeLa and BTC6 cells were labelled with [3H]inositol for 48–72 h, and cell lysates were then fractionated using differential centrifugation to generate a low-spin pellet fraction (P1) and a supernatant fraction (S1). The P1 fraction was further fractionated into unbroken cells, nuclei and membranes associated with nuclei, whereas S1 was further fractionated by a sucrose density gradient to generate six distinct fractions, as described in the Materials and methods section. The protocol was designed to minimize loss of material, volume of samples and time between lysis and lipid extraction. From each cellular lysate, nine fractions of equal volume were generated and analysed for either PIs or protein content. As no portion of the lysate was wasted (there are no washes involved, for example) we were able to calculate the total content of each lysate by adding up the contents from each fraction.
Figure 2(A) shows that in BTC6 cells 72% of the cellular PIs (closed bars) were present in the supernatant (S1), where cytosol as well as microsomes are expected to be present, and 28% in the low-spin pellet (P1). The majority of the PIs present in the low-spin pellet were Triton-soluble (23 out of 28) and only 2% were Triton-insoluble. In contrast, over 60% of the PIs from HeLa cells (Figure 2B) were in the Triton-soluble P1 fraction, 9% were Triton-insoluble and only 16% were present in the supernatant (S1 fraction).
Subcellular distribution of protein markers and lipids after differential centrifugation
The Triton-soluble P1 fraction was rich in β1-integrin (open bars), an integral membrane protein, and in calnexin (results not shown), an ER protein. The nuclear proteins lamin A/C and phospho-histone (grey bars) were most abundant in the Triton-insoluble P1 fraction, as expected. These results confirm that the Triton-soluble P1 fraction contains ER membranes associated with the nucleus, whereas the Triton-insoluble fraction contains the nucleus itself. As expected, the cytosolic protein ERK (hatched bars) and the membrane protein β1-integrin (open bars) were found in the S1 fraction, confirming that S1 contains the microsomal and cytosolic portions. The contrasting distribution of PIs between S1 and P1 fractions in HeLa and BTC6 cells indicates that the nucleus-associated ER network is much more prominent in HeLa cells than in BTC6 cells.
Figure 3 shows the molecular characterization of the six fractions obtained after sucrose density gradient separation of the S1 fractions from HeLa (Figures 3A–3C) and BTC6 cells (Figure 3D). Measurements of the sucrose density and distribution of total PIs along the gradient are shown in Figure 3(A). Fraction 1 of the gradient had very few proteins, whereas fraction 2 contained the bulk of the proteins (based on Ponceau staining, results not shown) and was enriched in cytosolic proteins, such as ERK (Figures 3B and 3D). Fraction 3 of the gradient contained Golgi and SER markers, which are relatively light organelles, whereas fractions 4, 5 and 6 contained mitochondria and RER (rough ER) markers, which are denser organelles (Figures 3B–3D). Plasma membrane markers were found to localize sharply in fraction 4 (Figures 3B and 3D). This distribution is consistent with the expected density of each of these organelles in sucrose . Remarkably, in BTC6 cells where the S1 fraction contained visible amounts of microsomes (based on the turbidity of the solution), discrete opaque bands were seen in fractions 2, 3 and 4. In summary, this characterization shows that our approach allows us to generate subcellular fractions enriched in plasma membrane or in organelles that are either lighter than plasma membrane (SER and Golgi) or heavier than plasma membrane (RER and mitochondria). The small error bars highlight the reproducibility of these experiments (note also the similarities between the two cell lines).
Distribution of protein markers and lipids after fractionation of the cytosolic/microsomal fraction through a 20–65% sucrose gradient
Overall PI distribution after differential centrifugation
Table 2 shows the distribution of each PI into P1-derived fractions and S1. In BTC6 cells, the microsomal fraction (S1) contained the majority of all PIs, whereas in HeLa cells, the nuclear-associated ER fraction (Triton-soluble P1) contained the majority of all PIs. Although each PI followed the same pattern, a higher portion of PtdIns, as compared with other phosphorylated PIs [especially PtdIns(4,5)P2], was found at the nuclear-associated ER fraction. This is consistent with the fact that PtdIns is primarily synthesized at the ER, where it can be phosphorylated into other PI forms or transported to other membrane compartments.
From the sum of the absolute counts from each individual PI peak in the various fractions (results not shown), we were able to determine that the ratio between all PIs were within the expected range, indicating that very little loss, if any, occurred due to post-lytic dephosphorylation. For example, in three independent experiments where BTC6 cells were fractionated, the total counts for the PtdIns5P peak in each experiment were 8.7×103, 2.9×103 and 7.9×103, and for PtdIns4P they were 3.6×105, 1.1×105 and 2.3×105 respectively. In HeLa cells, the total counts for the PtdIns5P peak in three independent experiments were 7.6×103, 11.7×103 and 13.5×103, and for PtdIns4P they were 4.1×105, 13.2×105 and 9.2×105 respectively. Therefore the total PtdIns5P content after BTC6 cell fractionation was 2.4, 2.6 and 3.4% of PtdIns4P, and the total content for HeLa cells was 1.8, 0.9 and 1.4% of PtdIns4P, which are within the expected range for these cells (compare with Table 1).
