Plasma Lipoproteins Promote the Release of Bacterial Lipopolysaccharide from the Monocyte Cell Surface*

When bacterial lipopolysaccharide (LPS) enters the bloodstream, it is thought to have two general fates. If LPS binds to circulating leukocytes, it triggers innate host defense mechanisms and often elicits toxic reactions. If instead LPS binds to plasma lipoproteins, its bioactivity is largely neutralized. This study shows that lipoproteins can also take up LPS that has first bound to leukocytes. When monocytes were loaded with [3H]LPS and then incubated in plasma, they released over 70% of the cell-associated [3H]LPS into lipoproteins (predominantly high density lipoprotein), whereas in serum-free medium the [3H]LPS remained tightly associated with the cells. The transfer reaction could be reproduced in the presence of pure native lipoproteins or reconstituted high density lipoprotein. Plasma immunodepletion experiments and experiments using recombinant LPS transfer proteins revealed that soluble CD14 significantly enhances LPS release from the cells, high concentrations of LPS-binding protein have a modest effect, and phospholipid transfer protein is unable to facilitate LPS release. Essentially all of the LPS on the monocyte cell surface can be released. Lipoprotein-mediated LPS release was accompanied by a reduction in several cellular responses to the LPS, suggesting that the movement of LPS from leukocytes into lipoproteins may attenuate host responses to LPS in vivo.

Gram-negative bacterial lipopolysaccharide (LPS) 1 (endotoxin) is one of the most potent and ubiquitous of the known bacterial signal molecules. Animals have sensitive mechanisms for recognizing the presence of LPS in tissues. Monocytes, macrophages, and neutrophils, which express the LPS binding receptor, CD14 (1,2), and its signal transducer, Toll-like receptor-4 (3), are particularly sensitive to LPS (4). They respond by producing inflammatory mediators that amplify and diversify the LPS signal, triggering host defenses that wall off and destroy invading bacteria. For reasons that are not entirely clear, however, the response to LPS can also be harmful, resulting in severe sepsis, septic shock, and even death. Mechanisms that regulate responses to LPS are therefore likely to be very important for the host.
HDL is the most abundant of the lipoproteins in human plasma and interstitial fluids. It can remove both phospholipids and unesterified cholesterol from cells (reviewed in Ref. 5), and HDL-mediated cellular cholesterol efflux and the subsequent delivery of HDL-cholesterol to the liver (reverse cholesterol transport) are thought to help protect animals from atherosclerosis. This study addresses another facet of the HDL particle: its ability to bind LPS and neutralize its biological activity. There is abundant evidence that HDL and other plasma lipoproteins play an important role in controlling host responses to LPS. Numerous studies have shown that complexes of LPS with HDL and other lipoproteins have little or no stimulatory activity, either in vitro or in vivo (6 -9), and there is now strong evidence that HDL and other lipoproteins can neutralize endotoxin in vivo (10 -17).
Two plasma lipid transfer proteins, LPS-binding protein (LBP) (18,19) and phospholipid transfer protein (PLTP) (20), promote the binding of purified LPS to lipoproteins, whereas only LBP can facilitate LPS binding to CD14 on cell membranes (mCD14) or soluble CD14 (sCD14) in plasma (1,21,22). sCD14 can rapidly transfer LPS to mCD14 on cells (23), and it also facilitates the activation of cells that do not express mCD14 (24). Although sCD14 can also transfer LPS to HDL (19), it has been said to contribute very little to the movement of LPS to lipoproteins in whole plasma (25).
In this study, we sought to determine whether HDL and plasma LPS transfer proteins could remove LPS from host cells. We found that HDL facilitates the release of cell-bound LPS, that sCD14 enhances this release, and that the removal of LPS from the cells attenuates proinflammatory responses. This new pathway of LPS traffic to lipoproteins may be important for regulating host responses to Gram-negative bacteria. These findings also raise the possibility that efflux of microbial ligands from macrophages to HDL contributes to the antiinflammatory potency of HDL in processes such as atherosclerosis.

