Dynamic Lipid-dependent Modulation of Protein Topology by Post-translational Phosphorylation

Abstract

Membrane protein topology and folding are governed by structural principles and topogenic signals that are recognized and decoded by the protein insertion and translocation machineries at the time of initial membrane insertion and folding. We previously demonstrated that the lipid environment is also a determinant of initial protein topology, which is dynamically responsive to post-assembly changes in membrane lipid composition. However, the effect on protein topology of post-assembly phosphorylation of amino acids localized within initially cytoplasmically oriented extramembrane domains has never been investigated. Here, we show in a controlled in vitro system that phosphorylation of a membrane protein can trigger a change in topological arrangement. The rate of change occurred on a scale of seconds, comparable with the rates observed upon changes in the protein lipid environment. The rate and extent of topological rearrangement were dependent on the charges of extramembrane domains and the lipid bilayer surface. Using model membranes mimicking the lipid compositions of eukaryotic organelles, we determined that anionic lipids, cholesterol, sphingomyelin, and membrane fluidity play critical roles in these processes. Our results demonstrate how post-translational modifications may influence membrane protein topology in a lipid-dependent manner, both along the organelle trafficking pathway and at their final destination. The results provide further evidence that membrane protein topology is dynamic, integrating for the first time the effect of changes in lipid composition and regulators of cellular processes. The discovery of a new topology regulatory mechanism opens additional avenues for understanding unexplored structure-function relationships and the development of optimized topology prediction tools.

Introduction

Establishing the principles governing the folding of a membrane protein is central to understanding the molecular basis for membrane proteins that display multiple topologies and a wide spectrum of membrane protein conformational and topological disorders (1–6). At least 10% of 470 known pathogenic mutations were predicted to result in a change in the topological organization of the mutant membrane protein (7). A fundamental objective in membrane biology is to understand and predict how protein sequence determines the number and orientation of transmembrane domains (TMDs)³ (8). Dogma dictates that membrane protein topology is determined and fixed at the time of initial assembly by the following: (i) membrane protein topogenic signals in accordance with Positive Inside Rule (9–12) by an unknown mechanism; (ii) the membrane insertion machinery; and (iii) interactions within a membrane protein. A role for membrane protein lipid environment in determining membrane protein organization has been minimized mainly due to proper folding of many membrane proteins in detergents and amphipols (13); however, why some membrane proteins misfold in detergents or when expressed in a foreign host has not been fully explained. Our ability to systematically control and temporally change membrane lipid composition in Escherichia coli challenged this long-standing dogma. We demonstrated that topogenic signals can be decoded by the membrane lipid profile during initial membrane protein insertion (14–17) and that membrane proteins can undergo TMD flipping after initial assembly in response to changes in the lipid environment, both in vivo and in vitro (18–21). Thus, membrane protein topological organization is not static but highly dynamic. The Charge Balance Rule (8), as an extension of the Positive Inside Rule (22–24), was recently proposed by us to account for the influence of lipid environment on the effective net charge of membrane protein extramembrane domains (EMDs) as an orientation determinant of TMDs.

Changes in lipid environment, also associated with differences in membrane properties, occur during cell division, membrane fission and fusion, vesicular protein trafficking, and lateral movement of proteins in membranes (25, 26). A TMD sequence analysis of membrane proteins with known organelle localization pointed out organelle specificity for TMD sequence and length (27). However, depending on their final destination, membrane proteins trafficking to their target organelle will encounter drastically different lipid environments. Therefore, it is likely that a sub-fraction of membrane proteins undergoes lipid-induced post-assembly changes in their topology (28).

Phosphokinase-specific phosphorylation is one the most widespread post-translational modifications in prokaryotes and eukaryotes. Although reversible phosphorylation is recognized to trigger significant conformational changes in a protein to regulate its function or mediate a signal transduction cascade, little is known about the consequences of post-translational phosphorylation of a membrane protein on its topological orientation. Only a limited number of examples have been reported where membrane protein phosphorylation was postulated to trigger a topological reorientation. The mechanism by which CD38 metabolizes adenosine 5′-diphosphate-ribose as an intracellular Ca²⁺-mobilizing messenger is still elusive. However, the ability of two interconverted populations of CD38 to co-exist was demonstrated recently (1). A phosphorylation-dephosphorylation cycle was suggested to control the orientation of the single TMD-spanning eukaryotic protein CD38 (1, 29), suggesting a new topology-dependent mechanism regulating the signaling activity of CD38. Phospholipid scramblase 1 (PLSC1), which is also phosphorylated, undergoes topological changes during cell differentiation (30). However, no direct evidence linking post-translational phosphorylation to TMD orientation has been reported.

We hypothesize that naturally occurring phosphorylation within EMDs of membrane protein can serve as a physiological signal, which in the context of different phospholipid environments has the potential to result in a membrane protein adopting a new TMD organization and possibly a new function. If phosphorylation can induce a post-assembly inversion of protein TMDs, a new mechanism for regulation of membrane protein function would be uncovered.

Herein, we have used a previously established fluorescence resonance energy transfer (FRET)-based assay (21) to monitor EMD flipping now induced by phosphorylation. Using E. coli lactose permease (LacY) derivative as a complex model 12 TMD model membrane protein, wherein phosphorylation sites were genetically engineered, we have determined the rate and extent of phosphorylation-induced TMD flipping in vitro. Using proteoliposomes of various lipid compositions mimicking E. coli membranes and eukaryotic organelles, we have investigated the lipid dependence of phosphorylation-induced topological switching. We now demonstrate that phosphorylation of a membrane protein, which changes the effective net charge of an EMD, can induce a re-orientation of adjacent TMDs relative to the plane of the lipid bilayer. The switch proceeds at a physiologically significant rate as we demonstrated for lipid-induced TMD flipping (21). Phosphorylation can introduce a new topogenic signal, which can be executed post-assembly outside of the translocon in a lipid- and organelle-dependent manner. Our results establish a novel post-translational mechanism for the dynamic regulation of the structure and function of membrane proteins, which is potentially widespread, thus expanding our knowledge of the roles played by membrane lipids in the establishment and maintenance of membrane protein topology.