Distribution of microsomal PtdIns5P
Figure 4 shows the distribution of microsomal (S1) PtdIns5P and PtdIns over the density gradient (Figure 4A for HeLa and Figure 4B for BTC6). The majority of the microsomal PtdIns5P (black squares) was found in fraction 4, which is enriched in plasma membrane. Importantly, a significant portion of PtdIns5P was also found in fraction 3 and in the Triton-soluble P1 fraction (Table 2), which are enriched in SER/Golgi markers. In BTC6 cells, 30% of the microsomal PtdIns5P was in fraction 3, whereas only 20% of PtdIns was present in this fraction. In HeLa cells, PtdIns5P was almost equally distributed between fractions 3 and 4, in a pattern that deviated from the distribution of PtdIns. Fraction 3 was the only fraction in which PtdIns5P distribution was higher than PtdIns distribution in both cell lines, suggesting that PtdIns5P is specifically enriched in SER/Golgi.
Distribution and relative concentration of PtdIns5P through the various subcellular fractions
To confirm this, we determined the concentration of PtdIns5P (relative to PtdIns) in each individual fraction and compared it with its concentration in the total cell. Figure 4(C) shows that in HeLa cells, a higher relative concentration of PtdIns5P is found in all microsomal fractions, but especially in fraction 3, where the local concentration of this lipid was more than twice its concentration in the total cell (broken line). In BTC6 cells (Figure 4D), the PtdIns5P enrichment in fraction 3 was also evident (1.35-fold enrichment). The relative PtdIns5P concentration in P1 fractions, which include membranes associated with the nucleus and the nuclear fraction, was lower than in the total cell. These results confirm that, although PtdIns5P is most abundant in plasma membrane, it is specifically enriched in SER and/or Golgi.
In order to determine the subcellular fractions where PtdIns5P is converted into PtdIns(4,5)P2, we generated PIP4K IIα, IIβ and IIγ knockdown HeLa cells by infection with retrovirus carrying pSuper vectors in which the shRNA sequence for targeting each PIP4K isoform was introduced. In all three PIP4K-knockdown cell lines, the basal PtdIns5P levels were higher than in the control cell line, but in the PIP4K IIγ knockdown, PtdIns5P levels were highest (1.8–2-fold higher than control, results not shown). In these cells, the PIP4K IIγ protein is 90% knocked-down and PIP4K IIβ is 50% knocked-down (Figure 5A). Figure 5(B) shows that in PIP4K IIγ-knockdown cells (open bars) the PtdIns5P concentration increased in the lighter fractions of the gradient, especially in fraction 3, which had almost twice as much PtdIns5P as the control cells (closed bars). These results indicate that the PtdIns5P pathway for PtdIns(4,5)P2 synthesis is active in SER/Golgi.
Relative concentrations of PtdIns5P in each subcellular fraction in control or PIP4K IIγ-knockdown HeLa cells
Distribution of microsomal PtdIns4P and PtdIns(4,5)P2
As with PtdIns5P, PtdIns4P accumulated in fraction 3 of the gradient when compared with PtdIns (especially in BTC6 cells, Figure 6B, triangles), but it was most abundant in plasma membrane. Although our results are contrary to the common understanding that the majority of PtdIns4P is present in the Golgi, they are consistent with immunocytochemistry findings reported by Hammond et al. , where the majority of PtdIns4P was detected at the plasma membrane.
Distribution of PIs and relative concentration of PtdIns3P through the various subcellular fractions
In our analysis, PtdIns(4,5)P2 distribution in HeLa cells was similar to the distribution of PtdIns4P (Figure 6A, diamonds). In BTC6 cells, however, PtdIns(4,5)P2 was found sharply in fraction 4 (Figure 6B, diamonds). The percentage of PtdIns(4,5)P2 in fraction 4 was almost 70%, whereas the percentages of PtdIns and other lipids in this fraction were approx. 50%, indicating that plasma membranes are enriched in PtdIns(4,5)P2. In fact, the pattern of PtdIns(4,5)P2 distribution closely resembled the distribution of β1-integrin in these cells (Figure 3D, black triangles). Hammond et al.  also report that PtdIns(4,5)P2 is predominantly at the plasma membrane.