EXPERIMENTAL PROCEDURES
Cells and Reagents-Human monocytic THP-1 cells were cultured as described previously (26). CD14 expression was induced either by culturing the cells in 1,25-dihyroxyvitamin D 3 (26,27) or by transforming the cells with human CD14 cDNA (23). Peripheral blood mononuclear cells (PBMC) were separated from heparinized blood of normal human donors by centrifugation over Histopaque 1077 (Sigma) (28). Plasma was obtained from normal human volunteers without the use of anticoagulants as described (29). Briefly, venous blood was mixed immedi-ately with streptokinase (Calbiochem) (150 units/ml blood), which accelerates fibrinolysis, and platelet-free plasma was separated by centrifugation. Recombinant human LPS-binding protein (rLBP) was partially purified on SP-Sepharose (30) from serum-free culture supernatants overlying Chinese hamster ovary cells (a gift of P. Tobias, Scripps Research Institute, La Jolla, CA) that expressed the protein (23). Purified histidine-tagged LBP (rLBP-His), expressed in baculovirus-infected insect cells, was also provided by P. Tobias. Purified recombinant human soluble CD14 1-356 (rsCD14) was provided by R. Thieringer and S. D. Wright (Merck). RPMI 1640 medium, Cellgro Complete serum-free medium, and G418 were from Mediatech (Herndon, VA). All other reagents were from Sigma unless otherwise stated.
Lipoproteins-Native lipoproteins were isolated from plasma by sequential ultracentrifugal flotation in potassium bromide (KBr) at densities of 1.019 (VLDL), 1.063 (LDL), and 1.21 (HDL) g/ml (31). For some experiments, the total lipoprotein fraction of plasma (0 -1.21 g/ml) was isolated. The lipoproteins were concentrated by recentrifugation and dialyzed thoroughly against cold 0.9% NaCl containing 0.25 mM EDTA (pH 8). Lyophylized R-HDL (32), a gift of Dr. P. Lerch (Swiss Red Cross Blood Transfusion Service, Bern, Switzerland), was reconstituted in distilled water and dialyzed for 24 h (2-3 changes) in the above buffer to remove free cholate.
Apolipoprotein, Cholesterol, and Phospholipid Assays-Apolipoproteins A-I and B were measured by turbidometric assay kits from Sigma. Total (esterified and unesterified) cholesterol and choline-containing phospholipids were measured using kits from Sigma and Wako Chemical Co. (Richmond, VA), respectively.
Immunodepletion of Plasma LPS Transfer Proteins-Anti-human LBP monoclonal antibody 18G4 (a gift of P. Tobias, Scripps Research Institute) or nonimmune mouse IgG was covalently bound to Avidchrom Hydrazide F Gel beads (Unisyn Technologies, Hopkinton, MA) according to the manufacturer's instructions. Fresh normal human plasma or serum was mixed with sedimented mAb-conjugated beads on a rocking platform overnight at 0 -4°C. The beads were removed by centrifugation, and LBP was measured in the depleted and control sera by ELISA; the capture antibody (mAb IE8), the detection antibody (biotinylated mAb 18G4), and the LBP standard solution (rLBP-His) were provided by P. Tobias. Plasma LBP activity was measured by the ability of whole undiluted plasma to promote binding of [ 3 H]LPS to mCD14 on THP-1 cells; [ 3 H]LPS (2.5 or 10 ng) was incubated with plasma (50 l) and cells (3.5 ϫ 10 5 ) for 5 or 10 min at 37°C, the cells were washed with cold PBS, and cell-associated [ 3 H]LPS was measured (see below). Because control antibody beads or unconjugated beads also removed up to 90% of the LBP due to nonspecific adhesion, the precolumn serum was used as a control. The beads removed only 20% of PLTP activity and 54% of the sCD14. sCD14 was removed by incubating 500 l of plasma with anti-CD14 mAb 60bca (20 g) for 1 h. The mAb was then captured and removed by adding streptavidin-coated magnetic beads (Dynal, Lake Success, NY) that were saturated with biotin-SP-conjugated rabbit anti-mouse IgG (H ϩ L) (Jackson Immunoresearch Laboratories, West Grove, PA). The plasma was mixed with the beads on a rocking platform at 0 -4°C (three times with 1.5 mg of beads for 1 h each and once with 1.5 mg of beads for 16 h), and the beads were removed by brief centrifugation. Control plasma was produced in the same way, except that a nonimmune control antibody was used (MOPC-21, IgG1; Sigma). Plasma sCD14 was measured by ELISA (R & D Systems, Minneapolis, MN). The low sCD14 levels in the immunodepleted plasma were not due interference by residual antibody, because the addition of 60bca to plasma samples before and during ELISA had no inhibitory effect on the assay. PLTP activity was neutralized by incubating plasma with polyclonal rabbit anti-human PLTP. Nonimmune rabbit IgG was added as a control. Plasma PLTP activity was measured as described (33).