Results

Phosphorylation of a Membrane Protein Can Lead to a Change in Its Topological Arrangement

We sought to determine the effects of one of the most common post-translational modifications of EMDs on membrane protein dynamic organization, as a function of the lipid environment. We have developed a set of in vitro approaches to monitor membrane protein topology, both dynamically and at steady state, along with assessment of membrane properties. Our model membrane protein is LacY, a paradigm for secondary transporters throughout nature, for which we have characterized the effect of membrane anionic surface charge density contributed by lipid head groups and net charge of EMDs on topology. Assembly of LacY in vivo (16) or in vitro (20) in membranes or proteoliposomes, respectively, containing only anionic phospholipids (i.e. lacking phosphatidylethanolamine (PE)) results in the inversion of the topology of the N-terminal six-TMD bundle with respect to the C-terminal five-TMD bundle. In addition, TMD VII exits the membrane and is exposed to the normally periplasmic side of LacY. The same lipid-dependent inversion of LacY TMDs was verified in vitro using atomic force microscopy (31).

We previously demonstrated that changing the net charge of cytoplasmic EMDs C2/C4/C6 of LacY (Fig. 1A) from +2/+2/+2 (WT LacY) to −2/−2/−2 results in topological inversion of the N-terminal six-TMD bundle of LacY in wild type E. coli cells, whereas a change to −2/0/−2 did not result in inversion (19). These findings pointed us toward the use of two different LacY charge templates, −2/+2/+2 and −2/0/−2, the latter being closer to the topological switch threshold based on net charge of its EMDs. The charge templates −2/0/−2 (which contains the following mutations: K69E, K74E, K131E, K211E, and K218E) and −2/+2/+2 (which contains the following mutations: K69E and K74E) have been previously described (19).

FIGURE 1.

Engineering phosphorylation of a model membrane protein. A, topology of LacY initially assembled in proteoliposomes made of E. coli total lipids is shown (adapted from Ref. 61, reprinted with permission from American Association for the Advancement of Science). Cytoplasmically oriented EMDs C2 (red line), C4 (orange line), and C6 (blue line), N terminus (NT), and C terminus (CT) are indicated. The C6 domain containing engineered kinase phosphorylation sites is detailed in the sequence alignments presented below. The residues mutated in the charge templates are indicated by black circles in domains C2, C4, and C6. Important residues for LacY function and structure stability are indicated as follows: residues marked with red and yellow circles are involved in substrate binding and proton translocation, respectively; residue Glu-269, represented by a blue circle, is involved in both substrate binding and proton transfer. B, engineered PDK1 and LKB1 kinase sequences in LacY EMD C6, either alone (PDK1 or LKB1) or in tandem (PDK1/LKB1), allows us to perform phosphorylation-induced charge alterations in both LacY charge templates (−2/+2/+2 and −2/0/−2). Top row or bottom row indicates the result of Western blottings using anti-phosphoserine or anti-LacY antibody, respectively. These modifications can be reversed by addition of phosphatase and liposomes solubilization (+).

Native LacY does not contain any known phosphorylation site; therefore, we engineered phosphoinositide-dependent protein kinase 1 (PDK1) and liver kinase B1 (LKB1) EMD sites based on the lack of charged amino acids in their kinase consensus sequences. In proteoliposomes, native LacY presents an inside-out orientation (cytoplasmic domains outside) relative to whole cells. Therefore, we engineered phosphorylation site(s) into EMD C6, which will be accessible to the outside of the proteoliposomes composed of native lipid composition (Fig. 1A). The ability of PDK1 and LKB1 kinases (used either alone or in combination) to phosphorylate LacY was verified by immunoblotting using an antibody specific for phospho-Ser/Thr (Fig. 1B).

First, we sought to determine the effect of phosphorylation on steady state LacY topology when reconstituted in proteoliposomes made of E. coli total lipids. Employing a single cysteine located in EMD C6 (Fig. 1A), we used the substituted cysteine accessibility method for determining TMD orientation (SCAM^TM) to analyze LacY TMD orientation in proteoliposomes before and after addition of kinase(s) (Fig. 2A). 3-(N-maleimidylpropionyl)biocytin (MPB) is a biotinylated sulfhydryl-specific probe that is impermeable to proteoliposome membranes. Therefore, cysteines located on the outside of the proteoliposomes should be derivatized by MPB, whereas cysteines located in the lumen (inside) of the proteoliposomes should be protected, unless the membrane is solubilized before or during labeling (use of the detergent octyl glucoside (OG)). 4-Acetamido-4′-maleimidylstilbene-2,2′-disulfonic acid (AMS), another membrane-impermeable but non-biotinylated maleimide, was used to block exposed cysteines prior to MPB treatment, thus preventing biotinylation of outward-facing cysteines.

FIGURE 2.

Phosphorylation of a membrane protein can lead to a change in its topological arrangement. A, determination of the orientation of domain C6 of LacY with altered EMD net charge in proteoliposomes made of E. coli total lipids before and after kinase addition using SCAM^TM as described under “Experimental Procedures.” Proteoliposomes were exposed to MBP before or after addition of OG or after exposure to and removal of AMS followed by addition of OG. The deduced orientation of C6 is also shown. B, detection of topological rearrangements in LacY upon phosphorylation by FRET. Side view of LacY (Protein Data Bank ID code 2CFQ) depicts the diagnostic Trp replacements introduced individually in EMDs NT and C6 (magenta spheres) and the IAEDANS label at Cys-331 (blue sphere). Real time FRET measurements monitoring the topological switch of domains NT and C6 in LacY −2/0/−2 template are shown. Data represent the normalized fluorescence expressed as the ratio F_t/F₀, as described under “Experimental Procedures.” Experiments were repeated three to five times, and the data represent mean values ± S.E.