Distribution of microsomal PtdIns3P
Unexpectedly, we found that in HeLa cells the majority of the PtdIns3P (more than 50%) was in fraction 3, where only 20% of PtdIns is found (Figure 6A). In BTC6 cells (Figure 6B), although most PtdIns3P is in fraction 4, a significantly higher percentage of this lipid, as compared with PtdIns, was distributed into fraction 3 (33% compared with 21%). When the relative concentration of PtdIns3P was examined in various fractions, we found that the levels of PtdIns3P in HeLa cells were almost 4-fold higher in fraction 3 than in the total cell (Figure 6C). In BTC6 cells, PtdIns3P levels were almost 2-fold higher in fraction 3 than in the total cell (Figure 6D). Although PtdIns3P enrichment was only observed in fraction 3, we also found this lipid to be present in fractions 4, 5 and 6, where the endosome marker Rab5 is present (Figures 3C and 7A). In contrast with PtdIns5P, PtdIns3P was virtually absent from or very low in fraction 1, which is the lightest fraction of the gradient. PtdIns3P distribution in HeLa cells closely matched the distribution of the COPI (coatamer protein I) vesicle marker βCOP (compare with Figure 3B). These results indicate that PtdIns3P is most abundant in SER and/or Golgi, contrary to the belief that the majority of PtdIns3P is found in endosomes.
Distribution of protein markers for ER, Golgi and endosomes after fractionation of the BTC6 cell cytosolic/microsomal fraction through a 20–65% sucrose gradient (A) and the effect of Brefeldin A on PtdIns5P and PtdIns3P distribution (B)
Brefeldin A affects the distribution of PtdIns5P and PtdIns3P
Since PtdIns3P and PtdIns5P are enriched in fraction 3 of the gradient, we decided to thoroughly characterize this fraction using various Golgi and ER markers. Figure 7(A) shows that fraction 3 is enriched in SER and Golgi markers such as calnexin (70%), GM130 (55%) and βCOP (40%), whereas it only contains 10% of the endosomal markers Rab5 or EEA1. Although EEA1 is an early endosomal marker that binds to PtdIns3P, the majority of this protein was found in fraction 2 of the gradient, where cytosolic proteins are found. This is probably due to a weak association of EEA1 with PtdIns3P and Rab5, which may not be strong enough to last through the gradient centrifugation. The presence of mitochondrial and RER markers in fraction 3 were equal to or lower than 10% (results not shown, and Figures 3C and 3D). Interestingly, over 30% of the PIP4K IIγ protein was found in fraction 3 (black diamonds), consistent with the findings that knockdown of this enzyme increased PtdIns5P in this fraction (Figure 5B).
Next, we decided to examine the effect of pharmacological disruption of Golgi in the distribution of PtdIns3P and PtdIns5P. After 2 h of Brefeldin A treatment, the Golgi markers 58K and GM130 were completely dispersed into small vesicles throughout the cytosol, as determined by immunocytochemistry (results not shown). Figure 7(B) shows that, after Brefeldin A treatment, there was a subtle, but reproducible, decrease in the distribution of PtdIns5P and PtdIns3P into denser microsomal fractions and an increase in the presence of these lipids in lighter microsomal fractions (compare open symbols with closed symbols). Brefeldin A treatment did not affect the distribution of PtdIns4P or PtdIns(4,5)P2 (results not shown). These results suggest that disruption of the Golgi interferes with the distribution of PtdIns3P and PtdIns5P within the cells, confirming that these lipids are present in internal membranes involved in intracellular trafficking.
The studies of PI function in cells have been revolutionized by the use of fluorescent probes that recognize specific PIs and allow subcellular imaging of these lipids in real time. These analyses rely on the assumption that a particular probe can bind to its lipid partner regardless of the microenvironment where that lipid resides within the cell. It is clear, however, that the association between a PI and its protein probe in vivo can be influenced by several factors . PI-binding probes can simultaneously bind to lipids and proteins (reviewed in ). Often a PI probe can bind to more than one PI, and although it may have a preferential partner, the binding in vivo will be affected by the abundance of weaker partners. For instance, the PHD (plant homeodomain) of ING2, which was shown to bind to PtdIns5P, can also bind to PtdIns3P and PtdIns4P , which will probably affect the localization of this domain, since they are far more abundant than PtdIns5P in cells. In addition, the specificity of a probe may change depending on the chemical environment of a particular compartment. In order to avoid some of the pitfalls associated with live-cell imaging, some groups have used lipid-binding probes to determine the cellular distribution of PI in fixed cells . However, if a PI is bound to its protein partner in a specific subcellular compartment, it may not be available for binding to its probe. Such is the case for PtdIns(4,5)P2, which binds to many cytoskeletal proteins . Thus analyses of the subcellular localization of PI-binding probes cannot on their own provide a full understanding of the distribution of the various PIs in cells. In the present study we have combined subcellular fractionation techniques with HPLC analysis of cellular PIs to study the distribution of cellular PIs, without the complications of using PI probes. The newly developed HPLC method described makes possible the study of the distribution of cellular PtdIns5P, which has not been previously assessed. The similarities between our findings on PtdIns4P localization and the ones recently reported by Hammond et al.  using immunostaining with anti-PtdIns4P antibodies highlight the reliability of our approach and the importance of using direct methods for measuring PI distribution in cells concomitantly with cellular probes.