LPS Preparation, Cell Binding, and Release-Escherichia coli LCD25 [ 3 H]LPS (1.5 ϫ 10 6 dpm/g) was biosynthetically labeled and isolated as described (34). BODIPY-LPS was prepared from E. coli LCD25 LPS as described (28). For cell binding, the LPS was diluted in HNEB (20 mM HEPES buffer, pH 7.4, 150 mM NaCl, 0.1 mM EDTA, 0.3 mg/ml bovine serum albumin) (23) and mixed with rsCD14 (25 g protein/g LPS) and rLBP (0.1 g protein/g LPS), incubated for 10 min at 37°C, and diluted with serum-free medium (SFM) (Cellgro Complete serum-free medium, 20 mM HEPES, pH 7.4, and 0.3 mg/ml bovine serum albumin). THP-1 cells (7 ϫ 10 6 /ml) or PBMC (15-25 ϫ 10 6 /ml) were warmed to 37°C, the LPS-sCD14-LBP mixtures were added (50 ng LPS/ml final concentration), and the incubation was continued for 5 min at 37°C. Unbound LPS was then removed by centrifugation, and the cells were washed three times with ice-cold SFM. The cells were resuspended either in undiluted plasma containing 20 mM HEPES, pH 7.4, or in SFM with or without lipoproteins or LPS transfer proteins. 50 or 100 l of cell suspension was added to 1.5-ml microcentrifuge tubes and incubated at 37°C with gentle mixing to keep cells in suspension. The incubations were stopped with 0.3 ml of cold PBS, the cells were removed by centrifugation, and the radioactivity in the cells and supernatants was counted to measure bound and released [ 3 H]LPS. The mean fluorescence intensity of BODIPY-LPS that remained associated with the cells was measured by flow cytometry (23). In some experiments, surface-exposed and internalized fractions of cell-associated LPS were measured either by proteolytic release of surface-bound LPS by incubating the cells for 30 min in 0.02% proteinase K at 0 -4°C or by quenching of surface-exposed BODIPY-LPS with trypan blue as described (23,28). Under the conditions of our assays, less than 2% of the cell-associated [ 3 H]LPS was deacylated by cellular acyloxyacyl hydrolase (35) (not shown).
Cell counting and trypan blue permeability tests performed after the incubations revealed that [ 3 H]LPS release was not due to cell loss or alterations in plasma membrane permeability. Control and immunodepleted plasma samples were assayed in the absence or presence of 10 mM EDTA to inhibit complement fixation (36) (e.g. from anti-PLTP-PLTP immune complexes). A significant degree of membrane permeability (Ͼ10% permeable cells) was observed only in plasma samples that contained anti-PLTP or high concentrations of rsCD14; cell permeability induced by these reagents was completely inhibited by EDTA or by removal of plasma. EDTA did not alter the percentage of cellassociated LPS released into control or sCD14-depleted plasma.
The [ 3 H]LPS distribution among the major plasma lipoprotein classes was measured by ultracentrifugal flotation of undiluted plasma supernatants as described above or by size exclusion chromotography on a Superose 6 HR 10/30 column (Amersham Pharmacia Biotech) (37).