We found that the two LacY charge templates behave differently upon phosphorylation, following the Charge Balance Rule. Addition of a single phosphate group in the −2/+2/+2 template does not lead to any inversion, although we observe mixed topology (∼80% inversion) in the −2/0/−2 template. Addition of two phosphate groups resulted in mixed topology for the −2/+2/+2 template and close to full inversion for the −2/0/−2 template. According to the Charge Balance Rule (8), the zwitterionic net neutral lipid PE appears to dampen the translocation potential of negatively charged residues in cytoplasmic EMDs in favor of the retention potential of positively charged residues. The requirement for multiple phosphorylation events to induce near complete inversion of LacY EMD C6 is consistent with this previous observation.

Next, we determined the rate of phosphorylation-induced TMD flipping in vitro to establish whether flipping proceeds at a physiologically significant rate as we demonstrated for lipid-induced TMD flipping. All the kinetic measurements were performed using LacY −2/0/−2 charge templates containing a single phosphorylation site (LKB1) engineered in EMD C6. The previously established FRET-based assay (21) was used to monitor EMD flipping now induced by phosphorylation (Fig. 2B). The same established Trp residues, serving as donor for the FRET pair, were introduced in EMD NT or C6. V331C, whose topology is insensitive to lipid environment (17), was labeled by 5-((((2-iodoacetyl)amino)ethyl)amino)naphthalene-1-sulfonic acid (IAEDANS) in the C-terminal TMD bundle of newly engineered LacY (Fig. 2B) to serve as the acceptor for the FRET pair. In the case of NT and C6 EMDs, high FRET indicates native orientation (both donor and acceptor in close proximity on the same side of the membrane), whereas low FRET indicates inverted orientation of LacY (donor and acceptor located across the membrane from each other). All controls directly linking fluorescence changes with the extent of topological changes followed by SCAM^TM and lack of FRET signals from native Trp residues were previously established (21). Addition of excess kinase (so that phosphorylation is not rate-limiting) at 20 °C resulted in a flipping of the C6 EMD followed by a delayed yet faster flipping of the NT EMD on a time scale of seconds (Fig. 2B). The delayed flipping of EMD NT, compared with EMD C6, was previously observed for lipid-induced flipping. The faster phosphorylation-dependent flipping rate of EMD NT, compared with lipid-induced flipping, could be partially explained by the elimination of the charge pair Asp-68–Lys-131 (32), resulting in a decreased protein stability. Such a modification was not present in the LacY template used to investigate the lipid-induced protein flipping. We previously showed that lipid-induced flipping at a physiological temperature of 37 °C was too fast to measure. These results fully support our hypothesis, indicating that increasing the negative charge of an EMD by phosphorylation can induce topological changes in a membrane protein. Therefore, topological switching can be triggered by sufficient EMD charge modifications, independent of changes in the lipid environment or any other cellular factors.

Phosphorylation-induced Topological Switch of a Membrane Protein Is Influenced by the Organelle Lipid Environment

Membrane lipid composition drastically varies among eukaryotic organelles. Besides additional phospholipid classes over those found in bacteria, eukaryotic organelles also contain sphingolipids and cholesterol. These various lipid species are extremely diverse, both in chemical and physical properties, leading to striking differences in organelle membrane properties such as membrane thickness, fluidity, asymmetry or surface charge. We sought to identify the effect of organelle membrane properties on the rate and extent of phosphorylation-triggered TMD flipping. Therefore, we used LacY reconstituted in proteoliposomes with lipid composition mimicking five canonical organelles (mitochondria, endoplasmic reticulum (ER), Golgi, plasma membrane (PM), and endosomes) (25). The phospholipid content of these model organelles is presented in Fig. 3A, along with E. coli total phospholipids. The detailed lipid composition of the various model organelles (with cholesterol approximating physiological levels) is presented in Table 1.

FIGURE 3.

Properties of organelle-mimicking proteoliposomes. A, lipid composition (minus cholesterol) of the proteoliposomes mimicking eukaryotic organelles and E. coli membranes. B, Laurdan GP calibration using synthetic liposomes. Emission spectra (left) of Laurdan solubilized in 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) or sphingomyelin/cholesterol (2:1 molar ratio) large unilamellar vesicles at 20 °C using excitation light at 340 nm. Quantitative analysis (right) of the GP values, using Equation 2 as described under “Experimental Procedures.” Values are means ± S.D. of the data obtained from four replicates. C, membrane fluidity measurements using Laurdan GP. Black or gray bars indicate membranes containing or lacking cholesterol, respectively. In all cases, the data represent mean values ± S.E. from 3 to 4 experimental replicates.

TABLE 1.

Detailed lipid composition of the various model membranes used in this study

The following abbreviations are used: PI, phosphatidylinositol; PS, phosphatidylserine; CL, cardiolipin; neg, negative. Except for the E. coli sample where PE, PG, and CL from E. coli were used, care was taken to use phospholipid from similar eukaryotic sources (brain for PE, PC, PS, and SM; liver for PI; heart for CL, and soy for PG) to limit fatty acid heterogeneity among phospholipid classes.

Sample	Phospholipids							Chol/PL ratio
	PE	PC	PI	PS	PG	CL	SM
	mol %
E. coli	75	0	0	0	20	5	0	0
Mitochondria	30	47	5	1	5	12	0	0.1
ER	30	50	15	5	0	0	0	0.15
Golgi	18	55	8	4	0	0	15	0.2
Endosomes	27	55	2	1	0	0	15	0.5
PM	25	40	2.5	7.5	0	0	25	1
Mitochondria-No Chol	30	47	5	1	5	12	0	0
ER-No Chol	30	50	15	5	0	0	0	0
Golgi-No Chol	18	55	8	4	0	0	15	0
Endosomes-No Chol	27	55	2	1	0	0	15	0
PM-No Chol	25	40	2.5	7.5	0	0	25	0
Golgi-MoreNeg	18	45	8	14	0	0	15	0.2
Golgi-MoreChol	18	55	8	4	0	0	15	1
PM-MoreNeg	25	30	2.5	17.5	0	0	25	1
PM-LessChol	25	40	2.5	7.5	0	0	25	0.2