While studying PtdIns5P localization, we made the surprising observation that a significant portion of cellular PtdIns3P is found in SER and/or Golgi. Supporting our findings is the report that the beclin–PI3K complex localizes to the trans-Golgi network . Although most FYVE domains specifically localize to endosomes in live or fixed cells , one FYVE domain-containing protein, DFCP1 (double FYVE-containing protein 1), was found associated with the Golgi . It is possible that in SER/Golgi, PtdIns3P is tightly bound to unidentified proteins, making it inaccessible to its probes. Whether the subpopulation of PtdIns3P associated with SER and Golgi plays a role in endosomal and/or autophagosomal biogenesis or in a different membrane trafficking process needs to be investigated further.
Our results provide strong evidence that PtdIns5P is enriched in intracellular vesicles of the ER and Golgi and therefore it may have a role in intracellular trafficking. In fact, the portion of microsomal PtdIns5P that accumulated in SER/Golgi-containing fractions was higher than the portion of PtdIns4P, which is a known Golgi resident (compare Figures 4B, squares, with 6B, triangles). A role for PtdIns5P in basic cell function, such as intracellular trafficking, is supported by our observation that PtdIns5P is present at basal levels in all cells examined (Table 1).
Lecompte et al.  have predicted a role for PtdIns5P in membrane trafficking, based on analysis of the phylogenetic distribution of proteins implicated in PtdIns5P metabolism. They hypothesize that PtdIns5P is involved in trafficking from late endosomes to the plasma membrane. Indeed, our results show that PtdIns5P is found in several intracellular vesicles and plasma membrane, but in contrast with Lecompte et al.'s prediction, PtdIns5P seems to be particularly enriched in microsomal vesicles that are lighter than plasma membrane vesicles. This observation is consistent with PtdIns5P being particularly enriched in SER/Golgi-derived vesicles rather than endosomal vesicles. Therefore we propose a role for PtdIns5P in Golgi-mediated transport (from or to ER, to plasma membrane or to late endosomes). Consistent with this model is the fact that PIP4K IIγ was found to localize at the ER  and at the Golgi  (Figure 5B and D. Sarkes and L.E. Rameh, unpublished work). In addition, the BTC6 insulinoma cell line, which is specialized in insulin secretion, was found to have the highest PtdIns5P levels of all the cells analysed (Table 1) and also to have high levels of PIP4K IIγ expression (results not shown). Interestingly, the Dictyostelium phosphatase PLIP, which was shown to preferentially use PtdIns5P as a substrate, localizes to the Golgi, suggesting the presence of PtdIns5P in the Golgi of Dictyostelium cells .
Several groups have now suggested a role for PtdIns5P in signalling. PtdIns5P has been implicated in insulin-induced PI3K signalling and Akt activation  and in TCR-activated phosphorylation of Dok-1 and Dok-2 , responses that occur at the plasma membrane level. PtdIns5P has also been implicated in stress-induced p53 acetylation mediated by ING2 in the nucleus . It is unclear at this point whether PtdIns5P present at the ER/Golgi plays a role in the cellular responses to extracellular signals. We are currently studying the subcellular localization of inducible PtdIns5P, which will help us answer this and other questions regarding the biological function of PtdIns5P in cells.
coatamer protein I
Dulbecco's modified Eagle's medium
early endosome antigen 1
fetal bovine serum
cis-Golgi matrix protein of 130 kDa
PTEN (phosphatase and tensin homologue deleted on chromosome 10)-like PI 5-phosphatase
Deborah Sarkes and Lucia Rameh designed, performed and analysed the experiments. Lucia Rameh conceived the project and wrote the manuscript.
We thank Dr Kent Nybakken, Dr Martin Duenwald, Dr Sara Wilcox-Adelman, Dr Jeff Miller and Dr Lynne Coluccio [Boston Biomedical Research Institute (BBRI), Watertown, MA, U.S.A.] for reagents and antibodies and Dr Charles Plant (Baystate Medical Center, Springfield, MA, U.S.A.) and Dr Ashley Mackey [Boston Biomedical Research Institute (BBRI), Watertown, MA, U.S.A.] for editing the manuscript prior to submission. We also thank Dr Lewis Cantley (Harvard Medical School, Boston, MA, U.S.A.) for insightful suggestions and discussions.
This work was supported by the National Institute for Diabetes, Digestive and Kidney Diseases [grant number DK R01-63219].