RESULTS
Human plasma rapidly removes LPS from monocytes. LPS binds rapidly to mCD14 on monocytes and THP-1 cells; maximal binding occurs within a few minutes. In contrast, bound LPS dissociates from the cells very slowly in the absence of serum, even when the cells are treated with inhibitors that prevent LPS internalization (23). In the experiments shown in Fig. 1, CD14-transfected THP-1 cells were loaded with [ 3 H]LPS for 5 min at 37°C, washed to remove the unbound LPS, and then reincubated in either SFM or human plasma. As expected, [ 3 H]LPS bound to the cells rapidly and then dissociated extremely slowly from cells that were suspended in SFM. When plasma was added after the cells were loaded with LPS, in contrast, the cells rapidly released most of the LPS they had bound. Plasma-mediated LPS release was not due to cell loss or membrane permeability, as trypan blue exclusion by the cells was not affected (not shown).
Cell-derived LPS Binds to Plasma Lipoproteins-We next asked whether the released [ 3 H]LPS was transferred to plasma lipoproteins. After the [ 3 H]LPS-loaded cells had been incubated with plasma for 30 min, the major lipoprotein classes (VLDL, LDL, and HDL) in the culture supernatants were separated by ultracentrifugation in KBr at different densities (d) to isolate VLDL, IDL, and chylomicrons (d Յ 1.019 g/ml), LDL (1.019 Ͻ d Յ 1.063 g/ml), and HDL (1.063 Ͻ d Յ 1.21 g/ml). The recoveries of [ 3 H]LPS, total cholesterol, and phospholipids were 74 Ϯ 9, 86 Ϯ 3, and 87 Ϯ 7%, respectively (mean Ϯ S.D.; n ϭ 4). After correcting for recovery of lipoprotein cholesterol, 80% of the released LPS was found in lipoproteins (the majority was in HDL), whereas only 5.6% was in nonlipoprotein fractions (d Ͼ 1.21 g/ml), and 14% was not accounted for (Table I). In control experiments, free [ 3 H]LPS did not float at any KBr density in the absence of plasma, even when the LPS was prebound to sCD14 or LBP or released from cells by protease treatment.
Having shown that at least 80% of the released [ 3 H]LPS was bound to lipoproteins, we also separated the lipoproteins according to their particle sizes on a Superose 6 gel filtration column (Fig. 2); most of the cell-derived [ 3 H]LPS co-eluted with HDL and LDL, the majority being in the HDL fractions. Whether the distribution of lipoprotein-bound [ 3 H]LPS was measured by centrifugal flotation (Table I) or gel filtration (Fig.  2), it correlated closely with phospholipid content but not with total cholesterol. Taken together, these experiments showed that HDL was the dominant acceptor for cell-bound LPS in normal plasma (HDL Ͼ LDL Ͼ Ͼ VLDL), and the distribution among the fractions correlated with the phospholipid content of the lipoproteins. In experiments not shown, HDL was also the dominant acceptor when free [ 3 H]LPS was added to plasma.
Purified Plasma Lipoproteins Promote the Release of Cellassociated LPS-We next studied LPS release from THP-1 cells into serum-free medium to which pure lipoproteins were added, with or without added recombinant LPS transfer proteins. The results show that nHDL (Fig. 3, A-C) or R-HDL (Fig. 3, D-F) promoted LPS release from the cells without a requirement for soluble LPS transfer proteins. R-HDL promoted more rapid LPS release than did nHDL, in keeping with the ability of R-HDL to rapidly adsorb free LPS (32). In experiments not shown, nLDL promoted LPS release with kinetics that were similar to those of nHDL. Adding rsCD14 to nHDL (Fig. 3A) or R-HDL (Fig. 3D) accelerated the release of cell-associated LPS. Testing different concentrations of sCD14 showed that maximal release was obtained at approximately 5 g of sCD14/ml (not shown). rLBP also accelerated the release of cell-associated LPS in the presence of R-HDL (Fig. 3E), with maximal release at 0.03 g of rLBP/ml, whereas the same concentration of rLBP had no effect in the presence of nHDL (not shown). However, when much higher concentrations (10 g of rLBP-His/ml) were added, the LBP slightly enhanced LPS release from the cells in the presence of nHDL (Fig. 3B). The low activity of the rLBP-His in this assay was not due to its inability to transfer LPS; 1 g of rLBP-His/ml completely restored the ability of LBP-depleted plasma to promote [ 3 H]LPS binding to mCD14 on cells, and the activities of the rLBP and rLBP-His were similar (not shown). PLTP had no effect in the presence of either lipoprotein (Fig. 3, C and F) at concentrations (up to 20 g rPLTP/ml) that were many fold higher than normal plasma PLTP concentrations. The lack of effect of rPLTP was not due to a lack of LPS transfer activity. In experiments not shown, less than 1 g of rPLTP/ml transferred 80% of free [ 3 H]LPS (200 ng/ml) to native lipoproteins, whereas rPTLP was unable to transfer much smaller amounts (ϳ20 ng/ml) of cell-bound [ 3 H]LPS to HDL (Fig. 3, C and F) or LDL (not shown). rPLTP did not contribute to the release of LPS when it was added together with rsCD14 or rLBP (not shown).