As an indicator of lipid-dependent changes in cellular membrane properties, we examined the impact of proteoliposome lipid composition on membrane fluidity using Laurdan general polarization (GP) (33), calibrated using reference liposomes of known lipid composition (Fig. 3B). The more positive GP values indicate higher membrane order, whereas negative GP values correspond to disordered membranes with higher water content (i.e. membrane packing defects). The results (Fig. 3C), indicating the impact of lipid composition with particular emphasis on cholesterol, are in good agreement with previous measurements performed using Laurdan on membranes isolated from various prokaryotic and eukaryotic organisms (34). The first striking observation is the difference in membrane fluidity between model eukaryotic organelle membranes, with higher disorder observed in model ER and mitochondrial membranes. Membrane order increases along the secretory pathway (ER < Golgi < PM), as well as in endosomes. Interestingly, E. coli membranes appear more ordered than mitochondrial membranes (with or without cholesterol), despite their similar phospholipid composition, pointing out the differences in physicochemical properties of PE and phosphatidylcholine (PC). The importance and pleiotropic effects of cholesterol are illustrated by the fluidity differences observed when omitting cholesterol from the organelles mimicking liposomes. Large membrane order increases (higher GP) are observed in organelles originally containing the lowest cholesterol to phospholipid ratios (mitochondria and ER), although the opposite is observed (GP decrease) in organelles originally containing high cholesterol to phospholipid ratios (Golgi, endosomes, and PM). These latter organelles also contain significant amounts of sphingomyelin, known to stiffen the membrane when incorporated into synthetic lipid membranes (35). Additionally, there is evidence that sphingolipids and cholesterol are located in an intimate association in specific sub-domains of membranes (36, 37). These properties could contribute to the higher order observed in membranes containing sphingomyelin and cholesterol.

Next, we determined the rate of phosphorylation-induced TMD flipping in proteoliposomes mimicking organelle lipid composition, with (Fig. 4A) and without (Fig. 4B) cholesterol. We used LacY −2/0/−2 charge template with a single phosphorylation site (LKB1) and a Trp replacement engineered in EMD C6. Linear drift stemming from potential photobleaching during the fluorescence measurement was accounted for by measuring the FRET signal of the various LacY templates used but omitting the kinase treatment. The linear drift observed in these measurements was then subtracted from the experimental determination of phosphorylation-induced LacY flipping. The kinetics of TMD flipping clearly illustrate the dependence of phosphorylation-induced topological reorientation on membrane lipid composition. Three patterns were observed in the presence and the absence of cholesterol. Biphasic kinetics were observed with first a fast (Golgi, PM, mitochondria with or without cholesterol), intermediate (ER with cholesterol), or slow (endosomes with or without cholesterol) flipping rate, all followed by slower second phase flipping kinetics. Single exponential fits of the kinetics observed allowed the determination of initial flipping rates (Fig. 4C), expressed as 1/τ (higher value indicates faster rate). Table 2 presents the details of the parameters generated by first order exponential fits of the data. TMD flipping rates were in the same range as that observed with E. coli lipids, with an increase in the flipping rates for mitochondria < ER < Golgi < PM that contained cholesterol. Only endosomes exhibited strikingly slower rates. In the absence of cholesterol, flipping rates increased for mitochondria although they decreased for the ER. No significant changes were observed for membranes mimicking Golgi, PM, or endosomes with or without cholesterol. These results indicate for the first time that membrane protein topology rearrangements can be triggered in any organelle membrane upon phosphorylation, with rates fast enough to be of physiological relevance. They also clearly indicate the influence of organelle lipid composition.

FIGURE 4.

Phosphorylation-induced topological switch of a membrane protein is influenced by the organelle lipid environment. Real time FRET measurements monitoring phosphorylation-induced topological switch of EMD C6 in LacY −2/0/−2 template reconstituted in proteoliposomes containing (A) or lacking (B) cholesterol are shown. Data represent the averaged normalized fluorescence (orange, red, blue, gray, and black lines) expressed as the ratio F_t/F₀, as described under “Experimental Procedures.” Data fits are also indicated, either by a black or white lines superimposed on normalized fluorescence. C, bar graphs displaying flipping rates determined from single exponential fit of the data shown in A and B. Black or gray bars indicate membranes containing or lacking cholesterol, respectively. D, bar graphs displaying amounts of correct topology determined at steady state (30 min post phosphorylation performed at 20 °C) using SCAM^TM. Black or gray bars indicate membranes containing or lacking cholesterol, respectively. In all cases, the data represent mean values ± S.E. from 3 to 4 experimental replicates.

TABLE 2.

Kinetic parameters for the real time measurement of protein flipping

Detailed parameters extrapolated from first order exponential decay fits to the data presented in Figs. 4 and 5, according to the following equation: y = y₀ + Ae^−(x/τ), where y₀ is the Y offset; A is the amplitude; τ is the “time” constant; neg is negative. In all cases, the data represent mean values ± S.E. from 3 to 4 experimental replicates.