No significant release of cell-associated LPS occurred in the presence of rLBP or rPLTP alone (Fig. 3, B, C, E, and F). In contrast, when sCD14 was added alone (Fig. 3, A and D), the cells released 20 -30% of their LPS within 7 min and maintained the same percentage of released LPS throughout the time course of the incubation. Maximal LPS release did not occur unless lipoproteins were added. These data suggest that an equilibrium is rapidly established between cell-associated LPS and soluble LPS-sCD14 complexes, whereas the addition of a stronger LPS acceptor (HDL) shifts the equilibrium toward the lipoprotein particles and allows the bulk of the LPS to be removed from the cells.   Fig. 1. 200 l of the resulting plasma was separated on a Superose 6 column, and the fractions were assayed for phospholipids (g/fraction), total cholesterol (g/fraction), and [ 3 H]LPS (dpm divided by 10) per fraction. Recoveries of phospholipids, cholesterol, and [ 3 H]LPS were 92, 91, and 80%, respectively. HDL peak fractions contained apoA-I but not apoB, whereas LDL peak fractions contained apoB but not apoA-I (not shown). The experiment was repeated with plasma from a different donor with similar results.

FIG. 2. Chromatographic separation of plasma [ 3 H]LPS released by cells. Whole human plasma was incubated with [ 3 H]LPSloaded THP-1 cells for 30 min as in
sCD14 Enhances the Release of LPS into Human Plasma-To test whether normal plasma levels of the three LPS transfer proteins are important for accelerating LPS release from cells into whole plasma, we immunodepleted or neutralized LBP, PLTP, or sCD14 using specific antibodies. Immunodepletion removed 97 Ϯ 4 (n ϭ 8) and 99.6 Ϯ 0.3% (n ϭ 5) of the immunoreactive plasma sCD14 and LBP, respectively. As expected, LPS binding assays showed a marked reduction in the ability of LBP-depleted plasma to promote As shown in Fig. 4, reducing sCD14 levels significantly decreased the release of cell-associated [ 3 H]LPS into plasma, whereas reducing LBP or PLTP levels had no effect. Neutralization of PLTP in LBP-depleted plasma also had no effect on LPS release from cells (not shown). The amount of cell-associated [ 3 H]LPS released into sCD14-depleted plasma was the same as that released into serum-free medium to which were added approximately normal concentrations of plasma lipoproteins (isolated by KBr density flotation (d Ͻ 1.21 g/ml)); the isolated lipoprotein fraction was essentially devoid of sCD14, LBP, and PLTP (not shown). Normal activity was restored to sCD14-depleted plasma by adding physiologic concentrations (2 g/ml) of rsCD14 (Fig. 4), whereas adding rPLTP (10 g/ml) or rLBP (0.5 g/ml) had no effect, and adding rLBP-His (10 g/ml) only slightly increased LPS transfer.
In the Presence of Lipoproteins, Monocytes Release Almost All Cell-Surface LPS-As noted above, the fraction of the total cell-associated LPS that was released from the cells in the presence of lipoproteins reached a plateau at approximately 70 -80%. We used two LPS internalization assays (23,28) to determine the cellular location of the remaining 20 -30% of the LPS. In the experiment shown in Fig. 5, the release of cellassociated LPS was measured in normal human monocytes (mononuclear cells). After various incubation times in the presence of R-HDL, the cells were washed and incubated with 0.02% proteinase K on ice to remove LPS from mCD14 or other protease-sensitive LPS-binding proteins on the cell surface. Before the cells were exposed to R-HDL (Fig. 5A, time 0), the protease released 78% of the total cell-associated [ 3 H]LPS. After 60 min in the presence of R-HDL ϩ rLBP, the cells released 73% of the [ 3 H]LPS into the medium (not shown), and protease treatment released only an additional 2% of the original cell-associated [ 3 H]LPS (Fig. 5A). The results suggest that R-HDL promoted the release of nearly all of the surface-bound (protease-sensitive) LPS from the cells, whereas the remaining cell-associated LPS was protease-resistant (Fig. 5B). Because surface-exposed LPS that is inserted into the membrane or bound to a protease-resistant protein would be counted with the internalized LPS in this assay, we measured cell-associated BODIPY-LPS by flow cytometry before and after quenching the fluorescence of the surface-exposed BODIPY-LPS with trypan blue. Before rewarming the cells (Fig. 5C, 0 -4°C), the surface and internal fractions of LPS were 80 and 20%, respectively. After incubation with R-HDL, surface LPS decreased dramatically, whereas there was little change in internalized LPS. Similar results were obtained using normal human monocytes (not shown). These data suggest that virtually all of the cellsurface LPS is rapidly released in the presence of lipoprotein particles, leaving only internalized LPS.
HDL Attenuates Signal Responses in LPS-loaded Cells-The cells were incubated for 5 min at 37°C with LPS concentrations that were sufficient to stimulate cytokine responses. The unbound LPS was removed by washing, the cells were incubated in fresh serum-free medium with or without R-HDL or nHDL, and cytokines released by the cells were measured. IL-8, which is produced rapidly in response to LPS (27), was measured in the medium after incubation for 2 h, whereas the cells were incubated for 16 h before measuring cytokines that are produced more slowly (TNF-␣, IL-1␤, and IL-6). As shown in Table II, in both freshly isolated PBMC and THP-1 cells, R-HDL reduced IL-8 release to approximately 45% of control levels. nHDL was less inhibitory than R-HDL in this assay, in keeping with the ability of R-HDL to promote more rapid LPS release, yet nHDL was an effective inhibitor of cytokine responses that occurred over a longer time period. nHDL reduced TNF-␣, IL-1␤, and IL-6 release by LPS-loaded PBMC to 49 -61% of control levels; this inhibitory effect appeared to be LPS-specific because the cellular response to recombinant IL-1B (rIL-1␤) was not inhibited. Similar degrees of response inhibition occurred after cells had been exposed to various stimulatory concentrations of LPS. Because the measurements of LPS release described above were performed using higher LPS concentrations, we asked whether the rate of LPS release was different when lower amounts of LPS had bound to the cells. We found that after exposure to only 3 ng/ml [ 3 H]LPS, the percentage of cell-associated LPS released in the presence of nHDL and sCD14 in 1 h was only slightly greater than the percentage of LPS released by cells that had been exposed to 50 ng/ml [ 3 H]LPS (data not shown). Although the specific activity of our [ 3 H]LPS precluded cell binding and release measurements at lower concentrations, we believe that the measurements of [ 3 H]LPS movement presented in Figs. 1-5 can be used to describe the LPS release that occurs at low levels of cellbound LPS (Table II). per tube was 520 dpm) and incubated with R-HDL (0.1 mg of apoA-I/ml) with or without rLBP (0.03 g/ml) for the indicated times. After washing, the cells were treated with cold protease solution as described under "Experimental Procedures." A shows the percentage of the total cell-associated [ 3 H]LPS that could be released by proteolysis of cellsurface proteins, and B shows the protease-resistant fraction. Percentages were calculated from the original cell-associated [ 3 H]LPS. C, CD14-transfected THP-1 cells were loaded with BODIPY-LPS as described under "Experimental Procedures." Surface and internalized fractions of cell-associated LPS were determined by measuring internal LPS (presence of trypan blue) and total LPS (absence of trypan blue) by flow cytometry before (0 -4°C) or after reincubating (30 min at 37°C) the washed cells in rLBP (0.03 g/ml) and rsCD14 (2 g/ml). Surface LPS was calculated (total minus internal). The experiments were repeated with similar results. Symbols and error bars show mean Ϯ range of duplicate determinations.