Membrane	Offset (± S.E.)	A (± S.E.)	τ (± S.E.)	R2 (± S.E.)
E. coli EMD C6	0.2897 ± 0.0216	0.5419 ± 0.0351	155.4834 ± 6.6179	0.9991 ± 0.0004
With cholesterol
PM	0.3224 ± 0.0433	0.6625 ± 0.0424	177.1626 ± 11.5476	0.9996 ± 0.0001
Golgi	0.4082 ± 0.0108	0.5820 ± 0.0132	167.8669 ± 8.3230	0.9996 ± 0.0001
Endosomes	0.7966 ± 0.0313	0.2040 ± 0.0335	460.7849 ± 45.5187	0.9968 ± 0.0005
ER	0.6346 ± 0.0107	0.3478 ± 0.0118	130.5466 ± 5.5144	0.9902 ± 0.0053
Mitochondria	0.5818 ± 0.0141	0.4056 ± 0.0160	152.1759 ± 1.9554	0.9981 ± 0.0006
No cholesterol
PM	0.3224 ± 0.0433	0.5015 ± 0.0462	145.5874 ± 7.6305	0.9987 ± 0.0003
Golgi	0.5466 ± 0.0017	0.4455 ± 0.0069	145.7686 ± 3.5911	0.9955 ± 0.0019
Endosomes	0.9007 ± 0.0069	0.1065 ± 0.0080	169.3154 ± 18.6300	0.9837 ± 0.0045
ER	0.6259 ± 0.0079	0.4013 ± 0.0085	293.6767 ± 17.2230	0.9987 ± 0.0001
Mitochondria	0.5105 ± 0.0514	0.4674 ± 0.0550	135.4257 ± 13.2839	0.9960 ± 0.0014
Further dissection
Golgi-MoreChol	0.4498 ± 0.0157	0.5427 ± 0.0240	160.2730 ± 2.6659	0.9981 ± 0.0002
PM-LessChol	0.7596 ± 0.0118	0.2512 ± 0.0098	226.4869 ± 6.7455	0.9963 ± 0.0006
Golgi-MoreNeg	0.6643 ± 0.0131	0.3122 ± 0.0132	121.7762 ± 1.3187	0.9940 ± 0.0002
PM-MoreNeg	0.6359 ± 0.0110	0.3482 ± 0.0117	133.1973 ± 3.9983	0.9908 ± 0.0055

We also determined the extent of topology inversion of LacY in these various proteoliposomes using SCAM^TM. The results (Fig. 4D) present the percentage of correct membrane protein topology (compared with the initial 100% correct orientation). Interestingly, there appears to be no direct correlation between the flipping rates and the amount of correct topology at steady state (R² = 0.26, Fig. 5A), as indicated by the very low amount of correct topology in proteoliposomes made of E. coli lipids compared with the other organelle-mimicking compositions. However, when excluding the E. coli lipid sample, a possible inverse correlation between rate and extent of flipping can be drawn (R² = 0.68, Fig. 5B), further demonstrating the complexity of the effects played by membrane lipids on membrane protein topological stability.

FIGURE 5.

Visualizing possible correlations between flipping rates, topology, membrane fluidity, and membrane lipid composition. Flipping rates as a function of all data points. Flipping rates as a function of the amounts of correct topology (A–D), membrane fluidity (E), sphingomyelin amounts (F), anionic lipids amounts (G), and cholesterol to phospholipid ratio (H). A, data points include E. coli lipids and organelle-mimicking membranes (mitochondria, Golgi, endoplasmic reticulum, plasma membrane, and endosomes), with or without cholesterol. B, same as A, omitting the E. coli lipid data point. C, all data points. D, all minus E. coli lipid data point. E–H, all data points. A–D and G, lines indicate linear fitting of the data, with the R² value indicated.

Impact of Cholesterol and Anionic Lipid Content on Phosphorylation-induced Topological Switch of a Membrane Protein

To dissect the direct effect of individual lipid classes in more detail, we performed additional determinations of flipping rates and final topology, along with membrane fluidity assessment, using proteoliposomes made of lipids mimicking Golgi and PM with alterations in their anionic lipids and cholesterol content. These two organelles were chosen based on their similar content in anionic lipids (10%) but moderate and drastic differences in sphingomyelin content (15 and 25%, respectively) and cholesterol to phospholipid ratio (0.2 and 1, respectively). We performed (i) variations in the cholesterol amount in original Golgi (Fig. 6, Golgi-MoreChol) and PM (Fig. 6, PM-LessChol) membranes and (ii) increase in the amount of anionic lipids (Fig. 6, MoreNeg). The results are presented in Fig. 6, along with previous determinations for Golgi and PM with or without cholesterol.

FIGURE 6.

Impact of cholesterol and anionic lipid contents on phosphorylation-induced topological switch of a membrane protein. A and B, real time FRET measurements monitoring phosphorylation-induced topological switching of EMD C6 in LacY −2/0/−2 template reconstituted in proteoliposomes. A, Golgi-mimicking liposomes; B, plasma membrane-mimicking liposomes. Data represent the averaged normalized fluorescence (orange, red, blue, gray, and black lines) expressed as the ratio F_t/F₀, as described under “Experimental Procedures.” Data fits are also indicated, either by black or white lines superimposed on normalized fluorescence. C, bar graphs display flipping rates (black bars) and amounts of correct topology (gray bars). Flipping rates (×10⁵ s⁻¹) were determined from a single exponential fit of the data shown in A and B. Amounts of correct topology were determined at steady state (30 min post-phosphorylation performed at 20 °C) using SCAM^TM. D, membrane fluidity measurements using Laurdan GP. In all cases, the data represent mean values ± S.E. from 3 to 4 experimental replicates.

Real time measurements of TMD flipping performed in the various modified Golgi-mimicking liposomes all display a biphasic profile with similar initial rates and minor differences observed after 100 s (Fig. 6A) in the slower phase. Measurements performed in various modified PM-mimicking liposomes (Fig. 6B) exhibit differences in both early and late kinetics. Single exponential fits allowed determination of initial flipping rates (Fig. 6C, black bars, × 10⁵ s⁻¹), further illustrating these differences. Table 2 presents the details of the parameters generated by first order exponential fits of the data. Steady state topology (Fig. 6C, gray bars) and membrane fluidity (Fig. 6D) were also determined. Although increased anionic lipid content in both Golgi- and PM-mimicking membranes led to significantly lower flipping rates, only the steady state inverse topology in more negative PM-mimicking liposomes was significantly increased, which could be due to the maintenance of high membrane order in PM-mimicking membranes (Fig. 6D). These results are not in perfect agreement with the Charge Balance Rule, because cholesterol affects simultaneously electrostatic properties and the fluidity of the lipid bilayer, thereby highlighting the contribution of additional lipid-dependent factors to membrane protein topological switching. Indeed, in cases where modulations of anionic lipids or cholesterol content led to significant alterations in membrane fluidity, any direct correlation between flipping rates, topology, and lipid composition was impossible to draw.