TABLE II HDL reduces cytokine responses in LPS-loaded cells
For IL-8 response assays, cells were pulsed with LPS (0.03-0.1 ng/ml for PBMC or 0.3-1 ng/ml for THP-1 (1,25-dihydroxyvitamin D 3 -differentiated cells), washed, and reincubated in SFM containing R-HDL, rLBP (0.03 g/ml), and sCD14 (5 g/ml) as described in Fig. 5. For TNF␣, IL-1␤, and IL-6 response assays, PBMC were pulsed with LPS (0.3-3 ng/ml), washed, and reincubated in medium with or without nHDL (1 mg of apoA-I/ml) in the absence of LPS transfer proteins for 16 h as described under "Experimental Procedures." In control experiments, the cells were incubated with rIL-1␤ (5-10 ng/ml) in the absence of LPS to stimulate IL-6 production. Cytokines were measured in the culture supernatants by ELISA. The data are shown as percentage of control values (100%) obtained from the same LPS-loaded cells that were incubated in the absence of HDL. Means Ϯ S.D. are shown for n experiments done in duplicate or triplicate.

DISCUSSION
Although numerous studies have shown that HDL and other lipoproteins can bind and neutralize the biological activity of free LPS, lipoprotein-mediated removal of LPS from cell membranes has not been previously described. We found not only that LPS can leave the monocyte surface and bind lipoproteins, but also that exposing LPS-loaded cells to HDL significantly reduces the production of several cytokines (TNF-␣, IL-1␤, IL-6, and IL-8). Inhibition of LPS responses occurred when cell responses to a non-LPS agonist, IL-1␤, were not affected. The findings thus suggest that removing LPS from the cell surface in the presence of HDL attenuates cellular responses.
Our data suggest that prolonged contact of the cells with LPS must be maintained in order to produce LPS signals of maximal strength and duration. This hypothesis is supported by other studies. For example, prolonged expression of LPS-induced inflammatory genes (TNF-␣, IL-1␤, and tissue factor) in whole blood requires continuous exposure to LPS (38), and in vivo therapeutic intervention with Limulus anti-LPS factor (39) or CD14 antibodies (40) given after LPS challenge can rescue animals from the lethal effects of LPS. Doran et al. (41) also reported that the addition of R-HDL particles to whole blood attenuated LPS-induced TNF-␣ production even when the HDL was added up to 1 h after the LPS. Although Gallay et al. (42) showed that brief exposure (5-10 min) of monocytes to LPS at concentrations of 10 -100 ng/ml followed by incubation in 10% autologous plasma was sufficient to stimulate maximal responses, they found that lower LPS concentrations (1 ng/ml) required longer exposure (e.g. 1 h). When cells are exposed to very high concentrations of LPS, lipoproteins may not be able remove a sufficiently large fraction of the LPS to decrease cell responses. The LPS concentrations used in our experiments were Յ3 ng/ml (Table II), and both R-HDL and nHDL reduced LPS-induced cytokine production.
Because the kinetics of LPS binding to monocytes in whole plasma are similar to those that occur in plasma-free medium containing rLBP and rsCD14 (not shown), this suggests that a large fraction (up to 50%) of the plasma LPS transiently associates with the cells before it moves into lipoproteins. The degree of cell stimulation may thus be related to the persistence of the interaction of CD14-bound LPS with signaling receptors (Toll-like receptor-4) on the cell surface. On the other hand, if LPS interactions with membrane lipids and endocytosis are important for signaling (43), then the degree of cell stimulation could be related to flow of LPS into these sites.