Identification of Patterns among the Various Model Organelles Using Principal Component Analysis

The complex interplay between lipid composition, membrane physicochemical properties, and membrane protein topology switching led us to perform principal component analysis (PCA) using the three parameters experimentally measured to identify potential clusters with shared characteristics previously unexamined. PCA reduces the dimensionality of a data set by identifying directions (i.e. principal components) along which the variation in the data is maximal. By using just a few components, large data sets can be represented by relatively few variables. Furthermore, PCA allows plotting a data set to visually assess similarities and differences between samples and to determine whether samples can be grouped (or “clustered”) by using additional algorithms. Together, these advantages make PCA a powerful tool to explore high-dimensional data sets, as in the case for our study, which contains complex samples with multiple variables. The principal components of a data set are uncorrelated and thus represent different aspects of the samples. We systematically tried different combinations of components when visualizing the data set, because much information can be lost in two- or three-dimensional visualizations. We chose a PCA biplot with samples plotted in two dimensions using their projections onto the first two principal components (PC1 and PC2), because these two components allow for the explanation of 99% of the variation in the data.

PCA is a first step, before the actual data analysis by clustering or classification of samples. Cluster analysis allows grouping samples in sets. Samples within a cluster are as similar as possible, whereas samples from different clusters are as dissimilar as possible. We used k medoids clustering or partitioning around medoids (PAM). k means clustering is the simplest and the most commonly used partitioning method for splitting a dataset into a set of k groups (i.e. clusters). It requires the analyst to specify the number of optimal clusters to be generated from the data, by defining clusters in which the total intra-cluster variation is minimized. In the PAM algorithm, k representative objects (medoids) are chosen as cluster centers, and objects are assigned to the center (medoid = cluster) with which they have minimum dissimilarity. Some of the interesting features of PAM are that it is robust to outliers as the centroids of the clusters are data objects and that one can determine the number of clusters by exploring the average silhouette value. We further validated our clustering using a variant of PAM named CLARA (clustering large applications). The same clustering results were obtained, allowing us to rule out any stochastic organization of the data readout. We identified the number of clusters (k value) by using the percentage of variance explained (F test) as a function of the number of clusters. Furthermore, the decision to choose four clusters was made based on the high dissimilarities between E. coli lipids and the lipid composition of cluster 3. Our extended works on LacY lipid-dependent topological (re)orientation in E. coli membrane, both in vivo and in vitro, encouraged us to maintain this data point as a single cluster, despite the presence of a single data point.

Therefore, the best clustering solution (Fig. 7A) was obtained using a cluster number of 4, with liposomes made of E. coli lipids singled out as cluster 1. The detailed composition of the four clusters is presented in Fig. 7, along with the PCA details. The experimental characteristics and the main lipid composition features for the various clusters (averaged from the various organelle-mimicking liposomes they contain) are presented in Fig. 7, B–D. Cluster 1 exhibits the highest content in anionic lipids (25%). Although this does not translate into faster flipping rates, cluster 1 also exhibits the lowest amounts of correct topology. This observation is in good agreement with our proposed Charge Balance Rule, which predicts an increase in inversion of topology with increased membrane anionic surface density. This is further illustrated by cluster 4, which exhibits the smallest amounts of anionic lipids and the least amount of inversion. Although the amounts of PE + PC appear constant across all clusters (making >70% of the phospholipids), it is important to note that PE amounts in E. coli lipids (75%) are significantly higher than in clusters 2–4 (∼20–30%). Although both PE and PC participate in dampening the negative charge of membranes containing anionic lipids, some physicochemical properties specific to PE (increased membrane curvature and headgroup hydrogen bonding properties), and not related to membrane fluidity features, might explain the increased membrane protein inversion observed for cluster 1.

FIGURE 7.

Identification of patterns among the various model organelles using principal component analysis. A, PCA plot for the various mimicked organelles performed using the three measured parameters (membrane fluidity, flipping rate, and amount of correct topology). The two first principal components are plotted. Individual grouping inferred by pam with k = 4. The detailed composition of each cluster is indicated on the right, as well as a summary of the analysis indicating the contribution of each component and standard deviations. B, experimentally determined properties of the identified clusters. C, lipid characteristics of the identified clusters. D, detailed lipid composition of the clusters identified using PCA. PI, phosphatidylinositol; PS, phosphatidylserine; CL, cardiolipin. Values are means ± S.E. of the data obtained from 3 to 4 replicates.

The PCA dissection also demonstrates that the increased complexity in lipids observed in eukaryotic membranes translates into a more complex lipid-dependent phenomenon, where the Charge Balance Rule explains only part of the results. The PCA allowed us to conclude that knowing the membrane fluidity and flipping rate allowed us to explain the separation between different model organelles. Strong contributions also stem from cholesterol and sphingomyelin, either alone or in combination. Interestingly, when comparing clusters 1, 3, and 4, the effect of lipid composition differences is not reflected in membrane fluidity changes. The intimate interaction of cholesterol and sphingomyelin, which is described as one of the bases for lipid raft formation, could hamper the single effects of cholesterol or sphingomyelin on membrane fluidity, resulting in more ordered membrane (positive GP) when present together. However, cluster 2 presents with very low amounts of cholesterol and ∼8% of sphingomyelin, which results in more disordered membrane (negative GP). Similar flipping rates are observed in clusters 2 and 3; however, higher amounts of correct topology are observed in disordered membranes, illustrating the importance of the collective physicochemical properties of the lipids making up the organelles in establishing and maintaining membrane protein topology.