Although plasma HDL was the dominant acceptor of cellbound LPS in our experiments, LDL also bound a significant proportion of the released LPS. Many lipoproteins (VLDL, LDL, and chylomicrons) and lipid emulsions can bind and neutralize LPS (12, 44 -46). Some studies have found that LDL is able to neutralize LPS better than HDL (47), but when several lipoprotein classes were compared according to their phospholipid content, LPS neutralizing ability was similar in all lipoprotein classes (14). Moreover, increasing the phospholipid to apolipoprotein ratios in reconstituted HDL particles increased their ability to neutralize LPS (15,41). We found that LPS movement into lipoproteins correlated with their phospholipid content. Thus, phospholipids may be important for both release and neutralization of cell-associated LPS. Phospholipids are the primary constituents of the HDL particle surface, and, in most individuals, the large numbers of HDL particles present in plasma provide the greatest effective surface area of any of the circulating lipoproteins. We therefore hypothesize that the abundance of phospholipid on HDL particle surfaces contributes to its superior ability to bind released LPS.
The  (50)) bind native HDL and LDL as well as modified lipoproteins, and recent studies show that SR-BI promotes the efflux of unesterified cholesterol from cells to HDL or LDL (51,52). Because human monocytes express both SR-BI and CD36 (50,53), we are investigating the roles of these receptors in cellular "efflux" of LPS. Just as cholesterol efflux from cell membranes is thought to occur both by receptormediated and non-receptor-mediated processes (48), it seems likely that LPS efflux may also occur by both kinds of mechanisms.
We found that sCD14 enhanced the release of cell-bound LPS in the presence of isolated nHDL or plasma, whereas LBP and PLTP had little or no effect. Although sCD14 accelerated LPS transfer from cells to both R-HDL and nHDL, we found unexpectedly that rLBP strongly enhanced LPS release in the presence of R-HDL. LBP (18,54) is normally found attached to native HDL particles in plasma, but it is lost when HDL is isolated using KBr, and we were unable to detect it by ELISA in our preparations of nHDL (data not shown). Although traces of contaminating LBP might contribute to the ability of nHDL to release cell-surface LPS (and thus to the inability of rLBP to increase LPS release much above that seen with nHDL alone), this explanation thus seems unlikely.
Our results suggest that the interactions between LPS, lipoproteins, plasma transfer proteins, and leukocytes may be more complex than has previously been suspected (Fig. 6). Whereas both purified and native LPS were known to bind to leukocytes and lipoproteins, these were generally thought to be irreversible interactions that either inactivated the LPS (lipoprotein binding) or led to cellular activation (cell binding and internalization). It now seems that the LPS-leukocyte interaction is probably quite dynamic, with plasma transfer proteins facilitating both the binding and release of LPS from cells when lipoproteins are present. It is also possible that before LPSlipoprotein complexes are cleared from the circulation, the LPS can move from lipoproteins to cells; the ability of the lipoproteins to sequester LPS effectively may be due to the large size of the lipoprotein "acceptor sink." The concentrations of circulating lipoproteins and LPS transfer proteins may thus be very important determinants of the fate of LPS molecules in plasma.
Our findings raise the possibility that plasma lipoproteins may also be important acceptors for other host cell-associated FIG. 6. LPS traffic in blood. LPS is released from bacterial membranes and binds to leukocytes (principally monocytes and neutrophils that express CD14) and activates them (1), or binds to plasma lipoprotein particles (principally HDL), which sequester (inactivate) the LPS and clear it from the circulation (2). Lipoproteins promote the release of LPS from leukocyte surfaces and attenuate signal responses (3). bacterial lipids. For example, a recent study showed that plasma lipoproteins can bind and neutralize lipoteichoic acids, which are thought to be important stimulatory agonists in Gram-positive bacteria (55). Recent findings also raise the possibility that LPS and other microbial agonists may contribute to the inflammatory reaction in the vessel wall that occurs during atherosclerosis (56,57). In particular, both LPS and LPS-containing microbes (such as Chlamydia pneumoniae) can induce macrophages to form foam cells (58 -60). In addition to the known antioxidant properties of HDL (61,62) and its ability to promote cholesterol efflux from cells, our findings raise the possibility that its antiatherogenic potency may also be related to its ability to accept LPS (and possibly other CD14binding ligands) from macrophages.