Discussion

The folding and final topology of a polytopic membrane protein are determined by a complex interplay (24, 38, 39) between the ribosome (40), the translocon (41, 42), short range (43, 44) and long range intra-protein tertiary interactions (17, 19, 45), and short range lipid-protein interactions (46, 47), which are governed by specific structural (48) and biosynthetic assembly rules (49, 50). Protein sequence and translocation machineries have early on been described as determinants of membrane protein topology, whereas lipid composition was only more recently demonstrated to play a critical role. Similarly, post-translational modifications resulting in EMD charge modifications (e.g. phosphorylation) have been increasingly integrated in topology prediction software (51, 52). The importance of phosphorylation in membrane protein structure and function is not limited to eukaryotes, as evidenced by the identification of about 370 E. coli proteins containing consensus phosphorylation sites (53, 54), several of these being membrane proteins involved in normal cell division, which requires PE in E. coli (55, 56). Altogether, our results clearly demonstrate that phosphorylation of a membrane protein can induce its topological rearrangement independent of any other cellular factors and with kinetics of phosphorylation-induced membrane protein topological switching fast enough to be of physiological significance. In eukaryotes, the rate of N-terminal translocation of a phosphorylated EMD was estimated to be 1.6 times the rate of translation, which is on a minute time scale (57). A priori one might expect that phosphorylation of an EMD would stabilize topological organization of a membrane protein as has been suggested (58). However, our results run counter to this assumption.

Phosphorylation-induced TMD flipping can be explained and predicted by the Charge Balance Rule, which incorporates the effect of lipid-protein interactions on the potency of charged residues as topological signals governed by the Positive-Inside Rule. We now extend this rule to charge modifications resulting from post-translational modifications of membrane proteins, such as phosphorylation, methylation, or acylation of amino acids. Our data suggest that phosphorylation reduces the net positive charge of an EMD and favors a change in orientation of an adjacent TMD. The role of PE in diluting the net negative charge contributed by anionic lipids has already been described (8, 19) and extended to other zwitterionic and net neutral lipids (e.g. PC and neutral glycerol-based glycolipids) (8, 15). We extend here the Charge Balance Rule to eukaryotic lipid composition and present experimental evidence for the role played by sphingomyelin and cholesterol in these lipid-protein interactions. Cholesterol does not possess any charge, and thus it can act simply as a diluent of negative charge surface density (18) supporting the Charge Balance Rule (8). In vivo, cholesterol can also trigger conformational changes within the translocon or restrict the freedom of its lateral gate by affecting the fluidity of the lipid bilayer (59, 60). Our in vitro results in a translocon-free proteoliposomes system allow excluding these effects.

Because many proteins are subjected to multiple phosphorylation events, the degree of phosphorylation could regulate the ratio of inverted to non-inverted EMDs, as we observed (Fig. 2A). Indeed, human CD38, a single span membrane protein with a large C-terminal domain containing the active site, displays two topologies in various cell compartments dependent on the physiological state of the cell (1, 29). The switch from C-terminal “out to in” has been postulated to serve as a regulatory mechanism for accessibility to cytoplasmic substrates. When positively charged amino acid residues in the N-terminal domain were mutated to neutral or negatively charged residues (mimicking protein phosphorylation), protein localization in the cell changed, and the proportion of C-terminal-in configuration increased (29). Phosphorylation of three serines in the N-terminal segment was suggested as a regulator of CD38 topology and function. Using a complex model membrane protein made of 12 TMDs, we demonstrate that phosphorylation-induced topological rearrangements are not limited to simple single span membrane proteins. Our study demonstrates the validity of such a regulatory mechanism, extending the Charge Balance Rule to a eukaryotic context. Although lipid-protein charge interactions still play a major role, our results expand to both the individual and collective properties of the lipids making up the organelle membranes also playing critical roles in topological switching. Additionally, we show that both the rate and extent of membrane protein topological switching are lipid-dependent. Our results particularly emphasize the differences among key membrane protein synthesis and trafficking organelles with a significant increase in flipping rates in organelles along the secretory pathway (ER < Golgi < PM), consistent with their differences in membrane properties. Using additional organelle-mimicking lipid composition, we have dissected in more detail the effect of cholesterol and anionic lipids on membrane protein topological switching. We show that both individual and collective properties of the lipids making up the organelle membranes play a role in defining the rates and extents of topological switching. Phosphorylation events introduce a new topogenic signal within EMDs that could result in different structural organization of a protein within the changing lipid environment along the organelle trafficking pathway. These findings have crucial implications in membrane protein trafficking and proper targeting, as postulated for human CD38 (1) and phospholipid scramblase 1 (30), and could improve membrane protein topology prediction tools.

Experimental Procedures

Reagents

Anti-phosphoserine/threonine antibody (catalog no. ab17464) was purchased from Abcam. All lipids were purchased from Avanti Polar Lipids. Except for E. coli mimicking membranes where PE, PG, and CL from E. coli were used, care was taken to use phospholipids from similar eukaryotic sources (brain for PE, PC, PS, and SM; liver for PI; heart for CL; and soy for PG) to limit fatty acid heterogeneity among phospholipid classes. LKB1 and PDK1 kinases were purchased from Sigma. Laurdan was purchased from Thermo Fisher Scientific. The sources of all other reagents are described in Ref. 21.

Plasmids Construction

Plasmid pT7-5/C-less LacY (ampR, ColE1 replicon), encoding lactose permease (LacY) in which Cys residues are replaced by Ser, was used to construct plasmids expressing LacY derivatives under OP_lac regulation and containing multiple amino acid replacements in a derivative of LacY (−2/+2/+2 and −2/0/−2 in EMDs C2/C4/C6) encoding a single Cys replacement at His-205 in otherwise Cys-less LacY (19). The LacY charge templates −2/0/−2 (which contains the following mutations: K69E, K74E, K131E, K211E, and K218E) and −2/+2/+2 (which contains the following mutations: K69E and K74E) have been previously described (19). These latter plasmids were used to generate the templates to engineer phosphorylation sites in EMD C6 (Fig. 1A) to be used for SCAM^TM assays. The amino acid sequences “FAAFSY,” “LATY,” and “FAAFSYLATY” were inserted between amino acids 195 and 196 to engineer PDK1, LKB1, and PDK1/LKB1 phosphorylation sites, respectively. Plasmid pT7-5/C-less LacY/V331C (ampR, ColE1 replicon) encoding a single Cys replacement at Val-331 in otherwise Cys-less LacY, where Trp replacements were made in EMDs NT and C6 at positions 14 and 205, respectively (21), were used to generate the templates containing engineered phosphorylation sites in EMD C6 (Fig. 1A) to be used for real time monitoring of TMD flipping by FRET. The QuikChange Lightning site-directed mutagenesis kit was used to construct all plasmid derivatives.

Growth Conditions and Protein Purification

Strain AL95 (pss93::kanR lacY::Tn9) with plasmid pDD72GM (pssA⁺ genR and pSC101 temperature-sensitive replicon) was used as the host for expression of LacY derivatives in a wild type lipid environment. All LacY derivatives were engineered with a His₆ tag at the C terminus to facilitate purification and were expressed under control of OP_lac by growth of cells in the presence of 1 mm isopropyl β-thiogalactoside (IPTG). Cells were grown in LB-rich medium containing ampicillin (100 μg/ml) to an A₆₀₀ of 0.6, induced by addition of IPTG and grown until cell arrest occurred. Purification of LacY derivatives was carried out at 4 °C or on ice as described previously (20). Protein content during purification and the concentration of protein in proteoliposomes were determined by the micro-BCA protein assay, according to the manufacturer's instructions.

IAEDANS Labeling

Trp replacement derivatives of LacY/V331C were purified from PE-containing cells because previous results demonstrated that proteoliposome lipid composition and not the cell source lipid composition determines LacY organization. Labeling of LacY with IAEDANS was carried out as described previously (21). The labeling ratio was quantified by comparing the absorption spectrum of labeled protein with unlabeled protein using extinction coefficients of ϵ₃₃₄ = 5,700 m⁻¹ · cm⁻¹ for IAEDANS. Typical labeling efficiency was above 85%.

Preparation of Proteoliposomes

Proteoliposomes were formed by reconstitution of protein into small unilamellar liposomes of various lipid compositions, as described previously (20), by incorporation of non-IAEDANS-labeled or IAEDANS-labeled LacY into preformed liposomes.

Protein Topology Mapping

Topological determination (SCAM^TM) of TMDs is based on the controlled membrane permeability of the thiol-specific reagent MPB and its reactivity with diagnostic cysteine residues in EMDs of LacY as described previously (20). Using LacY with a single cysteine engineered in EMD C6, the degree of mixed or dual topologies co-existing within the same proteoliposome membrane was assessed with a two-step labeling protocol by pre-blocking putative external cysteines in intact liposomes with transparent non-permeant AMS before the standard biotinylating procedure with MPB and quantification by densitometry.

Membrane Fluidity Assessment

We monitored the bilayer fluidity-dependent fluorescence spectral shift of Laurdan due to dipolar relaxation phenomena (33). The proteoliposome samples were first resuspended in PBS buffer (pH 7.4) and adjusted to A₆₀₀ ∼0.1 in a final volume of 1 ml. The samples were then split between two tubes, one for Laurdan labeling and one for unlabeled control. The labeled samples were incubated with 100 μm Laurdan (stock solution in DMSO) at 37 °C for 1 h prior to florescence measurements. Determinations were carried out using a QuantaMaster model QM3-SS (Photon Technology International), a cuvette-based fluorescence spectrometer equipped with a Peltier TE temperature controller. The measurements were all performed at 20 °C, the same temperature used to monitor dynamic topology (see below). Data were collected and analyzed using Felix 32 software. Fluorescence measurements were conducted in a final volume of 2 ml in PBS after sample equilibration at 20 °C for 10 min. An excitation wavelength of 360 nm was used, and emission spectra were recorded between 400 and 550 nm, both with a 0.5-nm bandpass filter. Measurements of unlabeled membranes were performed in the same conditions, and the respective spectra were subtracted from the Laurdan-labeled sample. GP from emission spectra was calculated using Equation 1,

$$\text{GP}=\left({\text{I}}_{440}-{\text{I}}_{490}\right)/\left({\text{I}}_{440}+{\text{I}}_{490}\right)$$ (Eq. 1)

where I₄₄₀ and I₄₉₀ are the fluorescence intensities at emission wavelengths of 440 nm (gel phase, L_β) and 490 nm (liquid crystalline phase, Lα), respectively.

Real Time Topological Switch Monitoring by FRET

The various fluorescence measurements were performed using a QuantaMaster model QM3-SS (Photon Technology International), a cuvette-based fluorescence spectrometer. Using a Peltier TE temperature controller, the sample was held at a constant 20 °C. Data were collected and analyzed using Felix 32 software. Real time protein flipping was monitored at an excitation wavelength of 295 nm (for Trp) and at an emission wavelength of 475 nm (for IAEDANS), both with a 1-nm bandpass filter. The proteoliposomes (containing 1 μm IAEDANS-labeled LacY, at a lipid to protein ratio (w/w) of 500) were first equilibrated at 20 °C. Phosphorylation was then triggered by addition of 0.5 μm LKB1 kinase under constant stirring at 20 °C. We used the FRET values observed before phosphorylation (F₀) to normalize the FRET values (F_t), using Equation 2.

$$F_{\text{norm}}=F_t/F_0$$ (Eq. 2)

In all cases, before starting kinetics measurements, care was taken to measure and adjust Trp fluorescence before phosphorylation to start with similar fluorescence signal from the donor.

Principal Component Analysis, Statistical Data Analysis, and Fitting

We conducted PCA on model organelles using membrane fluidity, flipping rate, and amount of correct topology. Hierarchical clustering was performed on model organelles identified by PCA using Euclidean or correlation distance metric. For data analysis, the time course of F_norm was averaged across experiments and fitted to a single exponential function beginning when triggering phosphorylation of LacY using Origin 8.6 (OriginLab Corp.). In all figures, solid lines indicate these fits, and error bars mark the upper and lower confidence limits (95%) for the fitted parameters. In all graphs, we used the flipping rate (determined as 1/τ) as a representation of the relative timing of each event.

Author Contributions

H. V., D. M. M., A. K., V. J., M. B., and W. D. designed the experiments and interpreted the results. H. V. performed the experiments. H. V., M. B., and W. D. wrote the paper with input from all authors.